1
00:00:04,671 --> 00:00:06,973
GOOD MORNING

2
00:00:07,040 --> 00:00:08,475
EVERYBODY, THIS MEETING OF THE

3
00:00:08,541 --> 00:00:11,077
NLM BOARD OF SCIENTIFIC

4
00:00:11,144 --> 00:00:11,945
COUNSELORS IS BEING CALLED

5
00:00:12,011 --> 00:00:13,980
TOGETHER AS WE HAVE A QUORUM OF

6
00:00:14,047 --> 00:00:14,414
MEMBERS.

7
00:00:14,481 --> 00:00:17,050
THE MEETING IS BEING VIDEOCAST

8
00:00:17,117 --> 00:00:22,122
AT THE NIH WEBSITE AND WILL BE

9
00:00:22,188 --> 00:00:22,689
AVAILABLE FOR VIEWING WITH

10
00:00:22,756 --> 00:00:23,857
EXCEPTION TO THE EXECUTIVE

11
00:00:23,923 --> 00:00:24,124
SESSIONS.

12
00:00:24,190 --> 00:00:27,327
SO I WOULD LIKE TO WELCOME

13
00:00:27,394 --> 00:00:28,228
EVERYONESH EVERYONE TO THIS

14
00:00:28,294 --> 00:00:30,196
SESSION AND I WILL NOW TURN IT

15
00:00:30,263 --> 00:00:33,133
OVER TO ACTING NLM DIRECTOR

16
00:00:33,199 --> 00:00:33,767
STEVE SHERRY.

17
00:00:33,833 --> 00:00:34,567
>> THANK YOU.

18
00:00:34,634 --> 00:00:35,502
WELCOME, EVERYONE.

19
00:00:35,568 --> 00:00:37,003
I APPRECIATE ALL OF YOU FOR

20
00:00:37,070 --> 00:00:38,638
JOINING US AND THANK EACH OF YOU

21
00:00:38,705 --> 00:00:40,273
FOR THE IMPORTANT ROLE YOU PLAY

22
00:00:40,340 --> 00:00:42,342
IN ADVANCING RESEARCH HERE AT

23
00:00:42,409 --> 00:00:43,309
THE NATIONAL LIBRARY OF

24
00:00:43,376 --> 00:00:43,777
MEDICINE.

25
00:00:43,843 --> 00:00:46,379
THIS IS THE APRIL 18th BOARD

26
00:00:46,446 --> 00:00:48,348
OF SCIENTIFIC COUNSELORS ACTING

27
00:00:48,415 --> 00:00:54,988
SCIENTIFIC DIRECTORS UPDATE.

28
00:00:55,054 --> 00:00:57,524
NEXT SLIDE, PLEASE.

29
00:00:57,590 --> 00:01:02,362
SO AS WE ALL KNOW NLM IS WORKING

30
00:01:02,429 --> 00:01:04,264
CONSTANTLY FOR DISCOVERY AND

31
00:01:04,330 --> 00:01:06,166
DATA POWERED HEALTH THROUGH

32
00:01:06,232 --> 00:01:07,066
ENHANCED RESEARCH, DISSEMINATION

33
00:01:07,133 --> 00:01:08,701
AND ENGAGEMENT AND BUILDING A

34
00:01:08,768 --> 00:01:10,003
WORK FOR FOR DATA DRIVEN

35
00:01:10,069 --> 00:01:10,770
RESEARCH AND HEALTH.

36
00:01:10,837 --> 00:01:11,538
NEXT SLIDE, PLEASE.

37
00:01:11,604 --> 00:01:20,346
AND THIS IS WELL ALIGNED WITH

38
00:01:20,413 --> 00:01:21,047
DR. MONICA BERTAGNOLLI'S VISION,

39
00:01:21,114 --> 00:01:22,816
SHE TOOK THE HELM THIS PAST

40
00:01:22,882 --> 00:01:24,350
NOVEMBER, AND WITH A NEW

41
00:01:24,417 --> 00:01:25,852
DIRECTOR COMES A NEW VISION, SO

42
00:01:25,919 --> 00:01:29,656
I WANTED TO SHARE INSIGHT FOR

43
00:01:29,722 --> 00:01:31,357
THE DOCTOR PROVIDING EXCITING

44
00:01:31,424 --> 00:01:33,593
ROLES IN THE ROLE THAT NLM WILL

45
00:01:33,660 --> 00:01:36,563
PLAY IN THE NIH STORY.

46
00:01:36,629 --> 00:01:39,732
LET'S BEGIN WITH CHALLENGES IN

47
00:01:39,799 --> 00:01:43,536
PUBLIC HEALTH THAT

48
00:01:43,603 --> 00:01:50,143
DR. BERTAGNOLLI BELIEVES WILL BE

49
00:01:50,210 --> 00:01:51,211
QUITE IMPORTANT.

50
00:01:51,277 --> 00:01:53,580
--WHAT'S MORE IS THAT OUR LIFE

51
00:01:53,646 --> 00:01:55,648
EXPECTANCY IS DIPPED SINCE 2014,

52
00:01:55,715 --> 00:01:57,050
A TREND THAT CANNOT BE FULLY

53
00:01:57,116 --> 00:02:00,220
EXPLAINED BY THE COVID-19

54
00:02:00,286 --> 00:02:03,156
PANDEMIC.

55
00:02:03,223 --> 00:02:04,657
AS DR. BERTAGNOLLI ABOUT THIS

56
00:02:04,724 --> 00:02:06,092
TROUBLING DATA, WE ARE CLEARLY

57
00:02:06,159 --> 00:02:09,028
NOT ON THE RIGHT TRACK.

58
00:02:09,095 --> 00:02:10,530
ANOTHER PLACE WE'RE NOT ON THE

59
00:02:10,597 --> 00:02:13,233
RIGHT TRACK IS THAT DESPITE OUR

60
00:02:13,299 --> 00:02:14,234
POOR LIFE EXPECTANCY AMONG OUR

61
00:02:14,300 --> 00:02:16,035
PEERS, THE U.S. IS SPENDING FAR

62
00:02:16,102 --> 00:02:16,903
MORE ON HEALTHCARE.

63
00:02:16,970 --> 00:02:19,806
WE ARE MORE ALIGNED WITH OTHER

64
00:02:19,873 --> 00:02:21,574
WEALTHY COUNTRIES IN THE 1970S

65
00:02:21,641 --> 00:02:23,676
BUT AS THIS FIGURE FROM 2020

66
00:02:23,743 --> 00:02:24,844
SHOWS OUR INCREASED SPENDING IN

67
00:02:24,911 --> 00:02:26,579
THIS AREA HAS NOT RESULTED IN

68
00:02:26,646 --> 00:02:27,647
BETTER HEALTH.

69
00:02:27,714 --> 00:02:30,650
IN 2018 THE U.S. HAD ROUGHLY

70
00:02:30,717 --> 00:02:32,619
$10,000   PERCAPITTA OF HEALTH

71
00:02:32,685 --> 00:02:35,321
EXPENDITURES AND A AVERAGE

72
00:02:35,388 --> 00:02:36,956
HEALTH LIFE EXPECTANCY OF 90

73
00:02:37,023 --> 00:02:37,524
YEARS OLD.

74
00:02:37,590 --> 00:02:39,526
WE CAN CONTRAST THAT TO YAP AN

75
00:02:39,592 --> 00:02:41,361
WITH A AVERAGE LIFE EXPECTANCY

76
00:02:41,427 --> 00:02:44,097
OF 84 YEAR ANDS ONLY ABOUT $45

77
00:02:44,163 --> 00:02:46,566
PER CAPITA IN HEALTH

78
00:02:46,633 --> 00:02:47,767
EXPENDITURES, THE DOCTOR AND I

79
00:02:47,834 --> 00:02:50,503
AGREEN CELLS THAT NIH AND MLM

80
00:02:50,570 --> 00:02:52,105
CAN PLAY AN IMPORTANT ROLE IN

81
00:02:52,171 --> 00:02:54,440
REVERSING THIS TREND AND HELP

82
00:02:54,507 --> 00:02:56,609
BRING DATA THAT LINK NEW

83
00:02:56,676 --> 00:02:58,511
RESOURCES AND UTILIZATION TO

84
00:02:58,578 --> 00:03:02,415
MEANINGFUL CLINICAL OUTCOMES.

85
00:03:02,482 --> 00:03:04,150
NEXT SLIDE, PLEASE EMPLOY BUT AS

86
00:03:04,217 --> 00:03:06,753
POINTED OUT THE LINK BETWEEN

87
00:03:06,819 --> 00:03:07,654
BIOMEDICAL EVIDENCE AND CLINICAL

88
00:03:07,720 --> 00:03:08,988
DECISION MAKE BEING BEING BE

89
00:03:09,055 --> 00:03:10,890
MADE STRONGER SHE GIVES AN

90
00:03:10,957 --> 00:03:12,892
EXAMPLE FROM HER OWN FIELD IN

91
00:03:12,959 --> 00:03:14,394
ONCOLOGY WHERE RESEARCH HAS

92
00:03:14,460 --> 00:03:15,728
SHOWN THAT THERE'S BEEN LITTLE

93
00:03:15,795 --> 00:03:18,197
TO NO CHANGE IN PAST DECADE IN

94
00:03:18,264 --> 00:03:19,999
THE NATURE OF EVIDENT USED TO

95
00:03:20,066 --> 00:03:21,234
DEVELOP CLINICAL CARE

96
00:03:21,301 --> 00:03:24,103
GUIDELINES, THIS ANALYSIS OF THE

97
00:03:24,170 --> 00:03:25,939
2019 NATIONAL COMPREHENSIVE

98
00:03:26,005 --> 00:03:27,206
CANCER NETWORK GUIDELINES FOUND

99
00:03:27,273 --> 00:03:29,509
THAT ONLY 7% OF THE

100
00:03:29,576 --> 00:03:30,710
RECOMMENDATIONS CAN THAT PART IN

101
00:03:30,777 --> 00:03:33,446
THE LITTLE BLUE WEDGE THERE ARE

102
00:03:33,513 --> 00:03:34,781
SUPPORTED BY RANDOMIZED CLINICAL

103
00:03:34,847 --> 00:03:35,214
TRIALS.

104
00:03:35,281 --> 00:03:36,316
THE GREAT MAJORITY OF GUIDELINES

105
00:03:36,382 --> 00:03:38,551
THAT WE USE TO DELIVER CARE,

106
00:03:38,618 --> 00:03:40,286
WERE DERIVE FRIDAY LEVEL 2

107
00:03:40,353 --> 00:03:41,487
EVIDENCE WHICH IS GENERALLY

108
00:03:41,554 --> 00:03:43,690
PROVIDED BY CAREFULLY CONTROLLED

109
00:03:43,756 --> 00:03:45,191
OBSERVATIONAL TRIALS EARLIER

110
00:03:45,258 --> 00:03:46,192
PHASED RANDOMIZED TRIALS BUT NOT

111
00:03:46,259 --> 00:03:51,731
THE BEST WORK THAT WE CAN DO

112
00:03:51,798 --> 00:03:54,767
WHICH IS A RANDOMIZED CLINICAL

113
00:03:54,834 --> 00:03:56,569
CONTROL.

114
00:03:56,636 --> 00:03:57,003
NEXT SLIDE, PLEASE.

115
00:03:57,070 --> 00:03:58,571
NOT ONLY DO WE HAVE A GAP

116
00:03:58,638 --> 00:03:59,872
BETWEEN THE EVIDENCE AND ITS

117
00:03:59,939 --> 00:04:01,374
IMPACT ON CLINICAL CARE, BUT WE

118
00:04:01,441 --> 00:04:04,510
ALSO HAVE A CONCERNING GAP IN

119
00:04:04,577 --> 00:04:05,111
THE EVIDENCE ITSELF.

120
00:04:05,178 --> 00:04:06,813
WE KNOW THERE ARE SERIOUS

121
00:04:06,879 --> 00:04:08,114
DISPARITIES IN ACCESS TO

122
00:04:08,181 --> 00:04:10,183
CLINICAL TRIALS THAT MIRROR

123
00:04:10,249 --> 00:04:11,784
DISPARITIES AND HEALTHCARE

124
00:04:11,851 --> 00:04:12,418
DELIVERY OVERALL.

125
00:04:12,485 --> 00:04:14,587
WE ARE MISSING DATA, ON A VERY

126
00:04:14,654 --> 00:04:16,155
LARGE AND IMPORTANT GROUP OF

127
00:04:16,222 --> 00:04:18,091
PATIENTS INCLUDING THOSE LISTED

128
00:04:18,157 --> 00:04:18,458
HERE.

129
00:04:18,524 --> 00:04:21,761
THE 1S THAT TYPICALLY DO NOT GET

130
00:04:21,828 --> 00:04:23,162
TARGETED FOR ENROLLMENT IN CARE

131
00:04:23,229 --> 00:04:27,100
SUCH AS THE ELDERLY, MINORITY,

132
00:04:27,166 --> 00:04:29,168
UNINSURED GROUPS, RURAL, THOSE

133
00:04:29,235 --> 00:04:30,269
WHO HAVE EXCESSIVE COMORBID

134
00:04:30,336 --> 00:04:31,437
CONDITIONS ARE JUST EXAMPLES OF

135
00:04:31,504 --> 00:04:33,072
PATIENT WHAT'SY NEED TO DO

136
00:04:33,139 --> 00:04:34,974
BETTER AND TRY TO ENGAGE IN

137
00:04:35,041 --> 00:04:37,977
CLINICAL RESEARCH.

138
00:04:38,044 --> 00:04:39,112
NEXT SLIDE, PLEASE.

139
00:04:39,178 --> 00:04:40,847
WHILE THESE CHALLENGES ARE

140
00:04:40,913 --> 00:04:42,615
SERIOUS, NIH IS COMMITTED TO

141
00:04:42,682 --> 00:04:44,417
ADOPTING PRINCIPLES THAT CAN

142
00:04:44,484 --> 00:04:45,618
MEANINGFULLY TURN THE TIDE

143
00:04:45,685 --> 00:04:46,786
TOWARDS BETTER HEALTH FOR ALL

144
00:04:46,853 --> 00:04:47,153
PEOPLE.

145
00:04:47,220 --> 00:04:48,421
NEXT SLIDE, PLEASE.

146
00:04:48,488 --> 00:04:50,156
THIS INCLUDES LOOKING BEYOND THE

147
00:04:50,223 --> 00:04:52,692
MOMENT OF SCIENTIFIC DISCOVERY,

148
00:04:52,759 --> 00:04:53,259
WHILE DORPH SECERTAINLY THE

149
00:04:53,326 --> 00:04:54,727
FOUNDATION OF OUR WORK HERE AT

150
00:04:54,794 --> 00:04:58,564
NLM AND NE H BROADLY,

151
00:04:58,631 --> 00:04:59,499
DR. BERTAGNOLLI HAS REMINDED US

152
00:04:59,565 --> 00:05:01,267
THAT THE WORK IS NOT FINISHED

153
00:05:01,334 --> 00:05:02,902
UNTIL THE SCIENTIFIC ADVANCES

154
00:05:02,969 --> 00:05:04,270
ACTUALLY REACH THE PEOPLE WHO

155
00:05:04,337 --> 00:05:07,840
NEED THEM THAT'S THROUGH LIVING

156
00:05:07,907 --> 00:05:09,275
LONGER, HEALTHIER LIVES.

157
00:05:09,342 --> 00:05:10,043
NEXT SLIDE, PLEASE.

158
00:05:10,109 --> 00:05:13,713
TO MAKE THIS HAPPEN,

159
00:05:13,780 --> 00:05:14,380
DR. BERTAGNOLLI, EMPHASIZES WE

160
00:05:14,447 --> 00:05:17,850
NEED TO RELY ON ALL COMPONENTS

161
00:05:17,917 --> 00:05:19,152
OF NIH RESEARCH, BASIC SCIENCE

162
00:05:19,218 --> 00:05:19,919
TO IMPLEMENTATION RESEARCH IN

163
00:05:19,986 --> 00:05:20,653
THE COMMUNITY.

164
00:05:20,720 --> 00:05:22,522
OUR PURSUIT OF BETTER HEALTH

165
00:05:22,588 --> 00:05:24,657
MUST BE ROBUST AND COMPREHENSIVE

166
00:05:24,724 --> 00:05:28,294
AND AS DR. BERTAGNOLLI SAYS,

167
00:05:28,361 --> 00:05:29,929
PEOPLE ARE OUR PARTNERS IN

168
00:05:29,996 --> 00:05:32,131
DISCOVERY, WHETHER THEY'RE

169
00:05:32,198 --> 00:05:33,099
RESEARCHERS, PARTICIPANTS OR

170
00:05:33,166 --> 00:05:34,367
TRIAL CLINICIANS, THEY WANT TO

171
00:05:34,434 --> 00:05:35,902
SEE A COMMUNITY AS THE THIRD

172
00:05:35,968 --> 00:05:38,938
PART OF THE TRIAD FOR RESEARCH

173
00:05:39,005 --> 00:05:40,339
ENTERPRISE HERE AT NIH.

174
00:05:40,406 --> 00:05:42,075
NEXT SLIDE, PLEASE.

175
00:05:42,141 --> 00:05:43,476
AND FINALLY, OF PARTICULAR

176
00:05:43,543 --> 00:05:45,445
RELEVANCE TO NLM, SHE HAS

177
00:05:45,511 --> 00:05:47,580
EMPHASIZED WE HAVE A GREAT

178
00:05:47,647 --> 00:05:48,781
OPPORTUNITY TO ADEMBRACE NEW

179
00:05:48,848 --> 00:05:50,550
ADVANCES IN SCIENTIFIC METHODS

180
00:05:50,616 --> 00:05:52,185
LIKE DATABASE DATAAN LYTICS AND

181
00:05:52,251 --> 00:05:54,053
TO INSURE THESE ADVANCES SYRUP

182
00:05:54,120 --> 00:05:59,559
EVERYONE BY INCLUDING EVERYONE.

183
00:05:59,625 --> 00:06:00,593
DR. BERTAGNOLLI HAS NOTED THAT

184
00:06:00,660 --> 00:06:02,128
WE NOTE AT NLM THAT WHILE

185
00:06:02,195 --> 00:06:05,498
RESEARCHER VS BEEN USING

186
00:06:05,565 --> 00:06:06,299
ARTIFICIAL INTELLIGENCE AND

187
00:06:06,365 --> 00:06:08,401
MACHINE LEARNING FOR SOMETIME

188
00:06:08,468 --> 00:06:10,470
WHERE THE TECHNOLOGIES CAPTURED

189
00:06:10,536 --> 00:06:13,039
THE WORLD'S ATTENTION, AI IS

190
00:06:13,106 --> 00:06:14,107
ALREADY REVOLUTIONIZING CARE AND

191
00:06:14,173 --> 00:06:16,342
THE DOCTOR CALLED FOR A MORE

192
00:06:16,409 --> 00:06:17,844
SECURE AND SUSTAINABLE DATA

193
00:06:17,910 --> 00:06:19,112
SHARING INFRASTRUCTURE, NOW IS

194
00:06:19,178 --> 00:06:20,913
THE TIME TO DELIVER

195
00:06:20,980 --> 00:06:21,581
EVIDENCE-BASED DATA DRIVEN

196
00:06:21,647 --> 00:06:22,715
HEALTHCARE TO EVERY PATIENT AND

197
00:06:22,782 --> 00:06:24,684
TO EMBRACE THE REALTIME FEEDBACK

198
00:06:24,751 --> 00:06:27,887
THESE TECHNOLOGIES CAN OFFER US

199
00:06:27,954 --> 00:06:29,021
ALLOWING THE CLINIC TO INFORM

200
00:06:29,088 --> 00:06:30,556
THE RESEARCH AND NOT JUST THE

201
00:06:30,623 --> 00:06:32,925
OTHER WAY AROUND.

202
00:06:32,992 --> 00:06:36,662
NEXT SLIDE, PLEASE.

203
00:06:36,729 --> 00:06:37,230
DR. BERTAGNOLLI, PROVIDED A

204
00:06:37,296 --> 00:06:39,532
SAMPLE OF THE FEEDBACK LOOP

205
00:06:39,599 --> 00:06:41,334
BETWEEN CLINICAL AND BASIC

206
00:06:41,400 --> 00:06:42,835
RESEARCH, WHERE THEY INFORM ON

207
00:06:42,902 --> 00:06:44,537
EACH OTHER AND WAS FROM A

208
00:06:44,604 --> 00:06:46,606
DIRECTOR'S BLOG, WE KNOW THAT AS

209
00:06:46,672 --> 00:06:49,308
MANY AS 4 OUT OF EVERY 5 PEOPLE

210
00:06:49,375 --> 00:06:50,476
WITH AUTOIMMUNE DISEASES ARE

211
00:06:50,543 --> 00:06:52,011
FEMALE BUT WE DON'T FULLY

212
00:06:52,078 --> 00:06:53,880
HAPPENED WHY THIS DISPARITY

213
00:06:53,946 --> 00:06:55,715
OCCURS, RECEIPTLY AN NIH FUBDED

214
00:06:55,782 --> 00:06:57,116
TEAM AT STANFORD FOUND THIS MAY

215
00:06:57,183 --> 00:06:59,852
BE RELATED TO MOLECULES CALLED

216
00:06:59,919 --> 00:07:03,422
EXIST WHICH ARE ENCODED ON THE

217
00:07:03,489 --> 00:07:04,957
X-CHROMOSOME AND TRANSCRIBED

218
00:07:05,024 --> 00:07:07,226
INTO NONLONG CODING STRETCHES OF

219
00:07:07,293 --> 00:07:09,762
RNA ONLY WHEN THERE ARE 2 X

220
00:07:09,829 --> 00:07:10,596
CHROMOSOMES.

221
00:07:10,663 --> 00:07:15,768
THESE X6 MOLECULES WIND

222
00:07:15,835 --> 00:07:17,570
THEMSELVES AROUND THE CHROMOSOME

223
00:07:17,637 --> 00:07:19,472
TO SHUOF THE DOWN TRANSCRIPTION,

224
00:07:19,539 --> 00:07:21,073
AS A REULTRAVIOLET B, A WOMAN

225
00:07:21,140 --> 00:07:22,775
HAS VERY DIFFERENT PATTERNS THAN

226
00:07:22,842 --> 00:07:25,912
A MAN WHO ONLY HAS 1

227
00:07:25,978 --> 00:07:26,245
X-CHROMOSOME.

228
00:07:26,312 --> 00:07:29,182
THE STANFORD TEAM LED BY HOWARD

229
00:07:29,248 --> 00:07:31,517
AND DIANEA, FOUND THIS FEMALE

230
00:07:31,584 --> 00:07:33,119
PATTERN OF TRANSCRIPTION CAN

231
00:07:33,186 --> 00:07:35,454
RESULT INCLYSMS OF DMA, RNA AND

232
00:07:35,521 --> 00:07:37,390
PRO10S THAT IN TURN TRIGGER

233
00:07:37,456 --> 00:07:38,991
AUTOIMMUNE RESPONSES, TD

234
00:07:39,058 --> 00:07:41,227
RESEARCHERS MAKE 2 STRAINS THAT

235
00:07:41,294 --> 00:07:43,663
CAN PRODUCE TO SEE IF THIS WOULD

236
00:07:43,729 --> 00:07:46,866
MEASURABLY INCREASE THEIR RISK

237
00:07:46,933 --> 00:07:49,535
OF AUTOIMMUNITY DISEASE, 1 XIST

238
00:07:49,602 --> 00:07:51,637
WAS ACTIVATED IN MALE MICE THAT

239
00:07:51,704 --> 00:07:52,538
WERE PROTECT FRIDAY AUTOIMMUNITY

240
00:07:52,605 --> 00:07:54,874
AND ONCE THIS WAS DONE, THE

241
00:07:54,941 --> 00:07:56,542
RESEARCHERS FOUND THESE MICE

242
00:07:56,609 --> 00:07:57,543
BECAME MORE SUSCEPTIBLE TO

243
00:07:57,610 --> 00:07:59,345
DEVELOPING A WILL YOU PLEASEUS

244
00:07:59,412 --> 00:08:00,146
LIKE CONDITION, THERE'S MUCH

245
00:08:00,213 --> 00:08:01,647
MORE TO UNDERSTAND ABOUT THIS

246
00:08:01,714 --> 00:08:03,282
CONNECTION BUT IT'S ELEGANT WORK

247
00:08:03,349 --> 00:08:04,884
THAT SERVES AS 1 OF MANY

248
00:08:04,951 --> 00:08:07,286
EXAMPLES TO SHOW HOW CLINICAL

249
00:08:07,353 --> 00:08:08,588
OBSERVATIONS CAN INSPIRE BASIC

250
00:08:08,654 --> 00:08:10,656
SCIENCE AND HOW FINDINGS AT THE

251
00:08:10,723 --> 00:08:13,159
BENCH CAN ELIMINATE UPON

252
00:08:13,226 --> 00:08:15,628
CLINICAL RESEARCH AND PRACTICE.

253
00:08:15,695 --> 00:08:18,631
NEXT SLIDE, PLEASE.

254
00:08:18,698 --> 00:08:19,799
DR. BERTAGNOLLI, AS NIH DIRECTOR

255
00:08:19,866 --> 00:08:21,634
IS TO SUPPORT RESEARCH THAT

256
00:08:21,701 --> 00:08:24,704
INTEGRATES WITH HEALTHCARE AND

257
00:08:24,770 --> 00:08:26,372
REVITALIZES OUR OW PROACH TO

258
00:08:26,439 --> 00:08:27,273
DATA INTEGRATION, SHARING AND

259
00:08:27,340 --> 00:08:29,775
DATA USE.

260
00:08:29,842 --> 00:08:30,076
NEXT SLIDE.

261
00:08:30,142 --> 00:08:32,011
BY INTEGRATING WITH HEALTHCARE,

262
00:08:32,078 --> 00:08:32,845
SHE MEANS CONNECTING RESEARCH

263
00:08:32,912 --> 00:08:36,282
WITH THE POINTS OF CARE

264
00:08:36,349 --> 00:08:37,683
PEOPLE--THE POINTS OF CARE THAT

265
00:08:37,750 --> 00:08:40,186
MOST PEOPLE ARE LIKELY TO

266
00:08:40,253 --> 00:08:40,486
ENCOUNTIER.

267
00:08:40,553 --> 00:08:42,521
SHE EMPHASIZED WE HAVE NOT YET

268
00:08:42,588 --> 00:08:45,558
ESTABLISHED AN NIH WIDE CLINICAL

269
00:08:45,625 --> 00:08:47,059
INFRASTRUCTURE FOR THE PRIMARY

270
00:08:47,126 --> 00:08:48,361
CARE SETTING, PRIMARY CARE

271
00:08:48,427 --> 00:08:49,528
DELIVERER WHO SEES EVERYTHING

272
00:08:49,595 --> 00:08:51,464
THAT IS HAPPENING WITH THEIR

273
00:08:51,530 --> 00:08:52,398
PATIENTS HEALTH.

274
00:08:52,465 --> 00:08:53,466
SHE EMPHASIZES THAT THIS FOCUS

275
00:08:53,532 --> 00:08:56,102
IS A WAY TO MEET PATIENTS WHERE

276
00:08:56,168 --> 00:08:58,437
THEY ARE, PARTICULARLY THOSE IN

277
00:08:58,504 --> 00:08:59,939
UNDERSERVED COMMUNITIES OR IN

278
00:09:00,006 --> 00:09:01,540
RURAL AND URBAN REGIONS ALIKE.

279
00:09:01,607 --> 00:09:04,010
THIS MAY LOOK LIKE ADOPTING

280
00:09:04,076 --> 00:09:05,945
INNOVATIVE STUDY DESIGNS THAT

281
00:09:06,012 --> 00:09:06,646
ADDRESS COMMON HEALTH ISSUES

282
00:09:06,712 --> 00:09:09,982
THAT CAN BE DONE IN A BUSY

283
00:09:10,049 --> 00:09:11,384
PRIMARY CARE OFFICE OR COMMUNITY

284
00:09:11,450 --> 00:09:13,786
HEGHT CENTER, THIS CAN ALSO LOOK

285
00:09:13,853 --> 00:09:15,154
LIKE USING THE COMMUNITY HEALTH

286
00:09:15,221 --> 00:09:16,455
RECORD NOT ONLY AS A SOURCE OF

287
00:09:16,522 --> 00:09:19,558
DATA BUT AS A WAY FOR PATIENTS

288
00:09:19,625 --> 00:09:20,893
TO DISSEMINATE THE RESEARCH

289
00:09:20,960 --> 00:09:22,995
FINDINGS THEMSELVES.

290
00:09:23,062 --> 00:09:24,931
NEXT SLIDE.

291
00:09:24,997 --> 00:09:25,932
DR. BERTAGNOLLI, HAS ALSO

292
00:09:25,998 --> 00:09:27,967
EXPRESSED THE NEED TO EXPAND BY

293
00:09:28,034 --> 00:09:30,503
O RESEARCH USE TO INFORM NEW

294
00:09:30,569 --> 00:09:32,171
RESEARCH IN IMPROVED HEALTH

295
00:09:32,238 --> 00:09:33,639
OUTCOMES, SHE EMPHASIZES THIS

296
00:09:33,706 --> 00:09:34,540
MEANS LEVERAGES INSIGHTS FROM

297
00:09:34,607 --> 00:09:35,608
THE SOCIAL SCIENCES AND CLINICAL

298
00:09:35,675 --> 00:09:39,211
CARE AS WELL AS BASIC SCIENCE.

299
00:09:39,278 --> 00:09:42,882
THIS WAS POLED IN ADVANCE

300
00:09:42,949 --> 00:09:44,617
DIAGNOSTIC, WEARABLES AND ALL

301
00:09:44,684 --> 00:09:45,918
OTHERS OF NIH FIELD RESEARCH

302
00:09:45,985 --> 00:09:48,988
THAT CAN HELP MOVE THE NEEDLE ON

303
00:09:49,055 --> 00:09:49,622
PEOPLE'S HEALTH.

304
00:09:49,689 --> 00:09:51,924
NIH WILL EMPLOY A FEDERATED

305
00:09:51,991 --> 00:09:53,326
ARCHITECTURE FOR DATA USE AND

306
00:09:53,392 --> 00:09:55,261
HOSTING AND TO ENABLE LOW COST

307
00:09:55,328 --> 00:09:57,063
ACCESS USING OPEN INDUSTRY

308
00:09:57,129 --> 00:09:58,297
STANDARDS, IMPORTANTLY NIH WILL

309
00:09:58,364 --> 00:10:00,299
TAKE A SUSTAINABLE APPROACH TO

310
00:10:00,366 --> 00:10:01,400
THOSE ADVANCES BY BROADENING

311
00:10:01,467 --> 00:10:02,635
ACCESS TO THESE TOOLS AND

312
00:10:02,702 --> 00:10:03,836
INVESTING IN EDUCATION CAN

313
00:10:03,903 --> 00:10:05,972
WORKFORCE DEVELOPMENT TO PREPARE

314
00:10:06,038 --> 00:10:07,540
A NEW GENERATION OF DATA

315
00:10:07,606 --> 00:10:09,642
SCIENTISTS.

316
00:10:09,709 --> 00:10:11,043
NEXT SLIDE, PLEASE.

317
00:10:11,110 --> 00:10:16,582
I AM ENERGIZED BY THE WAY IN

318
00:10:16,649 --> 00:10:17,850
WHICH DR. BERTAGNOLLI'S

319
00:10:17,917 --> 00:10:19,218
STRATEGIES ALIEP WITH NLM AND

320
00:10:19,285 --> 00:10:20,086
OUR VISION.

321
00:10:20,152 --> 00:10:21,520
FROM OUR DAT DRIVEN RESEARCH TO

322
00:10:21,587 --> 00:10:22,722
TRAINING PROGRAMS WE HAVE

323
00:10:22,788 --> 00:10:23,823
EXCITING OPPORTUNITIES TO

324
00:10:23,889 --> 00:10:25,391
SUPPORT THE NIH WIDE GOAL ANDS

325
00:10:25,458 --> 00:10:26,559
PRIORITIES THAT I'VE SHARED

326
00:10:26,625 --> 00:10:30,129
HERE.

327
00:10:30,196 --> 00:10:30,963
NEXT SLIDE, PLEASE.

328
00:10:31,030 --> 00:10:32,031
SO THINKING BEYOND THE VISION

329
00:10:32,098 --> 00:10:36,335
AND THE PRIORITIES OF NIH TO OUR

330
00:10:36,402 --> 00:10:37,603
FUNDING SITUATION, JUST AS AN

331
00:10:37,670 --> 00:10:40,072
UPDATE FOR EVERYONE THAT THE

332
00:10:40,139 --> 00:10:41,674
FINAL FISCAL YEAR 2024

333
00:10:41,741 --> 00:10:42,975
APPROPRIATIONS BILL WAS SIGNED

334
00:10:43,042 --> 00:10:45,511
INTO LAW ON MARCH 23rd, 2024,

335
00:10:45,578 --> 00:10:48,381
THE RESULT AS YOU SEE HERE, FOR

336
00:10:48,447 --> 00:10:52,585
NLM, IS A FLAT BUDGET, RELATIVE

337
00:10:52,651 --> 00:10:57,089
TO FISCAL YEAR 2023 LEVELS.

338
00:10:57,156 --> 00:10:58,624
NEXT SLIDE, PLEASE.

339
00:10:58,691 --> 00:11:02,228
IN OTHER NEWS, ON

340
00:11:02,294 --> 00:11:03,329
DECEMBER 7th, 2023, A MAJORITY

341
00:11:03,396 --> 00:11:05,197
OF THE VOTING NIH FELLOWS

342
00:11:05,264 --> 00:11:06,499
UUNITTED, VOTED TO BE

343
00:11:06,565 --> 00:11:09,301
REPRESENTED BY THE INTERNATIONAL

344
00:11:09,368 --> 00:11:12,505
UNION OF UNITED AUTOMOBILE

345
00:11:12,571 --> 00:11:13,906
AEROSPACE AND AGRICULTURAL

346
00:11:13,973 --> 00:11:19,111
IMPLEMENT WORKERS OF AMERICA OR

347
00:11:19,178 --> 00:11:22,415
THE UAW UNION, THE FLLA

348
00:11:22,481 --> 00:11:26,185
CERTIFIED THE BARGAINING UNIT ON

349
00:11:26,252 --> 00:11:29,922
DECEMBER 15th 2023, THE UAW

350
00:11:29,989 --> 00:11:30,890
UNIT COVERS APPROXIMATELY 5000

351
00:11:30,956 --> 00:11:32,491
FELLOWS ACROSS THE NIH.

352
00:11:32,558 --> 00:11:34,126
NIRE H IS LOOKING FORWARD TO

353
00:11:34,193 --> 00:11:35,961
PARTNER WITH THE UAW TO

354
00:11:36,028 --> 00:11:37,930
NEGOTIATE A HISTORIC COLLECTIVE

355
00:11:37,997 --> 00:11:40,132
BARGAINING AGREEMENT FOR THE

356
00:11:40,199 --> 00:11:41,167
PARTIES, COLLECTIVE BARGAINING

357
00:11:41,233 --> 00:11:45,638
WILL BE ONGOING THROUGHOUT 2024

358
00:11:45,704 --> 00:11:47,807
NEXT SLIDE, PLEASE.

359
00:11:47,873 --> 00:11:52,078
NLM HAS ALSO KICKED OFF AN AI

360
00:11:52,144 --> 00:11:53,379
GENERATED PILOT WITH 10 USE

361
00:11:53,446 --> 00:11:56,148
CASES ACROSS NLM USING

362
00:11:56,215 --> 00:11:58,984
MICROSOFT'S OPEN AI AND AZURE

363
00:11:59,051 --> 00:11:59,518
SAFE PLAYGROUND.

364
00:11:59,585 --> 00:12:01,353
AS SEEN HERE THERE'S A VARIETY

365
00:12:01,420 --> 00:12:02,988
OF USE CASES RANGING FROM

366
00:12:03,055 --> 00:12:05,124
PRODUCT EFFICIENCY AND CUSTOMER

367
00:12:05,191 --> 00:12:06,158
EXPERIENCE THROUGH PRODUCTION

368
00:12:06,225 --> 00:12:08,260
SERVICES THAT ARE WRITING CODE

369
00:12:08,327 --> 00:12:10,262
AND AUTOMATING DATA

370
00:12:10,329 --> 00:12:11,831
VISUALIZATION TASKS, WE'RE

371
00:12:11,897 --> 00:12:13,799
LOOKING SYSTEMATICALLY AT HOW TO

372
00:12:13,866 --> 00:12:16,368
REDUCE BIAS IN OUR WORK FLOWS

373
00:12:16,435 --> 00:12:17,837
AND TASKS, EXAMPLES HERE OF A

374
00:12:17,903 --> 00:12:20,206
GRANT FORT FOLIO OR EDITING UMLS

375
00:12:20,272 --> 00:12:23,576
AND THEN USING IT IN RESEARCH

376
00:12:23,642 --> 00:12:27,480
DISCOVERY ITSELF THROUGH SOME

377
00:12:27,546 --> 00:12:33,352
IRP PROGRAMS, HERE SHOWN BY

378
00:12:33,419 --> 00:12:35,321
DR. ZHIYONG LU, AND ALSO

379
00:12:35,387 --> 00:12:36,355
UNDERSTANDING THE DYNAMICS OF

380
00:12:36,422 --> 00:12:37,890
OUR OUR SERVICES CAN BE USE

381
00:12:37,957 --> 00:12:38,624
INDEED REALTIME.

382
00:12:38,691 --> 00:12:39,492
DURING THIS PROJECTS OUR STAFF

383
00:12:39,558 --> 00:12:41,427
ARE GAINING NEW KNOWLEDGE,

384
00:12:41,494 --> 00:12:42,394
COLLABORATING ACROSS NLM AND

385
00:12:42,461 --> 00:12:44,296
NIRE H TO GAIN INSIGHT INTOS

386
00:12:44,363 --> 00:12:45,531
PRACTICAL USES OF LARGE LANGUAGE

387
00:12:45,598 --> 00:12:47,967
MODELS AND OVER THE PAST FEW

388
00:12:48,033 --> 00:12:48,767
MONTHS THEY'VE EXPERIMENTED WITH

389
00:12:48,834 --> 00:12:50,236
NEW TOOLS AND ARE IN THE PROCESS

390
00:12:50,302 --> 00:12:51,670
OF DEVELOPING PROOF OF CONCEPT

391
00:12:51,737 --> 00:12:53,706
THAT WILL APPLY TO THEIR USE

392
00:12:53,772 --> 00:12:53,973
CASES.

393
00:12:54,039 --> 00:12:55,374
THE NEXT STAGE OF THESE WORKS

394
00:12:55,441 --> 00:12:57,610
WILL BE MEASURING RESULTS THAT

395
00:12:57,676 --> 00:12:59,145
INSURE SAFE, UNBIASED AND

396
00:12:59,211 --> 00:13:01,413
RELIABLE RESULTS USING THIS

397
00:13:01,480 --> 00:13:05,284
FRAMEWORK FOR RISK MITIGATION

398
00:13:05,351 --> 00:13:06,218
AND BIAS.

399
00:13:06,285 --> 00:13:06,719
NEXT SLIDE, PLEASE.

400
00:13:06,785 --> 00:13:08,420
FINALLY I WOULD LIKE TO SHARE AN

401
00:13:08,487 --> 00:13:09,889
UPDATE ON OUR EFFORTS TO CREATE

402
00:13:09,955 --> 00:13:12,424
A CULTURE OF CONTINUOUS

403
00:13:12,491 --> 00:13:13,526
INNOVATION AT NLM.

404
00:13:13,592 --> 00:13:14,827
NLM CONTINUED TO MAKE GREAT

405
00:13:14,894 --> 00:13:16,595
PROGRESS AND ESTABLISH A FORWARD

406
00:13:16,662 --> 00:13:17,930
LOOKING MODEL OF IPT GREATER

407
00:13:17,997 --> 00:13:19,798
NOVATION THAT IS APPLIED

408
00:13:19,865 --> 00:13:20,633
CONSISTENTLY ACROSS THE LIBRARY

409
00:13:20,699 --> 00:13:22,968
AND ALL OUR DIVISIONS OF

410
00:13:23,035 --> 00:13:24,236
OPERATIONS, NEXT SLIDE, PLEASE.

411
00:13:24,303 --> 00:13:25,871
EVERY DAY I SEE COUNTLESS

412
00:13:25,938 --> 00:13:27,406
EXAMPLES OF STAFF BRINGING

413
00:13:27,473 --> 00:13:28,574
FORWARD NOVEL APPROACHES, IDEAS

414
00:13:28,641 --> 00:13:31,677
AND ALTERNATIVE METHODS THAT

415
00:13:31,744 --> 00:13:33,112
SHOWCASE HOW INNOVATIVE WE ARE

416
00:13:33,179 --> 00:13:34,280
HERE, THIS ONGOING COMMITMENT TO

417
00:13:34,346 --> 00:13:36,715
INNOVATION IS SEEN IN OUR

418
00:13:36,782 --> 00:13:37,349
DAY-TO-DAY OPERATIONAL DELIVERY

419
00:13:37,416 --> 00:13:38,751
AND OUR FORWARD LOOKING APPROACH

420
00:13:38,817 --> 00:13:40,152
AS WE REMAIN RESPONSIVE IN HOW

421
00:13:40,219 --> 00:13:42,087
WE SERVE THE PUBLIC, INNOVATION

422
00:13:42,154 --> 00:13:43,622
IS A COMMON THREAD THROUGHOUT

423
00:13:43,689 --> 00:13:46,091
THE 3 PILLARS OF OUR NLM

424
00:13:46,158 --> 00:13:47,726
STRATEGIC PLAN WE LOOKED AT A

425
00:13:47,793 --> 00:13:49,195
FEW MINUTES AGO, IT'S ALSO

426
00:13:49,261 --> 00:13:50,863
REINFORCED FROM THE VITIONZ

427
00:13:50,930 --> 00:13:53,199
THROUGH HHS AND NIH OVERALL WITH

428
00:13:53,265 --> 00:13:54,667
EXAMPLES LIKE CUSTOMER

429
00:13:54,733 --> 00:13:56,368
EXPERIENCE AND DIGITAL

430
00:13:56,435 --> 00:13:57,203
TRANSFORMATION, WITHIN NNLM, WE

431
00:13:57,269 --> 00:13:58,771
ARE LEADING IN MANY OF THESE

432
00:13:58,837 --> 00:14:01,807
SPACES AS WE BUILD OUT A NEW

433
00:14:01,874 --> 00:14:03,676
DIGITAL STRATEGY AND SHOWCASE

434
00:14:03,742 --> 00:14:05,010
THE APPLICATION OF DATA SCIENCE

435
00:14:05,077 --> 00:14:06,779
THROUGH MACHINE LEARNING AND

436
00:14:06,845 --> 00:14:07,913
ARTIFICIAL INTELLIGENCE, SO

437
00:14:07,980 --> 00:14:09,014
WE'RE BUILDING A NEW STRUCTURE

438
00:14:09,081 --> 00:14:11,650
AND PROCESS THAT WILL CREATE

439
00:14:11,717 --> 00:14:12,751
CONSISTENCY THROUGH NLM THROUGH

440
00:14:12,818 --> 00:14:13,786
A COMMON STRUCTURE AND PROCESS

441
00:14:13,852 --> 00:14:15,054
AND I WILL SHOW MORE ABOUT THIS

442
00:14:15,120 --> 00:14:18,123
ON THE NEXT COUPLE OF SLIDES.

443
00:14:18,190 --> 00:14:19,325
NEXT SLIDE, PLEASE.

444
00:14:19,391 --> 00:14:21,393
WE HAVE CONTINUED TO MAKE GREAT

445
00:14:21,460 --> 00:14:22,494
PROGRESS IN ESTABLISHING THIS

446
00:14:22,561 --> 00:14:25,297
FORWARD LOOKING APPROACH FOR

447
00:14:25,364 --> 00:14:28,033
INNOVATION THAT IS APPLIED

448
00:14:28,100 --> 00:14:30,236
CONSISTENTLY ACROSS THE NLM,

449
00:14:30,302 --> 00:14:32,037
THIS PIVOTS US ACROSS SEPARATE

450
00:14:32,104 --> 00:14:34,240
ACTIONS AT CONTINUOUS MODEL AND

451
00:14:34,306 --> 00:14:36,075
INTENTIONAL KNOW CONSIST EPT

452
00:14:36,141 --> 00:14:38,277
ACROSS UNITS AND EMBEDDED ACROSS

453
00:14:38,344 --> 00:14:39,578
ALL LEVELS OF THE ORGANIZE EGG,

454
00:14:39,645 --> 00:14:41,213
WITH THE DEVELOPMENT OF THE

455
00:14:41,280 --> 00:14:42,414
ROADMAP OF KEY ACTIVITIES AND

456
00:14:42,481 --> 00:14:44,283
MILESTONES REQUIRED FOR THIS

457
00:14:44,350 --> 00:14:46,485
CULTURE OF CONTINUOUS INNOVATION

458
00:14:46,552 --> 00:14:48,053
DEPLOATED BACK IN DECEMBER, WE

459
00:14:48,120 --> 00:14:51,357
HAVE NOW STARTED TO BUILD ACROSS

460
00:14:51,423 --> 00:14:52,524
4 MAJOR COMPONENTS, GOVERNANCE

461
00:14:52,591 --> 00:14:54,293
WHICH IS TO CREATE A STRUCTURE

462
00:14:54,360 --> 00:14:55,594
AND DECISION MAKING PROCESS TO

463
00:14:55,661 --> 00:14:57,129
APPROVE NOVEL IDEAS FROM ALL

464
00:14:57,196 --> 00:14:59,598
LEVELS OF STAFF ACROSS NLM, THE

465
00:14:59,665 --> 00:15:01,567
IDEA LIFE CYCLE WHICH IS TO BELL

466
00:15:01,634 --> 00:15:02,868
AN END-TO-END PROCESS THAT TAKES

467
00:15:02,935 --> 00:15:04,470
IDEAS FROM INITIAL CONCEPT,

468
00:15:04,536 --> 00:15:05,938
CONNECTED TO A PROBLEM

469
00:15:06,005 --> 00:15:07,740
STATEMENT, FOR THE KRAGZ OF NEW

470
00:15:07,806 --> 00:15:09,208
PROTOTYPES AND OR ALTERNATIVE

471
00:15:09,275 --> 00:15:10,809
METHODS TO IMPROVE OUR DELIVERY

472
00:15:10,876 --> 00:15:11,944
TO THE PUBLIC.

473
00:15:12,011 --> 00:15:14,513
THIRD IS MEASUREMENT, IT DEFINES

474
00:15:14,580 --> 00:15:15,581
STANDARD MEASURES TO EVALUATE

475
00:15:15,648 --> 00:15:17,349
THE RESULTS FROM OUR INNOVATION

476
00:15:17,416 --> 00:15:18,183
COMMITMENTS AND INFORM

477
00:15:18,250 --> 00:15:19,585
CONTINUOUS LEARNING AND

478
00:15:19,652 --> 00:15:21,520
IMPROVEMENT, AND FOURTH HIS

479
00:15:21,587 --> 00:15:23,689
ORGANIZATIONAL READINESS TO

480
00:15:23,756 --> 00:15:25,190
COMMUNICATE, ENGAGE AND

481
00:15:25,257 --> 00:15:26,125
REINFORCE CONTINUOUS EDUCATION

482
00:15:26,191 --> 00:15:27,493
ACROSS ALL LEVELS OF NLM.

483
00:15:27,559 --> 00:15:29,094
WE ARE ON TRACK TO HAVE AN

484
00:15:29,161 --> 00:15:31,163
INITIAL BUILD FOR EACH OF THESE

485
00:15:31,230 --> 00:15:33,165
COMPONENTS BY THE END OF JUNE

486
00:15:33,232 --> 00:15:35,367
WITH KEY ACTIVITIES CONTINUING

487
00:15:35,434 --> 00:15:37,603
INTO THE REMAINDER OF FISCAL

488
00:15:37,670 --> 00:15:40,105
YEAR 24 AND YOU THROUGHOUT

489
00:15:40,172 --> 00:15:41,540
FISCAL YEAR 2025, NEXT SLIDE,

490
00:15:41,607 --> 00:15:43,842
PLOAZ AS WE BUILD OUT THE

491
00:15:43,909 --> 00:15:45,444
CONTINUOUS INNOVATION LIFE CYCLE

492
00:15:45,511 --> 00:15:47,112
WE REMAIN COMMITTED TO CREATING

493
00:15:47,179 --> 00:15:48,213
A SUFFICIENT, SUPPORTIVE AND

494
00:15:48,280 --> 00:15:49,948
INCLUSIVE EXPERIENCE FOR ALL

495
00:15:50,015 --> 00:15:51,617
STAFF AT EVERY LEVEL ACROSS THE

496
00:15:51,684 --> 00:15:51,817
NLM.

497
00:15:51,884 --> 00:15:55,020
WE RECOGNIZE THAT CREATING A

498
00:15:55,087 --> 00:15:57,356
CULTURE OF INNOVATION IS A

499
00:15:57,423 --> 00:15:58,824
RESPONSIBILITY THAT LIES WITH

500
00:15:58,891 --> 00:16:00,726
EACH OF US, THEREFORE ANYONE AT

501
00:16:00,793 --> 00:16:03,028
NLM WITH SUBMIT AN INNOVATION

502
00:16:03,095 --> 00:16:04,630
IDEA, LEADERSHIP HAS BEEN

503
00:16:04,697 --> 00:16:06,265
THOUGHTFUL IN ITS VISIONING TO

504
00:16:06,332 --> 00:16:08,233
CREATE A PROCESS FOR NOVEL IDEAS

505
00:16:08,300 --> 00:16:10,069
THAT IS TRANSPARENT AND

506
00:16:10,135 --> 00:16:11,704
EQUITABLE REGARDLESS OF YOUR

507
00:16:11,770 --> 00:16:12,638
POSITION, OFFICE OR DIVISION,

508
00:16:12,705 --> 00:16:14,473
THERE WILL BE A SINGLE LOCATION

509
00:16:14,540 --> 00:16:16,308
TO SUBMIT YOUR IDEAS AND VIEW

510
00:16:16,375 --> 00:16:18,210
OTHER INNOVATION IDEAS FROM YOUR

511
00:16:18,277 --> 00:16:18,677
COLLEAGUES.

512
00:16:18,744 --> 00:16:20,546
IDEAS WILL BE EVALUATED THROUGH

513
00:16:20,612 --> 00:16:21,714
COMMON AND CONSISTENT CRITERIA,

514
00:16:21,780 --> 00:16:23,682
TO DETERMINE HOW AN INNOVATION

515
00:16:23,749 --> 00:16:25,684
IDEA WILL MOVE FORWARD, AND AS

516
00:16:25,751 --> 00:16:27,519
IDEAS MOVE FORWARD, A DEDICATED

517
00:16:27,586 --> 00:16:29,154
TEAM WILL BE ASSIGNED TO HELP

518
00:16:29,221 --> 00:16:30,622
BUILD AND REFINE THE INNOVATION

519
00:16:30,689 --> 00:16:34,927
IDEA AND CREATE A PROTOTYPE.

520
00:16:34,993 --> 00:16:37,596
ANC LAUNCHING INNOVATION ASSET,

521
00:16:37,663 --> 00:16:39,164
IT WILL BE--COMMUNICATED ACROSS

522
00:16:39,231 --> 00:16:42,167
THE NLM TO RECOGNIZE THE IDEA

523
00:16:42,234 --> 00:16:43,135
OWNER AND ASSIGNED TEAM MEMBERS

524
00:16:43,202 --> 00:16:45,003
FOR THEIR EFFORTS AND DEDICATION

525
00:16:45,070 --> 00:16:47,539
FOR FURTHERING OUR DELIVERY TO

526
00:16:47,606 --> 00:16:49,375
THE BIOMEDICAL COMMUNITY.

527
00:16:49,441 --> 00:16:50,075
NEXT SLIDE, PLEASE.

528
00:16:50,142 --> 00:16:51,043
WITH THAT, I WANT TO THANK YOU

529
00:16:51,110 --> 00:16:53,679
FOR YOUR TIME AND YOUR ATTENTION

530
00:16:53,746 --> 00:16:55,347
TODAY, I'M VERY PROUD OF THE

531
00:16:55,414 --> 00:16:57,750
INCREDIBLE WORK COMING OUT OF

532
00:16:57,816 --> 00:16:58,751
THE NLM'S INTRAMURAL RESEARCH

533
00:16:58,817 --> 00:16:59,852
PROGRAM AND I LOOK FORWARD FROM

534
00:16:59,918 --> 00:17:01,019
HEARING FROM OUR INVESTIGATORS

535
00:17:01,086 --> 00:17:04,022
AS WELL AS NLM SCIENTIFIC

536
00:17:04,089 --> 00:17:04,623
DIRECTOR DR. RICHARD SHERMAN,

537
00:17:04,690 --> 00:17:09,928
RICHARD I WILL NOW PASS THIS

538
00:17:09,995 --> 00:17:10,963
OVER TO YOU.

539
00:17:11,029 --> 00:17:21,507
>> GREAT, THANK YOU STEVE.

540
00:17:33,819 --> 00:17:35,154
ALL RIGHT, WELCOME EVERYONE.

541
00:17:35,220 --> 00:17:37,423
I HAD A FEW COMMENTS I WANTED TO

542
00:17:37,489 --> 00:17:41,360
MAKE AT THE BEGINNING OF OUR

543
00:17:41,427 --> 00:17:42,194
MEETING TODAY--

544
00:17:42,261 --> 00:17:43,695
>> YOU'RE SHARING THE PRESENTER

545
00:17:43,762 --> 00:17:47,433
VIEW, WHICH IS FINE JUST FYI.

546
00:17:47,499 --> 00:17:49,067
NOKAY, LET ME SWAP THIS.

547
00:17:49,134 --> 00:17:54,006
>> THERE YOU GO?

548
00:17:54,072 --> 00:17:54,673
THANKS, PETER.

549
00:17:54,740 --> 00:17:57,075
LET ME START BY OFFERING MY

550
00:17:57,142 --> 00:18:01,380
SINCEREST THANK TO YOU DR. KAING

551
00:18:01,447 --> 00:18:03,348
AND DR. ONOMA CHAD O, TODAY WILL

552
00:18:03,415 --> 00:18:06,251
BE THEIR LAST MEETING THEY WILL

553
00:18:06,318 --> 00:18:06,585
BE ATTENDING.

554
00:18:06,652 --> 00:18:09,621
WE REALLY APPRECIATE YOUR

555
00:18:09,688 --> 00:18:09,855
SERVICE.

556
00:18:09,922 --> 00:18:11,757
I KNOW THAT SERVING ON THESE

557
00:18:11,824 --> 00:18:13,392
KINDS OF BOARDS REQUIRES

558
00:18:13,459 --> 00:18:17,763
SIGNIFICANT  EFFORT AND YOUR

559
00:18:17,830 --> 00:18:19,398
CONTRIBUTIONS TO THE SUCCESS OF

560
00:18:19,465 --> 00:18:20,432
THE INTRAMURAL RESEARCH PROGRAM

561
00:18:20,499 --> 00:18:21,867
HERE IS GREATLY APPRECIATED SO I

562
00:18:21,934 --> 00:18:26,271
JUST WANT TO OFFER MY SINCEREST

563
00:18:26,338 --> 00:18:26,939
THANKS.

564
00:18:27,005 --> 00:18:28,707
AND WE HAVE IDENTIFIED 2 NEW

565
00:18:28,774 --> 00:18:33,445
BOARD MEMBERS WHO WILL BE

566
00:18:33,512 --> 00:18:35,180
JOINING US FOR THE NEXT MEETING

567
00:18:35,247 --> 00:18:38,484
WHICH IS DR. MONA SING WHICH IS

568
00:18:38,550 --> 00:18:40,519
A PROFESSOR AT SCIENCE AT

569
00:18:40,586 --> 00:18:42,020
PRINCETON UNIVERSITY AND

570
00:18:42,087 --> 00:18:44,289
DR. WILLIAM HOGAN WHO IS THE

571
00:18:44,356 --> 00:18:45,991
DIRECTOR OF A NEW DATA SCIENCE

572
00:18:46,058 --> 00:18:48,627
INSTITUTE AT THE MEDICAL COLLEGE

573
00:18:48,694 --> 00:18:51,129
OF WISCONSIN, SO WE LOOK FORWARD

574
00:18:51,196 --> 00:18:53,665
TO HAVING THESE 2 NEW BOSC

575
00:18:53,732 --> 00:18:57,603
MEMBERS JOIN US AT OUR NEXT

576
00:18:57,669 --> 00:18:57,836
MEETING.

577
00:18:57,903 --> 00:18:59,104
ONE OF THE THINGS WE HAVE BEEN

578
00:18:59,171 --> 00:19:01,006
WORKING ON OVER THE LAST SEVERAL

579
00:19:01,073 --> 00:19:04,176
MONTHS IS THAT WE ARE DEVELOPING

580
00:19:04,243 --> 00:19:05,777
A STRAY JECTORY STIGIC PLAN FOR

581
00:19:05,844 --> 00:19:09,414
THE NATIONAL LIBRARY OF

582
00:19:09,481 --> 00:19:10,048
MEDICINES INTRAMURAL RESEARCH

583
00:19:10,115 --> 00:19:12,885
PROGRAM EMPLOY SO WE HAVE

584
00:19:12,951 --> 00:19:14,686
ASSEMBLED A COMMITTEE OF

585
00:19:14,753 --> 00:19:17,356
INVESTIGATORS WITHIN THE IRP TO

586
00:19:17,422 --> 00:19:21,760
BEGIN THAT PROCESS, AND WE'VE

587
00:19:21,827 --> 00:19:22,327
RECENTLY ENGAGED CONSULTANT

588
00:19:22,394 --> 00:19:24,997
CONTRACTOR WHO IS HELPING KIND

589
00:19:25,063 --> 00:19:26,298
OF SHEPHERD US THROUGH THIS

590
00:19:26,365 --> 00:19:26,532
PROCESS.

591
00:19:26,598 --> 00:19:28,100
ONE OF THE PLANNED ACTIVITIES IS

592
00:19:28,166 --> 00:19:31,637
THAT WE ARE GOING TO BE HAVING A

593
00:19:31,703 --> 00:19:33,539
2 DAY RETREAT AT THE BEGINNING

594
00:19:33,605 --> 00:19:34,706
OF UNIWHERE WE ARE REALLY GOING

595
00:19:34,773 --> 00:19:36,975
TO BE DIVING INTO THE DETAILS OF

596
00:19:37,042 --> 00:19:39,578
THE STRATEGIC PLANNING PROCESS.

597
00:19:39,645 --> 00:19:43,415
AND THAT WILL INCLUDE DOING

598
00:19:43,482 --> 00:19:45,384
THINGS LIKE A SWAT ANALYSIS

599
00:19:45,450 --> 00:19:46,718
WHERE WE IDENTIFY THE STRENGTHS

600
00:19:46,785 --> 00:19:50,155
OF THE EXISTING PROGRAM AND

601
00:19:50,222 --> 00:19:53,692
POSSIBLE WEAKNESSES AND THEN

602
00:19:53,759 --> 00:19:57,696
IDENTIFY OPPORTUNITIES FOR

603
00:19:57,763 --> 00:19:59,131
EXPANDING OUR RESEARCH PROGRAM

604
00:19:59,197 --> 00:20:00,732
TO ADDRESS SOME OF THE

605
00:20:00,799 --> 00:20:02,401
WEAKNESSES WE IDENTIFY.

606
00:20:02,467 --> 00:20:03,802
WE WOULD VERY MUCH APPRECIATE

607
00:20:03,869 --> 00:20:05,637
INPUT FROM AND THE PERSPECTIVE

608
00:20:05,704 --> 00:20:09,942
OF THE BOARD OF SCIENTIFIC

609
00:20:10,008 --> 00:20:11,076
COUNSELORS IN THIS PROCESS AND

610
00:20:11,143 --> 00:20:16,014
SO IN ORDER TO DO THAT, WE HAVE

611
00:20:16,081 --> 00:20:19,618
SET ASIDE A DEDICATED SESSION

612
00:20:19,685 --> 00:20:22,287
DURING THAT WORKSHOP ON

613
00:20:22,354 --> 00:20:23,989
JUNE--MONDAY JUNE 3rd FROM

614
00:20:24,056 --> 00:20:28,393
1:30 TO 315, WHERE WE WOULD LIKE

615
00:20:28,460 --> 00:20:30,529
TO INVITE BOSC MEMBERS TO

616
00:20:30,596 --> 00:20:34,299
PARTICIPATE IN THIS EVALUATION

617
00:20:34,366 --> 00:20:34,666
PROCESS.

618
00:20:34,733 --> 00:20:37,803
THIS IS AN OPTIONAL--YOU KNOW

619
00:20:37,869 --> 00:20:40,238
ATTENDING THIS SESSION IS

620
00:20:40,305 --> 00:20:41,440
OPTIONAL FOR BOSC MEMBERS BUT WE

621
00:20:41,506 --> 00:20:43,141
WOULD VERY MUCH LIKE YOU TO BE

622
00:20:43,208 --> 00:20:46,411
ABLE TO ATTEND IF POSSIBLE.

623
00:20:46,478 --> 00:20:48,380
SINCE MANY OF YOU MAY HAVE

624
00:20:48,447 --> 00:20:51,650
CONFLICTS FOR THAT TIME, WE WILL

625
00:20:51,717 --> 00:20:57,055
ALSO PLAN ON SETTING ASIDE TIME

626
00:20:57,122 --> 00:20:59,057
TO GET STRATEGIC--FOR STRATEGIC

627
00:20:59,124 --> 00:21:01,026
DISCUSSIONS WITH BOSC MEMBERS

628
00:21:01,093 --> 00:21:03,362
DURING FUTURE BOSC MEETINGS AND

629
00:21:03,428 --> 00:21:06,365
ALSO IF WE MAY ALSO WANT TO

630
00:21:06,431 --> 00:21:08,467
SCHEDULE SOME INDIVIDUAL

631
00:21:08,533 --> 00:21:09,201
ONE-ON-ONE SESSIONS FOR PEOPLE

632
00:21:09,267 --> 00:21:11,403
WHO WOULD NOT BE ABLE TO ATTEND

633
00:21:11,470 --> 00:21:15,140
THAT SESSION DURING THE RETREAT.

634
00:21:15,207 --> 00:21:17,009
SO JUST IN SUMMARY, YOU KNOW WE

635
00:21:17,075 --> 00:21:19,544
ARE GOING TO HAVE A DEDICATED

636
00:21:19,611 --> 00:21:22,280
SESSION FOR BOARD OF SCIENTIFIC

637
00:21:22,347 --> 00:21:24,016
COUNSELOR MEMBERS ON MONDAY

638
00:21:24,082 --> 00:21:26,018
JUNE 3rd FROM 1:30 TO 3:15,

639
00:21:26,084 --> 00:21:28,286
AND WE WOULD LOVE TO HAVE YOU

640
00:21:28,353 --> 00:21:34,660
ATTEND BUT ATTENDANCE IS

641
00:21:34,726 --> 00:21:35,794
OPTIONAL.

642
00:21:35,861 --> 00:21:37,095
SO DURING STEVE'S COMMENTS, HE

643
00:21:37,162 --> 00:21:40,098
MENTIONED THE FACT THAT IN

644
00:21:40,165 --> 00:21:42,134
DECEMBER, THE IRP TRAINEES VOTED

645
00:21:42,200 --> 00:21:44,169
TO ESTABLISH A COLLECTIVE

646
00:21:44,236 --> 00:21:46,104
BARGAINING UNIT THAT WILL BE

647
00:21:46,171 --> 00:21:49,808
ADMINISTERED BY THE UAW.

648
00:21:49,875 --> 00:21:53,412
THE RULES FOR ENGAGEMENT WITH

649
00:21:53,478 --> 00:21:55,814
OUR TRAINEES THROUGH THIS

650
00:21:55,881 --> 00:21:57,249
COLLECTIVE BARGAINING UNIT ARE

651
00:21:57,315 --> 00:21:58,684
CURRENTLY BEING WORKED OUT SO

652
00:21:58,750 --> 00:22:00,852
WE'RE STILL NOT SURE EXACTLY HOW

653
00:22:00,919 --> 00:22:03,955
WE'RE GOING TO BE ENGAGING WITH

654
00:22:04,022 --> 00:22:08,293
OUR TRAINEES IN THIS UNIONIZED

655
00:22:08,360 --> 00:22:10,862
ENVIRONMENT BUT IN THE MEAN

656
00:22:10,929 --> 00:22:12,964
TIME, WE WANT TO BE VERY CAREFUL

657
00:22:13,031 --> 00:22:16,702
ABOUT HOW WE ARE INTERACTING

658
00:22:16,768 --> 00:22:17,903
WITH OUR TRAINEES AND SINCE

659
00:22:17,969 --> 00:22:21,840
THERE WILL BE SESSIONS DURING

660
00:22:21,907 --> 00:22:23,475
THIS PARTICULAR MEETING WHERE WE

661
00:22:23,542 --> 00:22:25,811
WILL HEAR POSTER PRESENTATIONS

662
00:22:25,877 --> 00:22:29,381
FROM SOME OF OUR TRAINEES AND SO

663
00:22:29,448 --> 00:22:30,816
WE JUST WANT TO MAKE SURE THAT

664
00:22:30,882 --> 00:22:31,917
THE BOARD IS AWARE OF THE FACT

665
00:22:31,983 --> 00:22:35,887
THAT WE ARE IN THE MIDDLE OF

666
00:22:35,954 --> 00:22:37,289
THIS UNIONIZATION PROCESS, AND

667
00:22:37,355 --> 00:22:39,791
THAT YOUR INTERACTIONS WITH THE

668
00:22:39,858 --> 00:22:42,194
TRAINEES SHOULD REALLY FOCUS ON

669
00:22:42,260 --> 00:22:43,028
UNDERSTANDING THE TRAINING

670
00:22:43,095 --> 00:22:46,531
ENVIRONMENT THAT'S PROVIDED BY

671
00:22:46,598 --> 00:22:46,932
THE MENTOR.

672
00:22:46,998 --> 00:22:49,768
IT SHOULD NOT REALLY BE, YOU

673
00:22:49,835 --> 00:22:52,104
KNOW CRITIQUING THE WORK OF THE

674
00:22:52,170 --> 00:22:54,940
TRAINEE, I MEAN, PROVIDING

675
00:22:55,006 --> 00:22:56,341
SCIENTIFIC FEEDBACK IS CERTAINLY

676
00:22:56,408 --> 00:22:57,809
WELCOMED, BUT WE QUUOF THE WANT

677
00:22:57,876 --> 00:22:59,377
TO MAKE SURE THAT--JUST WANT TO

678
00:22:59,444 --> 00:23:01,113
MAKE SURE THAT WE'RE NOT TALKING

679
00:23:01,179 --> 00:23:03,949
ABOUT THINGS LIKE THEIR WORK

680
00:23:04,015 --> 00:23:05,283
ENVIRONMENT AND QUESTIONS LIKE

681
00:23:05,350 --> 00:23:06,918
THAT WHICH WOULD BE COVERED

682
00:23:06,985 --> 00:23:09,321
UNDER 1 OF THESE COLLECTIVE

683
00:23:09,387 --> 00:23:09,821
BARGAINING AGREEMENTS.

684
00:23:09,888 --> 00:23:11,523
SO YOU KNOW AS I'M SURE YOU WERE

685
00:23:11,590 --> 00:23:14,726
GOING TO DO ANYWAY, FOCUSING IN

686
00:23:14,793 --> 00:23:18,730
ON THEIR SCIENCE AND RESEARCH IS

687
00:23:18,797 --> 00:23:20,665
PERFECTLY ACCEPTABLE AND REALLY

688
00:23:20,732 --> 00:23:23,568
CONSIDERING YOU KNOW HOW THEIR

689
00:23:23,635 --> 00:23:25,170
WORK IS A REFLECTION OF THE

690
00:23:25,237 --> 00:23:27,205
MENTORING THAT THEY'RE RECEIVING

691
00:23:27,272 --> 00:23:31,243
FROM THE INVESTIGATOR IS REALLY

692
00:23:31,309 --> 00:23:36,648
THE FOCUS OF THAT ASSESSMENT.

693
00:23:36,715 --> 00:23:38,717
AND THEN FINALLY, I JUST WANTED

694
00:23:38,784 --> 00:23:40,619
TO REMIND PEOPLE THAT 1 OF THE

695
00:23:40,685 --> 00:23:43,021
INVESTIGATORS WHO WILL BE

696
00:23:43,088 --> 00:23:51,029
PRESENTING TODAY, IS DR. DINA

697
00:23:51,096 --> 00:23:52,264
DEMNER-FUSHMAN, SHE IS AT THE

698
00:23:52,330 --> 00:23:53,565
TENURE TAIJ IN HER TRACK WHERE

699
00:23:53,632 --> 00:23:55,867
HE WILL BE CONSIDERING HER

700
00:23:55,934 --> 00:23:57,169
PROMOTION TO A TENURED

701
00:23:57,235 --> 00:24:00,405
INVESTIGATOR, SO WE WANT TO ASK

702
00:24:00,472 --> 00:24:02,908
THAT THE BOARD CONSIDERS THAT

703
00:24:02,974 --> 00:24:10,248
PROMOTION TO TENURE AS PART OF

704
00:24:10,315 --> 00:24:11,283
THEIR EVALUATION PROCESS EMPLOY

705
00:24:11,349 --> 00:24:13,118
SO THOSE ARE ALL THE COMMENTS

706
00:24:13,185 --> 00:24:15,487
THAT I WANTED TO COVER THIS

707
00:24:15,554 --> 00:24:15,754
MORNING.

708
00:24:15,821 --> 00:24:17,756
DAVID, I DON'T KNOW IF THERE'S

709
00:24:17,823 --> 00:24:18,723
ANYTHING ELSE FROM YOUR

710
00:24:18,790 --> 00:24:20,292
PERSPECTIVE THAT WE WANTED TO

711
00:24:20,358 --> 00:24:21,293
DISCUSS WITH THE BOARD BEFORE

712
00:24:21,359 --> 00:24:26,097
THEY GET STARTED WITH THEIR

713
00:24:26,164 --> 00:24:28,200
INVESTIGATOR PRESENTATIONS?

714
00:24:28,266 --> 00:24:32,137
>> NO, THERE WAS 1 QUESTION FROM

715
00:24:32,204 --> 00:24:37,442
[INDISCERNIBLE] ABOUT ARE THEY

716
00:24:37,509 --> 00:24:38,510
ALREADY UNIONIZED, SO MY ANSWER

717
00:24:38,577 --> 00:24:40,912
WAS YES, THEY WERE UNIONIZED

718
00:24:40,979 --> 00:24:42,447
LAST YEAR BUT COLLECTIVE

719
00:24:42,514 --> 00:24:45,217
BARGAINING IS BEING WORKED ON AT

720
00:24:45,283 --> 00:24:46,151
THE PRESENT.

721
00:24:46,218 --> 00:24:46,585
NRIGHT.

722
00:24:46,651 --> 00:24:46,785
YEAH.

723
00:24:46,852 --> 00:24:53,158
ALL RIGHT, THANK YOU.

724
00:24:53,225 --> 00:24:55,560
>> AND WE'RE REGARDING AS ALMOST

725
00:24:55,627 --> 00:24:57,262
COMPLETE.

726
00:24:57,329 --> 00:24:58,063
>> YES, THEY HAVEUNIZED AUTOJUST

727
00:24:58,129 --> 00:25:00,298
THE TERMS OF THE COLLECTIVE

728
00:25:00,365 --> 00:25:01,266
BARGAINING AGREEMENT ARE RIGHT

729
00:25:01,333 --> 00:25:06,371
NOW, SO WE DON'T KNOW WHAT THOSE

730
00:25:06,438 --> 00:25:07,939
TERMS ARE YET.

731
00:25:08,006 --> 00:25:13,278
ARE THERE OTHER QUESTIONS?

732
00:25:13,345 --> 00:25:14,713
OKAY, IF NOT, WELL IF NOT I WILL

733
00:25:14,779 --> 00:25:23,655
TURN IT BACK OVER TO PETER.

734
00:25:23,722 --> 00:25:27,492
>> YEAH, I HAD A QUESTION,

735
00:25:27,559 --> 00:25:30,128
REGARDING THE PUBLIC NATURE OF

736
00:25:30,195 --> 00:25:31,129
THIS MEETING, IS THERE LIKE A

737
00:25:31,196 --> 00:25:35,166
WAY THAT WE CAN KNOW WHEN--THERE

738
00:25:35,233 --> 00:25:36,568
WAS AN INSTANCE IN THE PAST IT

739
00:25:36,635 --> 00:25:38,403
WAS PUBLIC AND WE DIDN'T KNOW IT

740
00:25:38,470 --> 00:25:38,904
WAS PUBLIC.

741
00:25:38,970 --> 00:25:40,906
SO WHICH PARTS OF THIS SESSION

742
00:25:40,972 --> 00:25:43,108
ARE OPEN TO THE PUBLIC WILL BE A

743
00:25:43,174 --> 00:25:47,279
GOOD QUESTION TO CLARIFY.

744
00:25:47,345 --> 00:25:50,215
>> MAYBE I CAN BRING UP THE

745
00:25:50,282 --> 00:25:52,284
AGENDA AND MAYBE THAT WILL BE

746
00:25:52,350 --> 00:25:52,651
HELPFUL.

747
00:25:52,717 --> 00:25:54,986
>> I CAN ANSWER THAT PRETTY

748
00:25:55,053 --> 00:25:55,854
QUICKLY AND EASILY.

749
00:25:55,921 --> 00:26:01,726
IT'S EVERYTHING IS PUBLIC EXCEPT

750
00:26:01,793 --> 00:26:03,895
FOR WHEN THE BSC IS IN COMMITTEE

751
00:26:03,962 --> 00:26:10,669
WITH A SUBSET OF NLM IRP

752
00:26:10,735 --> 00:26:11,436
MEMBERS.

753
00:26:11,503 --> 00:26:13,638
SO WHEN YOU--ON THE AGENDA, I'M

754
00:26:13,705 --> 00:26:16,841
LOOKING AT THE AGENDA, I IF I

755
00:26:16,908 --> 00:26:21,913
CAN FIND IT, HERE IT IS, SO

756
00:26:21,980 --> 00:26:29,454
WHEREVER IT SAYS CLOSED SESSION,

757
00:26:29,521 --> 00:26:31,289
THAT'S NOT PUBLIC.

758
00:26:31,356 --> 00:26:33,191
>> THANK YOU.

759
00:26:33,258 --> 00:26:36,828
>> MAYBE I BETTER BE MORE CLEAR

760
00:26:36,895 --> 00:26:38,129
ON THAT SINCE THE QUESTION CAME

761
00:26:38,196 --> 00:26:42,901
UP, IN MY FUTURE AGENDAS.

762
00:26:42,968 --> 00:26:46,304
>> ALL RIGHT, SO, THANK YOU VERY

763
00:26:46,371 --> 00:26:46,638
MUCH RICHARD.

764
00:26:46,705 --> 00:26:47,939
WHAT WE SHOULD I THINK GO AHEAD

765
00:26:48,006 --> 00:26:53,178
AND DO NOW IS MOVE TO THE

766
00:26:53,244 --> 00:26:54,946
PRESENTATION BY DR. LU, SO WE'RE

767
00:26:55,013 --> 00:26:56,348
A FEW MINUTES EARLY BUT I WOULD

768
00:26:56,414 --> 00:27:06,758
SAY LET'S GO AHEAD.

769
00:27:09,928 --> 00:27:20,105
>> OKAY, GOOD MORNING EVERYONE.

770
00:27:20,171 --> 00:27:23,341
CAN YOU SEE MY SLIDE?

771
00:27:23,408 --> 00:27:23,608
>> YES.

772
00:27:23,675 --> 00:27:26,044
>> YOU CAN HEAR ME OKAY, AS

773
00:27:26,111 --> 00:27:28,113
WELL?

774
00:27:28,179 --> 00:27:29,547
>> YES.

775
00:27:29,614 --> 00:27:30,949
GREAT.

776
00:27:31,016 --> 00:27:35,153
LET'S GET STARTED AGAIN, MY NAME

777
00:27:35,220 --> 00:27:37,455
IS ZHIYONG LU, I AM A SENIOR

778
00:27:37,522 --> 00:27:39,591
INVESTIGATOR AT THE IRP, MY LAB

779
00:27:39,657 --> 00:27:41,793
IS INTERESTED IN USING AI AND

780
00:27:41,860 --> 00:27:42,927
MISSION LEARNING FOR BIOMEDICAL

781
00:27:42,994 --> 00:27:44,763
TEXT AND IMAGE PROCESSING ON A

782
00:27:44,829 --> 00:27:45,797
DAILY BASIS.

783
00:27:45,864 --> 00:27:47,799
WE PRIMARILY WORK WITH MILLIONS

784
00:27:47,866 --> 00:27:49,934
OF PARTICLES IN THE BIOMEDICAL

785
00:27:50,001 --> 00:27:51,102
LITERATURE WHEN ACCESS IS

786
00:27:51,169 --> 00:27:52,804
ALLOWED, WE ARE ALSO WORKING

787
00:27:52,871 --> 00:27:54,439
WITH PATIENT DATA INCLUDING

788
00:27:54,506 --> 00:27:55,473
ELECTRONIC MEDICAL RECORDS AS

789
00:27:55,540 --> 00:27:56,474
WELL AS CLINICAL IMAGES IN

790
00:27:56,541 --> 00:27:58,810
PARTICULAR IN THE AREAS OF

791
00:27:58,877 --> 00:28:00,111
OPERATING GLOBALLYTAL MOLOGY AND

792
00:28:00,178 --> 00:28:03,014
RADIOLOGY BECAUSE WE HAVE GREAT

793
00:28:03,081 --> 00:28:05,583
CLINICAL COLLABORATIONOT NIESH H

794
00:28:05,650 --> 00:28:06,751
CAMPUS, SO USING AI, AND MISSION

795
00:28:06,818 --> 00:28:08,186
LEARNING WE WORK ON A WIDE

796
00:28:08,253 --> 00:28:10,488
VARIETY OF DIFFERENT PROBLEMS,

797
00:28:10,555 --> 00:28:13,024
INCLUDING IMPROVING INFORMATION

798
00:28:13,091 --> 00:28:13,658
ACCESS, BIOMEDICAL LITERATURE,

799
00:28:13,725 --> 00:28:15,026
WHICH IS VERY TIED TO THE

800
00:28:15,093 --> 00:28:16,795
MISSION OF NATIONAL LIBRARY OF

801
00:28:16,861 --> 00:28:17,929
MEDICINE, BEYOND THAT, WE ARE

802
00:28:17,996 --> 00:28:22,534
ALSO INTERESTED IN USING TO

803
00:28:22,600 --> 00:28:24,502
EXTRACT, KNOWLEDGE FROM

804
00:28:24,569 --> 00:28:25,904
UNOBSTRUCTED TEXT AND TURNING

805
00:28:25,970 --> 00:28:27,305
THAT INTO STRUCTURED AND

806
00:28:27,372 --> 00:28:28,139
COMPUTABLE INFORMATION AND LAST

807
00:28:28,206 --> 00:28:31,242
BUT NOT LEAST, USING AI IN THE

808
00:28:31,309 --> 00:28:32,677
LARGE MODELS WE'RE ABOUTED IN

809
00:28:32,744 --> 00:28:35,513
TAKING THE PATIENT DATA

810
00:28:35,580 --> 00:28:37,348
INCLUDING MOST IMAGE AND TEXT IN

811
00:28:37,415 --> 00:28:39,250
PULTY MODAL FASHION FOR MISSION

812
00:28:39,317 --> 00:28:41,753
DIAGNOSIS AND PROGNOSIS ALONG

813
00:28:41,820 --> 00:28:42,587
WITH OTHER APPLICATIONS.

814
00:28:42,654 --> 00:28:45,356
SO FOR TODAY'S PRESENTATION, I

815
00:28:45,423 --> 00:28:47,092
WILL BE PRESENTING SOME OF THE

816
00:28:47,158 --> 00:28:49,661
PROGRESS WE MADE IN EACH OF

817
00:28:49,727 --> 00:28:50,161
THESE AREAS.

818
00:28:50,228 --> 00:28:55,567
SO I WILL TART WITH

819
00:28:55,633 --> 00:28:58,136
[INDISCERNIBLE], SO ON AVERAGE

820
00:28:58,203 --> 00:29:01,172
DAY WE HAVE 5.2 MILLION USERS,

821
00:29:01,239 --> 00:29:01,706
3 MILLION SEARCHERS AND

822
00:29:01,773 --> 00:29:03,141
9 MILLION PAGE VIEWS SO EVEN

823
00:29:03,208 --> 00:29:04,676
THOUGH AI IS MISSION LEARNING IS

824
00:29:04,742 --> 00:29:06,711
A HOT TOPIC THESE DAYS BUT SINCE

825
00:29:06,778 --> 00:29:09,414
DAY 1, AT NCBI AND NLM WHICH IS

826
00:29:09,481 --> 00:29:11,015
ABOUT OVER 17 YEARS NOW, WE'VE

827
00:29:11,082 --> 00:29:14,119
ALWAYS BEEN INTERESTED IN USING

828
00:29:14,185 --> 00:29:16,654
AI IN THE MISSION TO IMPROVE PUB

829
00:29:16,721 --> 00:29:17,155
MED SEARCH.

830
00:29:17,222 --> 00:29:19,424
SO HERE JUST A LIST OF SEARCH

831
00:29:19,491 --> 00:29:21,392
FEATURES THAT ASSIST AND MADE

832
00:29:21,459 --> 00:29:22,627
AVAILABLE THROUGH COMPUTATIONAL

833
00:29:22,694 --> 00:29:25,363
MISSION LEARNING TECHNIQUES, SO,

834
00:29:25,430 --> 00:29:26,998
DIFFERENT FROM THE COMMERCIAL AI

835
00:29:27,065 --> 00:29:28,733
SYSTEMS WHAT WE HAVE TRIED TO DO

836
00:29:28,800 --> 00:29:31,136
IS TRIED VERY HARD TO EXPLAIN TO

837
00:29:31,202 --> 00:29:32,904
OUR USERS HOW THE DETAILS OF

838
00:29:32,971 --> 00:29:38,610
NUTS AND BOLTS OF THESE AI

839
00:29:38,676 --> 00:29:39,310
GENERATED SEARCH ASSISTANCE.

840
00:29:39,377 --> 00:29:42,914
SO WE HAVE PRETTY MUCH--EVERY

841
00:29:42,981 --> 00:29:45,016
FOR EACH OF THE SEARCH FEATURE

842
00:29:45,083 --> 00:29:46,885
WE HAVE A PAPER ASSOCIATED AND

843
00:29:46,951 --> 00:29:48,419
IN THE PRESENTATION FOLLOWED BY

844
00:29:48,486 --> 00:29:52,624
OUR TRAINEES, YOU WILL HEAR 1 BY

845
00:29:52,690 --> 00:29:54,225
ASHLEY SHIN, A POST BACK IN MY

846
00:29:54,292 --> 00:29:55,860
GROUP, SHE WILL BE PRECEPTING TO

847
00:29:55,927 --> 00:30:00,331
YOU A NEW PROPOSAL THAT WE THINK

848
00:30:00,398 --> 00:30:01,332
WILL INCREASE TRANSPARENCY OF

849
00:30:01,399 --> 00:30:04,269
READING ARTICLE WHICH IS A

850
00:30:04,335 --> 00:30:06,104
WIDELY USED FEATURE IN PUB MED.

851
00:30:06,171 --> 00:30:07,639
BECAUSE WE BELIEVE THE MORE YOU

852
00:30:07,705 --> 00:30:10,308
KNOW ABOUT A SYSTEM OR FEATURE

853
00:30:10,375 --> 00:30:12,076
THE BETTER OUR USERS WILL MAKE

854
00:30:12,143 --> 00:30:12,577
USE OF THIS.

855
00:30:12,644 --> 00:30:14,679
BUT AS MUCH AS I LOVE PUB MED

856
00:30:14,746 --> 00:30:17,882
BUT IN THE END IT IS A KEY WORD

857
00:30:17,949 --> 00:30:19,884
BASED SYSTEM, SO THAT MEANS FOR

858
00:30:19,951 --> 00:30:21,119
SIPTAL INFORMATION SEARCH NEEDS

859
00:30:21,186 --> 00:30:22,554
SUCH AS FINDING THE PAPER ANDS

860
00:30:22,620 --> 00:30:26,291
ONLY THE PAPERS OF A PARTICULAR

861
00:30:26,357 --> 00:30:30,228
TOPIC, THIS TOPIC COULD BE COULD

862
00:30:30,295 --> 00:30:32,030
COVID-19, GWAS, CANCER

863
00:30:32,096 --> 00:30:33,031
DISPARITY, HOPEFULLY OUR USERS

864
00:30:33,097 --> 00:30:35,266
WILL COME UP WITH A LONG LIST OF

865
00:30:35,333 --> 00:30:37,135
SEARCH TERMS, THEY TRY TO BE

866
00:30:37,202 --> 00:30:37,835
BOTH COMPREHENSIVE AND PRECISE

867
00:30:37,902 --> 00:30:40,038
BUT YOU SEE THESE 2 THINGS ARE

868
00:30:40,104 --> 00:30:42,373
TRADE OFF OF EACH OTHER, NOT TO

869
00:30:42,440 --> 00:30:43,608
MENTION CERTAIN TOPICS SUCH AS

870
00:30:43,675 --> 00:30:44,943
COVID, THERE'S NO TERM THAT KEY

871
00:30:45,009 --> 00:30:47,378
WORD YOU CAN USE BECAUSE THE

872
00:30:47,445 --> 00:30:49,647
TERM COVID-19 WAS NOT COINED AND

873
00:30:49,714 --> 00:30:52,383
USED IN THE LITERATURE IN THE

874
00:30:52,450 --> 00:30:54,452
RESEARCH COMMUNITY UNTIL SUMMER

875
00:30:54,519 --> 00:30:54,719
2020.

876
00:30:54,786 --> 00:30:57,789
SO, ALL THE PAPERS THAT BEFORE

877
00:30:57,855 --> 00:31:00,625
THAT, ARE BASICALLY USING ALL

878
00:31:00,692 --> 00:31:01,826
DIFFERENT WAYS DESCRIBING THE

879
00:31:01,893 --> 00:31:06,497
COVID RESEARCH, SO IT WAS VERY

880
00:31:06,564 --> 00:31:07,999
CHALLENGING IN THE BEGINNING OF

881
00:31:08,066 --> 00:31:09,267
THE PANDEMIC TO ELECTRIC FOR

882
00:31:09,334 --> 00:31:10,101
PAPERS ABOUT COVID RESEARCH AND

883
00:31:10,168 --> 00:31:11,502
THAT OF COURSE BASICALLY OUR

884
00:31:11,569 --> 00:31:13,304
MOTIVATION TO CREATE A CENTRAL

885
00:31:13,371 --> 00:31:16,874
REPOSITORY FOR ALL THE PAPERS WE

886
00:31:16,941 --> 00:31:18,543
CALLED LATE COVID AND IT PLAYED

887
00:31:18,610 --> 00:31:21,646
AN IMPORTANT REALM FOR NOT ONLY

888
00:31:21,713 --> 00:31:23,014
SCIENTIFIC LITERATURE BUT ALSO

889
00:31:23,081 --> 00:31:24,215
THE GENERAL PUBLIC AND THIS WAS

890
00:31:24,282 --> 00:31:25,683
TAKEN WHEN MY FORMER COLLEAGUE

891
00:31:25,750 --> 00:31:30,855
AND POST DOC TRAINEE,

892
00:31:30,922 --> 00:31:32,190
DR. [INDISCERNIBLE] WHO IS

893
00:31:32,257 --> 00:31:33,491
STARTING HIS LAB AT YALE

894
00:31:33,558 --> 00:31:33,925
UNIVERSITY.

895
00:31:33,992 --> 00:31:35,326
NOW WE START THIS PROJECT WITH

896
00:31:35,393 --> 00:31:37,128
JUST 2 POST DOCS AND MYSELF.

897
00:31:37,195 --> 00:31:39,063
SO WE STARTED PROJECT, WE

898
00:31:39,130 --> 00:31:41,933
BASICALLY HAVE TO MANUALLY LOOK

899
00:31:42,000 --> 00:31:44,235
AT ALL THE PAPERS TO TAKE IN TO

900
00:31:44,302 --> 00:31:45,536
SEE IF THEY'RE RELEVANT TO COVID

901
00:31:45,603 --> 00:31:47,305
RESEARCH AND WHAT KIND OF TOPICS

902
00:31:47,372 --> 00:31:50,975
THEY MENTION, SO WE STARTED, WE

903
00:31:51,042 --> 00:31:54,812
WORKED ON THESE 2 PROBLEMS

904
00:31:54,879 --> 00:31:55,780
MANUALLY FOR SEVERAL MONTHS OF

905
00:31:55,847 --> 00:31:58,783
IN THE BEGINNING OF THE PANDEMIC

906
00:31:58,850 --> 00:32:00,852
BUT AFTERRY REALIZE THIS IS NOT

907
00:32:00,918 --> 00:32:01,786
SUSTAINABLE BECAUSE THE RAPID

908
00:32:01,853 --> 00:32:03,488
GROWTH OF THE LITERATURE REALLY

909
00:32:03,554 --> 00:32:05,490
JUST OUTPAYS THE TIME WE HAVE

910
00:32:05,556 --> 00:32:07,592
AND THE 2 POST DOC CAN'T REALLY

911
00:32:07,659 --> 00:32:10,695
AFFORD, BUT LUCKILY WE HAVE DONE

912
00:32:10,762 --> 00:32:11,929
THIS MANUAL CURATION PROCESS, SO

913
00:32:11,996 --> 00:32:14,065
WE USED THAT FOR DEVELOPING SOME

914
00:32:14,132 --> 00:32:16,801
OF THE AI ALGORITHMS USING

915
00:32:16,868 --> 00:32:19,871
TRANSFORMER MODELS TO ASSIST US

916
00:32:19,937 --> 00:32:21,039
TO TRIAGE PAPERS, AND

917
00:32:21,105 --> 00:32:22,674
SOPHISTICATED SO THAT MOVED TO

918
00:32:22,740 --> 00:32:23,441
SOMETHING I CALLED AI ASSIST IN

919
00:32:23,508 --> 00:32:25,176
THE SENSE THAT WE WERE AROUND

920
00:32:25,243 --> 00:32:28,913
THE AI SYSTEM, BUT THEN EVERY

921
00:32:28,980 --> 00:32:30,448
ARTICLE WITH AI PREDICTIONS IS

922
00:32:30,515 --> 00:32:32,650
STILL BEING REVIEWED BY A HUMAN

923
00:32:32,717 --> 00:32:32,917
BEING.

924
00:32:32,984 --> 00:32:34,752
BUT THEN TIME GOES ON AND

925
00:32:34,819 --> 00:32:37,789
THERE'S EVEN MORE PAPERS AND

926
00:32:37,855 --> 00:32:39,490
THEN, EVEN THAT BECOMES NOT

927
00:32:39,557 --> 00:32:40,792
SUSTAINABLE WHAT WE HAVE THEN

928
00:32:40,858 --> 00:32:43,795
TURNED INTO IS THAT AI DRIVEN,

929
00:32:43,861 --> 00:32:45,663
SO DURING AI ASSISTED PROCESS

930
00:32:45,730 --> 00:32:47,598
THAT WE NOT ONLY REVIEW PAPERS

931
00:32:47,665 --> 00:32:49,000
BUT WE ALSO CONTINUE TO

932
00:32:49,067 --> 00:32:51,569
CALIBRATE OUR AI SYSTEM, SO IN

933
00:32:51,636 --> 00:32:53,471
THE AI DRIVEN, WHERE IF AI

934
00:32:53,538 --> 00:32:54,972
PREDICTION IS HAVING CONFIDENCE

935
00:32:55,039 --> 00:32:57,975
VERY HIGH, WE WILL NOT HUMAN

936
00:32:58,042 --> 00:33:02,246
REVIEW IT, BUT IF AI IS BORDER

937
00:33:02,313 --> 00:33:03,915
LINE, OR NOT SURE, WE WILL DO

938
00:33:03,981 --> 00:33:05,950
HUMAN REVIEW AND THAT COUNTS AS

939
00:33:06,017 --> 00:33:07,819
20%, AND YOU IN THE END, YOU

940
00:33:07,885 --> 00:33:09,120
KNOW WHY IT'S DONE AND MOST OF

941
00:33:09,187 --> 00:33:11,956
THE TRAINEES LEFT AND STARTED,

942
00:33:12,023 --> 00:33:12,890
NEW POSITIONS ELSEWHERE WE

943
00:33:12,957 --> 00:33:14,759
TURNED THAT INTO A HUNDRED

944
00:33:14,826 --> 00:33:18,363
MACHINE CURATION, SO THIS IS A

945
00:33:18,429 --> 00:33:19,997
GREAT EXAMPLE I THINK TO SHOW

946
00:33:20,064 --> 00:33:24,202
HOW AI CAN BE USED FOR DATABASED

947
00:33:24,268 --> 00:33:27,038
CURATION AND THROUGH THIS 3

948
00:33:27,105 --> 00:33:29,240
YEARS OF PANDEMIC, COVID REACHED

949
00:33:29,307 --> 00:33:30,908
MANY, MANY USERS FROM ALL OF THE

950
00:33:30,975 --> 00:33:32,944
WORLDS, WE HAVE OVER 60 MILLION

951
00:33:33,010 --> 00:33:34,312
HITS FROM ALL OVER THE WORLD,

952
00:33:34,379 --> 00:33:38,182
ALL OVER THE COUNTRIES INCLUDING

953
00:33:38,249 --> 00:33:42,286
PEOPLE WORKS IN ANT ARCTICCA,

954
00:33:42,353 --> 00:33:45,490
THIS IS A GREAT TOOL FOR NEXT

955
00:33:45,556 --> 00:33:47,658
GENERATION SCIENTIST.

956
00:33:47,725 --> 00:33:50,328
SO WE HAVE LILLY XU WHO WAS A

957
00:33:50,395 --> 00:33:51,496
SUMMER INTERN IN 2022 AND WORKED

958
00:33:51,562 --> 00:33:52,830
ON THIS PROJECT AND IN SEPTEMBER

959
00:33:52,897 --> 00:33:57,168
WE HELPED HER PREPARE FOR AN

960
00:33:57,235 --> 00:33:59,470
APPLICATION FOR AMIA HIGH SCHOOL

961
00:33:59,537 --> 00:34:01,205
SCHOLARSHIP AND WE'RE LATER TOLD

962
00:34:01,272 --> 00:34:02,740
IS MORE COMPETITIVE COMPARED TO

963
00:34:02,807 --> 00:34:03,841
THE REGULAR PAPER SUBMISSIONS

964
00:34:03,908 --> 00:34:05,176
AND LUKELY SHE GOT THAT IS THIS

965
00:34:05,243 --> 00:34:08,146
IS A PICTURE OF HER PRESENTING

966
00:34:08,212 --> 00:34:09,280
AT THE AMIA ANNUAL CONFERENCE IN

967
00:34:09,347 --> 00:34:11,149
NOVEMBER AND SHE'S NOW AN

968
00:34:11,215 --> 00:34:12,049
UNDERGRAD AT HARVARD.

969
00:34:12,116 --> 00:34:16,654
SO SO FAR I TALKED ABOUT THIS,

970
00:34:16,721 --> 00:34:18,189
THEY ARE HELPING YOU FIND THE

971
00:34:18,256 --> 00:34:21,459
PAPERS THAT ARE INTERESTING BUT

972
00:34:21,526 --> 00:34:23,127
GIVENNING AVERAGE OF KEY WORDS,

973
00:34:23,194 --> 00:34:23,795
OFTEN TIME IS NONAPOPTOTIC THE

974
00:34:23,861 --> 00:34:26,297
THAT THEY DON'T HAVE THE PAPERS,

975
00:34:26,364 --> 00:34:27,799
THEY HAVE TOO MANY PAPERS

976
00:34:27,865 --> 00:34:30,268
RETURNED BY THE SEARCH ENGINES,

977
00:34:30,334 --> 00:34:31,436
HUNDREDS OF PAPERS, OR THOUSANDS

978
00:34:31,502 --> 00:34:33,638
OF PAPERS SO THE PROBLEM IS THAT

979
00:34:33,704 --> 00:34:35,039
NO 1, NO INDIVIDUAL RESEARCHER

980
00:34:35,106 --> 00:34:36,307
CAN ACTUALLY TAKE THE TIME AND

981
00:34:36,374 --> 00:34:39,076
READ ALL THESE PAPERS, NOT TO

982
00:34:39,143 --> 00:34:39,977
MENTION MAKING THE CONNECTIONS

983
00:34:40,044 --> 00:34:41,913
OR THE KNOWLEDGE IN THE

984
00:34:41,979 --> 00:34:42,246
LITERATURE.

985
00:34:42,313 --> 00:34:44,615
SO THIS IS SORT OF THE--TO MEET

986
00:34:44,682 --> 00:34:46,350
THE MILLION DOLLAR QUESTION, CAN

987
00:34:46,417 --> 00:34:48,419
WE TEACH COMPUTERS TO ACTUALLY

988
00:34:48,486 --> 00:34:50,054
READ BECAUSE COMPUTERS CAN BE

989
00:34:50,121 --> 00:34:51,122
REALLY GOOD AT THAT TO READ ALL

990
00:34:51,189 --> 00:34:53,191
THESE PAPERS AND THEN MAKING THE

991
00:34:53,257 --> 00:34:54,926
HIDDEN CONNECKS THAT IT WOULD

992
00:34:54,992 --> 00:34:59,730
OTHERWISE NOT BE AVAILABLE TO

993
00:34:59,797 --> 00:35:02,500
HUMAN SCIENTISTS FOR NEW PROCESS

994
00:35:02,567 --> 00:35:03,201
GENERATION FOR DISCOVERY.

995
00:35:03,267 --> 00:35:07,939
SO FOR THIS WE GO TAKE BABY

996
00:35:08,005 --> 00:35:13,010
STEPS  INCLUDING TRYING TO LOOK

997
00:35:13,077 --> 00:35:16,481
THEA THE CONCEPTS AND WHAT IS

998
00:35:16,547 --> 00:35:18,015
THE RELATION OF THE CONCEPTS.

999
00:35:18,082 --> 00:35:20,184
SO LET ME GIVE AN EXAMPLE, HERE

1000
00:35:20,251 --> 00:35:21,419
A TITHES ARTICLE WHERE THERE'S A

1001
00:35:21,486 --> 00:35:23,254
MENTION OF A DISEASE, GENE AND

1002
00:35:23,321 --> 00:35:25,556
VARIANT BUT WE LIKE THE COMPUTER

1003
00:35:25,623 --> 00:35:27,325
TO RECOGNIZE THE TEXT WORDS,

1004
00:35:27,391 --> 00:35:29,827
THESE ARE REFERRED TO DEC GENE

1005
00:35:29,894 --> 00:35:31,696
AND VARIANT BUT ALSO TO MAP THAT

1006
00:35:31,762 --> 00:35:36,901
AND GROUND THAT INTO

1007
00:35:36,968 --> 00:35:38,703
STANDARDIZED ONTOLOGY AND

1008
00:35:38,769 --> 00:35:39,437
IDENTIFIERS OF [INDISCERNIBLE].

1009
00:35:39,504 --> 00:35:40,738
BUT SEE EVEN FOR THIS PARTICULAR

1010
00:35:40,805 --> 00:35:42,840
TASK IT'S NOT SO TRIVIAL BECAUSE

1011
00:35:42,907 --> 00:35:44,675
HUMAN LANGUAGE IS VERY RICH BUT

1012
00:35:44,742 --> 00:35:47,178
AT THE SAME TIME IT'S VERY

1013
00:35:47,245 --> 00:35:48,045
AMBIGUOUS, THERE ARE MANY

1014
00:35:48,112 --> 00:35:49,013
DIFFERENT WAYS THAT PEOPLE CAN

1015
00:35:49,080 --> 00:35:51,516
USE TO REFER TO THE SAME THING

1016
00:35:51,582 --> 00:35:53,351
AND BIOLOGISTS OR SCIENTISTS

1017
00:35:53,417 --> 00:35:56,020
INCLUDING MYSELF ARE NOT HARSHLY

1018
00:35:56,087 --> 00:35:57,722
KNOW THAT WE TYPICAL LYE DON'T

1019
00:35:57,788 --> 00:36:00,191
AGREE TO USE THE SAME TERM, EVEN

1020
00:36:00,258 --> 00:36:02,326
THOUGH THERE IS A NOMENCLATURE

1021
00:36:02,393 --> 00:36:04,695
FOR MANY THINGS, BUT AT THE SAME

1022
00:36:04,762 --> 00:36:06,097
TIME, THE SAME TERM CAN BE USED

1023
00:36:06,163 --> 00:36:09,500
TO REFER TO MANY DIFFERENT

1024
00:36:09,567 --> 00:36:13,237
CONCEPTS AS WELL, IN DIFFERENT

1025
00:36:13,304 --> 00:36:14,872
CONTEXT, THIS IS EVEN MORE

1026
00:36:14,939 --> 00:36:16,807
CHALLENGING IN THE EXAMPLE I

1027
00:36:16,874 --> 00:36:18,643
SHOWED YOU, THERE'S A GENE

1028
00:36:18,709 --> 00:36:19,577
DISEASE VARIANT NEXT TO EACH

1029
00:36:19,644 --> 00:36:21,779
OTHER BUT IN REALITY THERE ARE

1030
00:36:21,846 --> 00:36:24,348
MULTIPLE DECS, MULTIPLE GENES

1031
00:36:24,415 --> 00:36:26,951
AND VARIANTS ALL SCATTERED

1032
00:36:27,018 --> 00:36:27,752
ACROSS THE TEXT.

1033
00:36:27,818 --> 00:36:30,254
NOT TO MENTION TIMES HAVE YOU

1034
00:36:30,321 --> 00:36:31,822
OMISSIONS SO THIS MAKES THIS

1035
00:36:31,889 --> 00:36:35,560
MORE CHALLENGING SO THIS IS A

1036
00:36:35,626 --> 00:36:36,827
BIOMEDICAL INFORMATION

1037
00:36:36,894 --> 00:36:37,495
EXTRACTION, INFORMATION

1038
00:36:37,562 --> 00:36:38,963
EXTRACTION, AND THIS IS NOT

1039
00:36:39,030 --> 00:36:40,097
SOMETHING--THIS IS 1 OF THE

1040
00:36:40,164 --> 00:36:42,567
OLDEST PROBLEMS IN OUR FIELD AND

1041
00:36:42,633 --> 00:36:43,734
THE WHOLE COMMUNITY IS VERY

1042
00:36:43,801 --> 00:36:46,604
INTERESTED AND THE WORK ON THIS

1043
00:36:46,671 --> 00:36:48,739
PARTICULAR PROBLEM AND AND OUR

1044
00:36:48,806 --> 00:36:50,241
ROLE IS TO WORK WITH THE

1045
00:36:50,308 --> 00:36:52,009
COMMITTEE TO REALLY MOVING

1046
00:36:52,076 --> 00:36:54,812
FORWARD ON THIS SO IF THERE'S

1047
00:36:54,879 --> 00:36:55,613
SOMETHING CALLED BIOCREATIVE

1048
00:36:55,680 --> 00:36:57,548
WHICH IS I COMMUNITY WIDE

1049
00:36:57,615 --> 00:37:02,019
CHALLENGE EVENT FOR EVALUATING

1050
00:37:02,086 --> 00:37:05,323
BIOMEDICAL NLP SYSTEM,Y THIS ,-S

1051
00:37:05,389 --> 00:37:07,425
1 OF THE LONGEST RUNS EVENTS NOW

1052
00:37:07,491 --> 00:37:09,927
BECAUSE IT START INDEED 2004 AND

1053
00:37:09,994 --> 00:37:11,896
THROUGH THESE THIS SERIES, WE'RE

1054
00:37:11,963 --> 00:37:14,098
REALLY PUTTING UP SOME OF THE

1055
00:37:14,165 --> 00:37:15,066
TASKS FOR BENCHMARKING DIFFERENT

1056
00:37:15,132 --> 00:37:16,400
SYSTEMS AND 1 OF THE

1057
00:37:16,467 --> 00:37:18,803
CONTRIBUTION SYSTEM TO CREATE

1058
00:37:18,869 --> 00:37:19,570
THE CHALLENGE, TO CREATE THE

1059
00:37:19,637 --> 00:37:21,005
DAILY BASIS THEA SETS THAT CAN

1060
00:37:21,072 --> 00:37:22,640
BE USED NOT ONLY DURING THE

1061
00:37:22,707 --> 00:37:24,408
CHALLENGE BUT ALSO AFTER THE

1062
00:37:24,475 --> 00:37:26,043
CHALLENGE, SO, THERE'S A WEBSITE

1063
00:37:26,110 --> 00:37:27,111
CALLED PAPERS WITH CODON, I

1064
00:37:27,178 --> 00:37:29,180
DON'T KNOW IF YOU ARE FAMILIAR

1065
00:37:29,246 --> 00:37:32,750
WITH THE INDEX OR DIFFERENT DATA

1066
00:37:32,817 --> 00:37:37,955
SETS USED IN NLP, AI AND MISSION

1067
00:37:38,022 --> 00:37:39,523
LEARNING AND WHEN I FILTERED THE

1068
00:37:39,590 --> 00:37:41,525
TASK, I FIND ON TOP OF THE LIST,

1069
00:37:41,592 --> 00:37:43,661
HUNDREDS OF DATA SETS, SOME OF

1070
00:37:43,728 --> 00:37:46,631
THE TOP USED 1S INCLUDING THE 1S

1071
00:37:46,697 --> 00:37:48,699
WE HAVE DEVELOPED THROUGH THE

1072
00:37:48,766 --> 00:37:50,001
BIOCREATIVE AND THIS IS IN

1073
00:37:50,067 --> 00:37:50,935
COMPARISON TO MANY OTHER DATA

1074
00:37:51,002 --> 00:37:53,004
SETS THAT ARE USE INDEED THE

1075
00:37:53,070 --> 00:37:54,538
JOURNAL DOMAIN MUCH BROADLY AND

1076
00:37:54,605 --> 00:37:55,773
OUR LATEST CONTRIBUTION INCLUDES

1077
00:37:55,840 --> 00:38:00,077
A NEW DATA SET THAT WE CREATED

1078
00:38:00,144 --> 00:38:00,978
CALLED A BIO[INDISCERNIBLE]

1079
00:38:01,045 --> 00:38:02,313
WHICH I BELIEVE IS STILL THE

1080
00:38:02,380 --> 00:38:05,650
LARGEST DATA SET FOR BIOMEDICAL

1081
00:38:05,716 --> 00:38:09,320
RELATION INSTRUCTION TASKS, AND

1082
00:38:09,387 --> 00:38:13,424
THROUGH OUR COLLABARATORS

1083
00:38:13,491 --> 00:38:14,558
WITH--IN THE BIOCREATED 8 EVENT

1084
00:38:14,625 --> 00:38:17,561
LAST YEAR, WE HAVE HAD A

1085
00:38:17,628 --> 00:38:19,964
WONDERFUL TASK AND ALSO RECEIVED

1086
00:38:20,031 --> 00:38:22,466
ATTENTION AND PARTICIPATION FROM

1087
00:38:22,533 --> 00:38:23,401
MANY DIFFERENT GROUPS FROM

1088
00:38:23,467 --> 00:38:26,270
ACROSS THE WORLD.

1089
00:38:26,337 --> 00:38:27,805
AND OUR RESPONSE TO THIS IS

1090
00:38:27,872 --> 00:38:29,106
SOMETHING CALLED A

1091
00:38:29,173 --> 00:38:30,975
[INDISCERNIBLE] WHICH IS

1092
00:38:31,042 --> 00:38:33,244
SOMETHING THAT WE SEPARATELY

1093
00:38:33,310 --> 00:38:34,845
DEVELOPED [INDISCERNIBLE] SYSTEM

1094
00:38:34,912 --> 00:38:37,915
FOR DISEASES FOR GENES, STORY

1095
00:38:37,982 --> 00:38:38,749
DISEASES, FOR CHEMICALS AND

1096
00:38:38,816 --> 00:38:41,318
DRUGS AND THEN WE APPLY THESE

1097
00:38:41,385 --> 00:38:42,920
TAGGERS TO THE ENTIRE MILLIONS

1098
00:38:42,987 --> 00:38:44,789
OF ARTICLES IN PUB MED, IN THE

1099
00:38:44,855 --> 00:38:46,757
BACK END AND THEN WE AGGREGATE

1100
00:38:46,824 --> 00:38:49,293
RESULTS IN THE BACK END AND

1101
00:38:49,360 --> 00:38:51,328
RANDOMIZE RESULTS IN THE FRONT

1102
00:38:51,395 --> 00:38:53,731
FOR VISUALIZATION.

1103
00:38:53,798 --> 00:38:56,300
SO THE PUBITATOR TEST IS NOW

1104
00:38:56,367 --> 00:38:58,002
OVER 10 YEARS OLD AND THROUGH

1105
00:38:58,069 --> 00:39:00,137
THAT PROCESS WE HAVE ADDED

1106
00:39:00,204 --> 00:39:03,274
APIs, WE HAVE ADDED FTP, WE

1107
00:39:03,340 --> 00:39:04,942
EXPANDED FROM MINING RESULT ANDS

1108
00:39:05,009 --> 00:39:05,776
ANNOTATIONS FROM THE ABSTRACTION

1109
00:39:05,843 --> 00:39:09,146
TO THE FULL TEXT AND WE HAVE

1110
00:39:09,213 --> 00:39:10,347
CONTINUOUSLY IMPROVING ITS

1111
00:39:10,414 --> 00:39:11,515
TAGGING ACCURACY THROUGH THE

1112
00:39:11,582 --> 00:39:13,484
LATEST AND MOST ADVANCED MISSION

1113
00:39:13,551 --> 00:39:17,021
LEARNING AI TECHNIQUES AND I AM

1114
00:39:17,088 --> 00:39:18,355
HAPPY TO ANNOUNCE THAT YOU ARE

1115
00:39:18,422 --> 00:39:19,690
AMONG THE FIRST GROUP TO KNOW

1116
00:39:19,757 --> 00:39:22,760
THAT WE ARE GOING TO RELEASE THE

1117
00:39:22,827 --> 00:39:24,195
THIRD VERSION, THE 3.0 VERSION

1118
00:39:24,261 --> 00:39:30,034
OF THE PUP TATOR THIS YEAR AND

1119
00:39:30,101 --> 00:39:31,102
IT WILL RELEASE RESULT THIS IS

1120
00:39:31,168 --> 00:39:34,138
YEAR AND THIS IS AACCREDITTED TO

1121
00:39:34,205 --> 00:39:36,741
2 STAFF SCIENTISTS, DRS. WI AND

1122
00:39:36,807 --> 00:39:39,110
LEEMAN AND ALSO IN THE

1123
00:39:39,176 --> 00:39:40,611
PRESENTATION LATER ON BY

1124
00:39:40,678 --> 00:39:43,414
TRAINEES YOU WILL HEAR A SHORT

1125
00:39:43,481 --> 00:39:44,915
PRESENTATION BY SONIA, A SUMMER

1126
00:39:44,982 --> 00:39:49,386
INTERN HAOF THE YEAR, SHE

1127
00:39:49,453 --> 00:39:52,123
CLOSELY WORKED DRRKS REEMAN, 1

1128
00:39:52,189 --> 00:39:55,359
OF THE PROJECTS RELATED TO

1129
00:39:55,426 --> 00:39:56,861
PUPTATOR, 3, AND IT'S BEEN USE

1130
00:39:56,927 --> 00:39:59,330
INDEED A VARIETY OF DIFFERENT

1131
00:39:59,396 --> 00:39:59,563
CASES.

1132
00:39:59,630 --> 00:40:03,033
WE NOTED THEM HERE, WE KNOW, WE

1133
00:40:03,100 --> 00:40:09,106
LOOK AT PAPERS WHO CITED

1134
00:40:09,173 --> 00:40:14,478
PUBITATOR, AND ALSO THESE WHO

1135
00:40:14,545 --> 00:40:15,780
USE THEM AND ASKING FOR MORE.

1136
00:40:15,846 --> 00:40:18,382
SO IN TERMS OF USAGE, API IS THE

1137
00:40:18,449 --> 00:40:21,719
WINNER EACH THOUGH MAY MAKE

1138
00:40:21,786 --> 00:40:22,887
DIFFERENT WAYS PUBITATOR CAN BE

1139
00:40:22,953 --> 00:40:24,088
ACCESS THE OVER THE LAST COUPLE

1140
00:40:24,155 --> 00:40:26,957
YEARS WE HAVE OVER 1 BRILLION

1141
00:40:27,024 --> 00:40:30,928
API REQUESTS FROM MANY DIFFERENT

1142
00:40:30,995 --> 00:40:31,962
INSTITUTES AND WERERS ACROSS THE

1143
00:40:32,029 --> 00:40:33,697
WORLD AND LIKE TO MENTION, ALSO

1144
00:40:33,764 --> 00:40:35,566
HIGHLIGHT THAT FOR MANY

1145
00:40:35,633 --> 00:40:42,406
DATABASES INCLUDING FOR EXAMPLE,

1146
00:40:42,473 --> 00:40:45,509
COMPARATIVE TOXIC GENOMICS, CDG,

1147
00:40:45,576 --> 00:40:48,279
SWITZERLAND INSTITUTE OF

1148
00:40:48,345 --> 00:40:50,214
BIO-INFORMATICS, SWISS PRODUCT,

1149
00:40:50,281 --> 00:40:54,285
GWAS CATALOG, PHARMA GKB, ALL

1150
00:40:54,351 --> 00:40:58,889
THESE USING PUBTATOR, AND THEY

1151
00:40:58,956 --> 00:41:00,558
PUT IT IN THEIR SYSTEMS.

1152
00:41:00,624 --> 00:41:03,460
IT COVERS A WIDE VARIETY OF

1153
00:41:03,527 --> 00:41:04,528
CONTENTS, DISEASES, YEENS,

1154
00:41:04,595 --> 00:41:06,030
CHEMICALS, SO ON AND O FORTH AND

1155
00:41:06,096 --> 00:41:07,264
WE ALSO GET SPECIAL REQUESTS FOR

1156
00:41:07,331 --> 00:41:10,768
PEOPLE WHO ARE INTERESTED IN

1157
00:41:10,835 --> 00:41:11,569
VARIATION, SEQUENCE VARIATION

1158
00:41:11,635 --> 00:41:12,770
BECAUSE THEY MENTION A LOT OF

1159
00:41:12,837 --> 00:41:14,371
THINGS GET THEM LOST IF WE DON'T

1160
00:41:14,438 --> 00:41:17,441
TEXT MINE INFORMATION IN THE

1161
00:41:17,508 --> 00:41:18,142
SEPARATE [INDISCERNIBLE].

1162
00:41:18,209 --> 00:41:22,546
SO WE TREASES THAT CALL AND THE

1163
00:41:22,613 --> 00:41:25,149
REQUESTS BY KREAING A SPEN OFF,

1164
00:41:25,216 --> 00:41:28,419
WE CALL IT [INDISCERNIBLE] AND

1165
00:41:28,485 --> 00:41:30,754
SO NOT ONLY THE TEXT LINE HAD AN

1166
00:41:30,821 --> 00:41:33,691
ABSTRACT FOR TEXT BUT IN

1167
00:41:33,757 --> 00:41:34,925
[INDISCERNIBLE] WE GO 1 STEP

1168
00:41:34,992 --> 00:41:36,126
FURGT XER TEXT MINE THE

1169
00:41:36,193 --> 00:41:37,361
INFORMATION IN THE SEPARATE

1170
00:41:37,428 --> 00:41:38,229
MATERIAL BECAUSE THAT'S WHERE

1171
00:41:38,295 --> 00:41:39,597
THE INFORMATION IS REALLY

1172
00:41:39,663 --> 00:41:40,664
ENRICHED.

1173
00:41:40,731 --> 00:41:43,701
NOW WE DON'T GENERALLY DO THAT

1174
00:41:43,767 --> 00:41:46,003
FOR PUB TATOR BECAUSE AS YOU

1175
00:41:46,070 --> 00:41:47,271
KNOW THE SUPPLEMENTAL IT RECALL

1176
00:41:47,338 --> 00:41:48,806
IS REALLY NOISE EXPE WORKING

1177
00:41:48,873 --> 00:41:51,909
TOGETHER WITH THE COLLEAGUES

1178
00:41:51,976 --> 00:41:54,778
FROM NCBI, [INDISCERNIBLE]

1179
00:41:54,845 --> 00:41:57,982
GROUP, WE ARE MAKING SURE THAT

1180
00:41:58,048 --> 00:42:00,918
THEIR USERS ARE HAVING READILY

1181
00:42:00,985 --> 00:42:04,488
ACCESS TO THE LITVAR CONTENT AS

1182
00:42:04,555 --> 00:42:04,722
WELL.

1183
00:42:04,788 --> 00:42:06,156
SO NOW LET ME SWITCH GEAR AND

1184
00:42:06,223 --> 00:42:07,558
TALK TO ABOUT OUR PASSION IN

1185
00:42:07,625 --> 00:42:08,626
MEDICAL IMAGING AND I MENTIONED

1186
00:42:08,692 --> 00:42:10,461
AT THE VERY BEGINNING WE'RE

1187
00:42:10,527 --> 00:42:12,496
WORKING ON PARTICULAR

1188
00:42:12,563 --> 00:42:13,464
SPECIALTIES IN OPHTHALMOLOGY AND

1189
00:42:13,530 --> 00:42:15,699
RADIOLOGY BUT I WILL USE

1190
00:42:15,766 --> 00:42:17,534
OPHTHALMOLOGY AS AN EXAMPLE

1191
00:42:17,601 --> 00:42:19,370
TODAY AND IN PARTICULAR DISEASE

1192
00:42:19,436 --> 00:42:21,639
CALLED AGE RELATED MACULAR

1193
00:42:21,705 --> 00:42:22,539
DEGENERATION, AMD FOR SHORT.

1194
00:42:22,606 --> 00:42:25,075
THIS IS A LEADING CAUSE OF

1195
00:42:25,142 --> 00:42:29,580
BLINDNESS FOR ELDERLY POPULATION

1196
00:42:29,647 --> 00:42:31,048
AND IT'S PROGRESSED WITH--IT

1197
00:42:31,115 --> 00:42:33,150
PROGRESSES WITH AGE AND THE

1198
00:42:33,217 --> 00:42:34,084
EARLY STAGE INTERMEDIATE STAGE

1199
00:42:34,151 --> 00:42:35,619
AND LATE STAGE AS YOU CAN SEE,

1200
00:42:35,686 --> 00:42:37,922
IN THE LATE STAGE OR

1201
00:42:37,988 --> 00:42:39,790
INTERMEDIATE STAGE, WHERE THEY

1202
00:42:39,857 --> 00:42:41,325
REALLY GET IMPAIRED AND

1203
00:42:41,392 --> 00:42:42,159
UNFORTUNATELY THERE'S NO CURE

1204
00:42:42,226 --> 00:42:43,427
FOR THIS DISEASE AT THIS POINT

1205
00:42:43,494 --> 00:42:44,495
ALTHOUGH THE RESEARCH AND

1206
00:42:44,561 --> 00:42:47,164
CLINICAL TRIALS HAVE SHOWED,

1207
00:42:47,231 --> 00:42:50,401
RANDOMIZED TRIALS HAS SHOWED

1208
00:42:50,467 --> 00:42:51,902
THAT HIGH DOSE ZINC CAN HELP

1209
00:42:51,969 --> 00:42:54,004
SLOW DOWN THE PROGRESSION AT THE

1210
00:42:54,071 --> 00:42:55,039
RIGHT TIME, BUT ONLY AT THE

1211
00:42:55,105 --> 00:42:55,606
RIGHT TIME.

1212
00:42:55,673 --> 00:42:56,874
THIS IS LOOKING AT THE PACK OF

1213
00:42:56,941 --> 00:42:58,509
THE PATIENT'S EYE AND THEY

1214
00:42:58,575 --> 00:43:02,212
IDENTIFY 2 MAIN RISK FACTORS, 1

1215
00:43:02,279 --> 00:43:04,014
IS CALLED [INDISCERNIBLE] WHICH

1216
00:43:04,081 --> 00:43:06,350
IS FATTY DEPOSIT AND THE OTHER

1217
00:43:06,417 --> 00:43:07,518
IS PIGMENT CHARGES AND THERE ARE

1218
00:43:07,584 --> 00:43:12,089
2 FORMS OF LATE AMD,

1219
00:43:12,156 --> 00:43:13,123
[INDISCERNIBLE], GA FOR SHORT,

1220
00:43:13,190 --> 00:43:14,992
REFERRED TO AS THE DRY FORM AND

1221
00:43:15,059 --> 00:43:21,065
THERE'S THE WET FORM, CORNEAL

1222
00:43:21,131 --> 00:43:22,299
VASCULARIZATION, SO THIS

1223
00:43:22,366 --> 00:43:23,801
REQUIRES SPECIAL TRAINING, YOU

1224
00:43:23,867 --> 00:43:26,570
DON'T GET THIS BY A GENERAL

1225
00:43:26,637 --> 00:43:27,204
OPHTHALMOLOGIST.

1226
00:43:27,271 --> 00:43:28,672
YOU NEED A RETINAL SPECIALIST

1227
00:43:28,739 --> 00:43:30,541
AND EACH THAT IT IS VERY TIME

1228
00:43:30,607 --> 00:43:33,877
CONSUMING AND LABOR INTENSIVE IN

1229
00:43:33,944 --> 00:43:35,479
THE PROCESS, NOT TO MENTION IN

1230
00:43:35,546 --> 00:43:36,814
LANE PARTS OF THE WORLD THERE'S

1231
00:43:36,880 --> 00:43:40,551
LACK OF ACCESS FOR SUCH RETINAL

1232
00:43:40,617 --> 00:43:43,253
SPECIALISTS SO WORKING WITH

1233
00:43:43,320 --> 00:43:44,989
LONG-TERM COLLABORATOR, A

1234
00:43:45,055 --> 00:43:47,224
SPECIALIST IN AMD FROM NATIONAL

1235
00:43:47,291 --> 00:43:48,258
EYE INSTITUTE, WE SET UP

1236
00:43:48,325 --> 00:43:51,261
SOMETHING TO DISCUSS WHAT WE CAN

1237
00:43:51,328 --> 00:43:55,032
DO USING AI TO ADDRESS OR TO

1238
00:43:55,099 --> 00:43:58,135
IMPROVE THIS PROCESS BY AI TO

1239
00:43:58,202 --> 00:43:59,103
ASSISTING AMD DIAGNOSIS AND

1240
00:43:59,169 --> 00:44:01,805
PROGNOSIS AND THE TIME IS REALLY

1241
00:44:01,872 --> 00:44:03,474
RIGHT BECAUSE WE ALSO--SHE IS

1242
00:44:03,540 --> 00:44:06,810
ALSO THE LEADING PI FOR 2 VERY

1243
00:44:06,877 --> 00:44:10,114
LARGE CLINICAL TRIALS CALLED

1244
00:44:10,180 --> 00:44:11,915
ACREDS 1 AND 2, THIS IS THE

1245
00:44:11,982 --> 00:44:13,751
LARGEST FOR AMD AND 1 OF THE

1246
00:44:13,817 --> 00:44:14,985
LARGEST IN THE EYE SPECIALTY.

1247
00:44:15,052 --> 00:44:17,287
SO HERE WE HAVE, 10S OF

1248
00:44:17,354 --> 00:44:19,857
THOUSANDS OF IMAGES AND THOUSAND

1249
00:44:19,923 --> 00:44:20,824
PARTICIPANTS, WHAT'S EACH NICER

1250
00:44:20,891 --> 00:44:23,027
FOR AI OF COURSE IS ALL THESE

1251
00:44:23,093 --> 00:44:25,662
IMAGES HAVE BEEN MANUELLY GRADED

1252
00:44:25,729 --> 00:44:27,197
BY READING CENTER SPECIALISTS

1253
00:44:27,264 --> 00:44:30,467
AND THIS DATA IS LONGITUDINAL SO

1254
00:44:30,534 --> 00:44:32,302
WE CAN NOT ONLY DO DIAGNOSIS BUT

1255
00:44:32,369 --> 00:44:34,805
WE CAN DO PROGNOSIS AND RISK

1256
00:44:34,872 --> 00:44:35,572
[INDISCERNIBLE] AND THAT'S WHAT

1257
00:44:35,639 --> 00:44:39,443
WE DID WITH PIE FORMER POST DOC

1258
00:44:39,510 --> 00:44:42,379
[INDISCERNIBLE] WHO LEFT MY LAB

1259
00:44:42,446 --> 00:44:47,051
IN LATE 2020 AND IS NOW--WE'RE

1260
00:44:47,117 --> 00:44:48,619
STILL WORKS WITH [INDISCERNIBLE]

1261
00:44:48,685 --> 00:44:50,921
BUT THROUGH HIS NEW TRAINEES, ON

1262
00:44:50,988 --> 00:44:53,323
THESE 2 PROBLEMS BUT IN MY OWN

1263
00:44:53,390 --> 00:44:57,795
LAB WE HAVE SORT OF ALSO

1264
00:44:57,861 --> 00:44:59,696
SWITCHED, AND 1 STEP FORWARD AND

1265
00:44:59,763 --> 00:45:00,964
LOOKING AT SOMETHING IN MORE

1266
00:45:01,031 --> 00:45:02,399
DEPTH AND MORE [INDISCERNIBLE]

1267
00:45:02,466 --> 00:45:04,101
THE WAY WHERE WE TRY TO DISCOVER

1268
00:45:04,168 --> 00:45:06,437
WHAT ARE THE THINGS THAT MATTERS

1269
00:45:06,503 --> 00:45:10,307
IF ARE AMD PROGNOSIS THAT MAKING

1270
00:45:10,374 --> 00:45:12,042
PEOPLE FOR INTERMEDIATE AND TO

1271
00:45:12,109 --> 00:45:15,379
LATE STAGE, SO THERE IS

1272
00:45:15,446 --> 00:45:16,447
SOMETHING CALLED RPD FOR SHORT,

1273
00:45:16,513 --> 00:45:19,516
SO PEOPLE KNOW THIS, LONG TIME

1274
00:45:19,583 --> 00:45:21,819
AGO, BUT SOMETHING, THE EVIDENCE

1275
00:45:21,885 --> 00:45:23,487
OF RPD IS JUST NOT STRONG ENOUGH

1276
00:45:23,554 --> 00:45:24,388
BECAUSE THIS IS SOMETHING VERY

1277
00:45:24,455 --> 00:45:28,158
HARD TO BE SEEN ON THE COLOR

1278
00:45:28,225 --> 00:45:29,326
PHOTO, COLOR FROM THIS

1279
00:45:29,393 --> 00:45:32,162
PHOTOGRAPHS AND COMPARED TO

1280
00:45:32,229 --> 00:45:33,597
CLASSIC WHICH IS MUCH MORE

1281
00:45:33,664 --> 00:45:35,632
EASILY TO BE SPOTTED IN THESE

1282
00:45:35,699 --> 00:45:36,867
COLOR PHOTOS SO AS CAN YOU SEE

1283
00:45:36,934 --> 00:45:39,002
IN THE RIGHT FIGURE WHERE WE

1284
00:45:39,069 --> 00:45:40,737
SHOW EACH INDIVIDUAL DOTS, THESE

1285
00:45:40,804 --> 00:45:41,972
ARE BASICALLY HUMAN SPECIALISTS,

1286
00:45:42,039 --> 00:45:46,610
AS YOU CAN SEE, THESE--THEIR

1287
00:45:46,677 --> 00:45:48,712
PERFORMANCE FOR SPOTTING RPD IS

1288
00:45:48,779 --> 00:45:50,214
VERY CLOSE TO THE

1289
00:45:50,280 --> 00:45:51,915
[INDISCERNIBLE] LINE WHICH IS

1290
00:45:51,982 --> 00:45:52,883
RANDOM GUESSING AND WHILE YOU

1291
00:45:52,950 --> 00:45:56,520
CAN SEE THAT AI METHOD THAT WE

1292
00:45:56,587 --> 00:45:57,488
DEVELOPED IS CONSIDERED MUCH

1293
00:45:57,554 --> 00:45:59,089
BETTER AND MUCH BETTER THAN

1294
00:45:59,156 --> 00:46:00,557
HUMAN BEINGS IN THIS CASE, AI

1295
00:46:00,624 --> 00:46:02,292
CAN SEE SOMETHING THAT HUMANS

1296
00:46:02,359 --> 00:46:03,560
CAN REALLY NOT SEE SOMETHING

1297
00:46:03,627 --> 00:46:05,028
VERY WELL EMPLOY SO USING THIS

1298
00:46:05,095 --> 00:46:09,032
AI SYSTEM, WE BASICALLY

1299
00:46:09,099 --> 00:46:10,100
REPROCESSED THE HISTORICAL

1300
00:46:10,167 --> 00:46:11,735
ACREDS 1 DATA SET WHERE COLOR

1301
00:46:11,802 --> 00:46:14,838
PHOTOS IS THE ONLY DATA MODALITY

1302
00:46:14,905 --> 00:46:16,206
THAT'S AVAILABLE AT THE TIME SO

1303
00:46:16,273 --> 00:46:17,441
WHEN WE REPROCESS THAT, WE CAN

1304
00:46:17,508 --> 00:46:20,077
CHECK TO SEE THE PRESENCE, AND

1305
00:46:20,144 --> 00:46:23,647
ABSENCE OF RPD AND WITH THAT,

1306
00:46:23,714 --> 00:46:26,216
EVIDENCE ENHANCED THROUGH THE

1307
00:46:26,283 --> 00:46:30,621
HARD WORK OF DR. TIAMAN KEENAN,

1308
00:46:30,687 --> 00:46:35,459
HE JUST STARTED HIS NEW

1309
00:46:35,526 --> 00:46:39,396
INDIVIDUAL, INDEPENDENT LAB AS A

1310
00:46:39,463 --> 00:46:41,965
STANDARD [INDISCERNIBLE] AT AND

1311
00:46:42,032 --> 00:46:43,300
HE'S ALSO ADJUNCT INVESTIGATOR

1312
00:46:43,367 --> 00:46:44,568
AT NLM, WE DISCOVER THAT

1313
00:46:44,635 --> 00:46:46,303
EVIDENCE IS MOWPTING AND STRONG

1314
00:46:46,370 --> 00:46:50,607
ENOUGH TO SHOW THAT RPD IS A

1315
00:46:50,674 --> 00:46:52,376
THIRD MACULAR RISK FACTOR TO AMD

1316
00:46:52,442 --> 00:46:57,114
AND NOW WITH THAT KNOWLEDGE, WE

1317
00:46:57,181 --> 00:46:59,082
HAVE ANOTHER PAPER, JUST SENT

1318
00:46:59,149 --> 00:47:02,052
THIS WEEK AND IT'S COMING OUT TO

1319
00:47:02,119 --> 00:47:04,154
SHOW THIS WORK, AND THOU WE KNOW

1320
00:47:04,221 --> 00:47:05,789
IN ADDITION TO 2 OLD RISK

1321
00:47:05,856 --> 00:47:06,723
FACTORS, THIS THIRD RISK FACTOR

1322
00:47:06,790 --> 00:47:09,660
HOW CAN WE USE THIS INTO THE

1323
00:47:09,726 --> 00:47:10,961
DIAGNOSIS PRACTICAL FOR

1324
00:47:11,028 --> 00:47:11,261
PHYSICIANS.

1325
00:47:11,328 --> 00:47:13,597
AND YOU CAN DO THIS, WE CAN DO

1326
00:47:13,664 --> 00:47:15,666
THIS FOR AMD, WE DID THE SAME

1327
00:47:15,732 --> 00:47:17,801
THING VERY SIMILAR FOR CATARACT

1328
00:47:17,868 --> 00:47:18,769
EYE DISEASES, AS WELL AS FOR

1329
00:47:18,835 --> 00:47:19,670
DISEASES THAT ARE IN THE OTHER

1330
00:47:19,736 --> 00:47:21,905
PART OF THE PART, NOTHING TO DO

1331
00:47:21,972 --> 00:47:23,974
WITH THE EYES, SO FOR EXAMPLE,

1332
00:47:24,041 --> 00:47:25,676
IN COLLABORATION WITH THE TEAM

1333
00:47:25,742 --> 00:47:28,111
IN EUROPE, WE SHOWED THAT WE CAN

1334
00:47:28,178 --> 00:47:30,314
IN THE RETINAL SCAN TO PREDICT

1335
00:47:30,380 --> 00:47:32,583
HEART ATTACK, WE HAVE WORK BY

1336
00:47:32,649 --> 00:47:36,119
ANOTHER WORK, TO PREDICT COMPANY

1337
00:47:36,186 --> 00:47:39,156
FUNCTION  DECLINE.

1338
00:47:39,223 --> 00:47:40,691
I ALSO WANT TO EMPHASIZE THESE 2

1339
00:47:40,757 --> 00:47:42,226
PAPERS ARE EXAMPLES WHERE WE

1340
00:47:42,292 --> 00:47:44,294
DEVELOP AI SYSTEMS AND THEN WE

1341
00:47:44,361 --> 00:47:45,729
HELP OUR COLLABORATORS OR OUR

1342
00:47:45,796 --> 00:47:46,997
COLLABORATORS HELPING US TO

1343
00:47:47,064 --> 00:47:48,432
EVALUATE ON THEIR OWN DATA SET,

1344
00:47:48,498 --> 00:47:50,067
SO THIS IS VERY IMPORTANT TO

1345
00:47:50,133 --> 00:47:51,068
MAKING SURE THAT AI ARE

1346
00:47:51,134 --> 00:47:52,569
DEVELOPED IN 1 SET OF DATA SET,

1347
00:47:52,636 --> 00:47:56,940
YOU KNOW FOR EXAMPLE, AREGS IS

1348
00:47:57,007 --> 00:47:57,874
PRIMARILY CAUCASIAN, AND IT CAN

1349
00:47:57,941 --> 00:47:59,643
WORK VERY WELL IN THE ASIAN

1350
00:47:59,710 --> 00:48:02,346
POPULATION, SO IN THE LAST 10

1351
00:48:02,412 --> 00:48:03,580
MINEUES I WILL TALK TO SOMETHING

1352
00:48:03,647 --> 00:48:05,349
ABOUT THAT WE ARE VERY EXCITED

1353
00:48:05,415 --> 00:48:06,650
ABOUT AND WE THINK WE ARE GOING

1354
00:48:06,717 --> 00:48:08,585
TO DO MORE IN THE NEAR FUTURE

1355
00:48:08,652 --> 00:48:11,221
AND IN THE NEXT 4 YEARS, AND THE

1356
00:48:11,288 --> 00:48:14,591
NEXT 1 WHICH IS WAS ANNOUNCED IN

1357
00:48:14,658 --> 00:48:15,525
NOVEMBER, 2022 AND EVERYBODY HAS

1358
00:48:15,592 --> 00:48:20,030
THEIR TAKE ON THIS, AND I'M SURE

1359
00:48:20,097 --> 00:48:21,665
KNOW THAT HOW POPULAR IT IS, NOT

1360
00:48:21,732 --> 00:48:23,400
ONLY AMONG OUR RESEARCH, LIKE

1361
00:48:23,467 --> 00:48:25,235
OURSELVES, BUT ALSO AMONG THE

1362
00:48:25,302 --> 00:48:28,405
GENERAL PUBLIC.

1363
00:48:28,472 --> 00:48:31,675
SOME OF THE EARLY WIDE WHAT OF

1364
00:48:31,742 --> 00:48:34,044
CHAT GPT, IT HAS   PERFORMERS

1365
00:48:34,111 --> 00:48:36,613
COMPARABLE TO THE HUMAN PATTERN

1366
00:48:36,680 --> 00:48:38,615
COPE IN MANY YEARS AND EVEN MORE

1367
00:48:38,682 --> 00:48:40,717
AMAZING IS WITH LESS THAN 6

1368
00:48:40,784 --> 00:48:42,419
MONTHS IT'S PERFORMANCE HAS

1369
00:48:42,486 --> 00:48:44,521
REACHED CLOSE TO HUMAN EXPERT

1370
00:48:44,588 --> 00:48:44,821
PERFORMANCE.

1371
00:48:44,888 --> 00:48:46,690
YOU MAY HAVE HEARD THIS ABOUT IN

1372
00:48:46,757 --> 00:48:49,793
THE NEWS THAT CHAT GPT WAS ABLE

1373
00:48:49,860 --> 00:48:51,862
TO DIAGNOSE, MAKE THE CORRECT

1374
00:48:51,928 --> 00:48:54,364
DIAGNOSIS FOR A BOY WHO WENT OUT

1375
00:48:54,431 --> 00:48:56,833
TO SEE 17 DOCTORS OVER 3 YEARS

1376
00:48:56,900 --> 00:48:59,336
FOR A CHRONIC PAIN WITH NO

1377
00:48:59,403 --> 00:49:01,838
DIAGNOSIS, SO JUST ABOUT THIS

1378
00:49:01,905 --> 00:49:03,740
TIME LAST YEAR, MAYBE IN LATE

1379
00:49:03,807 --> 00:49:06,043
MARCH WE WROTE OUR VERY FESTER

1380
00:49:06,109 --> 00:49:08,345
PAPER, I RELATED TO CHAT GPT AND

1381
00:49:08,412 --> 00:49:11,515
YOU KNOW A PROBLEM THAT IS VERY

1382
00:49:11,581 --> 00:49:13,016
MUCH IMPORTANT IN OUR MIND WHICH

1383
00:49:13,083 --> 00:49:14,685
IS BACK TO THE LITERATURE

1384
00:49:14,751 --> 00:49:17,154
SEARCH, SO WE WERE ASKING

1385
00:49:17,220 --> 00:49:19,122
OURSELVES, YOU KNOW IF CHAT GPT

1386
00:49:19,189 --> 00:49:24,227
IS SO POPULAR AND SO SMART, WITH

1387
00:49:24,294 --> 00:49:26,229
PEOPLE JUST USING CHAT GPT TO

1388
00:49:26,296 --> 00:49:27,964
ASK MEDICAL QUESTIONS, DURING

1389
00:49:28,031 --> 00:49:30,267
THE PUB MED RESEARCH AND THEN

1390
00:49:30,334 --> 00:49:32,269
READING THE PAPERS, AND SO WE

1391
00:49:32,336 --> 00:49:35,405
SET OUT AND ASK CHAT GPT, CHA IS

1392
00:49:35,472 --> 00:49:37,708
THE LONG TYPICAL OR MOST COMMON

1393
00:49:37,774 --> 00:49:39,676
SYMPTOM ESPECIALLY WITH LONG

1394
00:49:39,743 --> 00:49:40,110
COVID.

1395
00:49:40,177 --> 00:49:41,511
CHAT GPT GIVE THOSE SYMPTOMS

1396
00:49:41,578 --> 00:49:43,980
WHICH ISN'T SO IMPRESSIVE TO US

1397
00:49:44,047 --> 00:49:46,717
BUT CHAT GPT SEEMS COMPETENT FOR

1398
00:49:46,783 --> 00:49:48,485
ANSWERS BY PROVIDING THE

1399
00:49:48,552 --> 00:49:50,787
EVIDENCE, ARTICLES OR RESEARCH

1400
00:49:50,854 --> 00:49:52,322
THAT BACK IN THESE SYMPTOMS,

1401
00:49:52,389 --> 00:49:54,725
HOWEVER, IF WE COP SEPASTE IT,

1402
00:49:54,791 --> 00:49:56,927
THESE TITLES AND PUT THAT INTO

1403
00:49:56,993 --> 00:49:59,963
PUB MED OR SEARCH ENGINES, WE

1404
00:50:00,030 --> 00:50:01,598
FIND NO SEARCH ARTICLE EXISTED.

1405
00:50:01,665 --> 00:50:03,133
THIS IS SOMETHING, OBVIOUS LE

1406
00:50:03,200 --> 00:50:04,935
EVERYBODY KNOWS THIS FABRICATING

1407
00:50:05,001 --> 00:50:07,971
INFORMATION ARE ALSO KNOWN AS

1408
00:50:08,038 --> 00:50:08,305
HALLUCINATION.

1409
00:50:08,372 --> 00:50:10,474
THESE--THESE ARE REAL THOUGH,

1410
00:50:10,540 --> 00:50:12,075
BUT [INDISCERNIBLE] IF WE PUT IT

1411
00:50:12,142 --> 00:50:13,143
INTO PUB MED, THEY HAVE

1412
00:50:13,210 --> 00:50:14,611
SOMETHING TO DO WITH COVID

1413
00:50:14,678 --> 00:50:15,679
RESEARCH.

1414
00:50:15,746 --> 00:50:17,414
SO THIS IS--SO WE WROTE THIS

1415
00:50:17,481 --> 00:50:19,516
PERSPECTIVE PAPER LAST YEAR AND

1416
00:50:19,583 --> 00:50:21,084
WE BASICALLY SAID THAT HUMAN

1417
00:50:21,151 --> 00:50:22,753
VERIFICATION AS THE LAST STEP IS

1418
00:50:22,819 --> 00:50:25,088
STILL VERY IMPORTANT, I THINK

1419
00:50:25,155 --> 00:50:26,823
EVEN FOR TODAY, THIS PERSPECTIVE

1420
00:50:26,890 --> 00:50:29,359
STILL IS VALID BECAUSE EVEN

1421
00:50:29,426 --> 00:50:32,062
TODAY, I DON'T THINK THAT THE

1422
00:50:32,129 --> 00:50:34,631
LARGE [INDISCERNIBLE] CHAT GPT

1423
00:50:34,698 --> 00:50:35,198
HAS COMPLETELY ADDRESSED OR

1424
00:50:35,265 --> 00:50:37,167
SOLVED ISSUE THAT WE MENTIONED

1425
00:50:37,234 --> 00:50:37,901
IN THAT ARTICLE.

1426
00:50:37,968 --> 00:50:40,036
SO THIS FITS IN 1 OF THE THEMES

1427
00:50:40,103 --> 00:50:45,876
THAT WE ARE  INVESTIGATING LARGE

1428
00:50:45,942 --> 00:50:47,878
MODELS THAT RELATES TO

1429
00:50:47,944 --> 00:50:48,178
EVALUATION.

1430
00:50:48,245 --> 00:50:49,446
NOT ONLY WE EVALUATE HOW IT

1431
00:50:49,513 --> 00:50:52,783
WORKS FOR LITERATURE SEARCH OR

1432
00:50:52,849 --> 00:50:55,552
MEDICAL LITERATURE BUT WE ALSO

1433
00:50:55,619 --> 00:50:58,288
EVALUATE NLP TASKS THROUGH OUR

1434
00:50:58,355 --> 00:51:00,090
COLLABORATION WITH A GROUP AT

1435
00:51:00,157 --> 00:51:00,524
YALE.

1436
00:51:00,590 --> 00:51:03,193
WE ALSO HAVE DONE SOMETHING WITH

1437
00:51:03,260 --> 00:51:05,295
EVALUATION WITH THE LATEST GPT

1438
00:51:05,362 --> 00:51:07,230
WORLD VISION THAT INCLUDES BOTH

1439
00:51:07,297 --> 00:51:08,832
THE TEXT OR IMAGE ALTOGETHER

1440
00:51:08,899 --> 00:51:11,334
WHICH YOU WILL HEAR A

1441
00:51:11,401 --> 00:51:14,771
PRESENTATION BY DR. JIN IN MY

1442
00:51:14,838 --> 00:51:18,542
GROUP WHO IS A NEWLY JOINED POST

1443
00:51:18,608 --> 00:51:20,610
DOC FELLOW, HE HAS DONE A LOT OF

1444
00:51:20,677 --> 00:51:22,646
WORK INVOLVED IN THIS SPACE, SO

1445
00:51:22,712 --> 00:51:24,548
NOT ONLY ARE WE THINKING ABOUT

1446
00:51:24,614 --> 00:51:26,416
VALUE BUT HOW WE CAN USE OUR

1447
00:51:26,483 --> 00:51:29,519
DOMAIN EXPERTISE TO IMPROVE THE

1448
00:51:29,586 --> 00:51:30,520
STANDARD LARGE LANGUAGE MODELS.

1449
00:51:30,587 --> 00:51:32,289
YOU KNOW PEOPLE TALK ABOUT AND

1450
00:51:32,355 --> 00:51:34,458
KNOW THAT RETRIEVAL OF THE

1451
00:51:34,524 --> 00:51:35,659
GENERATION OR RAC FOR SHORT IS

1452
00:51:35,725 --> 00:51:36,827
ABLE TO INCLUDE STANDARD BUT

1453
00:51:36,893 --> 00:51:38,662
THERE ARE THINGS THAT ARE

1454
00:51:38,728 --> 00:51:39,963
SPECIAL IN OUR COMMITTEE, YOU

1455
00:51:40,030 --> 00:51:41,798
KNOW WHAT KIND OF DOCUMENTS YOU

1456
00:51:41,865 --> 00:51:45,268
WANT TO DO THE RETRIEVAL, WHAT

1457
00:51:45,335 --> 00:51:47,103
RETRIEVER YOU WANT TO USE.

1458
00:51:47,170 --> 00:51:48,505
THESE ARE MORE DEPTH RESEARCH

1459
00:51:48,572 --> 00:51:52,542
PROGRAMS WE HAVE STARTED TO

1460
00:51:52,609 --> 00:51:52,843
INVESTIGATE.

1461
00:51:52,909 --> 00:51:58,715
WE HAVE ALSO DOWN THE WORK TO

1462
00:51:58,782 --> 00:52:01,151
TEACH CHAT GPT TO USING TOOLS,

1463
00:52:01,218 --> 00:52:04,421
AND WE USE CHAT GPT TO USE THE

1464
00:52:04,488 --> 00:52:07,190
NCBI TOOLS AND BY DOING THAT, IT

1465
00:52:07,257 --> 00:52:08,291
BECOMES MUCH SMARTER IN

1466
00:52:08,358 --> 00:52:09,526
ANSWERING QUESTIONS RELATED TO

1467
00:52:09,593 --> 00:52:10,961
THE PROTEIN GENES AND THINGS

1468
00:52:11,027 --> 00:52:13,763
THAT IN OUR DOMAIN, AND THIS IS

1469
00:52:13,830 --> 00:52:14,664
OTHERWISE NOT AVAILABLE BECAUSE

1470
00:52:14,731 --> 00:52:16,233
I DON'T THINK THESE DOMAIN

1471
00:52:16,299 --> 00:52:19,469
DATABASE HAS BEEN USED FOR

1472
00:52:19,536 --> 00:52:24,541
TRAINING CHAT GPT.

1473
00:52:24,608 --> 00:52:25,442
SO UNDERSTANDING THE POTENTIAL

1474
00:52:25,509 --> 00:52:27,711
AS WELL AS THE RISK OF LARGE

1475
00:52:27,777 --> 00:52:29,479
MODELS, WE ALSO THINK HOW CAN WE

1476
00:52:29,546 --> 00:52:31,815
REALLY TAKE ADVANTAGE OF THE

1477
00:52:31,882 --> 00:52:32,816
GENERATIVE AI IS ITS CAPABILITY

1478
00:52:32,883 --> 00:52:37,854
FOR SOME OF THE MOST IMPORTANT

1479
00:52:37,921 --> 00:52:39,256
PROVINCE IN OUR FIELD, SO WE'VE

1480
00:52:39,322 --> 00:52:41,024
DONE SOME OF OUR PROJECTS BUT

1481
00:52:41,091 --> 00:52:43,026
FOR THE TIME BEING I WILL ONLY

1482
00:52:43,093 --> 00:52:45,061
TALK ABOUT 1 WHICH IS CALLED

1483
00:52:45,128 --> 00:52:48,331
TRIAL GPT, WHICH IS REALLY

1484
00:52:48,398 --> 00:52:48,632
EXCITING.

1485
00:52:48,698 --> 00:52:51,201
IT'S SAID THE DPOAL IS TO SPEED

1486
00:52:51,268 --> 00:52:52,602
UP THE PATIENT TRIAL MATCHING

1487
00:52:52,669 --> 00:52:54,004
PROCESS WHICH IS A REAL WORLD

1488
00:52:54,070 --> 00:52:55,872
PROBLEM AT THE NE H AND ALSO

1489
00:52:55,939 --> 00:52:57,073
MORE BROADLY, ABOUT YOU WE ARE

1490
00:52:57,140 --> 00:52:59,709
STARTING WITH THE NIH, SPECIALLY

1491
00:52:59,776 --> 00:53:01,878
IN OUR CLOSE COLLABORATION WITH

1492
00:53:01,945 --> 00:53:03,613
CLINICIANS AT THE NCI WHO

1493
00:53:03,680 --> 00:53:04,948
ACTUALLY RUN THOSE CLINICAL

1494
00:53:05,015 --> 00:53:05,315
TRIALS.

1495
00:53:05,382 --> 00:53:09,853
SO THE STATUS QUO, FOR THE

1496
00:53:09,920 --> 00:53:11,955
MOMENT IS THESE--THERE'S A TEAM

1497
00:53:12,022 --> 00:53:13,023
OF--THERE'S OFFICE IN THE TEAM

1498
00:53:13,089 --> 00:53:15,058
OF NURSES WHO WILL LOOK AT THE

1499
00:53:15,125 --> 00:53:16,826
PATIENT INFORMATION ON THE 1 END

1500
00:53:16,893 --> 00:53:19,696
AND TRY TO MANUALLY TO MATCH

1501
00:53:19,763 --> 00:53:20,964
WITH DIFFERENT TRENDS THAT ARE

1502
00:53:21,031 --> 00:53:23,233
RUNNING AT THE NCI OR CLINICAL

1503
00:53:23,300 --> 00:53:25,201
CENTER AT NIH, AND THIS PROCESS

1504
00:53:25,268 --> 00:53:28,438
AS WE TOLD BY OUR PHYSICIANS IS

1505
00:53:28,505 --> 00:53:32,075
VERY DIFFICULT OR NOT--IT'S

1506
00:53:32,142 --> 00:53:33,510
ERROR PRONE AND OFTEN TIMES THE

1507
00:53:33,577 --> 00:53:34,811
PHYSICIANS HAVE TO STEP IN AND

1508
00:53:34,878 --> 00:53:37,447
SPEND THEIR TIME TO DO JUST THIS

1509
00:53:37,514 --> 00:53:38,481
TRIAGE PROCESS EMPLOY SO WHAT WE

1510
00:53:38,548 --> 00:53:42,018
ARE TRYING TO DO IS TO IMPROVE

1511
00:53:42,085 --> 00:53:44,187
EFFICIENCY AS WELL AS PEOPLE

1512
00:53:44,254 --> 00:53:45,655
HAVE ACTESS TO PATIENTS IN ALL

1513
00:53:45,722 --> 00:53:46,856
DIFFERENT COMMUNITIES IN THIS

1514
00:53:46,923 --> 00:53:48,124
REGARD, EVEN THOUGH WE START

1515
00:53:48,191 --> 00:53:51,595
WIDE ONCOLOGY BUT THIS APPROACH

1516
00:53:51,661 --> 00:53:53,797
WE DEVELOP IS GENERALIZABLE AND

1517
00:53:53,863 --> 00:53:56,600
WE WERE INVITED TO WRITE A COVER

1518
00:53:56,666 --> 00:53:58,368
PAGE FOR TRIAL GPT FOR THE

1519
00:53:58,435 --> 00:53:58,835
AMERICAN NEUROLOGICAL

1520
00:53:58,902 --> 00:54:01,137
ASSOCIATION EMPLOY SO IN A NUT

1521
00:54:01,204 --> 00:54:03,206
SHELL WHAT TRIAL GPT DOES IS ON

1522
00:54:03,273 --> 00:54:05,508
THE 1 END WE HAVE PATIENT NOTES

1523
00:54:05,575 --> 00:54:07,410
AND ON THE OTHER TRIAL WE HAVE

1524
00:54:07,477 --> 00:54:07,811
CANDIDATES.

1525
00:54:07,877 --> 00:54:09,379
SO WE TAKE THE TRADITIONAL

1526
00:54:09,446 --> 00:54:11,114
METHOD, TAKE A BLACK BOX AND

1527
00:54:11,181 --> 00:54:13,583
TELL YOU THERE'S A MATCH,

1528
00:54:13,650 --> 00:54:14,618
WHETHER THERE'S PATIENT A GOOD

1529
00:54:14,684 --> 00:54:16,519
OR BETTER MATCH FOR THIS TRIAL,

1530
00:54:16,586 --> 00:54:18,922
BUT INSTEAD WE WILL HAVE

1531
00:54:18,989 --> 00:54:20,523
INTERMEDIATE STEP WHERE WE LOOK

1532
00:54:20,590 --> 00:54:23,827
AT EACH INDIVIDUAL INCLUSION,

1533
00:54:23,893 --> 00:54:25,528
EXCLUSION CRITERIA, WE HAVE

1534
00:54:25,595 --> 00:54:28,164
ASKED CHAT GPT TO EVALUATE BASED

1535
00:54:28,231 --> 00:54:30,533
ON THE PATIENT INFORMATION AND

1536
00:54:30,600 --> 00:54:32,068
WE ASKED CHAT GPT, NOT ONLY CAN

1537
00:54:32,135 --> 00:54:33,970
YOU TELL ME THE ANSWER, WHETHER

1538
00:54:34,037 --> 00:54:36,706
IT'S ELELIMINATEDDIBLE OR NOT

1539
00:54:36,773 --> 00:54:38,208
ELIGIBLE FOR EACH PERSON, BUT

1540
00:54:38,274 --> 00:54:39,042
ALSO EXPLAIN WHERE IS THE

1541
00:54:39,109 --> 00:54:41,578
EVIDENCE IN THE PATIENT NOTES,

1542
00:54:41,645 --> 00:54:42,412
THE RELEVANT SENTENCE AND WHAT

1543
00:54:42,479 --> 00:54:43,613
IS THE RATIONAL TO EXPLAIN THIS

1544
00:54:43,680 --> 00:54:45,348
AND BY DOING THIS, THIS IS

1545
00:54:45,415 --> 00:54:46,750
REALLY ADDING THE TRANSZ

1546
00:54:46,816 --> 00:54:47,884
APPEARANCE AND ALSO

1547
00:54:47,951 --> 00:54:49,085
EXPLAINABILITY OF THE AI SYSTEM

1548
00:54:49,152 --> 00:54:53,256
FOR THE POTENTIAL USERS, ABOUT

1549
00:54:53,323 --> 00:54:55,592
YOU IN ADDITION THIS ALSO HELPS

1550
00:54:55,659 --> 00:54:58,128
INCLUDING TO PUT CHAT GPT IN A

1551
00:54:58,194 --> 00:55:00,930
CONTROLLED FASHION THAT YOU CAN

1552
00:55:00,997 --> 00:55:01,331
MITIGATE POSSIBLE

1553
00:55:01,398 --> 00:55:03,800
HALLUCINATIONS, SO IN DOING

1554
00:55:03,867 --> 00:55:05,168
THAT, WE HAVE--WE HAVE BUNCH

1555
00:55:05,235 --> 00:55:06,569
MARK, ON 3 DIFFERENT DATA SETS

1556
00:55:06,636 --> 00:55:09,005
BECAUSE THIS IS A KNOWN PROBLEM,

1557
00:55:09,072 --> 00:55:12,976
WHERE WE HAVE 200 PATIENTS AND

1558
00:55:13,043 --> 00:55:14,344
20,000 TRIALS AND STATE OF ART

1559
00:55:14,411 --> 00:55:16,646
METHOD WHAT'SY HAVE LOOKED AT

1560
00:55:16,713 --> 00:55:18,114
HAVE ACHIEVED ONLY 55% ACCURACY

1561
00:55:18,181 --> 00:55:20,717
WHILE IN OUR CASE, TRIAL GPT IS

1562
00:55:20,784 --> 00:55:23,687
81% WHICH IS BEASKLY ON PARWITH

1563
00:55:23,753 --> 00:55:24,187
HUMAN [INDISCERNIBLE].

1564
00:55:24,254 --> 00:55:26,556
BECAUSE EVEN IF YOU ASK HUMANS

1565
00:55:26,623 --> 00:55:28,224
YOU MAY NOT BE GETTING THE

1566
00:55:28,291 --> 00:55:29,959
PERFECT ANSWER AND GETTING THAT

1567
00:55:30,026 --> 00:55:32,595
VERY ENCOURAGING RESULTS IN THE

1568
00:55:32,662 --> 00:55:33,997
OFFLINE EVALUATION WE HAVE

1569
00:55:34,064 --> 00:55:35,165
STARTED WE HAVE STARTED OUR

1570
00:55:35,231 --> 00:55:36,800
STUDY WHERE WE HAVE PHYSICIANS

1571
00:55:36,866 --> 00:55:39,002
LOOKING AT THE RESULTS WITH AND

1572
00:55:39,069 --> 00:55:40,403
WITHOUT AI ASSISTANCE AND WE

1573
00:55:40,470 --> 00:55:42,205
SHOWED THAT WITH THAT AI SYSTEM,

1574
00:55:42,272 --> 00:55:45,508
WE CAN DROP 40% OF TIME WITHOUT

1575
00:55:45,575 --> 00:55:50,547
ANY SACRIFICE IN ACCURACY.

1576
00:55:50,613 --> 00:55:52,215
SO I MENTIONED OTHER FUTURE

1577
00:55:52,282 --> 00:55:55,185
DIRECTIONS 1 WE HAVE MOVING ON

1578
00:55:55,251 --> 00:55:56,686
FROM TRADITIONAL HISTORICAL

1579
00:55:56,753 --> 00:56:00,724
COLOR, PHOTOGRAPHS TO THE MORE

1580
00:56:00,790 --> 00:56:02,258
ADVANCED OCT IMAGES IN

1581
00:56:02,325 --> 00:56:03,426
OPHTHALMOLOGY, I DIDN'T TALK

1582
00:56:03,493 --> 00:56:06,029
MUCH ABOUT RADIOLOGY BUT WE HAVE

1583
00:56:06,096 --> 00:56:06,863
THIS WONDERFUL COLLABBUATION

1584
00:56:06,930 --> 00:56:08,865
WITH DR. RON SUMMERS FROM THE

1585
00:56:08,932 --> 00:56:09,532
CLINICAL CENTER RADIOLOGY

1586
00:56:09,599 --> 00:56:10,700
DEPARTMENT WHERE HE ALSO HAS

1587
00:56:10,767 --> 00:56:15,438
BEEN A CO MENTOR FOR SEVERAL OF

1588
00:56:15,505 --> 00:56:18,308
MY TRAINEES AND THE MOST RECENT

1589
00:56:18,374 --> 00:56:19,843
1 IS DR. ZHU WHO HAS A PASSION

1590
00:56:19,909 --> 00:56:22,479
FOR WORKING ON CT IMAGES

1591
00:56:22,545 --> 00:56:23,880
TOGETHER WITH THE TEXT DATA.

1592
00:56:23,947 --> 00:56:26,082
SO WE ARE NOW WORKING ON 1

1593
00:56:26,149 --> 00:56:28,318
PROJECT, WE WILL POTENTIALLY

1594
00:56:28,384 --> 00:56:29,719
RELEASE A MULTIMODAL DATA SET

1595
00:56:29,786 --> 00:56:32,122
FOR THE CT DATA SET, THERE'S

1596
00:56:32,188 --> 00:56:33,389
NOTHING LIKE THIS AVAILABLE FOR

1597
00:56:33,456 --> 00:56:35,291
OUR RESEARCH COMMITTEE AT THE

1598
00:56:35,358 --> 00:56:35,525
MOMENT.

1599
00:56:35,592 --> 00:56:37,093
THERE'S A LOT INCLUDING THE 1

1600
00:56:37,160 --> 00:56:40,363
THAT RON AND I WORKED ON IN THE

1601
00:56:40,430 --> 00:56:42,465
RELEASE CALLED NIH CHECKS THE

1602
00:56:42,532 --> 00:56:44,067
RATE, IN 2017, WE RELEASED THAT

1603
00:56:44,134 --> 00:56:49,105
VERY LARGE AND THAT ACTUALLY WAS

1604
00:56:49,172 --> 00:56:51,141
VERY WELL RECEIVED BUT I HAVE A

1605
00:56:51,207 --> 00:56:53,676
HIGH HOPE THAT THIS CT WILL BE

1606
00:56:53,743 --> 00:56:55,278
ANOTHER HIGH IMPACT PROJECT AS

1607
00:56:55,345 --> 00:56:55,478
WELL.

1608
00:56:55,545 --> 00:56:56,980
AMONG OTHER THINGS WE TALKED

1609
00:56:57,046 --> 00:56:58,181
ABOUT THE NUMBER 1 PRIORITY FOR

1610
00:56:58,248 --> 00:57:03,219
AI SYSTEM, I THINK HAS TO BE THE

1611
00:57:03,286 --> 00:57:04,721
SAFETY, SO FOR THAT, THIS IS THE

1612
00:57:04,788 --> 00:57:08,691
TOPIC FOR MY Ph D STUDENT,

1613
00:57:08,758 --> 00:57:10,126
[INDISCERNIBLE] AT THE COMPUTER

1614
00:57:10,193 --> 00:57:13,196
SCIENCE AT UMD, COLLEGE PARK, IN

1615
00:57:13,263 --> 00:57:15,565
THE YEAR AND HALF HE'S WITH US,

1616
00:57:15,632 --> 00:57:17,433
HE HAS DONE INTERESTING RESEARCH

1617
00:57:17,500 --> 00:57:22,238
INCLUDING ARE VEALING THE RACIAL

1618
00:57:22,305 --> 00:57:24,574
BIAS OF LARGE [INDISCERNIBLE]

1619
00:57:24,641 --> 00:57:28,278
PRODLES, AND WE'VE SHOWN HOW

1620
00:57:28,344 --> 00:57:34,083
STRICT AND STRAIGHT FORWARD TO

1621
00:57:34,150 --> 00:57:35,485
GENERATE SOME ADVERSARIAL OR

1622
00:57:35,552 --> 00:57:37,954
HARMFUL OUTCOME, YOU MAY HAVE

1623
00:57:38,021 --> 00:57:45,528
HEARD HOW  AI GENERATED DAMAGING

1624
00:57:45,595 --> 00:57:46,963
STATEMENTS AND PICTURES BUT

1625
00:57:47,030 --> 00:57:48,198
SHOWING POTENTIAL RISK ANDS

1626
00:57:48,264 --> 00:57:49,799
BIASES, NOT ENOUGH IN MY MIND

1627
00:57:49,866 --> 00:57:52,101
FOR WE HAVE TO DEVELOP NOVEL ART

1628
00:57:52,168 --> 00:57:54,637
WITH THEM AND THIS IS THE

1629
00:57:54,704 --> 00:57:55,972
TECHNICAL CONTRIBUTION PART OF

1630
00:57:56,039 --> 00:57:57,373
HIS THESIS.

1631
00:57:57,440 --> 00:58:01,811
JUST ON WRAPPING UP, SO NLM HAS

1632
00:58:01,878 --> 00:58:03,746
BEEN A LEADER FOR AI AND

1633
00:58:03,813 --> 00:58:06,282
IMPROVING USER EXPERIENCE, WE

1634
00:58:06,349 --> 00:58:07,417
HOPE WE CAN CONTINUE TO

1635
00:58:07,483 --> 00:58:08,852
CONTRIBUTE TO THIS MISSION

1636
00:58:08,918 --> 00:58:10,587
THROUGH VARIOUS PROGECS

1637
00:58:10,653 --> 00:58:11,754
INCLUDING PUTTING AI IN REAL

1638
00:58:11,821 --> 00:58:12,856
WORLD APPLICATIONS THAT ARE VERY

1639
00:58:12,922 --> 00:58:15,425
CLOSE TIED TO THE MISSION OF

1640
00:58:15,491 --> 00:58:16,793
NLM, INCLUDING THE LITERATURE

1641
00:58:16,860 --> 00:58:19,062
SEARCH AND INDEX, THIS HAS TO BE

1642
00:58:19,128 --> 00:58:21,364
1 NLM PROJECT MEANING WE HAVE TO

1643
00:58:21,431 --> 00:58:23,766
ENGAGE AND INVOLVE A LOT OF

1644
00:58:23,833 --> 00:58:26,035
DIFFERENT PARTS OF NLM DIVISIONS

1645
00:58:26,102 --> 00:58:26,536
AND COLLEAGUES.

1646
00:58:26,603 --> 00:58:29,873
BUT ALSO WE ARE INTERESTED IN

1647
00:58:29,939 --> 00:58:31,741
DOING SOMETHING THAT SEE MORE AI

1648
00:58:31,808 --> 00:58:35,011
IN THE POTENTIAL CLINICAL SPACE

1649
00:58:35,078 --> 00:58:36,646
WHERE JUST TO SEE THE THINGS WE

1650
00:58:36,713 --> 00:58:39,349
HAVE DONE WITH AMD FOR EXAMPLE,

1651
00:58:39,415 --> 00:58:41,351
THIS WILL EVOLVE, NOT ONLY THE

1652
00:58:41,417 --> 00:58:44,387
NLM BUT ALSO OTHER ICs AT NIH

1653
00:58:44,454 --> 00:58:47,123
AND ALSO FDA, SO EVEN THOUGH,

1654
00:58:47,190 --> 00:58:48,424
I'M THE RESENTER TODAY BUT

1655
00:58:48,491 --> 00:58:51,160
REALLY THIS IS THE WORK FROM ALL

1656
00:58:51,227 --> 00:58:53,596
THE PEOPLE HARD WORKING AND

1657
00:58:53,663 --> 00:58:54,898
TALENTED GROUP MEMBERS FROM MY

1658
00:58:54,964 --> 00:58:55,331
GROUP.

1659
00:58:55,398 --> 00:58:57,567
AND AS WELL AS MY COLLABORATORS,

1660
00:58:57,634 --> 00:59:00,069
SO I WILL STOP HERE AND HAPPY TO

1661
00:59:00,136 --> 00:59:10,413
TAKE QUESTIONS.

1662
00:59:18,154 --> 00:59:19,022
THANK YOU.

1663
00:59:19,088 --> 00:59:20,924
NTHANK YOU VERY MUCH FOR YOUR

1664
00:59:20,990 --> 00:59:22,792
PRESENTATION, THIS IS OUTSIDE OF

1665
00:59:22,859 --> 00:59:23,893
MY AREA EXPECTATIONS PETTER EASE

1666
00:59:23,960 --> 00:59:25,795
SO I HAVE NAIVE QUESTIONS FOR

1667
00:59:25,862 --> 00:59:28,231
THE NATURAL LANGUAGE PROCESSING

1668
00:59:28,298 --> 00:59:29,866
MODELS OR ALGORITHMS YOU USE TO

1669
00:59:29,933 --> 00:59:30,867
POWER THOSE LITERATURE SEARCHES,

1670
00:59:30,934 --> 00:59:33,202
HOW DO YOU VALIDATE THOSE?

1671
00:59:33,269 --> 00:59:34,971
I'M WONDERING ABOUT WHAT THE

1672
00:59:35,038 --> 00:59:36,039
GROUND TRUTH IS.

1673
00:59:36,105 --> 00:59:37,573
I'M WONDERING ABOUT AMD WHERE

1674
00:59:37,640 --> 00:59:39,609
YOU HAVE THE CLINICALLY

1675
00:59:39,676 --> 00:59:41,544
VALIDATED IMAGES, AS WELL THERE

1676
00:59:41,611 --> 00:59:42,812
ANALOGOUS GROUND TRUTH FOR

1677
00:59:42,879 --> 00:59:43,613
NATURAL LANGUAGE.

1678
00:59:43,680 --> 00:59:44,547
>> YEAH, YEAH, GOOD QUESTION.

1679
00:59:44,614 --> 00:59:47,016
THANK YOU FOR THAT QUESTION.

1680
00:59:47,083 --> 00:59:49,919
SO YES, SO, IT VARIES FOR

1681
00:59:49,986 --> 00:59:51,187
DIFFERENT PROBLEMS THAT WE ARE

1682
00:59:51,254 --> 00:59:54,824
TALKING ABOUT, SO, I JUST GIVE

1683
00:59:54,891 --> 00:59:56,526
YOU SOME EXAMPLES WHERE WE

1684
00:59:56,592 --> 01:00:00,997
EVALUATE THE SYSTEM IN GENERAL.

1685
01:00:01,064 --> 01:00:01,864
SO WE GENERALLY

1686
01:00:01,931 --> 01:00:02,298
CERTAINLY--CERTAINLY

1687
01:00:02,365 --> 01:00:04,834
VALENTINEDUATE IN 2 PARTS.

1688
01:00:04,901 --> 01:00:06,169
FIRST PART WE WILL HAVE SOME

1689
01:00:06,235 --> 01:00:08,404
DPROWND TRUTHS THAT ARE MANUELLY

1690
01:00:08,471 --> 01:00:11,808
ANNOTATED BY HUMAN EXPERT

1691
01:00:11,874 --> 01:00:15,845
SUBJECTS SO IF WE'RE TALKING

1692
01:00:15,912 --> 01:00:17,280
ABOUT RELATED ARTICLES SO WE

1693
01:00:17,347 --> 01:00:18,948
WOULD KNOW THAT WHICH ARTICLE IS

1694
01:00:19,015 --> 01:00:21,217
RELEVANT TO THIS ARTICLE AND

1695
01:00:21,284 --> 01:00:25,121
THESE PAIRS OF ARTICLES HAS BEEN

1696
01:00:25,188 --> 01:00:26,656
MANUALLY ANNOTATE FRIDAY HUMAN

1697
01:00:26,723 --> 01:00:28,391
SUBJECT HUMAN BEINGS AND TIMES

1698
01:00:28,458 --> 01:00:30,493
THESE MANUELLY GENERATE OR

1699
01:00:30,560 --> 01:00:33,096
MANUELLY ANNOTATED DATA SETS OR

1700
01:00:33,162 --> 01:00:38,401
GROUND TRUTH ARE EITHER BY US,

1701
01:00:38,468 --> 01:00:39,235
BY OURSELVES, OR--WELL, LET ME

1702
01:00:39,302 --> 01:00:41,270
PUT IT THIS WAY, IF THERE'S SUCH

1703
01:00:41,337 --> 01:00:42,872
DATA SET AVAILABLE BY OTHER

1704
01:00:42,939 --> 01:00:45,408
RESEARCHERS, WE WOULD USE THEIR

1705
01:00:45,475 --> 01:00:47,076
DATA SETS, FIRST, OKAY, IF

1706
01:00:47,143 --> 01:00:48,845
SOMETHING AVAILABLE BUT THIS IS

1707
01:00:48,911 --> 01:00:50,079
IMPORTANT, WE WILL GENERATE AND

1708
01:00:50,146 --> 01:00:52,315
DO THIS, OF COURSE THIS HAS TO

1709
01:00:52,382 --> 01:00:53,950
BE DONE BLIND WAY, WE CANNOT

1710
01:00:54,017 --> 01:00:56,319
HAVING PEOPLE TO DO THIS AND

1711
01:00:56,386 --> 01:00:57,153
THEN EVALUATE.

1712
01:00:57,220 --> 01:00:58,921
SO BUT THE PROBLEM WITH THIS IS,

1713
01:00:58,988 --> 01:01:00,089
WELL NOT THE PROBLEM BUT WE

1714
01:01:00,156 --> 01:01:02,392
ALWAYS DO THIS AT THE FIRST STEP

1715
01:01:02,458 --> 01:01:05,628
AND WE SHOW THAT OUR ALGORITHM

1716
01:01:05,695 --> 01:01:07,163
WORKS WELL ON THIS EVALUATION

1717
01:01:07,230 --> 01:01:09,365
DATA SET BUT SOMETIMES THIS DATA

1718
01:01:09,432 --> 01:01:10,500
SET ARE RELATIVELY SMALL

1719
01:01:10,566 --> 01:01:12,568
COMPARED TO MILLIONS OF PAPERS,

1720
01:01:12,635 --> 01:01:14,437
RIGHT IN SO THE NEXT TEP IS IF

1721
01:01:14,504 --> 01:01:17,740
WE SEE ENOUGH SIGNALS, ENOUGH,

1722
01:01:17,807 --> 01:01:20,576
YOU KNOW, WE HAVE ENOUGH

1723
01:01:20,643 --> 01:01:22,211
CONFIDENCE THAT OUR RHYTHM WORKS

1724
01:01:22,278 --> 01:01:24,447
BETTER THAN IN THESE OFFLINE

1725
01:01:24,514 --> 01:01:27,183
BENCH MARK AND DATA SET IN TERMS

1726
01:01:27,250 --> 01:01:30,353
OF, YOU KNOW F1 SCORE OR

1727
01:01:30,420 --> 01:01:33,289
ACCURACY OR RETRIEVAL RANKS AND

1728
01:01:33,356 --> 01:01:36,192
THEN WE STARTING TO IMPROVE AND

1729
01:01:36,259 --> 01:01:37,860
EVALUATE THIS IN AN ONLINE

1730
01:01:37,927 --> 01:01:39,195
FASHION, SO HERE THE EXAMPLE, I

1731
01:01:39,262 --> 01:01:42,698
WOULD USE IS FOR EXAMPLE, THE

1732
01:01:42,765 --> 01:01:44,801
RELEVANT SEARCH IN PUB SO ONCE

1733
01:01:44,867 --> 01:01:47,770
WE HAVE EVALUATED ALL THE

1734
01:01:47,837 --> 01:01:48,838
ALGORITHMS TO DEVELOP, ON

1735
01:01:48,905 --> 01:01:50,440
MISSION LEARNING AND BENCHMARK

1736
01:01:50,506 --> 01:01:52,008
AND DATA SETS THEN WE ARE

1737
01:01:52,075 --> 01:01:53,643
EVALUATING IN THE REAL WORLD

1738
01:01:53,709 --> 01:01:55,611
USERS, SO IN THAT CASE, WE DO

1739
01:01:55,678 --> 01:01:56,212
SOMETHING CALLED AB TESTING

1740
01:01:56,279 --> 01:02:00,550
WHICH IS SORT OF LIKE A CASE

1741
01:02:00,616 --> 01:02:04,687
CONTROL STUDY, IN BIOLOGY, WHERE

1742
01:02:04,754 --> 01:02:06,889
WE WOULD HAVE BOTH VERSION OF A

1743
01:02:06,956 --> 01:02:10,393
AND B WITH THE NEW ALGORITHM AND

1744
01:02:10,460 --> 01:02:11,394
WITHOUT THE NEW ALGORITHM AND

1745
01:02:11,461 --> 01:02:13,729
WHAT WE MEASURE IS CALLED A

1746
01:02:13,796 --> 01:02:14,864
CLICK THROUGH RATE.

1747
01:02:14,931 --> 01:02:17,600
SO OUR DIFFERENT HYPOTHESIS IS

1748
01:02:17,667 --> 01:02:18,367
THAT [INDISCERNIBLE] WOULD LIKE

1749
01:02:18,434 --> 01:02:20,136
TO SEE THAT MORE AND WILL CLICK

1750
01:02:20,203 --> 01:02:20,503
MORE.

1751
01:02:20,570 --> 01:02:23,372
SO THAT IS SOMETHING THAT WE

1752
01:02:23,439 --> 01:02:24,540
MONITOR AND THIS IS VERY MUCH

1753
01:02:24,607 --> 01:02:25,842
SIMILAR TO THE TECH COMPANIES,

1754
01:02:25,908 --> 01:02:27,610
YOU KNOW GOOGLE WHEN THEY

1755
01:02:27,677 --> 01:02:28,644
RELEASE NEW FEATURES, THEY DO

1756
01:02:28,711 --> 01:02:29,245
THIS KIND OF

1757
01:02:29,312 --> 01:02:30,046
CERTAINLY--CERTAINLY VALUATION

1758
01:02:30,113 --> 01:02:31,114
AS WELL, AND WE WERE VERY

1759
01:02:31,180 --> 01:02:33,950
CAUTIOUS ABOUT DOING THIS

1760
01:02:34,016 --> 01:02:34,951
PROCESS, WE WOULD ALWAYS

1761
01:02:35,017 --> 01:02:37,987
STARTING WITH 1% OF OUR USERS,

1762
01:02:38,054 --> 01:02:39,689
AND THEN ONLY 5% OF OUR USERS

1763
01:02:39,755 --> 01:02:40,756
AND IF THE SIGNAL IS CONTINUED

1764
01:02:40,823 --> 01:02:42,358
TO BE STRONG AND THE DIRECTION

1765
01:02:42,425 --> 01:02:44,961
THEY ARE LOOKING TO SEE, THEN WE

1766
01:02:45,027 --> 01:02:49,232
OPEN THAT TO 10% OR 25% TO

1767
01:02:49,298 --> 01:02:50,566
EVENTUALLY A HUNDRED PERCENT.

1768
01:02:50,633 --> 01:02:52,735
THIS IS HOW WE EVALUATE BECAUSE

1769
01:02:52,802 --> 01:02:54,103
WE UNDERSTAND THAT EVERY CHANGE

1770
01:02:54,170 --> 01:02:56,606
WE MAKE IN PUB MED CAN AFFECT

1771
01:02:56,672 --> 01:02:58,074
MILLIONS OF USERS SO THIS IS THE

1772
01:02:58,141 --> 01:02:59,342
GENERAL PIPELINE THAT WE DO.

1773
01:02:59,408 --> 01:03:02,879
EVEN AFTER WE ROLL OUT AN

1774
01:03:02,945 --> 01:03:05,581
ALGORITHM IN PUB MED, WE MONITOR

1775
01:03:05,648 --> 01:03:07,750
ITS USAGE AFTERWARDS AND WE

1776
01:03:07,817 --> 01:03:08,317
CONTINUE TO IMPROVE.

1777
01:03:08,384 --> 01:03:10,419
MEANING SOME OF THE MODELS WE

1778
01:03:10,486 --> 01:03:12,155
DEVELOP INITIALLY, FOR EXAMPLE,

1779
01:03:12,221 --> 01:03:13,055
THE NEW RESEARCH THAT WEB

1780
01:03:13,122 --> 01:03:14,557
CONNECTED DEVELOP AND PUT INTO

1781
01:03:14,624 --> 01:03:17,126
PUB MED WAS IN 2017, THAT MODEL

1782
01:03:17,193 --> 01:03:19,695
IS ACTUALLY CONTINUOUSLY

1783
01:03:19,762 --> 01:03:20,730
PERIODICALLY UPGRADED, UPDATED

1784
01:03:20,796 --> 01:03:22,832
EVERY 6 MONTHS SO WE YOU MAKING

1785
01:03:22,899 --> 01:03:25,134
SURE THE MODEL ACTUALLY FIT WITH

1786
01:03:25,201 --> 01:03:27,036
THE MOST RECENT REQUEST OF OUR

1787
01:03:27,103 --> 01:03:29,105
USERS, WITH THE LATEST DATA WE

1788
01:03:29,172 --> 01:03:29,305
HAVE.

1789
01:03:29,372 --> 01:03:31,007
SO THESE ARE SORT OF THE WAYS

1790
01:03:31,073 --> 01:03:41,484
THAT WE DO FOR THE NRP.

1791
01:03:41,918 --> 01:03:43,819
>> IT LOOKS LIKE THERE'S A BUNCH

1792
01:03:43,886 --> 01:03:48,958
OF QUESTIONS LINED UP SO--

1793
01:03:49,025 --> 01:03:50,293
>> THANK YOU FOR YOUR

1794
01:03:50,359 --> 01:03:54,964
PRESENTATION

1795
01:03:55,031 --> 01:03:56,065
PRESENTATION EMPLOY I HAD 1

1796
01:03:56,132 --> 01:03:57,800
TECHNICAL QUESTION AND THEN

1797
01:03:57,867 --> 01:04:00,069
LARGER QUESTION, YOU KNOW YOU

1798
01:04:00,136 --> 01:04:02,271
MENTIONED RAG AS GREAT

1799
01:04:02,338 --> 01:04:03,573
OPPORTUNITY FOR ESPECIALLY THE

1800
01:04:03,639 --> 01:04:07,109
KIND OF WORK THAT YOU ARE DOING

1801
01:04:07,176 --> 01:04:10,146
HOW DO YOU--HOW DO YOU ENVISION

1802
01:04:10,213 --> 01:04:11,480
USING RAG LIKE IS REFERENCE TIME

1803
01:04:11,547 --> 01:04:13,649
OR DO YOU THINK THERE'S AN

1804
01:04:13,716 --> 01:04:14,617
OPPORTUNITY GIVEN YOUR ACCESS TO

1805
01:04:14,684 --> 01:04:15,618
ALL THE INFORMATION THAT YOU

1806
01:04:15,685 --> 01:04:17,720
HAVE AND ALL THE DATA THAT YOU

1807
01:04:17,787 --> 01:04:19,655
HAVE TO THINK ABOUT RAG

1808
01:04:19,722 --> 01:04:23,025
APPROACHES AT THE TRAINING TIME

1809
01:04:23,092 --> 01:04:23,626
OF IT?

1810
01:04:23,693 --> 01:04:26,629
>> THAT'S A GOOD QUESTION.

1811
01:04:26,696 --> 01:04:29,098
SO, SO FAR WHAT WE DID AT THE

1812
01:04:29,165 --> 01:04:32,068
INFERENCE TIME, MORE ALIGNED

1813
01:04:32,134 --> 01:04:35,705
WITH OTHER, THE PRACTICE OF

1814
01:04:35,771 --> 01:04:37,206
USING RAG, NOW THIS COULD BE

1815
01:04:37,273 --> 01:04:39,742
CHANGING IF YOU'RE FAMILIAR WITH

1816
01:04:39,809 --> 01:04:41,777
THIS, BECAUSE THE NEWER VERSION

1817
01:04:41,844 --> 01:04:45,414
OF THE PLASMA CAN TAKE MUCH

1818
01:04:45,481 --> 01:04:47,516
LONGER CONTEXT THAN THE 32 K FOR

1819
01:04:47,583 --> 01:04:49,685
THE MOMENT, IT COULD BE MILLIONS

1820
01:04:49,752 --> 01:04:51,153
SO FOR THE MOMENT TO STUDY THAT

1821
01:04:51,220 --> 01:04:55,891
I WAS REFERRING TO, AND WHEN WE

1822
01:04:55,958 --> 01:04:56,959
BASICALLY EVALUATE DIFFERENT

1823
01:04:57,026 --> 01:04:57,994
KIND OF [INDISCERNIBLE] THAT WE

1824
01:04:58,060 --> 01:04:59,962
CAN USE AND WE CAN PROVIDE TO

1825
01:05:00,029 --> 01:05:01,597
THE MEDICAL,--FIRST OF ALL, LET

1826
01:05:01,664 --> 01:05:04,634
ME TAKE 1 BACK, SO WE EVALUATE

1827
01:05:04,700 --> 01:05:06,736
ON THE MEDICAL QA, DIFFERENT

1828
01:05:06,802 --> 01:05:15,745
MEDICAL QA DATA SETS AND THEN WE

1829
01:05:15,811 --> 01:05:16,545
COMPARED MULTIPLE LARGE IMAGE

1830
01:05:16,612 --> 01:05:18,948
WITH AND WITHOUT RAG AND THEN

1831
01:05:19,015 --> 01:05:20,916
USING RAG, WE ARE PARTICULARLY

1832
01:05:20,983 --> 01:05:22,918
LOOKING AT 2 DIFFERENT ASPECT, 1

1833
01:05:22,985 --> 01:05:26,789
ASPECT IS WHAT DOCUMS CAN BE

1834
01:05:26,856 --> 01:05:28,024
SUPPLIED FOR HELPING.

1835
01:05:28,090 --> 01:05:29,925
IT COULD BE PUB MED ARTICLES, IT

1836
01:05:29,992 --> 01:05:31,761
COULD BEING BOOKS, IT COULD BE

1837
01:05:31,827 --> 01:05:35,298
SOME OTHER MEDICAL TEXT, AND

1838
01:05:35,364 --> 01:05:36,332
THEN, THE OTHER--THE OTHER

1839
01:05:36,399 --> 01:05:39,201
DIMENSION WE LOOKED INTO IS WHAT

1840
01:05:39,268 --> 01:05:42,638
RETRIEVAL CAN BE USED FOR THIS

1841
01:05:42,705 --> 01:05:44,507
PURPOSE, IT COULD BE GENERAL,

1842
01:05:44,573 --> 01:05:47,009
N25, COULD BE MORE DENSE

1843
01:05:47,076 --> 01:05:49,812
RETRIEVERS SO ON AND O FORTH SO

1844
01:05:49,879 --> 01:05:50,579
BY COMBINING THIS INFORMATION

1845
01:05:50,646 --> 01:05:52,048
TOGETHER WE HAVE SOME GUIDANCE

1846
01:05:52,114 --> 01:05:53,916
AND SUGGESTION FOR SOME OF THE

1847
01:05:53,983 --> 01:05:56,786
BEST PRACTICE FOR RAG, NOW THIS

1848
01:05:56,852 --> 01:05:58,187
IS DEFINITELY, THINGS ARE STILL

1849
01:05:58,254 --> 01:06:00,589
CHANGING LIKE YOU KNOW WITH THE

1850
01:06:00,656 --> 01:06:02,224
MORE ADVANCED THAT MAYBE PEOPLE

1851
01:06:02,291 --> 01:06:02,825
WILL STILL DO SOMETHING

1852
01:06:02,892 --> 01:06:06,862
DIFFERENT BUT I THINK FOR THE

1853
01:06:06,929 --> 01:06:08,731
MOMENT, THAT THE FINDINGS THAT

1854
01:06:08,798 --> 01:06:10,099
WE UNCOVER INDEED THIS RESEARCH

1855
01:06:10,166 --> 01:06:11,067
WILL BE VERY INSIGHTFUL FOR

1856
01:06:11,133 --> 01:06:12,335
PEOPLE WHO ARE INTERESTED IN

1857
01:06:12,401 --> 01:06:15,204
USING RAG BECAUSE IT'S SO FAR AT

1858
01:06:15,271 --> 01:06:16,906
THE MOMENT IT SEEMS TO BE 1 WAY

1859
01:06:16,972 --> 01:06:18,407
TO GO, 1 OF THE IMPORTANT THINGS

1860
01:06:18,474 --> 01:06:22,144
WE CAN USE TO IMPROVE STANDARD

1861
01:06:22,211 --> 01:06:23,512
LARGE LANGUAGE MODELS FOR OUR

1862
01:06:23,579 --> 01:06:28,517
DOPAIN EMPLOY --DOMAIN.

1863
01:06:28,584 --> 01:06:31,120
BECAUSE I'M NOT A BELIEVER IN

1864
01:06:31,187 --> 01:06:33,089
CREATING LARGE RESOURCE MODELS

1865
01:06:33,155 --> 01:06:34,090
GIVEN THE TECHNICAL CHALLENGES

1866
01:06:34,156 --> 01:06:40,196
WE FACE IN THIS DOMAIN.

1867
01:06:40,262 --> 01:06:40,996
>> INTERESTING, THANK YOU.

1868
01:06:41,063 --> 01:06:42,365
MY SECOND QUESTION IS SHORT BUT

1869
01:06:42,431 --> 01:06:45,501
I'M CURIOUS ABOUT THE CHOICE OF

1870
01:06:45,568 --> 01:06:49,171
GPT, CHAT GPT, BUT NOWADAYS YOU

1871
01:06:49,238 --> 01:06:51,774
MENTIONED MIXED ROLE IS THERE

1872
01:06:51,841 --> 01:06:53,342
ANY INTEREST--LIKE WHAT'S A

1873
01:06:53,409 --> 01:06:55,511
REASON ON CHOOSE A COMMERCIAL

1874
01:06:55,578 --> 01:06:59,181
NONTRANSFER ENSEL LARGE LANGUAGE

1875
01:06:59,248 --> 01:07:00,516
MODEL VERSUS [INDISCERNIBLE] 1

1876
01:07:00,583 --> 01:07:03,386
IN YOUR AGENDA, I GUESS.

1877
01:07:03,452 --> 01:07:04,553
>> YEAH, YEAH, GREAT QUESTION.

1878
01:07:04,620 --> 01:07:06,288
SO, YOU KNOW AT THE

1879
01:07:06,355 --> 01:07:09,792
BEGENERATEDDING, WELL, THERE ARE

1880
01:07:09,859 --> 01:07:10,893
A COUPLE REASONS, 1 IS ERGS

1881
01:07:10,960 --> 01:07:13,529
PECIALLY AT THE BEGINNING, CHAT

1882
01:07:13,596 --> 01:07:15,398
GPT SEEMS TO BE   PERFORMING

1883
01:07:15,464 --> 01:07:16,966
BETTER THAN OPEN SOURCE MODELS,

1884
01:07:17,032 --> 01:07:19,068
THAT'S 1 OF THE REAPS, BUT I

1885
01:07:19,135 --> 01:07:24,507
GUESS, THE MOST IMPORTANT REASON

1886
01:07:24,573 --> 01:07:25,741
IS, DATA PRIVACY.

1887
01:07:25,808 --> 01:07:28,177
OKAY, SO WE WILL ACCOUNT FOR

1888
01:07:28,244 --> 01:07:30,212
EXAMPLE, TRIAL GPT PROJECT.

1889
01:07:30,279 --> 01:07:32,114
WE DEAL WITH SENSITIVE PRIVATE

1890
01:07:32,181 --> 01:07:34,617
PATIENT INFORMATION EMPLOY EVEN

1891
01:07:34,683 --> 01:07:36,585
THOUGH THESE ARE SEMI--THESE ARE

1892
01:07:36,652 --> 01:07:38,954
DEIDENTIFIED IN A WAY, BUT WE

1893
01:07:39,021 --> 01:07:42,324
THINK THIS IS MUCH MORE SAFE AND

1894
01:07:42,391 --> 01:07:43,459
CONTROLLED ENVIRONMENT BECAUSE

1895
01:07:43,526 --> 01:07:46,262
WE USE CHAT GPT NOT THROUGH THE

1896
01:07:46,328 --> 01:07:47,430
OPEN API, SO I DIDN'T MENTION

1897
01:07:47,496 --> 01:07:48,464
THAT, I SHOULD MENTION THAT.

1898
01:07:48,531 --> 01:07:50,866
WE DO EVERYTHING HERE WITH

1899
01:07:50,933 --> 01:07:51,400
MICROSOFT [INDISCERNIBLE]

1900
01:07:51,467 --> 01:07:54,003
ENVIRONMENT THROUGH THE CLUED

1901
01:07:54,069 --> 01:07:54,637
NIH [INDISCERNIBLE] INITIATIVE,

1902
01:07:54,703 --> 01:07:56,472
WHERE WE HAVE THIS HIPAA

1903
01:07:56,539 --> 01:07:58,040
COMPLIANT ENVIRONMENT THAT WE

1904
01:07:58,107 --> 01:08:00,476
CAN TEST AND RUN OUR RESEARCH

1905
01:08:00,543 --> 01:08:00,776
PROJECTS.

1906
01:08:00,843 --> 01:08:02,344
SO THIS IS SOMETHING THAT I

1907
01:08:02,411 --> 01:08:04,713
THINK TO US IS THE NUMBER 1 MOST

1908
01:08:04,780 --> 01:08:07,450
IMPORTANT THING TO DO AND

1909
01:08:07,516 --> 01:08:09,752
BECAUSE OF THAT WE CAN ACTUALLY

1910
01:08:09,819 --> 01:08:11,787
CREATE SOMETHING THAT IF

1911
01:08:11,854 --> 01:08:13,956
WE--LIKE I MENTIONED, CHAT GPT

1912
01:08:14,023 --> 01:08:15,424
AT THE MOMENT EXTENT AT THE

1913
01:08:15,491 --> 01:08:16,425
RESEARCH BUT IT'S SOMETHING THAT

1914
01:08:16,492 --> 01:08:18,093
COULD BE USED WITH OUR

1915
01:08:18,160 --> 01:08:19,395
PHYSICIANS AT NCI OR OTHER

1916
01:08:19,462 --> 01:08:21,764
ICs, THEN WE HAVE TO CREATE

1917
01:08:21,831 --> 01:08:26,535
SOMETHING THEY CAN USE IN A SAFE

1918
01:08:26,602 --> 01:08:27,269
CAN [INDISCERNIBLE] ENVIRONMENT

1919
01:08:27,336 --> 01:08:30,673
AND THIS HAS TO BE SOMETHING

1920
01:08:30,739 --> 01:08:32,107
LIKE AZURE, THE OPEN SOURCE 1S.

1921
01:08:32,174 --> 01:08:33,876
WE DO USE THAT, I DID NOT

1922
01:08:33,943 --> 01:08:35,778
MENTION THAT, WE DO USE THAT IN

1923
01:08:35,845 --> 01:08:38,514
MANY OF OUR PAPERS WHEN WE'RE

1924
01:08:38,581 --> 01:08:41,383
COMPARING WITH CHAT GPT 3.0, 5

1925
01:08:41,450 --> 01:08:46,589
AND 4, WE USE EXTENSIVELY FOR

1926
01:08:46,655 --> 01:08:48,991
EXAMPLE, THE [INDISCERNIBLE] 7 B

1927
01:08:49,058 --> 01:08:51,627
MODEL, FOR OUR EVALUATION AND

1928
01:08:51,694 --> 01:08:53,262
COMPARISON PURPOSE SO WE DO USE

1929
01:08:53,329 --> 01:09:03,839
OPEN SOURCE ONCE BUT NOT--NOT

1930
01:09:06,208 --> 01:09:11,113
SOMETHING THAT WE USE.

1931
01:09:11,180 --> 01:09:12,681
THANK YOU.

1932
01:09:12,748 --> 01:09:13,916
>> LUCILLA WAS NEXT AND THEN

1933
01:09:13,983 --> 01:09:18,621
CANNED WHAT AND THEN ME.

1934
01:09:18,687 --> 01:09:19,889
>> YEAH, THANKS FOR THE

1935
01:09:19,955 --> 01:09:20,756
QUESTION, MY QUESTION WAS

1936
01:09:20,823 --> 01:09:25,094
RELATED TO THE HIPAA

1937
01:09:25,160 --> 01:09:31,300
ENVIRONMENT, IS IT A SPECIAL

1938
01:09:31,367 --> 01:09:34,036
ARRANGEMENT IN THAT YOU'RE

1939
01:09:34,103 --> 01:09:35,471
PROVIDING THE DATA TO TRAIN THE

1940
01:09:35,538 --> 01:09:37,840
OPEN MODEL IN AI OR IS IT THE

1941
01:09:37,907 --> 01:09:41,277
CASE THAT YOU DON'T HAVE AN

1942
01:09:41,343 --> 01:09:45,748
ENVIRONMENT THAT OHM HAS THE

1943
01:09:45,814 --> 01:09:47,616
PRETRAINED MODEL IN THAT

1944
01:09:47,683 --> 01:09:50,853
EVERYTHING REMAINS WITHIN THE

1945
01:09:50,920 --> 01:09:52,388
NIH PLATFORM.

1946
01:09:52,454 --> 01:09:53,689
>> YEAH, GREAT QUESTION.

1947
01:09:53,756 --> 01:09:57,693
SO, THE 1 THAT WE'RE USING

1948
01:09:57,760 --> 01:09:58,527
THROUGH [INDISCERNIBLE], IT

1949
01:09:58,594 --> 01:10:01,063
GIVES US ACCESS TO THE LATEST

1950
01:10:01,130 --> 01:10:03,766
GPT MODELS BUT ALSO PROTECTS

1951
01:10:03,832 --> 01:10:05,834
JUST LIKE YOU'RE SAYING THAT

1952
01:10:05,901 --> 01:10:09,772
DATA THAT'S USED WILL NOT BE AS

1953
01:10:09,838 --> 01:10:11,407
I UNDERSTAND AS MICROSOFT AS YOU

1954
01:10:11,473 --> 01:10:14,843
TOLD US IS NOT GOING TO BE USED,

1955
01:10:14,910 --> 01:10:18,447
PUT BACK FOR OPEN AI FOR THEIR

1956
01:10:18,514 --> 01:10:19,715
MODEL TRAINEE, SO, IT'S

1957
01:10:19,782 --> 01:10:26,922
PROTECTED IN THAT SENSE.

1958
01:10:26,989 --> 01:10:28,557
WE HAVE FOR EXAMPLE, USED OTHER

1959
01:10:28,624 --> 01:10:30,125
DATA SETS LIKE THE MIMIC DATA

1960
01:10:30,192 --> 01:10:32,828
SET FOR EXAMPLE, AND AS YOU KNOW

1961
01:10:32,895 --> 01:10:35,230
THAT, YOU KNOW IN OF THE

1962
01:10:35,297 --> 01:10:37,232
RESEARCH IN EARLY DAYS, THEY

1963
01:10:37,299 --> 01:10:39,668
USED OPEN AI API AND USING

1964
01:10:39,735 --> 01:10:41,303
IMITATED DIRECTLY AND THAT WAS

1965
01:10:41,370 --> 01:10:43,439
ESSENTIALLY NOT A GOOD THING AND

1966
01:10:43,505 --> 01:10:44,573
OUR COMMITTEE SOON REALIZED WE

1967
01:10:44,640 --> 01:10:47,242
SHOULD NOT DO THAT SO USING THE

1968
01:10:47,309 --> 01:10:48,210
AI ENVIRONMENT ACTUALLY PROTECTS

1969
01:10:48,277 --> 01:10:49,945
US AND ALLOW US TO PURSUE OUR

1970
01:10:50,012 --> 01:10:57,419
RESEARCH PROJECT IN THAT WAY.

1971
01:10:57,486 --> 01:10:58,087
>> THANK YOU.

1972
01:10:58,153 --> 01:10:59,989
>> JUST A QUICK ADD ON TO THAT,

1973
01:11:00,055 --> 01:11:02,091
WE HAVE A SIMILAR RELATIONSHIP

1974
01:11:02,157 --> 01:11:04,493
AND IN ADDITION TO NOT USING IT

1975
01:11:04,560 --> 01:11:06,095
TO TRAIN, THEY ALSO DON'T RETAIN

1976
01:11:06,161 --> 01:11:09,498
ANY OF THE INFORMATION THAT'S

1977
01:11:09,565 --> 01:11:09,999
BEEN SENT.

1978
01:11:10,065 --> 01:11:12,368
SO THERE'S A NUMBER OF

1979
01:11:12,434 --> 01:11:14,036
PROTECTIONS THERE.

1980
01:11:14,103 --> 01:11:16,171
IT'S ENCRYPTED IN TRANSIT, IT IS

1981
01:11:16,238 --> 01:11:17,406
USED TRANSIENTLY, IT IS NOT

1982
01:11:17,473 --> 01:11:20,809
KEPT, NOT USED TO TRAIN.

1983
01:11:20,876 --> 01:11:28,283
SO IN THEORY IT IS SAFE, I SAY

1984
01:11:28,350 --> 01:11:29,451
IN THEORY BECAUSE--

1985
01:11:29,518 --> 01:11:30,653
>> --TRUST WHAT THEY TOLD US.

1986
01:11:30,719 --> 01:11:32,388
>> --WELL WE WERE IN A BIT OF

1987
01:11:32,454 --> 01:11:34,390
DISCUSSION WITH THEM TO VERIFY

1988
01:11:34,456 --> 01:11:36,659
THIS IN A NARR BIT OF DETAIL SO

1989
01:11:36,725 --> 01:11:43,332
I THINK, HOPEFULLY IT'S TRUE.

1990
01:11:43,399 --> 01:11:43,732
[INDISCERNIBLE]

1991
01:11:43,799 --> 01:11:46,435
>> YES, I HAVE A COUPLE

1992
01:11:46,502 --> 01:11:48,003
QUESTIONS, FIRST OF ALL

1993
01:11:48,070 --> 01:11:48,904
EXCELLENT PRESENTATION AND

1994
01:11:48,971 --> 01:11:49,338
RESEARCH IN GENERAL.

1995
01:11:49,405 --> 01:11:51,273
I HAVE A COUPLE QUESTIONS AS

1996
01:11:51,340 --> 01:11:55,577
TO--I'M A BIG FAN OF YOURS, SO I

1997
01:11:55,644 --> 01:11:59,081
COME MORE FROM THAT DIRECTION,

1998
01:11:59,148 --> 01:12:00,916
BUT ESPECIALLY IN THE FIELD OF

1999
01:12:00,983 --> 01:12:04,319
DETECTING VARIANT ANDS IMPACTING

2000
01:12:04,386 --> 01:12:07,156
AND IMPACTING DEVISION VARIANTS

2001
01:12:07,222 --> 01:12:10,392
AND SUCH, ARE YOU SORT OF THE

2002
01:12:10,459 --> 01:12:12,895
ENTITIES THAT PUB TATOR IS

2003
01:12:12,961 --> 01:12:13,495
RECOGNIZING, FOR INSTANCE I

2004
01:12:13,562 --> 01:12:14,963
NOTICE THERE ARE PATHWAYS THAT

2005
01:12:15,030 --> 01:12:17,066
THEY ARE NOW PART OF YOUR RECK

2006
01:12:17,132 --> 01:12:20,669
NIGS AND YOU HAVE GENE DISEASE

2007
01:12:20,736 --> 01:12:23,138
AND MUTATION?

2008
01:12:23,205 --> 01:12:25,207
ANOTHER QUESTION RELATED TO PUB

2009
01:12:25,274 --> 01:12:26,375
DATA ARE YOU--IT WASN'T VERY

2010
01:12:26,442 --> 01:12:29,044
CLEAR TO ME IN YOUR PRESENTATION

2011
01:12:29,111 --> 01:12:33,348
AND YOUR DOCUMENTS EXACTLY HOW

2012
01:12:33,415 --> 01:12:36,752
THE PROGRESS OF PUBTATOR 3 IN

2013
01:12:36,819 --> 01:12:37,786
RECOGNIZING RELATIONSHIPS

2014
01:12:37,853 --> 01:12:40,923
BETWEEN THOSE ENTITIES WHICH

2015
01:12:40,989 --> 01:12:42,357
IS--WE HAD TRIED AND I WOULD

2016
01:12:42,424 --> 01:12:43,692
LIKE TO DO, IT'S A VERY

2017
01:12:43,759 --> 01:12:46,895
DIFFICULT TASK AND I THINK IT'S

2018
01:12:46,962 --> 01:12:50,099
REALLY SIGNIFICANT FOR, YOU KNOW

2019
01:12:50,165 --> 01:12:50,799
UNDERSTANDING THOSE

2020
01:12:50,866 --> 01:12:54,570
RELATIONSHIPS IN MOLECULAR

2021
01:12:54,636 --> 01:12:55,104
VERSUS PHENOTYPICAL

2022
01:12:55,170 --> 01:12:55,437
RELATIONSHIPS.

2023
01:12:55,504 --> 01:12:56,605
SO THAT WOULD BE MORE MY

2024
01:12:56,672 --> 01:12:57,306
QUESTION WITH THAT.

2025
01:12:57,372 --> 01:12:59,007
AND THE THIRD QUESTION, JUST

2026
01:12:59,074 --> 01:13:03,812
RELATED, THIS IS ONLY MY FIRST,

2027
01:13:03,879 --> 01:13:05,781
THE PUB TATOR, IF YOU ARE ALSO

2028
01:13:05,848 --> 01:13:08,317
THINKING OF LINKING MORE, YOUR

2029
01:13:08,383 --> 01:13:10,652
PUB TATOR IN YOUR SEARCH AND

2030
01:13:10,719 --> 01:13:13,055
YOUR RESEARCH THERE WITH YOUR

2031
01:13:13,122 --> 01:13:18,427
BIOCREATIVE, YOU KNOW DEPOSITING

2032
01:13:18,494 --> 01:13:20,295
BENCHMARKING WE CAN USE ALSO FOR

2033
01:13:20,362 --> 01:13:22,030
OUR TRAINING OF, YOU KNOW

2034
01:13:22,097 --> 01:13:24,733
MACHINE LEARNING TOOLS.

2035
01:13:24,800 --> 01:13:26,635
SO YOU HAVE TAKEN--THEY PUB

2036
01:13:26,702 --> 01:13:32,875
TATOR AND ASK YOU ANOTHER NTHANK

2037
01:13:32,941 --> 01:13:33,041
YOU.

2038
01:13:33,108 --> 01:13:35,711
>> SO, I THINK THIS IS

2039
01:13:35,778 --> 01:13:37,446
REALLY--THANK YOU FOR LETTING ME

2040
01:13:37,513 --> 01:13:40,449
CLARIFY THAT I MAY NOT HAVE PUT

2041
01:13:40,516 --> 01:13:41,650
IN MY PRESENTATION IN RICHER

2042
01:13:41,717 --> 01:13:44,019
MATERIAL, SO AS YOU KNOW THAT

2043
01:13:44,086 --> 01:13:44,653
RELATIONSHIP [INDISCERNIBLE] IS

2044
01:13:44,720 --> 01:13:45,320
NOT A NEW THING.

2045
01:13:45,387 --> 01:13:47,356
A LOT OF US WHO WORKED ON THIS

2046
01:13:47,422 --> 01:13:49,992
KNOW ABOUT THIS AND THE REASON I

2047
01:13:50,058 --> 01:13:51,160
HESITATE AND INCLUDING OUR OWN

2048
01:13:51,226 --> 01:13:53,262
GROUP AND OTHERS GROUPS THAT ARE

2049
01:13:53,328 --> 01:13:55,097
ON THE CALL, THE REASON THAT WE

2050
01:13:55,164 --> 01:13:57,166
DIDN'T PUT THIS INTO PUB TATOR

2051
01:13:57,232 --> 01:13:59,501
IS BECAUSE I WAS RELUCTANT TO DO

2052
01:13:59,568 --> 01:14:01,703
THAT BAH OF THE RACERACY OF THE

2053
01:14:01,770 --> 01:14:03,338
RELATIONSHIP ISN'T THAT GOOD TO

2054
01:14:03,405 --> 01:14:04,740
BE HONEST, OKAY EMPLOY DEPENDING

2055
01:14:04,807 --> 01:14:06,175
ON THE KIND OF RELATIONS YOU'RE

2056
01:14:06,241 --> 01:14:07,676
TALKING ABOUT, IT COULD BE

2057
01:14:07,743 --> 01:14:10,913
RANGING FROM 80% TO REALLY 20%,

2058
01:14:10,979 --> 01:14:13,448
OKAY, SO THAT MADE ME REALLY

2059
01:14:13,515 --> 01:14:14,283
UNCOMFORTABLE PUTTING THAT INTO

2060
01:14:14,349 --> 01:14:16,485
THE EARLIER VERSION OF THE PUB

2061
01:14:16,552 --> 01:14:19,421
TATOR AND PART OF THE REASON IS

2062
01:14:19,488 --> 01:14:21,924
BECAUSE THERE WAS NO GREAT DATA

2063
01:14:21,990 --> 01:14:24,726
SET FOR SUPER VISING LEARNING

2064
01:14:24,793 --> 01:14:25,828
FOR TRAINING RELATION EXPRESSION

2065
01:14:25,894 --> 01:14:30,999
MODELS, AND THAT IS WHY WE, I

2066
01:14:31,066 --> 01:14:36,905
THINK 2022, 3 POST DOCS, 3 FTs

2067
01:14:36,972 --> 01:14:39,208
THEY SPENT 1 WHOLE YEAR AND

2068
01:14:39,274 --> 01:14:40,175
CURATED THE ABSTRACT WITH ALL

2069
01:14:40,242 --> 01:14:44,546
THE RELATIONS IN THE PAPERS SO

2070
01:14:44,613 --> 01:14:46,248
THAT WAS THE BIORED DATA SET I

2071
01:14:46,315 --> 01:14:48,750
WAS REFORRING TO AND AFTER THAT

2072
01:14:48,817 --> 01:14:50,252
EVEN THOUGH IT TURNED OUT TO THE

2073
01:14:50,319 --> 01:14:52,921
MOST LARGEST AND WITH THE MOST

2074
01:14:52,988 --> 01:14:53,956
COMPREHENSIVE COVERAGE WITH

2075
01:14:54,022 --> 01:14:54,656
DIFFERENT RELATIONSHIPS, NOT

2076
01:14:54,723 --> 01:14:58,894
JUST, YOU KNOW IT COULD BE

2077
01:14:58,961 --> 01:14:59,761
DRUG-DRUG INTERACTION, DRUG-DEC

2078
01:14:59,828 --> 01:15:02,564
RELATIONSHIP, COULD BE DRUG-GENE

2079
01:15:02,631 --> 01:15:03,599
RELATIONSHIP, RESEARCH HAS BEEN

2080
01:15:03,665 --> 01:15:07,703
USING IT BUT ALSO MENTIONED 600

2081
01:15:07,769 --> 01:15:09,238
IS NOT BIG ENOUGH.

2082
01:15:09,304 --> 01:15:09,905
WE WANT MORE.

2083
01:15:09,972 --> 01:15:20,449
SO TO ADDRESS THAT FROM OUR

2084
01:15:23,552 --> 01:15:25,187
COMMUNITY WE ADDED 400 SO NOW WE

2085
01:15:25,254 --> 01:15:26,021
MAKE A TOTAL 1000 AND NOW WE

2086
01:15:26,088 --> 01:15:27,089
THINK THIS IS REALEE GOOD AND

2087
01:15:27,155 --> 01:15:29,358
THE REASON IS THAT WE SEE THAT

2088
01:15:29,424 --> 01:15:30,893
AT LEAST IN OUR OWN BENCHMARKING

2089
01:15:30,959 --> 01:15:33,495
FOR A LOT OF THE NUMBERS, IT HAS

2090
01:15:33,562 --> 01:15:35,931
STARTED TO BECOME SATURATED.

2091
01:15:35,998 --> 01:15:37,633
IN OTHER WORDS, THE MORE DATA

2092
01:15:37,699 --> 01:15:39,835
YOU PUT INTO THAT IT MAY NOT BE

2093
01:15:39,902 --> 01:15:41,503
ACTUALLY HELPING YOU TO

2094
01:15:41,570 --> 01:15:42,004
DRAMATICALLY IMPROVE THE

2095
01:15:42,070 --> 01:15:43,171
PERFORMANCE ANY URGT IFER.

2096
01:15:43,238 --> 01:15:53,715
OKAY, SO NOW THE ACCURACY OF

2097
01:16:00,155 --> 01:16:02,491
BIORED, IS NOW MAKING ME MORE

2098
01:16:02,557 --> 01:16:05,060
COMFORTABLE TO ADD THAT TO THE

2099
01:16:05,127 --> 01:16:05,460
RESEARCH WEBSITE.

2100
01:16:05,527 --> 01:16:06,528
SO EVEN THAT AS YOU NOTICED OR I

2101
01:16:06,595 --> 01:16:08,363
SHOULD MENTION THATY WOO ARE

2102
01:16:08,430 --> 01:16:09,665
ONLY TEXT MINING RELATIONS IN

2103
01:16:09,731 --> 01:16:11,266
THE ABSTRACT BECAUSE WE TRIED

2104
01:16:11,333 --> 01:16:13,168
THE FULL TEXT, IT'S MUCH MORE

2105
01:16:13,235 --> 01:16:14,469
COMPLICATED AND ONCE WE DO THAT,

2106
01:16:14,536 --> 01:16:18,340
THE RESULTS ARE GOING BACKWARDS

2107
01:16:18,407 --> 01:16:20,842
WHICH MAKES ME, EVEN THOUGH

2108
01:16:20,909 --> 01:16:22,778
OTHER ENTITIES WE DO ABSTRACT

2109
01:16:22,844 --> 01:16:24,680
FOR FULL TEXT BUT FOR RELATIONS

2110
01:16:24,746 --> 01:16:26,949
WE ARE ONLY FOCUSING ON THE

2111
01:16:27,015 --> 01:16:27,416
ABSTRACT LEVEL.

2112
01:16:27,482 --> 01:16:30,619
THERE ARE ALWAYS OTHER THINGS WE

2113
01:16:30,686 --> 01:16:36,725
CAN TAG, GENE ONTOLOGY TERMS,

2114
01:16:36,792 --> 01:16:40,095
CELL PHENOTYPES, AND PATHWAYS,

2115
01:16:40,162 --> 01:16:41,630
WE'RE NOT--YOU KNOW, WITH THE

2116
01:16:41,697 --> 01:16:44,599
RESOURCES WE HAVE, WE'RE

2117
01:16:44,666 --> 01:16:46,301
GRADUALLY EXPANDING IN THESE

2118
01:16:46,368 --> 01:16:46,535
AREAS.

2119
01:16:46,601 --> 01:16:48,503
WE WELCOME INPUT FROM OUR

2120
01:16:48,570 --> 01:16:49,805
COMMUNITY WHICH 1S YOU THINK ARE

2121
01:16:49,871 --> 01:16:52,107
THE MOST IMPORTANT TO ADD, YOU

2122
01:16:52,174 --> 01:16:53,608
KNOW FOR EXAMPLE, PEOPLE HAVE

2123
01:16:53,675 --> 01:16:55,711
REACHED OUT TO US AND SAY, YOU

2124
01:16:55,777 --> 01:16:57,713
KNOW, PENE O TYPES, IS REALLY

2125
01:16:57,779 --> 01:17:02,584
IMPORTANT EMPLOY SO WE HAD A

2126
01:17:02,651 --> 01:17:07,222
PRONL ECTOMYOSIN TAGGING HUMAN

2127
01:17:07,289 --> 01:17:10,859
PHENOTYPES USING HUMAN PHENOTYPE

2128
01:17:10,926 --> 01:17:12,194
ONTOLOGY SO WE HAD SUCCESS

2129
01:17:12,260 --> 01:17:14,196
THERE, , SO THIS IS A HUMAN

2130
01:17:14,262 --> 01:17:15,897
DRICH COMMUNITY PROJECT WE WANT

2131
01:17:15,964 --> 01:17:17,299
TO ADD MORE BUT DON'T EXPECT US

2132
01:17:17,366 --> 01:17:18,033
TO DO EVERYTHING BECAUSE THAT

2133
01:17:18,100 --> 01:17:23,472
WILL BE AT THE COST OF THE

2134
01:17:23,538 --> 01:17:24,806
QUALITY UNFORTUNATELY, SO THAT'S

2135
01:17:24,873 --> 01:17:27,809
BASICALLY MY ANSWER TO YOU AT

2136
01:17:27,876 --> 01:17:29,611
THIS POINT.

2137
01:17:29,678 --> 01:17:30,645
>> EXCELLENT.

2138
01:17:30,712 --> 01:17:30,879
ANSWER.

2139
01:17:30,946 --> 01:17:31,380
THANK YOU.

2140
01:17:31,446 --> 01:17:34,883
>> WE HAVE 2 MORE MINUTES LEFT

2141
01:17:34,950 --> 01:17:37,452
BEFORE THE PRESENTATION FROM

2142
01:17:37,519 --> 01:17:38,186
YOUR TRAINEES DR. LU.

2143
01:17:38,253 --> 01:17:39,421
>> JUST 1 LITTLE QUESTION, I

2144
01:17:39,488 --> 01:17:41,323
THINK IT'S IMPORTANT FOR MY

2145
01:17:41,390 --> 01:17:42,958
EVALUATION IT'S ABOUT HOW YOU

2146
01:17:43,025 --> 01:17:44,760
ADDRESSING DIVERSITY IN TERMS OF

2147
01:17:44,826 --> 01:17:47,362
YOU KNOW THE ACTUAL RESEARCH?

2148
01:17:47,429 --> 01:17:49,965
I NOTICE YOU HAVE A LITTLE

2149
01:17:50,032 --> 01:17:51,500
STRONG DIVERSITY STATEMENT IN

2150
01:17:51,566 --> 01:17:54,102
YOUR PRESENTATION, IT WAS CLEAR

2151
01:17:54,169 --> 01:17:55,270
THAT YOU HAVE PEOPLE AND

2152
01:17:55,337 --> 01:17:57,372
RESOURCES THAT IN EMERGING ITS,

2153
01:17:57,439 --> 01:17:59,508
YOU DID MENTION VERY BRIEFLY

2154
01:17:59,574 --> 01:18:01,209
RACIAL BIAS AND I WOULD LIKE TO

2155
01:18:01,276 --> 01:18:03,945
HEAR YOUR ANSWER TO THAT

2156
01:18:04,012 --> 01:18:04,246
QUESTION.

2157
01:18:04,312 --> 01:18:04,646
THANK YOU.

2158
01:18:04,713 --> 01:18:06,848
NTHANK YOU FOR THAT QUESTION

2159
01:18:06,915 --> 01:18:10,886
MUCH SO AS I MENTIONED IN THE

2160
01:18:10,952 --> 01:18:13,055
WRITTEN REPORT, WE TAKE THIS

2161
01:18:13,121 --> 01:18:14,122
VERY SERIOUSLY AND WE REALLY

2162
01:18:14,189 --> 01:18:15,724
BELIEVE THE IMPORTANCE OF

2163
01:18:15,791 --> 01:18:19,127
DIVERSITY, IN BOTH RESEARCH AND

2164
01:18:19,194 --> 01:18:20,095
DAY-TO-DAY WORK, SO, I DON'T

2165
01:18:20,162 --> 01:18:22,464
KNOW HOW MUCH I CAN ADD ON TO

2166
01:18:22,531 --> 01:18:24,232
THIS BUT VERY QUICKLY TO

2167
01:18:24,299 --> 01:18:25,267
SUMMARIZE IN TERMS OF WE HAVE

2168
01:18:25,333 --> 01:18:28,503
DONE A LOT OF THINGS INCLUDING

2169
01:18:28,570 --> 01:18:33,108
MAKING SURE THAT OUR GROUP IS

2170
01:18:33,175 --> 01:18:34,543
VERY DIVERSE, EVERYBODY HAS TO

2171
01:18:34,609 --> 01:18:39,147
SAY, AND WE ALSO WORKING ON

2172
01:18:39,214 --> 01:18:42,851
VARIOUS INVOLVEMENT IN TERMS OF

2173
01:18:42,918 --> 01:18:43,418
CERTAINLY--CERTAINLY QUITY,

2174
01:18:43,485 --> 01:18:46,188
COMMITTEE AT THE NIH LEVEL AND

2175
01:18:46,254 --> 01:18:51,326
OUTSIDE INTERNATIONAL

2176
01:18:51,393 --> 01:18:51,827
PROFESSIONAL SOCIETIES.

2177
01:18:51,893 --> 01:18:53,929
WE HAVE, YOU KNOW IN SOME

2178
01:18:53,995 --> 01:18:56,198
INITTIAIVE IT IN TERMS OF GENDER

2179
01:18:56,264 --> 01:18:58,366
BALANCE, IN TERMS OF UNDERSERVED

2180
01:18:58,433 --> 01:19:01,670
COMMUNITY IN MY GROUP, AND LAST

2181
01:19:01,736 --> 01:19:02,304
BUT NOT LEAST, YOU SLEEP APNEA

2182
01:19:02,370 --> 01:19:04,206
AND OBESITYY THAT WE PUT THIS

2183
01:19:04,272 --> 01:19:05,240
INTO RESEARCH PROBLEMS AND WE

2184
01:19:05,307 --> 01:19:07,609
THINK THIS WILL ALSO HELP US TO

2185
01:19:07,676 --> 01:19:08,844
ADVANCE THE KNOWLEDGE ON MANY OF

2186
01:19:08,910 --> 01:19:11,713
THE THINGS THAT WE WORK IN

2187
01:19:11,780 --> 01:19:14,483
ESPECIALLY IN TODAY'S AI SPACE.

2188
01:19:14,549 --> 01:19:16,318
HAPPY TO TALK MORE IF WE NEEDED

2189
01:19:16,384 --> 01:19:20,989
DURING THE CLOSED SESSION.

2190
01:19:21,056 --> 01:19:21,289
>> GREAT.

2191
01:19:21,356 --> 01:19:22,991
THANK YOU VERY MUCH DR. LU, SO

2192
01:19:23,058 --> 01:19:27,963
WE NOW HAVE A CLOSED POSTER

2193
01:19:28,029 --> 01:19:30,265
SESSION WITH POST DOCTORAL

2194
01:19:30,332 --> 01:19:33,935
FELLOWS OF DR. LU?

2195
01:19:34,002 --> 01:19:34,369
>> [INDISCERNIBLE]

2196
01:19:34,436 --> 01:19:38,306
>> SORRY IN.

2197
01:19:38,373 --> 01:19:39,274
>> I THINK MARCILLA, WAS

2198
01:19:39,341 --> 01:19:39,508
WAITING.

2199
01:19:39,574 --> 01:19:41,209
>> I THINK THIS IS PUBLICLY

2200
01:19:41,276 --> 01:19:44,779
BEING BROADCAST, SO I TEXTED

2201
01:19:44,846 --> 01:19:46,281
THAT GRACILLIA CAN ASK THAT

2202
01:19:46,348 --> 01:19:47,949
DURING OUR 15 MINUTE CLOSED

2203
01:19:48,016 --> 01:19:50,051
SESSION WITH DR. LU.

2204
01:19:50,118 --> 01:19:51,720
ALL RIGHT, SO AGAIN WE ARE NOW

2205
01:19:51,786 --> 01:19:53,321
STARTING A CLOSED SESSION, THAT

2206
01:19:53,388 --> 01:19:56,057
POSTER SESSION WILL BE WITH POST

2207
01:19:56,124 --> 01:19:56,791
DOCTORAL STUDENTS AND THEN WE

2208
01:19:56,858 --> 01:19:58,860
WILL HAVE A CLOSED SESSION WITH

2209
01:19:58,927 --> 01:20:05,200
DR. LU AND THEN A BREAK THE OPEN

2210
01:20:05,267 --> 01:20:09,271
SESSION WITH DR. DEMNER-FUSHMAN,

2211
01:20:09,337 --> 01:20:10,472
WILL START AT 2:00 P.M. EASTERN

2212
01:20:10,539 --> 01:20:12,440
SO THE BOARD AND TRAINEES SHOULD

2213
01:20:12,507 --> 01:20:13,675
JOIN THE CLOSED SESSION BREAK

2214
01:20:13,742 --> 01:20:18,180
OUT ROOM NOW.

2215
01:20:18,246 --> 01:20:19,614
>> ALL RIGHT, WELCOME, EVERYBODY

2216
01:20:19,681 --> 01:20:25,253
TO OUR SECOND PUBLIC SESSION

2217
01:20:25,320 --> 01:20:31,026
WHERE DR. DINA DEMNER-FUSHMAN

2218
01:20:31,092 --> 01:20:33,128
WILL PRESENT ON HER RESEARCH.

2219
01:20:33,195 --> 01:20:36,932
NGOOD AFTERNOON, I'M PRESENTING

2220
01:20:36,998 --> 01:20:37,666
MY INTRAMURAL RESEARCH PROJECT

2221
01:20:37,732 --> 01:20:40,035
THAT I HOPE IS PROVIDING TIMELY

2222
01:20:40,101 --> 01:20:41,636
AND ACCESS TO RELIABLE HEGHT

2223
01:20:41,703 --> 01:20:44,172
RELATED INFORMATION FOR DECISION

2224
01:20:44,239 --> 01:20:46,508
SUPPORT AND EICATION.

2225
01:20:46,575 --> 01:20:47,709
AND THE OVERALL RESEARCH

2226
01:20:47,776 --> 01:20:50,245
OBJECTIVE FOR THIS PROJECT IS TO

2227
01:20:50,312 --> 01:20:53,181
ENABLE EASY ACCESS TO RELIABLE

2228
01:20:53,248 --> 01:20:53,882
HEALTH RELATED INFORMATION AND

2229
01:20:53,949 --> 01:20:59,221
AT THE SAME TIME ADVANCE THE

2230
01:20:59,287 --> 01:21:00,155
FOUNDATIONAL NLM RESEARCH

2231
01:21:00,222 --> 01:21:01,323
APPROACHES AND THESE OBLIGATIONS

2232
01:21:01,389 --> 01:21:04,092
YECTIVES TIE WITH THE HHS

2233
01:21:04,159 --> 01:21:07,429
STRATEGIC GOALS ON PROTECTING

2234
01:21:07,495 --> 01:21:11,266
HEALTHCARE AND ALSO SUSTAINING

2235
01:21:11,333 --> 01:21:15,403
AND ADVANCING SCIENCES.

2236
01:21:15,470 --> 01:21:20,275
SO, I WILL START WITH MODERATING

2237
01:21:20,342 --> 01:21:22,143
MY APPROACH THAT THAT HIGHLY

2238
01:21:22,210 --> 01:21:23,445
AVAILABLE INFORMATION DPLEVERY

2239
01:21:23,511 --> 01:21:28,817
AS A QUESTION ON STREAM PROCESS

2240
01:21:28,883 --> 01:21:31,253
AND THEN VERY BRIEFLY TOUCH ON

2241
01:21:31,319 --> 01:21:32,621
THE TRADITIONAL QUESTION AND

2242
01:21:32,687 --> 01:21:35,790
ANSWERING THAT WE ARE DOING IN

2243
01:21:35,857 --> 01:21:37,759
SOME YEARS BEFORE THE LARGE

2244
01:21:37,826 --> 01:21:38,426
LANGUAGE MODEL REVOLUTION AND

2245
01:21:38,493 --> 01:21:40,495
THEN I WILL TALK MORE ABOUT OUR

2246
01:21:40,562 --> 01:21:43,164
CURRENT RESEARCH WHICH IS

2247
01:21:43,231 --> 01:21:43,698
BASICALLY LEVERAGING THE

2248
01:21:43,765 --> 01:21:46,735
OPPORTUNITIES AND ADDRESSING THE

2249
01:21:46,801 --> 01:21:53,775
CHALLENGES, POSED BY THE LLMs.

2250
01:21:53,842 --> 01:21:56,011
SO 1 QUESTION ANSWERING, IT IS A

2251
01:21:56,077 --> 01:21:57,412
NATURAL WAY FOR HUMANS TO LEARN

2252
01:21:57,479 --> 01:22:03,251
AND SPECIFICALLY FOR OUR TARGET

2253
01:22:03,318 --> 01:22:04,552
AUDIENCES WHICH IS ANYONE WHO IS

2254
01:22:04,619 --> 01:22:06,388
WORKING ON HEALTH PROBLEMS OR

2255
01:22:06,454 --> 01:22:11,259
HAVING HEALTH ISSUES OR IS EAR

2256
01:22:11,326 --> 01:22:13,094
HAS INFORMATION NEEDS RELATED TO

2257
01:22:13,161 --> 01:22:14,529
THEIR HEALTH, SO THE RESEARCHERS

2258
01:22:14,596 --> 01:22:16,798
HERE SAY THAT SCIENCE BEGINS BY

2259
01:22:16,865 --> 01:22:20,068
ASKING QUESTIONS AND OUR JOB IS

2260
01:22:20,135 --> 01:22:23,438
TO FIND THEM THE ANSWERS THAT

2261
01:22:23,505 --> 01:22:27,375
ALREADY EXIST AND MAYBE SHOW THE

2262
01:22:27,442 --> 01:22:29,811
GAPS THAT ARE NOT COVERED YET.

2263
01:22:29,878 --> 01:22:32,747
SOMEWHERE AT THE TURN OF THE

2264
01:22:32,814 --> 01:22:33,948
CENTURY, PHYSICIANS HAD REALLY

2265
01:22:34,015 --> 01:22:37,619
MOVED STUDIES ON SHOWING HOW

2266
01:22:37,686 --> 01:22:40,255
MANY CLENICAL QUESTIONS OCCUR AT

2267
01:22:40,322 --> 01:22:44,926
BEDSIDE BUT THEY'RE UNANSWERED,

2268
01:22:44,993 --> 01:22:45,727
BECAUSE PHYSICIANS OFTEN KNOW

2269
01:22:45,794 --> 01:22:48,263
THAT THE ANSWERS MIGHT NOT BE

2270
01:22:48,330 --> 01:22:54,035
THERE, DON'T HAVE TIME AND SO

2271
01:22:54,102 --> 01:22:54,669
ON.

2272
01:22:54,736 --> 01:22:56,404
SO THAT WAS 23 YEARS AGO, AND IS

2273
01:22:56,471 --> 01:23:00,342
IT STILL THE CASE AND RECENT

2274
01:23:00,408 --> 01:23:01,876
TESTAMENT THAT IT IS MORE THE

2275
01:23:01,943 --> 01:23:04,045
CASE NOW BECAUSE THERE IS A LOT

2276
01:23:04,112 --> 01:23:05,780
MORE INFORMATION COMING

2277
01:23:05,847 --> 01:23:06,715
PHYSICIANS WAY, THERE'S ALSO A

2278
01:23:06,781 --> 01:23:08,016
LOT INFORMATION IN THE EHR,

2279
01:23:08,083 --> 01:23:14,422
THERE WAS NOT A PROBLEM IN THE

2280
01:23:14,489 --> 01:23:16,024
20th CENTURY SO NOW IT'S EVEN

2281
01:23:16,091 --> 01:23:17,859
MORE IMPORTANT TO THE GET THE

2282
01:23:17,926 --> 01:23:19,060
INFORMATION THOSE THAT NEED IT

2283
01:23:19,127 --> 01:23:22,831
AND FINALLY THE CONSUMERS WHICH

2284
01:23:22,897 --> 01:23:26,267
ARE A VERY MAJOR PART OF OUR

2285
01:23:26,334 --> 01:23:27,569
RESEARCH, ARE ALWAYS FINDING THE

2286
01:23:27,635 --> 01:23:29,671
ANSWERS TO THOSE QUESTIONS, BUT

2287
01:23:29,738 --> 01:23:30,805
UNFORTUNATELY, THEY HAVE NO

2288
01:23:30,872 --> 01:23:34,008
MEANS TO JUDGE IF THE ANSWER IS

2289
01:23:34,075 --> 01:23:37,045
TRUTHFUL AND IF THE ANSWER IS

2290
01:23:37,112 --> 01:23:39,681
CORRECT SO THAT'S WHY WE ARE

2291
01:23:39,748 --> 01:23:45,387
FOCUSING ON THE

2292
01:23:45,453 --> 01:23:46,221
QUESTION-ANSWERING APPROACHES

2293
01:23:46,287 --> 01:23:48,490
AND THE DPOALS FOR

2294
01:23:48,556 --> 01:23:50,125
QUESTION-ANSWERING SYSTEMS WERE

2295
01:23:50,191 --> 01:23:55,597
FORMULATED AGAIN OVER 20 YEARS

2296
01:23:55,663 --> 01:23:58,800
AGO AND JUST IN 4 BULLETS, IT'S

2297
01:23:58,867 --> 01:24:02,537
TO ALLOW A USER TO ANSWER

2298
01:24:02,604 --> 01:24:04,839
QUESTIONS IN EVERY DAY LANGUAGE,

2299
01:24:04,906 --> 01:24:08,910
AND ANSWER QUICKLY, ANSWER

2300
01:24:08,977 --> 01:24:09,844
SUCCINCTLY, AND PROVIDE

2301
01:24:09,911 --> 01:24:11,713
SUFFICIENT CONTEXT TO VALIDATE

2302
01:24:11,780 --> 01:24:12,647
THE ANSWER, WE HAVE TO

2303
01:24:12,714 --> 01:24:14,082
UNDERSTAND THE WHAT THE QUESTION

2304
01:24:14,149 --> 01:24:16,918
IS ABOUT AND FIND THEM IN

2305
01:24:16,985 --> 01:24:20,088
REALTIME AND AGGREGATE THEM AND

2306
01:24:20,155 --> 01:24:20,955
PROVIDE ENOUGH EVIDENCE TO

2307
01:24:21,022 --> 01:24:21,556
SUPPORT THE ANSWER.

2308
01:24:21,623 --> 01:24:25,026
AND THIS IS WHAT WE HAVE DEMON

2309
01:24:25,093 --> 01:24:27,162
TRAITED TO THE PREVIOUS BOARD 4

2310
01:24:27,228 --> 01:24:29,798
YEARS AGO.

2311
01:24:29,864 --> 01:24:35,370
WE WERE JUST FINISHED BUILDING A

2312
01:24:35,437 --> 01:24:36,004
PROTOTYPE QUESTION ANSWERING

2313
01:24:36,070 --> 01:24:37,939
SYSTEM THAT WAS USING A

2314
01:24:38,006 --> 01:24:40,875
TRADITIONAL INFORMATION

2315
01:24:40,942 --> 01:24:42,377
RESTREEFAL BASED APPROACH SO

2316
01:24:42,444 --> 01:24:45,513
HERE YOU SEE THAT THE QUESTION

2317
01:24:45,580 --> 01:24:47,515
IS KIND OF UNDERSTOOD, THE MAIN

2318
01:24:47,582 --> 01:24:48,650
POINTS OF THE QUESTION, WHERE

2319
01:24:48,716 --> 01:24:51,519
THE QUESTION IS ABOUT, AND WHAT

2320
01:24:51,586 --> 01:24:53,988
THE USER WILL DO WITH THAT

2321
01:24:54,055 --> 01:24:56,858
QUESTION, WILL THE ANSWER TO

2322
01:24:56,925 --> 01:24:58,827
THAT QUESTION REMAIN, THESE MAIN

2323
01:24:58,893 --> 01:25:00,395
COMPONENTS ARE USED TO SEARCH

2324
01:25:00,462 --> 01:25:03,331
THE DATABASES OF 2 TYPES, BUT

2325
01:25:03,398 --> 01:25:05,600
THEN 1 OF SIMILAR QUESTIONS THAT

2326
01:25:05,667 --> 01:25:07,068
WERE ALREADY ASKED AND ANSWERED

2327
01:25:07,135 --> 01:25:11,940
BY THE NIH INSTITUTES SO HERE IF

2328
01:25:12,006 --> 01:25:14,209
I FIND A QUESTION, IT ALREADY

2329
01:25:14,275 --> 01:25:15,710
HAS AN EXCELLENT ANSWER PROVIDED

2330
01:25:15,777 --> 01:25:19,547
BY THE SPECIALIST BUT ALSO

2331
01:25:19,614 --> 01:25:21,182
FINDING THE LITERATURE AND

2332
01:25:21,249 --> 01:25:23,151
ENEXTRACTING THE ANSWER FROM THE

2333
01:25:23,218 --> 01:25:28,590
LITERATURE AND SUMMARIZING AND

2334
01:25:28,656 --> 01:25:29,357
AGGREGATING THE ANSWERS.

2335
01:25:29,424 --> 01:25:31,793
SO SOMETIME RIGHT AFTER WE

2336
01:25:31,860 --> 01:25:34,128
PRESENTED THE SYSTEM, WE STARTED

2337
01:25:34,195 --> 01:25:36,397
SEEING THESE INCREDIBLE SUCCESS

2338
01:25:36,464 --> 01:25:37,665
STORIES FROM THE LARGE LANGUAGE

2339
01:25:37,732 --> 01:25:38,967
MODELS THAT WERE DOING EXACTLY

2340
01:25:39,033 --> 01:25:42,203
WHAT WE ARE TRYING TO DO, THEY

2341
01:25:42,270 --> 01:25:44,239
WERE ANSWERING QUESTIONS, SO IF

2342
01:25:44,305 --> 01:25:47,475
WE LOOK CLOSER, UNDER THE HOOD

2343
01:25:47,542 --> 01:25:48,343
OF THESE QUESTION-ANSWERING

2344
01:25:48,409 --> 01:25:52,881
SYSTEMS, THIS IS WHAT THE GOOGLE

2345
01:25:52,947 --> 01:25:56,217
TEAM DID WITH THEIR OWN SYSTEM

2346
01:25:56,284 --> 01:26:00,321
AND THEY ASSEMBLED THESE FIRST

2347
01:26:00,388 --> 01:26:01,389
MEDICAL QUESTION ANSWERING

2348
01:26:01,456 --> 01:26:02,056
BENCHMARK AND THEY ACTUALLY

2349
01:26:02,123 --> 01:26:08,162
REACHED OUT TO US AND ASKED IF

2350
01:26:08,229 --> 01:26:11,366
THIS LIVE QA AND MEDICAL QA

2351
01:26:11,432 --> 01:26:13,368
QUESTIONS THAT WE PROVIDED TO

2352
01:26:13,434 --> 01:26:14,435
THE NATIONAL INSTITUTE OF

2353
01:26:14,502 --> 01:26:16,704
STANDARD AND TECHNOLOGY FOR

2354
01:26:16,771 --> 01:26:18,540
THEIR TEXT RETRIEVAL CONFERENCE

2355
01:26:18,606 --> 01:26:19,007
EVALUATIONS.

2356
01:26:19,073 --> 01:26:21,075
IF THESE ARE REAL LIFE

2357
01:26:21,142 --> 01:26:22,243
QUESTIONS, AND WE SAID THAT YES,

2358
01:26:22,310 --> 01:26:28,082
WE TOOK THOSE FROM OUR CUSTOMERS

2359
01:26:28,149 --> 01:26:30,618
PURPOSES AND MED LINE PLUS LOGS

2360
01:26:30,685 --> 01:26:32,453
SO THESE ARE TRULY QUESTIONS

2361
01:26:32,520 --> 01:26:34,556
ASKED BY THE PUBLIC AND THEY

2362
01:26:34,622 --> 01:26:37,058
INCLUDED THESE IN THE BENCHMARK

2363
01:26:37,125 --> 01:26:39,294
AND THE DIFFERENCE BETWEEN THOSE

2364
01:26:39,360 --> 01:26:41,896
SUCCESS STORIES ON THE MEDICAL

2365
01:26:41,963 --> 01:26:43,865
EXAMS, AND THESE QUESTIONS WERE

2366
01:26:43,932 --> 01:26:48,703
THAT ASKED BY THE PUBLIC IS THAT

2367
01:26:48,770 --> 01:26:49,871
MEDICAL EXAM QUESTIONS IN MANY

2368
01:26:49,938 --> 01:26:51,606
CASES HERE ARE MULTIPLE CHOICE,

2369
01:26:51,673 --> 01:26:54,008
SO YOU JUST PICK 1 ANSWER THAT

2370
01:26:54,075 --> 01:26:57,645
IS CORRECT, WHEREAS THESE ARE

2371
01:26:57,712 --> 01:26:58,680
OPEN ENDED QUESTIONS, AND THEY

2372
01:26:58,746 --> 01:27:00,682
DO NOT HAVE EXISTING ANSWERS SO

2373
01:27:00,748 --> 01:27:02,684
TO PROVIDE THE GOLD STANDARD

2374
01:27:02,750 --> 01:27:03,918
ANSWERS, WE ACTUALLY GENERATE

2375
01:27:03,985 --> 01:27:06,387
THE ANSWERS USING SUBJECT MATTER

2376
01:27:06,454 --> 01:27:10,858
EXPERTS, AND THEN EVALUATING

2377
01:27:10,925 --> 01:27:12,226
THESE OPEN ENDED QUESTIONS IS

2378
01:27:12,293 --> 01:27:14,762
ALSO VERY HARD SO THE GOOGLE

2379
01:27:14,829 --> 01:27:17,065
TEAM HAD 9 CLINICIANS FROM

2380
01:27:17,131 --> 01:27:20,468
DIFFERENT COUNTRIES THAT WERE

2381
01:27:20,535 --> 01:27:22,437
ASSESSING THE ANSWERS OF

2382
01:27:22,503 --> 01:27:28,142
INTERWEAVING THE ANSWERS OF

2383
01:27:28,209 --> 01:27:29,410
THE--THAT WERE PRODUCED BY THE

2384
01:27:29,477 --> 01:27:31,779
HUMANS BUT ALSO WERE PRODUCED BY

2385
01:27:31,846 --> 01:27:33,081
THE LARGE LANGUAGE MODELS AND

2386
01:27:33,147 --> 01:27:36,718
WHEN THE MICROSOFT TEAM WAS

2387
01:27:36,784 --> 01:27:37,552
EVALUATING THEIR MEDICAL

2388
01:27:37,619 --> 01:27:39,087
QUESTION ANSWERING SYSTEM ON A

2389
01:27:39,153 --> 01:27:40,321
BENCHMARK, THEY ACTUALLY HAD TO

2390
01:27:40,388 --> 01:27:41,889
LEAVE OUT THESE QUESTIONS

2391
01:27:41,956 --> 01:27:43,458
BECAUSE IT'S JUST VERY EXPENSIVE

2392
01:27:43,524 --> 01:27:51,299
TO DO THESE KINDS OF

2393
01:27:51,366 --> 01:27:51,933
EVALUATIONS.

2394
01:27:52,000 --> 01:27:54,702
SO SWITCHING TO THESE NEW

2395
01:27:54,769 --> 01:27:56,471
TIEWPTS AND CHALLENGES BY THESE

2396
01:27:56,537 --> 01:27:58,406
LLMs, FIRST WE SEE THAT THEY

2397
01:27:58,473 --> 01:28:00,708
CAN DIRECTLY GENERATE THE ANSWER

2398
01:28:00,775 --> 01:28:03,578
BUT WE ALSO SEE THAT THE ANSWER

2399
01:28:03,645 --> 01:28:05,279
NEEDS SOME SUPPORT SO I WILL

2400
01:28:05,346 --> 01:28:07,448
TALK ABOUT SUPPORTING THE ANSWER

2401
01:28:07,515 --> 01:28:09,183
AND GROUNDING IT IN EVIDENT, WE

2402
01:28:09,250 --> 01:28:15,123
ALSO SEE THAT THERE IS A

2403
01:28:15,189 --> 01:28:16,958
MEANINGFUL PLAIN LANGUAGE AND WE

2404
01:28:17,025 --> 01:28:20,795
WILL TALK ABOUT THAT LATER AND

2405
01:28:20,862 --> 01:28:22,030
MULTIMODAL QUESTION ANSWERING.

2406
01:28:22,096 --> 01:28:23,264
SO HERE'S AN EXAMPLE OF WHAT WE

2407
01:28:23,331 --> 01:28:25,033
DO WHEN WE START A NEW TASK AND

2408
01:28:25,099 --> 01:28:27,101
NO DATA EXISTS FOR THAT, WE TAKE

2409
01:28:27,168 --> 01:28:30,738
THE QUESTION THAT WAS ASKED BY

2410
01:28:30,805 --> 01:28:32,907
THE CONSUMER AND A MEDICAL

2411
01:28:32,974 --> 01:28:34,842
STUDENT HAS GENERATED THIS GOLD

2412
01:28:34,909 --> 01:28:38,946
STANDARD AS WELL AS LOOKING AT

2413
01:28:39,013 --> 01:28:39,614
THE LATEST EVIDENCE.

2414
01:28:39,681 --> 01:28:41,416
SO NOW WE ASK THAT SAME QUESTION

2415
01:28:41,482 --> 01:28:45,086
ABOUT THE TOTAL HIP REPLACEMENT

2416
01:28:45,153 --> 01:28:47,488
WHICH THE MEDICAL STUDENT

2417
01:28:47,555 --> 01:28:49,791
INTERPRETED AS DO I NEED PAIN

2418
01:28:49,857 --> 01:28:53,494
MEDICATION TO IMPROVE MY

2419
01:28:53,561 --> 01:28:53,828
RECOVERY?

2420
01:28:53,895 --> 01:28:55,029
WE ASKED THE LARGE LANGUAGE

2421
01:28:55,096 --> 01:29:00,835
MODEL AND THE MODEL PROVIDES A

2422
01:29:00,902 --> 01:29:01,769
VERY LONG ANSWER.

2423
01:29:01,836 --> 01:29:03,004
INTERPRETING THE QUESTION AS,

2424
01:29:03,071 --> 01:29:07,041
HOW MUCH DOES PAIN MEDICATION

2425
01:29:07,108 --> 01:29:07,709
DECREASE--IMPROVE MY PAIN?

2426
01:29:07,775 --> 01:29:09,977
HOW MUCH DOES IT AFFECT THE PAIN

2427
01:29:10,044 --> 01:29:12,613
IN ASK WE ALSO SEE THESE

2428
01:29:12,680 --> 01:29:14,982
PECULIAR THINGS THAT EVERYONE IS

2429
01:29:15,049 --> 01:29:17,652
NOW FAMILIAR WITH, VERY SPECIFIC

2430
01:29:17,719 --> 01:29:18,986
SYSTEMATIC REVIEW OF 19 STUDIES

2431
01:29:19,053 --> 01:29:21,556
THAT IS NOT REFERENCED, AND A

2432
01:29:21,622 --> 01:29:24,592
REFERENCE TO A PAPER IN A

2433
01:29:24,659 --> 01:29:25,626
RELEVANT JOURNAL BUT AGAIN,

2434
01:29:25,693 --> 01:29:28,563
THERE IS NO WAY TO FIND THAT

2435
01:29:28,629 --> 01:29:32,033
PAPER GIVEN THESE ANSWER.

2436
01:29:32,100 --> 01:29:35,236
SO WE LOOK AT MANY OF THESE

2437
01:29:35,303 --> 01:29:37,205
ANSWERS AND IN GENERAL, WE FIND

2438
01:29:37,271 --> 01:29:39,273
THAT THE STRENGTHS OF THESE NEW

2439
01:29:39,340 --> 01:29:41,776
MODELS ARE PRODUCING THESE

2440
01:29:41,843 --> 01:29:43,978
EXTREMELY GRAMMATICAL AND FLUENT

2441
01:29:44,045 --> 01:29:45,379
ANSWERS, SEEMS TO BE

2442
01:29:45,446 --> 01:29:47,315
COMPREHENSIVE AND IN MANY

2443
01:29:47,381 --> 01:29:49,817
ANSWERS, NOT IN THIS 1, WE HAVE

2444
01:29:49,884 --> 01:29:53,054
SEEN THAT AT THE END, THE MODEL

2445
01:29:53,121 --> 01:29:58,159
RECOMMENDS SEEING THE HEALTHCARE

2446
01:29:58,226 --> 01:29:58,426
PROVIDER.

2447
01:29:58,493 --> 01:30:00,061
WE SEE THAT SOMETIMES THE MODEL

2448
01:30:00,128 --> 01:30:02,864
DOES NOT UNDERSTAND THE QUESTION

2449
01:30:02,930 --> 01:30:03,097
FULLY.

2450
01:30:03,164 --> 01:30:08,002
WE ALSO SEE A LOT OF HEDGING IN

2451
01:30:08,069 --> 01:30:08,970
SOME DIFFERENT ANSWERS.

2452
01:30:09,036 --> 01:30:10,304
THE ANSWERS WERE VERY LONG AND

2453
01:30:10,371 --> 01:30:13,407
OF COURSE WE HAVE SEEN THAT IT

2454
01:30:13,474 --> 01:30:16,878
NEEDS FACT CHECKING AND IN THIS

2455
01:30:16,944 --> 01:30:18,112
CASE SPECIFICALLY, WE ASK THE

2456
01:30:18,179 --> 01:30:24,118
MODEL TO PRODUCE THE ANSWER IN

2457
01:30:24,185 --> 01:30:28,055
PLAIN LANGUAGE BUT, THE LANGUAGE

2458
01:30:28,122 --> 01:30:33,761
WAS MORE SUBJECT MATTER EXPERT

2459
01:30:33,828 --> 01:30:34,195
ORIENTED.

2460
01:30:34,262 --> 01:30:36,597
SO CAN WE IMPROVE THAT AND ALSO

2461
01:30:36,664 --> 01:30:40,134
GOING BACK TO THAT IR-BASED

2462
01:30:40,201 --> 01:30:45,640
QUESTION ANSWERING BUT NOW IT'S

2463
01:30:45,706 --> 01:30:47,508
CALLED RETRIEVAL OF

2464
01:30:47,575 --> 01:30:48,309
[INDISCERNIBLE] GENERATION AND

2465
01:30:48,376 --> 01:30:51,712
WHAT HAPPENS HERE IS THE

2466
01:30:51,779 --> 01:30:53,815
QUESTION GOES TO SOME KNOWLEDGE

2467
01:30:53,881 --> 01:30:58,152
BASE IN OUR CASE PUB MED, WE

2468
01:30:58,219 --> 01:30:59,453
RETRIEVE RELEVANT DOCUMS AND WE

2469
01:30:59,520 --> 01:31:00,521
THEN PROVIDE THE MODEL WITH THE

2470
01:31:00,588 --> 01:31:02,390
QUESTION AND THE RELEVANT DOCK

2471
01:31:02,456 --> 01:31:04,892
YOU WANTS AND HERE WE ASK THE

2472
01:31:04,959 --> 01:31:06,627
MODEL TO WRITE AND ANSWER THAT

2473
01:31:06,694 --> 01:31:11,165
IS BASED ON THESE DOCUMENTS, AND

2474
01:31:11,232 --> 01:31:14,569
THEN, INDICATE IN THE ANSWER

2475
01:31:14,635 --> 01:31:15,703
WHICH DOCUMENTS EXACTLY

2476
01:31:15,770 --> 01:31:21,809
CONTRIBUTED TO THAT ANSWER.

2477
01:31:21,876 --> 01:31:24,278
SO IF WE LOOK CLOSER NOW, WE NOW

2478
01:31:24,345 --> 01:31:25,913
CAN EVALUATE WHETHER THE MODEL

2479
01:31:25,980 --> 01:31:30,184
USED THE RIGHT DOCUMENT  TO

2480
01:31:30,251 --> 01:31:34,455
SUPPORT ITS STATEMENT AND ALSO

2481
01:31:34,522 --> 01:31:37,491
WE CAN EVALUATE IF THE DOCUMENT

2482
01:31:37,558 --> 01:31:38,893
INDEED SUPPORTS THE STATEMENT.

2483
01:31:38,960 --> 01:31:44,665
SO HERE YOU SEE THAT IF WE READ

2484
01:31:44,732 --> 01:31:45,700
THE MODELS ANSWER FIRST, WE

2485
01:31:45,766 --> 01:31:47,101
MIGHT THINK THAT IT IS ACTUALLY

2486
01:31:47,168 --> 01:31:52,306
SUPPORTED BY THE DOCUM, SO THE

2487
01:31:52,373 --> 01:31:56,344
DOCUMENT DID SAY THAT A PAIN

2488
01:31:56,410 --> 01:31:57,311
DECREASES CONSTANTLY IN THE

2489
01:31:57,378 --> 01:32:00,014
FIRST WEEK AFTER SURGERY, BUT

2490
01:32:00,081 --> 01:32:04,452
THEN THE MODEL ALSO--THE GOLD

2491
01:32:04,518 --> 01:32:07,588
STANDARD ANSWER ALSO SAYS THAT

2492
01:32:07,655 --> 01:32:09,457
WHEN YOU START PHYSICAL THERAPY,

2493
01:32:09,523 --> 01:32:14,362
THE PAIN LEVELS GO UP WHEREAS

2494
01:32:14,428 --> 01:32:17,632
THE MODEL SAYS, THAT THE PAIN IS

2495
01:32:17,698 --> 01:32:19,533
DECREASING BECAUSE OF THE

2496
01:32:19,600 --> 01:32:20,368
PHYSICAL THERAPY.

2497
01:32:20,434 --> 01:32:21,702
SO BASICALLY WHEN YOU START THE

2498
01:32:21,769 --> 01:32:24,839
PAIN WILL GO DOWN EVEN MORE

2499
01:32:24,906 --> 01:32:28,075
WHICH OF COURSE MISLEAD XG COULD

2500
01:32:28,142 --> 01:32:29,477
POTENTIALLY BE HARMFUL BECAUSE

2501
01:32:29,543 --> 01:32:31,846
NOW THE PATIENT MAY THINK THAT

2502
01:32:31,913 --> 01:32:33,814
SOMETHING IS WRONG WITH PHYSICAL

2503
01:32:33,881 --> 01:32:37,451
THERAPY BECAUSE THE PAIN LEVEL

2504
01:32:37,518 --> 01:32:38,953
THAT WAS GOING UP.

2505
01:32:39,020 --> 01:32:41,822
SO LOOKING AT THESE POTENTIAL

2506
01:32:41,889 --> 01:32:43,057
HARMS THERE WAS SPECIFICALLY A

2507
01:32:43,124 --> 01:32:46,327
PART OF THAT GOOGLE EVALUATION,

2508
01:32:46,394 --> 01:32:48,729
THE MODELS DO PROVIDE HARMFUL

2509
01:32:48,796 --> 01:32:54,802
AND BY SOMEWHAT MORE OFTEN D&  D

2510
01:32:54,869 --> 01:32:56,070
CLINICIANS AND 1 EXAMPLE OF THE

2511
01:32:56,137 --> 01:32:58,172
HARM THAT IS NOT ONLY TO THAT

2512
01:32:58,239 --> 01:32:59,473
INDIVIDUAL PATIENT BUT ALSO TO

2513
01:32:59,540 --> 01:33:02,043
THE ORGANIZATION THAT PROVIDED

2514
01:33:02,109 --> 01:33:06,514
THE ANSWER, IS MAY HAVE LOST

2515
01:33:06,580 --> 01:33:08,649
HERE, THE BRITISH EATING

2516
01:33:08,716 --> 01:33:09,650
DISSEREDDER HELP LINE ANNOUNCED

2517
01:33:09,717 --> 01:33:14,522
THAT THEY ARE REPLACING THEIR

2518
01:33:14,588 --> 01:33:17,224
HELP LINE WITH THE CHAT BOT AND

2519
01:33:17,291 --> 01:33:18,693
ON JUNE 1st OF THAT SAME YEAR,

2520
01:33:18,759 --> 01:33:22,763
SO BASICALLY THE SAME DAY THE

2521
01:33:22,830 --> 01:33:25,399
ANSWERS PROVIDED BY THE CHAT BOT

2522
01:33:25,466 --> 01:33:26,767
WAS SO HARMFUL THAT THE SERVICE

2523
01:33:26,834 --> 01:33:29,603
WENT DOWN AND I CHECKED RECENTLY

2524
01:33:29,670 --> 01:33:30,938
THE IS STILL NOT RESTORED SO

2525
01:33:31,005 --> 01:33:39,480
THAT WAS LIKE A PERMANENT DAMAGE

2526
01:33:39,547 --> 01:33:41,515
TO THAT SERVICE.

2527
01:33:41,582 --> 01:33:43,451
SO WE DEFINITELY NEED TO

2528
01:33:43,517 --> 01:33:46,387
MITIGATE THESE ARMS AND 1 WAY IS

2529
01:33:46,454 --> 01:33:47,521
TO SOMEHOW LEVERAGE THE DOMAIN

2530
01:33:47,588 --> 01:33:49,290
KNOWLEDGE THAT WE NOW HAVE AND

2531
01:33:49,357 --> 01:33:54,362
WE  STARTED WITH LEVERAGING THE

2532
01:33:54,428 --> 01:33:56,364
MLS AND WITH 2 QUESTIONS, 1, HOW

2533
01:33:56,430 --> 01:33:58,132
MUCH OF THE MLS KNOWLEDGE IS

2534
01:33:58,199 --> 01:34:00,868
ALREADY ENCODED IN THE MODEL AND

2535
01:34:00,935 --> 01:34:03,371
IS THE MODEL CAPABLE OF

2536
01:34:03,437 --> 01:34:04,438
RETRIEVING THAT KNOWLEDGE IF

2537
01:34:04,505 --> 01:34:06,741
IT'S THERE, AND THE SECOND PART,

2538
01:34:06,807 --> 01:34:09,010
IF THE KNOWLEDGE IS NOT THERE,

2539
01:34:09,076 --> 01:34:12,146
OR IS ONLY PARTIALLY THERE, CAN

2540
01:34:12,213 --> 01:34:14,248
WE DIRECTLY PROVIDE THAT

2541
01:34:14,315 --> 01:34:16,450
KNOWLEDGE TO THE MODEL TO

2542
01:34:16,517 --> 01:34:18,085
IMPROVE PERFORMANCE AND OUR USE

2543
01:34:18,152 --> 01:34:22,556
CASE IS A DIFFERENTIAL SUPPORT

2544
01:34:22,623 --> 01:34:23,991
OF THE DIFFERENTIAL █DIAGNOSIS

2545
01:34:24,058 --> 01:34:26,260
WHICH IS 1 OF THE OLDEST

2546
01:34:26,327 --> 01:34:27,495
CLINICAL INFORMATIC TAVENS BUT

2547
01:34:27,561 --> 01:34:29,330
IS STILL VERY RELEVANT AND OUR

2548
01:34:29,397 --> 01:34:30,998
COLLABORATOR [INDISCERNIBLE] IS

2549
01:34:31,065 --> 01:34:32,533
WORKING ON THAT PROBLEM.

2550
01:34:32,600 --> 01:34:36,670
SO OUR COMBINED EFFORTS WERE

2551
01:34:36,737 --> 01:34:41,242
FIRST IN PROBING HOW MUCH OF THE

2552
01:34:41,308 --> 01:34:42,543
UMLS KNOWLEDGE IS ALREADY

2553
01:34:42,610 --> 01:34:43,611
CONTAINED IN THE LANGUAGE MODEL

2554
01:34:43,677 --> 01:34:45,479
EMPLOY SO TO THAT END WE ASKED

2555
01:34:45,546 --> 01:34:47,681
THE MODEL TO COMPLETE THE UMLS

2556
01:34:47,748 --> 01:34:48,015
PATH.

2557
01:34:48,082 --> 01:34:53,454
SO WE GIVE THE MODEL A CONCEPT

2558
01:34:53,521 --> 01:34:56,057
FROM THE UMLS, SUCH AS

2559
01:34:56,123 --> 01:35:00,327
ANAPHYLAXIS AND THEN THE EDGES

2560
01:35:00,394 --> 01:35:03,264
THAT ARE RELEVANT TO THE

2561
01:35:03,330 --> 01:35:03,898
DIFFERENTIAL DIAGNOSIS SUPPORT,

2562
01:35:03,964 --> 01:35:07,535
AND WE ASK THE MODEL TO COMPLETE

2563
01:35:07,601 --> 01:35:12,573
THE PATH ALONG THIS EDGE, WITH A

2564
01:35:12,640 --> 01:35:17,511
CONCEPT, WITH ANOTHER CONCEPT,

2565
01:35:17,578 --> 01:35:21,615
AND THE BEST MODEL FOR THAT TASK

2566
01:35:21,682 --> 01:35:23,818
WAS CHAT GPT BUT EVEN THAT MODEL

2567
01:35:23,884 --> 01:35:26,754
FAILED TO COMPLETE 9 MONTHS

2568
01:35:26,821 --> 01:35:28,589
ALTOGETHER AND FOR 22 OF 50

2569
01:35:28,656 --> 01:35:29,957
PATHS IT CAN ONLY PROVIDE LESS

2570
01:35:30,024 --> 01:35:33,627
THAN A THIRD OF THE RELEVANT

2571
01:35:33,694 --> 01:35:35,096
CONCEPTS.

2572
01:35:35,162 --> 01:35:36,597
SO DEFINITELY, EITHER BEING

2573
01:35:36,664 --> 01:35:39,066
MODELED DOES NOT HAVE ALL THE

2574
01:35:39,133 --> 01:35:40,868
KNOWLEDGE OR IT CANNOT RETRIEVE

2575
01:35:40,935 --> 01:35:47,808
IT IN THIS FORMAT.

2576
01:35:47,875 --> 01:35:49,743
SO FOR THE SECOND QUESTION, IF

2577
01:35:49,810 --> 01:35:54,281
WE END THAT KNOWLEDGE EXPLICITLY

2578
01:35:54,348 --> 01:35:57,918
WITH THE DATA THAT WE PROVIDE TO

2579
01:35:57,985 --> 01:36:03,891
THE MODEL ABOUT THE PATIENTS

2580
01:36:03,958 --> 01:36:05,526
SYMPTOMS, AND IF WE WILL LET

2581
01:36:05,593 --> 01:36:10,764
THEM IMPROVE THE RESULTS AND YOU

2582
01:36:10,831 --> 01:36:12,099
CAN SEE HERE THAT IN THE PAST TO

2583
01:36:12,166 --> 01:36:14,768
THE PROMPT AND THE DATA GIVEN TO

2584
01:36:14,835 --> 01:36:16,003
THE MODELS, SOMEWHAT IMPROVES

2585
01:36:16,070 --> 01:36:18,706
THE RESULTS, BUT THE RESULTS ARE

2586
01:36:18,772 --> 01:36:20,975
STILL NOT GREAT, IN THE

2587
01:36:21,041 --> 01:36:23,144
PREDICTIONS OF THE DIFFERENTIAL

2588
01:36:23,210 --> 01:36:24,879
DIAGNOSIS, AND YOU SEE HERE,

2589
01:36:24,945 --> 01:36:32,553
THAT WE USE THESE TRADITIONAL

2590
01:36:32,620 --> 01:36:33,787
LXICAL AND METRICS AND WE WILL

2591
01:36:33,854 --> 01:36:35,089
TALK LATER HOW WE ARE NOW

2592
01:36:35,156 --> 01:36:36,757
EXPERIENCING A BIT OF A CRISIS

2593
01:36:36,824 --> 01:36:38,259
ON USING THESE TRADITIONAL

2594
01:36:38,325 --> 01:36:40,127
METRICS THAT WE HAVE USED FOR

2595
01:36:40,194 --> 01:36:46,967
YEARS IN THESE EVALUATIONS.

2596
01:36:47,034 --> 01:36:50,804
SO SWITCHING TO THE PLAIN

2597
01:36:50,871 --> 01:36:51,872
LANGUAGE ADAPTATIONS, THE

2598
01:36:51,939 --> 01:36:54,475
<u>MITIGATION FOR THAT PROJECT %S 2</u>

2599
01:36:54,542 --> 01:36:55,809
FOLD, 1, EVEN BEFORE THE

2600
01:36:55,876 --> 01:36:58,312
PANDEMIC WE HAD THE PATIENTS WHO

2601
01:36:58,379 --> 01:36:59,847
ALREADY READ EVERYTHING THAT'S

2602
01:36:59,914 --> 01:37:01,749
AVAILABLE IN CONSUMER FRIENDLY

2603
01:37:01,815 --> 01:37:03,784
RESOURCES, ASKING, WHAT IS THE

2604
01:37:03,851 --> 01:37:08,556
LATEST RESEARCH ON MY CONDITION.

2605
01:37:08,622 --> 01:37:12,092
AND WE HAVE SEEN DURING A

2606
01:37:12,159 --> 01:37:16,397
PANDEMIC THAT THE CONSUMER

2607
01:37:16,463 --> 01:37:17,031
FRIENDLY RESOURCES WERE BEHIND

2608
01:37:17,097 --> 01:37:19,867
AND THEY COULD NONAPOPTOTIC THE

2609
01:37:19,934 --> 01:37:20,534
PROVIDE ANY RELEVANT INFORMATION

2610
01:37:20,601 --> 01:37:22,636
TO THE PATIENT SO THERE IS

2611
01:37:22,703 --> 01:37:25,506
CLEARLY A NEED TO PROVIDE PLAIN

2612
01:37:25,573 --> 01:37:27,775
LANGUAGE ANSWERS IN REALTIME,

2613
01:37:27,841 --> 01:37:29,677
THERE ARE 2 ONGOING EFFORTS

2614
01:37:29,743 --> 01:37:35,816
RIGHT NOW WITHIN THE COMMUNITY,

2615
01:37:35,883 --> 01:37:39,620
1 IS DIRECTLY SUMMARIZE THE

2616
01:37:39,687 --> 01:37:40,221
RESEARCH EVIDENT INTO THE

2617
01:37:40,287 --> 01:37:42,256
LANGUAGE THAT ANSWER FOR A

2618
01:37:42,323 --> 01:37:43,791
SPECIFIC QUESTION, WE CHOSE

2619
01:37:43,857 --> 01:37:46,160
ANOTHER WAY BECAUSE THIS WAY, IF

2620
01:37:46,227 --> 01:37:50,531
WE ARE FAITHFULLY TRANSLATING

2621
01:37:50,598 --> 01:37:51,832
THE SCIENTIFIC ABSTRACT INTO A

2622
01:37:51,899 --> 01:37:53,334
PLAIN LANGUAGE, THEN WE CAN USE

2623
01:37:53,400 --> 01:37:56,670
IT FOR DIFFERENT PURPOSES AND

2624
01:37:56,737 --> 01:38:00,140
ANSWER DIFFERENT QUESTIONS

2625
01:38:00,207 --> 01:38:00,441
POTENTIALLY.

2626
01:38:00,507 --> 01:38:04,778
SO THIS IS AN ONGOING EFFORT

2627
01:38:04,845 --> 01:38:06,180
WITHIN THE TRUCK EVALUATIONS,

2628
01:38:06,247 --> 01:38:09,016
IT'S IN THE SECOND YEAR AND

2629
01:38:09,083 --> 01:38:10,985
BRIAN [INDISCERNIBLE] WHO IS

2630
01:38:11,051 --> 01:38:12,553
PRESENTING IN THE PRIVATE

2631
01:38:12,620 --> 01:38:14,121
SESSION WILL TALK MORE ABOUT

2632
01:38:14,188 --> 01:38:14,588
THIS EFFORT.

2633
01:38:14,655 --> 01:38:18,025
AS WITH ALL OF OUR NEW TASKS,

2634
01:38:18,092 --> 01:38:22,496
THERE WERE NO CORRECTIONS THAT

2635
01:38:22,563 --> 01:38:25,499
HAD THESE SYMPTOMS BY SYMPTOMS

2636
01:38:25,566 --> 01:38:26,634
FOR PUB MED ABNORMALITIES TRACKS

2637
01:38:26,700 --> 01:38:28,736
SO HERE I'M SHOWING HOW WE

2638
01:38:28,802 --> 01:38:29,436
CREATED THE COLLECTION.

2639
01:38:29,503 --> 01:38:34,008
WE STARTED WITH THE POPULAR MED

2640
01:38:34,074 --> 01:38:35,142
LINE PLUS SEARCHES; WE SEARCHED

2641
01:38:35,209 --> 01:38:38,812
PUB MED FOR THE RELEVANT

2642
01:38:38,879 --> 01:38:40,614
ABSTRACTS AND THEN FOR EACH 1 OF

2643
01:38:40,681 --> 01:38:45,819
THOSE  ABSTRACTS, WE TRANSLATED

2644
01:38:45,886 --> 01:38:49,590
THESE SENTENCES INTO PLAIN

2645
01:38:49,657 --> 01:38:50,024
LANGUAGE.

2646
01:38:50,090 --> 01:38:52,960
USING THIS COLLECTION, WE THEN

2647
01:38:53,027 --> 01:38:56,530
HAD LAST YEAR, THIS COMMUNITY

2648
01:38:56,597 --> 01:39:00,000
EVALUATION AND I SHOW HERE THE

2649
01:39:00,067 --> 01:39:01,302
RESULTS OF THE LATEST NARRATION

2650
01:39:01,368 --> 01:39:03,170
WE HAVE DONE ON COLLECTION AND

2651
01:39:03,237 --> 01:39:08,042
THE TOP YOU SEE, THE RESULTS

2652
01:39:08,108 --> 01:39:09,710
SUBMITTED BY THE TEAMS THAT

2653
01:39:09,777 --> 01:39:10,344
PARTICIPATED IN THE CHALLENGE

2654
01:39:10,411 --> 01:39:12,413
AND IN THE BOTTOM, YOU SEE THE

2655
01:39:12,479 --> 01:39:14,748
NLM BASE LINE THAT WAS PRODUCED

2656
01:39:14,815 --> 01:39:17,618
BY BRIAN USING CHAT GPT AND I

2657
01:39:17,685 --> 01:39:19,620
BELIEVE HE IS PROVIDING THE

2658
01:39:19,687 --> 01:39:20,988
DETAILS ON HOW HE HAS DONE IT

2659
01:39:21,055 --> 01:39:26,026
AND THEN AT THE BOTTOM, YOU SEE

2660
01:39:26,093 --> 01:39:29,063
THE IMITATORS THAT WE USED FOR

2661
01:39:29,129 --> 01:39:29,963
HUMAN ADAPTATIONS, HUMAN

2662
01:39:30,030 --> 01:39:30,831
TRANSLATIONS OF THESE SENTENCES

2663
01:39:30,898 --> 01:39:35,536
AND YOU CAN SEE HERE THE MACHINE

2664
01:39:35,602 --> 01:39:37,705
IS DEFINITELY MORE FLUENT THAN

2665
01:39:37,771 --> 01:39:38,505
PEOPLE.

2666
01:39:38,572 --> 01:39:40,140
SO FLUENCY PROBLEM IS DEFINITELY

2667
01:39:40,207 --> 01:39:43,143
SOLVED AND 1 OF THE PROBLEMS

2668
01:39:43,210 --> 01:39:44,378
WITH THESE TRADITIONAL METRICS

2669
01:39:44,445 --> 01:39:48,916
IS THAT IT IS MEASURING FLUENCY.

2670
01:39:48,982 --> 01:39:56,156
NOW, WHEN IT COMES TO

2671
01:39:56,223 --> 01:39:57,524
FAITHFULNESS AND ACCURACY OF THE

2672
01:39:57,591 --> 01:40:00,361
ANSWERS, THEN AGAIN, SOME OF THE

2673
01:40:00,427 --> 01:40:04,498
APPROACH THAT BRIAN AS

2674
01:40:04,565 --> 01:40:05,699
DEVELOPED, THE NUMBERS LOOK VERY

2675
01:40:05,766 --> 01:40:06,967
GOOD BUT AGAIN, THEY ARE HIDING

2676
01:40:07,034 --> 01:40:09,603
SOMETHING THAT I SHOWED EARLIER

2677
01:40:09,670 --> 01:40:16,343
SO 1 OF THOSE ANSWERS COULD

2678
01:40:16,410 --> 01:40:17,111
POTENTIALLY BE HARMFUL.

2679
01:40:17,177 --> 01:40:21,281
NOW SWITCHING TO THE MULTIMODAL

2680
01:40:21,348 --> 01:40:22,116
QUESTION ANSWERING, THERE ARE

2681
01:40:22,182 --> 01:40:23,884
MANY CASES WHERE IT'S BETTER TO

2682
01:40:23,951 --> 01:40:26,720
SEE SOMETHING THAN TO READ A

2683
01:40:26,787 --> 01:40:28,455
LONG TEXTURAL DESCRIPTION, FOR

2684
01:40:28,522 --> 01:40:32,359
EXAMPLE, IF SOMEONENIES TO GIVE

2685
01:40:32,426 --> 01:40:35,129
FIRST AID, OR A

2686
01:40:35,195 --> 01:40:36,063
PATIENT'S--PATIENTS ARE

2687
01:40:36,130 --> 01:40:41,668
INTRUCTED TO DO

2688
01:40:41,735 --> 01:40:42,269
SELF-EXAMINATIONS ORIFICEICAL

2689
01:40:42,336 --> 01:40:43,804
THERAPY, SO THE GOAL IS TO SHOW

2690
01:40:43,871 --> 01:40:45,539
THEM HOW TO DO THESE THINGS AND

2691
01:40:45,606 --> 01:40:46,840
TO ACCOMPLISH THAT, WE FIRST

2692
01:40:46,907 --> 01:40:50,811
NEED TO FIND RELEVANT VIDEOS

2693
01:40:50,878 --> 01:40:52,179
THAT ACTUALLY CONTAIN INTRUKSS

2694
01:40:52,246 --> 01:40:54,248
AND THEN WITHIN THOSE VIDEOS WE

2695
01:40:54,314 --> 01:40:55,816
NEED TO FIND THE SNIPITS THAT

2696
01:40:55,883 --> 01:40:59,920
SHOW HOW TO DO SOMETHING, AND

2697
01:40:59,987 --> 01:41:04,992
WITHIN THESE SNIPITS, THE

2698
01:41:05,058 --> 01:41:07,327
PROCEDURES SOMETIMES COMBINE OUT

2699
01:41:07,394 --> 01:41:10,764
OF MANY STEPS.

2700
01:41:10,831 --> 01:41:13,000
AND OUR GOAL, FOR EXAMPLE, IF

2701
01:41:13,066 --> 01:41:15,169
SOMEONE SAYS HOW DO I STAND UP,

2702
01:41:15,235 --> 01:41:16,904
WALK AND SIT DOWN USING

2703
01:41:16,970 --> 01:41:17,504
CRUTCHES?

2704
01:41:17,571 --> 01:41:18,872
AND FOR EXAMPLE, THEY JUST WANT

2705
01:41:18,939 --> 01:41:21,208
TO KNOW HOW TO SIT DOWN, THEN WE

2706
01:41:21,275 --> 01:41:24,378
WANT TO IESHES DENTIFY THAT STEP

2707
01:41:24,445 --> 01:41:26,113
IN THE VIDEO AND WILL LABEL IT

2708
01:41:26,180 --> 01:41:27,147
AS SUCH.

2709
01:41:27,214 --> 01:41:31,018
SO THIS IS WHAT WE ARE WORKING

2710
01:41:31,084 --> 01:41:32,719
ON RIGHT NOW, AND THIS PAPER

2711
01:41:32,786 --> 01:41:34,822
THAT I'M PRESENTING WILL BE

2712
01:41:34,888 --> 01:41:40,294
PRESENTED NEXT MONTH AT THE

2713
01:41:40,360 --> 01:41:40,861
COMPUTATIONAL LINGUISTICS

2714
01:41:40,928 --> 01:41:42,062
CONFERENCE.

2715
01:41:42,129 --> 01:41:43,897
AS WITH ALL OTHER TASKS, THE

2716
01:41:43,964 --> 01:41:46,366
DATA SAID THAT WE CREATED

2717
01:41:46,433 --> 01:41:48,001
MANUALLY IN THE FIRST YEAR WAS

2718
01:41:48,068 --> 01:41:54,942
FAIRLY SMALL, SO WE LOOKED INTO

2719
01:41:55,008 --> 01:42:00,047
CREATING DATA SET FOR TRAINING

2720
01:42:00,113 --> 01:42:00,380
AUTOMATICALLY.

2721
01:42:00,447 --> 01:42:04,351
AND WE STARTED WITH THESE FILES

2722
01:42:04,418 --> 01:42:07,287
THAT WE CLASSIFIED AS MEDICAL

2723
01:42:07,354 --> 01:42:10,524
INTRUKSES AND WE FOUND MORE OF

2724
01:42:10,591 --> 01:42:14,161
THESE IN THIS YouTUBE

2725
01:42:14,228 --> 01:42:16,163
COLLECTIONS THEN WE SEGMENTED

2726
01:42:16,230 --> 01:42:18,565
THE INSTRUCTIONAL VIDEOS INTO

2727
01:42:18,632 --> 01:42:21,401
THESE TOPICS FOR EXAMPLE,

2728
01:42:21,468 --> 01:42:24,905
WALKING WITH CRUTCHES AND THEN,

2729
01:42:24,972 --> 01:42:29,676
EACH SEGMENT WAS REPRESENTED

2730
01:42:29,743 --> 01:42:31,912
USING TRANSFORMER ENCODER.

2731
01:42:31,979 --> 01:42:34,081
AND WE WILL--THIS EVIDENT WAS

2732
01:42:34,147 --> 01:42:35,215
LABELED AS BEGINNING FOR INSIDE

2733
01:42:35,282 --> 01:42:38,385
OR OUTSIDE OF THE ANSWER AND

2734
01:42:38,452 --> 01:42:40,754
THEN ON TOP OF THESE SEGMENTS

2735
01:42:40,821 --> 01:42:42,523
NOW THAT ARE THE ANSWER

2736
01:42:42,589 --> 01:42:45,826
SEGMENTS, THE QUESTION IS

2737
01:42:45,893 --> 01:42:47,961
GENERATED THAT COULD BE

2738
01:42:48,028 --> 01:42:53,367
POTENTIALLY ANSWERED BY THAT

2739
01:42:53,433 --> 01:42:54,902
SEGMENT.

2740
01:42:54,968 --> 01:42:55,936
SO TO [INDISCERNIBLE] THAT

2741
01:42:56,003 --> 01:43:00,807
COLLECTION HAS IMPROVED OUR

2742
01:43:00,874 --> 01:43:07,481
PERFORMANCE ON THAT VIDEO,

2743
01:43:07,548 --> 01:43:09,516
ANSWERED [INDISCERNIBLE] WE

2744
01:43:09,583 --> 01:43:11,251
DEVELOPED AN APPROACH THAT

2745
01:43:11,318 --> 01:43:14,922
LEVERAGES THE VIDEO FEATURES AND

2746
01:43:14,988 --> 01:43:16,623
THE CAPTIONS FEATURES AND THE

2747
01:43:16,690 --> 01:43:23,363
TIME STAMPS THIS IS CALLED THE

2748
01:43:23,430 --> 01:43:24,298
CYCLE CONSISTENT ANSWER

2749
01:43:24,364 --> 01:43:27,901
LOCALIZATION AND THE DIFFERENT

2750
01:43:27,968 --> 01:43:30,103
ANALOGIES, THE WAY IT IS, THE

2751
01:43:30,170 --> 01:43:32,272
GOLD STANDARD QUESTION SYSTEM

2752
01:43:32,339 --> 01:43:34,107
USED TO GENERATE AN ANSWER, THAT

2753
01:43:34,174 --> 01:43:38,211
GENERATED ANSWER IS FED INTO A

2754
01:43:38,278 --> 01:43:39,546
MODEL THAT INVOLVES ANOTHER

2755
01:43:39,613 --> 01:43:41,348
QUESTION WITH THAT WORRY AND

2756
01:43:41,415 --> 01:43:45,085
THAT OTHER MODEL IS GENERATING

2757
01:43:45,152 --> 01:43:47,220
THE ANSWER TO THE GENERATED

2758
01:43:47,287 --> 01:43:48,488
QUESTION AND DURING TRAINING ALL

2759
01:43:48,555 --> 01:43:53,827
OF THESE STEPS NEED TO BE

2760
01:43:53,894 --> 01:43:54,127
COORDINATED.

2761
01:43:54,194 --> 01:44:00,867
SO USING THIS MODEL AND USING

2762
01:44:00,934 --> 01:44:03,971
THESE ADDITIONAL DATA, WE CAN

2763
01:44:04,037 --> 01:44:05,739
SEE THAT ADDING THESE NEW DAILY

2764
01:44:05,806 --> 01:44:07,841
BASIS THEA, THE COLLECTION WAS

2765
01:44:07,908 --> 01:44:08,875
BUILT RIGHT, IT IMPROVES

2766
01:44:08,942 --> 01:44:09,476
PERFORMANCE SIGNIFICANTLY AND

2767
01:44:09,543 --> 01:44:12,412
THE METRIC HERE IS THE

2768
01:44:12,479 --> 01:44:14,581
INTERSECTION OF THE ANSWER

2769
01:44:14,648 --> 01:44:16,350
GENERATED BY THE MODEL WITH THE

2770
01:44:16,416 --> 01:44:20,554
GOLD STANDARD ANSWER, AND

2771
01:44:20,621 --> 01:44:22,122
THERE'S DIFFERENT BOUNDARIES ON

2772
01:44:22,189 --> 01:44:24,591
HOW MUCH OF INTERSECTION THERE

2773
01:44:24,658 --> 01:44:27,227
SHOULD BE SO THAT WE SEE THE

2774
01:44:27,294 --> 01:44:29,262
DIFFERENT NUMBERS FOR THESE

2775
01:44:29,329 --> 01:44:30,731
DEFINITE BOUNDARIES WITHIN THE

2776
01:44:30,797 --> 01:44:35,168
ANSWER AND OF COURSE, THE MODELS

2777
01:44:35,235 --> 01:44:37,037
TILL MAKE MISTAKES, AND SOME OF

2778
01:44:37,104 --> 01:44:40,841
THESE MISTAKES COULD BE

2779
01:44:40,907 --> 01:44:44,378
ADDRESSED WHEN WE ADD THE VISUAL

2780
01:44:44,444 --> 01:44:45,245
FEATURES FROM THE VIDEO FRAMES

2781
01:44:45,312 --> 01:44:53,654
TO THE TEXT, WE SEE THAT

2782
01:44:53,720 --> 01:44:54,688
PERFORMANCE GOES UP.

2783
01:44:54,755 --> 01:44:56,323
SO EVALUATION IS NOW A VERY OPEN

2784
01:44:56,390 --> 01:44:59,159
QUESTION BECAUSE THESE

2785
01:44:59,226 --> 01:45:02,629
TRADITIONAL LEXICAL MET  METRICT

2786
01:45:02,696 --> 01:45:05,732
WE SEE AND THESE METRICS SHOW

2787
01:45:05,799 --> 01:45:07,801
VERY GOOD RESULTS BUT WITHIN

2788
01:45:07,868 --> 01:45:10,771
THOSE RESULTS ARE THESE RARE

2789
01:45:10,837 --> 01:45:13,974
OCCASIONS, YOU KNOW, 5%, WHICH

2790
01:45:14,041 --> 01:45:16,610
COULD POTENTIALLY BE HARMFUL, SO

2791
01:45:16,677 --> 01:45:20,781
WE'RE VERY ACTIVELY WORKING WITH

2792
01:45:20,847 --> 01:45:23,450
THE NIST GROUP THAT CONDUCTS

2793
01:45:23,517 --> 01:45:25,519
THESE TEXT RETRIEVAL CONFERENCE

2794
01:45:25,585 --> 01:45:27,888
EVALUATIONS ON DEVELOPING NEW

2795
01:45:27,954 --> 01:45:29,389
METRICS AND OF COURSE THAT

2796
01:45:29,456 --> 01:45:30,724
ENROLLS A LOT OF MANUAL

2797
01:45:30,791 --> 01:45:32,959
JUDGMENTS ON THE WAY SO THAT WE

2798
01:45:33,026 --> 01:45:34,594
CAN COMPARE THE DIFFERENT

2799
01:45:34,661 --> 01:45:43,837
METRICS TO THE MANUEL JUDGMENTS.

2800
01:45:43,904 --> 01:45:45,305
THE INTERFACE I SHOWED YOU AT

2801
01:45:45,372 --> 01:45:47,941
THE BEGINNING IS NOW READY TO BE

2802
01:45:48,008 --> 01:45:50,744
REPLACED BY THE NEW INTERFACE

2803
01:45:50,811 --> 01:45:53,847
FOR THAT PRIVATE TYPE CONSUMER

2804
01:45:53,914 --> 01:45:56,016
HEALTH QUESTION ANSWERING

2805
01:45:56,083 --> 01:45:57,551
SYSTEM, YOU CAN SEE AT THE TOP

2806
01:45:57,617 --> 01:45:58,985
THE ANSWER IS GENERATED BY THE

2807
01:45:59,052 --> 01:46:00,721
MODEL AND THEN THE EVIDENCE TO

2808
01:46:00,787 --> 01:46:03,890
SUPPORT THE ANSWER, AND THE

2809
01:46:03,957 --> 01:46:08,028
VIDEOS THAT WE PULL IN FROM THE

2810
01:46:08,095 --> 01:46:10,831
OPEN EYE COLLECTION VIDEOS AND

2811
01:46:10,897 --> 01:46:11,064
IMAGES.

2812
01:46:11,131 --> 01:46:12,799
NOW TECHNICALLY IT'S FAST AND

2813
01:46:12,866 --> 01:46:15,469
READY TO BE PRESENTED BUT I

2814
01:46:15,535 --> 01:46:18,071
DON'T BELIEVE THAT WE ARE READY

2815
01:46:18,138 --> 01:46:19,406
TO HAND IT OVER RIGHT NOW

2816
01:46:19,473 --> 01:46:23,577
BECAUSE WE STILL NEED TO DO LOTS

2817
01:46:23,643 --> 01:46:28,081
OF USER EVALUATIONS, IS THAT

2818
01:46:28,148 --> 01:46:29,015
SOMETHING--IS THAT ANOTHER WAY

2819
01:46:29,082 --> 01:46:30,317
TO PRESENT SOMETHING TO THE USER

2820
01:46:30,383 --> 01:46:34,187
SO THAT THEY WILL BE ABLE TO

2821
01:46:34,254 --> 01:46:35,355
DETECT DISCREPANCIES AND JUDGE

2822
01:46:35,422 --> 01:46:36,723
WHAT'S WRONG WITH THE ANSWER, OR

2823
01:46:36,790 --> 01:46:40,627
IF THE ANSWER GOOD AND

2824
01:46:40,694 --> 01:46:40,894
SUPPORTED.

2825
01:46:40,961 --> 01:46:43,897
AND THAT IS GOING TO THAT FUTURE

2826
01:46:43,964 --> 01:46:47,934
WORK WHERE WE NEED TO VERY

2827
01:46:48,001 --> 01:46:48,502
[INDISCERNIBLE] COMPLETE AND

2828
01:46:48,568 --> 01:46:50,670
STRENGTH OF THE ANSWERS, EXPLORE

2829
01:46:50,737 --> 01:46:51,404
QUESTIONS THAT NEED

2830
01:46:51,471 --> 01:46:52,806
CLARIFICATION FOR EXAMPLE, THE

2831
01:46:52,873 --> 01:46:55,509
QUESTION THAT THE MODEL AND THE

2832
01:46:55,575 --> 01:46:57,878
MEDICAL STUDENT INTERPRETED

2833
01:46:57,944 --> 01:46:59,045
DIFFERENTLY, WE COULD TURN ARK

2834
01:46:59,112 --> 01:47:02,149
ROUND AND ASK THE USER IF THEY

2835
01:47:02,215 --> 01:47:04,818
REALLY WANT TO KNOW WHICH PAIN

2836
01:47:04,885 --> 01:47:06,052
MEDICATIONS ARE BETTER OR IF

2837
01:47:06,119 --> 01:47:08,922
THEY WANT TO KNOW HOW MUCH THEIR

2838
01:47:08,989 --> 01:47:13,693
RECOVERY WILL BE INFLUENCED BY

2839
01:47:13,760 --> 01:47:15,462
THE--BY PAIN MANAGEMENT AND THE

2840
01:47:15,529 --> 01:47:18,732
CONTINUING WORK ON THAT VISUAL

2841
01:47:18,799 --> 01:47:24,004
MODEL QUESTION ANSWERING, YOU

2842
01:47:24,070 --> 01:47:26,039
WILL ALSO SEE [INDISCERNIBLE]

2843
01:47:26,106 --> 01:47:31,344
PRESENTING THE WORK HE STARTED

2844
01:47:31,411 --> 01:47:32,646
ON EHR MODELING AND SOMETHING IT

2845
01:47:32,712 --> 01:47:36,316
MORE THAT THE CLINICIANS WANT TO

2846
01:47:36,383 --> 01:47:38,485
KNOW ABOUT THEIR CARE PROCESS

2847
01:47:38,552 --> 01:47:41,388
AND ALSO DEVELOPING RELIABLE AND

2848
01:47:41,454 --> 01:47:42,322
AFFORDABLE EVALUATION APPROACHES

2849
01:47:42,389 --> 01:47:46,827
AND AGAIN, MOVING CLOSER TO

2850
01:47:46,893 --> 01:47:50,597
TRANSFERRING THE TECHNOLOGY TO

2851
01:47:50,664 --> 01:47:52,165
OUR OTHER DIVISIONS, FOR

2852
01:47:52,232 --> 01:47:54,701
EXAMPLE, IN WHAT I'VE SHOWN

2853
01:47:54,768 --> 01:47:56,336
TODAY THAT FINDING THAT DANCE

2854
01:47:56,403 --> 01:48:00,307
RETRIEVAL, THAT WE USED FOR

2855
01:48:00,373 --> 01:48:01,975
FINDING RELEVANT VIDEOS IS READY

2856
01:48:02,042 --> 01:48:04,845
FOR TRANSFER, IT'S HARMLESS, THE

2857
01:48:04,911 --> 01:48:09,282
RESULTS ARE MUCH BETTER, BUT OF

2858
01:48:09,349 --> 01:48:10,984
COURSE, THE QUESTION ANSWERING

2859
01:48:11,051 --> 01:48:12,786
SYSTEM ITSELF IS NOT READY FOR

2860
01:48:12,853 --> 01:48:13,420
TRANSFER YET.

2861
01:48:13,486 --> 01:48:18,892
SO WITH THAT, OF COURSE, IT'S A

2862
01:48:18,959 --> 01:48:20,694
GROUP EFFORT ON THE TRAINEES

2863
01:48:20,760 --> 01:48:23,964
WILL BE PRESENTING THEIR WORK

2864
01:48:24,030 --> 01:48:31,905
RIGHT AFTER THIS SESSION BRIAN

2865
01:48:31,972 --> 01:48:33,340
AND SARVESH, THE CURRENT GROUP

2866
01:48:33,406 --> 01:48:34,040
CONTRIBUTES ARE LISTED ON THE

2867
01:48:34,107 --> 01:48:35,876
LEFT AND ON THE RIGHT THESE ARE

2868
01:48:35,942 --> 01:48:37,077
ALL THE FORMER TRAINEES THAT

2869
01:48:37,143 --> 01:48:39,012
CONTRIBUTED OVER THE YEARS TO

2870
01:48:39,079 --> 01:48:41,915
THE PROJECT AND A LOT OF WHAT

2871
01:48:41,982 --> 01:48:44,117
THEY HAVE KRIEWBED IS TILL BEING

2872
01:48:44,184 --> 01:48:48,955
USED FOR EXAMPLE, KATE MASTER

2873
01:48:49,022 --> 01:48:50,457
TON HAS DEVELOPED THESE REPORTS

2874
01:48:50,523 --> 01:48:53,093
OUT OF THE NETHERLANDS,

2875
01:48:53,159 --> 01:48:54,828
[INDISCERNIBLE] TO GET THE

2876
01:48:54,895 --> 01:48:57,230
LATEST QUESTIONS AND WE ARE

2877
01:48:57,297 --> 01:49:00,133
STILL USING THIS SERVICE

2878
01:49:00,200 --> 01:49:01,701
DEVELOPED BY KATE SOME YEARS AGO

2879
01:49:01,768 --> 01:49:12,312
AND SO THANK YOU AND I HOPE YOU

2880
01:49:14,881 --> 01:49:16,283
HAVE QUESTIONS.

2881
01:49:16,349 --> 01:49:18,685
NIF YOU CAN CAN YOU HIGHLIGHT

2882
01:49:18,752 --> 01:49:19,886
THE QUESTION FOR SOMEONE OUTSIDE

2883
01:49:19,953 --> 01:49:21,488
OF FIELD, WHAT IS

2884
01:49:21,554 --> 01:49:24,457
STATE-OF-THE-ART, WHAT IS NEW

2885
01:49:24,524 --> 01:49:26,059
AND WHAT IS KIND OF COMMON THING

2886
01:49:26,126 --> 01:49:35,435
BECAUSE IT'S VERY HARD FOR ME TO

2887
01:49:35,502 --> 01:49:35,635
JUDGE.

2888
01:49:35,702 --> 01:49:37,904
>> SO WHAT I'VE SHOWN HERE IS

2889
01:49:37,971 --> 01:49:43,476
COMPLETELY NEW, SO, YOU KNOW,

2890
01:49:43,543 --> 01:49:45,912
LIKE THIS MAGICAL VIDEO

2891
01:49:45,979 --> 01:49:46,980
INSTRUCTIONAL VIDEO QUESTION

2892
01:49:47,047 --> 01:49:49,683
ANSWERING, WE STARTED IT 2 YEARS

2893
01:49:49,749 --> 01:49:49,950
AGO.

2894
01:49:50,016 --> 01:49:52,485
THIS IS A BRAND NEW EFFORT.

2895
01:49:52,552 --> 01:49:54,454
THERE ARE NOW--WE HAVE SEEN--SO

2896
01:49:54,521 --> 01:49:58,892
THE FIRST YEAR, WE HAD 3 TEAMS

2897
01:49:58,959 --> 01:49:59,960
PARTICIPATING, THE SECOND YEAR,

2898
01:50:00,026 --> 01:50:03,563
WE HAD I THINK 5 OR 7, NOW WE

2899
01:50:03,630 --> 01:50:05,799
SEE THAT 8 TEAMS HAVE SIGNED UP

2900
01:50:05,865 --> 01:50:10,437
TO PARTICIPATE THIS YEAR.

2901
01:50:10,503 --> 01:50:11,838
THIS RETRIEVAL OF THE

2902
01:50:11,905 --> 01:50:15,809
GENERATIONAL ANSWERS IS BRAND

2903
01:50:15,875 --> 01:50:16,876
NEW AS WELL.

2904
01:50:16,943 --> 01:50:19,179
THE [INDISCERNIBLE] WORK I'VE

2905
01:50:19,245 --> 01:50:21,247
SHOWN PROBING THE UMLS

2906
01:50:21,314 --> 01:50:24,351
KNOWLEDGE, WE JUST SUBMITTED THE

2907
01:50:24,417 --> 01:50:26,486
PAPER TO JBI, SO IT'S UNDER

2908
01:50:26,553 --> 01:50:28,355
REVIEW FROM THE JBI, SO

2909
01:50:28,421 --> 01:50:29,155
EVERYTHING I'VE SHOWN TODAY

2910
01:50:29,222 --> 01:50:30,690
EXCEPT FOR THE VERY FIRST SLIDE

2911
01:50:30,757 --> 01:50:34,694
IS THE OLD SYSTEM, EVERYTHING

2912
01:50:34,761 --> 01:50:42,168
ELSE IS NEW.

2913
01:50:42,235 --> 01:50:42,736
NTHANK YOU.

2914
01:50:42,802 --> 01:50:44,504
>> I THINK I WAS NEXT.

2915
01:50:44,571 --> 01:50:46,239
SOIME VERY INTERESTED IN THIS

2916
01:50:46,306 --> 01:50:47,640
CONCEPT FOR POTENTIAL FOR HARM

2917
01:50:47,707 --> 01:50:49,376
THAT YOU MENTIONED, I FOUND IT

2918
01:50:49,442 --> 01:50:51,011
QUITE STRIKING FOR EXAMPLE, THE

2919
01:50:51,077 --> 01:50:52,178
SLIDE WHERE YOU HAD THE TABLE

2920
01:50:52,245 --> 01:50:54,280
AND YOU COULD HAVE 95% ACCURACY

2921
01:50:54,347 --> 01:50:56,850
BUT THERE COULD STILL BE 5%

2922
01:50:56,916 --> 01:50:58,485
POTENTIAL FOR HARM, SO I WAS

2923
01:50:58,551 --> 01:51:00,387
WONDERING, YOU KNOW IF YOU COULD

2924
01:51:00,453 --> 01:51:01,421
ELABORATE ON THAT A LITTLE BIT,

2925
01:51:01,488 --> 01:51:03,390
TELL US MORE ABOUT WHAT'S MEANT

2926
01:51:03,456 --> 01:51:04,591
BY POTENTIAL FOR HARM AND IN

2927
01:51:04,657 --> 01:51:07,794
PARTICULAR, HOW IS THAT DECIDED?

2928
01:51:07,861 --> 01:51:11,998
WHAT'S THE THRESHOLD THERE?

2929
01:51:12,065 --> 01:51:14,501
THAT'S A VERY STICKY QUESTION,

2930
01:51:14,567 --> 01:51:16,369
SO GOOGLE WAS BRAVE ENOUGH TO

2931
01:51:16,436 --> 01:51:19,039
SAY, WE WILL EVALUATE THAT IT BE

2932
01:51:19,105 --> 01:51:19,672
BRING HARM.

2933
01:51:19,739 --> 01:51:21,741
NO 1 ELSE WANTS TO TOUCH IT SO

2934
01:51:21,808 --> 01:51:23,943
IN OUR PLAIN LANGUAGE

2935
01:51:24,010 --> 01:51:27,013
ADAPTATIONS, WE'RE JUST STOPPING

2936
01:51:27,080 --> 01:51:33,620
AT THIS IS EXACTLY WHAT THE

2937
01:51:33,686 --> 01:51:35,288
ORIGINAL SAID BECAUSE YOU KNOW

2938
01:51:35,355 --> 01:51:40,560
THESE TRAINED CLINICIANS THAT

2939
01:51:40,627 --> 01:51:42,862
WE'RE DOING THESE

2940
01:51:42,929 --> 01:51:43,963
MULTI[INDISCERNIBLE] JUDGMENTS

2941
01:51:44,030 --> 01:51:45,532
WITH THE [INDISCERNIBLE]

2942
01:51:45,598 --> 01:51:46,766
AGREEMENT, WE ARE ABLE TO JUDGE

2943
01:51:46,833 --> 01:51:48,368
THAT THAT MIGHT BE POTENTIALLY

2944
01:51:48,435 --> 01:51:49,602
HARMFUL TO THE PATIENT, RIGHT IN

2945
01:51:49,669 --> 01:51:53,139
SO AS I SAID, WITH THIS ANSWER,

2946
01:51:53,206 --> 01:51:54,407
HOW HARMFUL IS IT IF I EXPECT

2947
01:51:54,474 --> 01:51:58,078
THAT MY PAIN WILL GO DOWN BUT IT

2948
01:51:58,144 --> 01:52:01,281
WILL GO UP INSTEAD, RIGHT?

2949
01:52:01,347 --> 01:52:03,516
SO IT MIGHT BE FOR DIFFERENT

2950
01:52:03,583 --> 01:52:05,752
PEOPLE, THAT MIGHT BE DEFINITE

2951
01:52:05,819 --> 01:52:11,758
RANGE OF HARMFULNESS, SO THIS IS

2952
01:52:11,825 --> 01:52:13,927
A VERY SUBJECTIVE WITH THE

2953
01:52:13,993 --> 01:52:17,630
EXCEPTION OF IF YOU KNOW

2954
01:52:17,697 --> 01:52:18,198
RECOMMENDS CRINGING BLEACH

2955
01:52:18,264 --> 01:52:20,266
ONLINE AND THEN PEOPLE GO AND

2956
01:52:20,333 --> 01:52:21,701
DRINK BLEACH, THAT'S REAL HARM

2957
01:52:21,768 --> 01:52:24,070
BUT THERE'S ALSO THESE POTENTIAL

2958
01:52:24,137 --> 01:52:27,107
HARM THAT IS KIND OF LIKE A WIDE

2959
01:52:27,173 --> 01:52:29,209
RANGE OF HARMFUL CONSEQUENCES,

2960
01:52:29,275 --> 01:52:32,612
AND YOU KNOW THAT HARM TO THE

2961
01:52:32,679 --> 01:52:34,247
COMPANY THAT PUT UP THE CHAT BOT

2962
01:52:34,314 --> 01:52:38,284
AND THIS IS WHAT WE SEE

2963
01:52:38,351 --> 01:52:39,119
UNFORTUNATELY IN THE COMMUNITY

2964
01:52:39,185 --> 01:52:41,521
THAT SORT OF EAGERNESS TO RUSH

2965
01:52:41,588 --> 01:52:44,057
FORWARD AND JUST USE THIS LARGE

2966
01:52:44,124 --> 01:52:48,995
LANGUAGE MODELS BECAUSE THEY'RE

2967
01:52:49,062 --> 01:52:50,430
SO WONDERFUL BUT MAN, THE WHOLE

2968
01:52:50,497 --> 01:52:52,632
SERVICE GOES DOWN THE NEXT DAY.

2969
01:52:52,699 --> 01:52:55,935
>> THANK YOU EMPLOY.

2970
01:52:56,002 --> 01:52:58,104
>> I CAN JUMP IN HERE, NICE

2971
01:52:58,171 --> 01:53:02,041
PRESENTATION ON YOUR WORK.

2972
01:53:02,108 --> 01:53:04,511
I WAS JUST WONDERING FROM THE

2973
01:53:04,577 --> 01:53:05,778
QUESTIONNAIRE PERSPECTIVE, ARE

2974
01:53:05,845 --> 01:53:06,679
YOU TAKING INTO ACCOUNT

2975
01:53:06,746 --> 01:53:08,815
INFORMATION ON THE LEVEL OF

2976
01:53:08,882 --> 01:53:09,816
EXPERTISE MAYBE BY LOOKING AT

2977
01:53:09,883 --> 01:53:12,051
THE TYPES OF QUESTIONS THAT ARE

2978
01:53:12,118 --> 01:53:14,220
ASKED SO FOR INSTANCE ISSUES CAN

2979
01:53:14,287 --> 01:53:16,122
YOU HAVE VERY UNINFORMED PATIENT

2980
01:53:16,189 --> 01:53:17,824
ANDS VERY INFORMED PATIENT IN

2981
01:53:17,891 --> 01:53:19,792
MAYBE MAKING YOUR DECISIONS, YOU

2982
01:53:19,859 --> 01:53:21,327
KNOW LOOKING AT THE HARMFULNESS

2983
01:53:21,394 --> 01:53:23,730
AND INFORMED PATIENT MAY BE ABLE

2984
01:53:23,796 --> 01:53:24,764
TO HANDLE A LITTLE BIT MORE

2985
01:53:24,831 --> 01:53:28,368
INFORMATION THAN AN UNINFORMED?

2986
01:53:28,434 --> 01:53:31,237
>> YEAH, SO FOR THE QUESTIONS

2987
01:53:31,304 --> 01:53:33,506
THAT ARE EXTERIOR FSM POSTED TO

2988
01:53:33,573 --> 01:53:35,008
FORUMS OR SENT TO CUSTOMER

2989
01:53:35,074 --> 01:53:36,276
SERVICES, IT IS SOMEWHAT EASIER

2990
01:53:36,342 --> 01:53:38,378
TO FIGURE OUT THE LEVEL OF

2991
01:53:38,444 --> 01:53:44,851
EDUCATION FOR THESE VERY SHORT

2992
01:53:44,918 --> 01:53:45,318
QUESTIONS.

2993
01:53:45,385 --> 01:53:46,386
SOMETIMES IT'S HARDER TO, YOU

2994
01:53:46,452 --> 01:53:49,189
KNOW IF PEOPLE ARE USING 3-5

2995
01:53:49,255 --> 01:53:52,125
WORDS TO ASK SOMETHING, YOU

2996
01:53:52,192 --> 01:53:54,194
COULD STILL DISTINGUISH SOME

2997
01:53:54,260 --> 01:53:59,866
YEARS AGO, KIRK ROBERTS A

2998
01:53:59,933 --> 01:54:03,136
TRAINEE, HE PUBLISHED QUESTIONS

2999
01:54:03,203 --> 01:54:04,671
HE WAS ABLE TO DETERMINE THE

3000
01:54:04,737 --> 01:54:05,905
REGISTER OF THE PERSON WHO WAS

3001
01:54:05,972 --> 01:54:07,006
ASKING THE QUESTION.

3002
01:54:07,073 --> 01:54:12,045
BUT OF COURSE WE DON'T TRACK

3003
01:54:12,111 --> 01:54:14,147
OUR--UML, DOES NOT TRACK THE

3004
01:54:14,214 --> 01:54:15,448
KNOWLEDGE, ABOUT WHO THE USER IS

3005
01:54:15,515 --> 01:54:18,785
AND WHAT THEY HAVE SEEN BEFORE,

3006
01:54:18,851 --> 01:54:19,953
SO BASICALLY EVERYTHING WE INFER

3007
01:54:20,019 --> 01:54:22,222
ABOUT THE USER IS FROM THE

3008
01:54:22,288 --> 01:54:31,531
QUESTION THAT THEY'VE ASKED.

3009
01:54:31,598 --> 01:54:34,567
>> I DON'T KNOW WHO WAS FIRST

3010
01:54:34,634 --> 01:54:36,369
BUT I WAS FIRST?

3011
01:54:36,436 --> 01:54:36,603
THANKS.

3012
01:54:36,669 --> 01:54:38,371
SO DINA, THANK YOU, GREAT

3013
01:54:38,438 --> 01:54:41,407
PRESENTATION AND GREAT WORK, AS

3014
01:54:41,474 --> 01:54:42,675
ALWAYS, NOT EXACTLY MY

3015
01:54:42,742 --> 01:54:43,309
[INDISCERNIBLE] QUESTION

3016
01:54:43,376 --> 01:54:46,379
ANSWERING FOR A LONG TIME BUT 1

3017
01:54:46,446 --> 01:54:50,617
THING I WAS THINKING IS I

3018
01:54:50,683 --> 01:54:53,419
NOTICED THAT YOU'RE USING LLAMA

3019
01:54:53,486 --> 01:54:55,722
270 B AS A WAY TO COMPARE AND

3020
01:54:55,788 --> 01:54:56,623
THE PERFORMANCE PRETTY GOOD SO

3021
01:54:56,689 --> 01:54:58,992
YOUR OPINION ON THE USE OF THIS

3022
01:54:59,058 --> 01:55:00,526
FOUNDATION MODELS THAT ARE OPEN

3023
01:55:00,593 --> 01:55:02,662
SOURCE VERSUS LIKE THE CHAT GPT?

3024
01:55:02,729 --> 01:55:04,664
IS MY FIRST PART OF MY QUESTION

3025
01:55:04,731 --> 01:55:05,965
AND I WILL WAIT FOR YOU TO

3026
01:55:06,032 --> 01:55:07,333
ANSWER THAT BEFORE DOING MY

3027
01:55:07,400 --> 01:55:08,067
SECOND PART.

3028
01:55:08,134 --> 01:55:12,672
SO GO AHEAD, PLEASE.

3029
01:55:12,739 --> 01:55:14,774
>> YES, SO WE DO SEE THE BEST

3030
01:55:14,841 --> 01:55:19,112
RESULTS WITH CHAT GPT BUT IT'S A

3031
01:55:19,178 --> 01:55:21,047
VERY KRSHES DECISION TO ALWAYS

3032
01:55:21,114 --> 01:55:24,851
DO A OPEN SOURCE FULLY AVAILABLE

3033
01:55:24,917 --> 01:55:25,985
MODEL IN PARALLEL TO SEE IF WE

3034
01:55:26,052 --> 01:55:31,557
CAN GET THE SAME RESULTS AND

3035
01:55:31,624 --> 01:55:35,194
BETTER RESULTS BECAUSE IT'S SORT

3036
01:55:35,261 --> 01:55:37,897
OF UNACCEPTABLE TO PROVIDE--TO

3037
01:55:37,964 --> 01:55:39,465
RELY ON SOMETHING LIKE A BLACK

3038
01:55:39,532 --> 01:55:41,000
BOX TO PROVIDE OUR SERVICES.

3039
01:55:41,067 --> 01:55:45,405
SO YOU KNOW WE ARE WORKING ON

3040
01:55:45,471 --> 01:55:46,005
DEMOCRATIZING THESE LARGE

3041
01:55:46,072 --> 01:55:47,774
LANGUAGE MODEL ANDS MAKING THEM

3042
01:55:47,840 --> 01:55:52,045
MORE AFFORDABLE AND MORE

3043
01:55:52,111 --> 01:55:52,779
ACCESSIBLE.

3044
01:55:52,845 --> 01:55:53,212
>> EXCELLENT.

3045
01:55:53,279 --> 01:55:53,546
EXCELLENT.

3046
01:55:53,613 --> 01:55:57,450
MY SECOND WOP IS SOMETIMES WE

3047
01:55:57,517 --> 01:56:02,055
TEND IN THIS MOMENT OF THE CHAT

3048
01:56:02,121 --> 01:56:05,725
GPT ERA TO OVERUSE RESOURCES,

3049
01:56:05,792 --> 01:56:08,027
MEANING, I'LL GO ASK CHAT GPT

3050
01:56:08,094 --> 01:56:09,329
THE DEFINITION OF SOMETHING,

3051
01:56:09,395 --> 01:56:10,997
WHERE GOOGLE WAS PERFECTLY FINE

3052
01:56:11,064 --> 01:56:12,899
WITH THAT, RIGHT?

3053
01:56:12,965 --> 01:56:14,701
SO LIKE THIS WHOLE PUSH COMING

3054
01:56:14,767 --> 01:56:16,869
IN THE BACKGROUND OF GREENNESS

3055
01:56:16,936 --> 01:56:17,770
AND ENVIRONMENTAL CONSCIOUSNESS

3056
01:56:17,837 --> 01:56:21,174
USING THIS BUT, IN THIS

3057
01:56:21,240 --> 01:56:24,010
PRACTICAL INSSTANCE OF VIDEOS AS

3058
01:56:24,077 --> 01:56:27,513
ANSWERS TO QUESTIONS, GOOGLE HAS

3059
01:56:27,580 --> 01:56:29,048
THE OPTION TO CHOOSE, RIGHT?

3060
01:56:29,115 --> 01:56:33,786
YOU PUT A SEARCH AND YOU

3061
01:56:33,853 --> 01:56:35,088
HAVE--SHOW UP, AND VIDEO AND WEB

3062
01:56:35,154 --> 01:56:39,325
EXPW SO ON, HOW DOES--WHEN IS

3063
01:56:39,392 --> 01:56:43,830
IT--WHAT DO WE GAIN OVER THAT,

3064
01:56:43,896 --> 01:56:46,933
QUICKLY AND I WANT TO SEE HOW

3065
01:56:46,999 --> 01:56:47,400
YOU SEE THAT?

3066
01:56:47,467 --> 01:56:50,436
>> DID I UNDERSTAND THE QUESTION

3067
01:56:50,503 --> 01:56:50,703
RIGHT?

3068
01:56:50,770 --> 01:56:54,207
IT'S LIKE YOU CAN GIVE THAT

3069
01:56:54,273 --> 01:56:55,408
GOOGLE AND WHY THEY'RE WORKING

3070
01:56:55,475 --> 01:56:56,809
ON RESEARCHING THIS STUFF?

3071
01:56:56,876 --> 01:56:58,111
>> LET ME RESTATE MY QUESTION,

3072
01:56:58,177 --> 01:57:01,981
LIKE WHAT DO WE GAIN BY USING

3073
01:57:02,048 --> 01:57:04,283
LARGE LANGUAGE APPROACH MODELS

3074
01:57:04,350 --> 01:57:06,619
MULTIMODAL FOR ANSWERING

3075
01:57:06,686 --> 01:57:10,089
QUESTIONS WITH POTENTIALLY

3076
01:57:10,156 --> 01:57:13,259
VIDEOS OR VERSUS WE ARE

3077
01:57:13,326 --> 01:57:14,527
[INDISCERNIBLE] SEARCH OF VIDEOS

3078
01:57:14,594 --> 01:57:15,728
IN YOUR EXAMPLE THAT ARE STILL

3079
01:57:15,795 --> 01:57:20,833
ON THE KREEN, RIGHT, OF HOW TO

3080
01:57:20,900 --> 01:57:22,902
EXAMINE LYMPHNODINGS SO IT'S A

3081
01:57:22,969 --> 01:57:24,070
QUESTION THAT'S A SEARCH IN

3082
01:57:24,137 --> 01:57:25,271
GOOGLE AND YOU CLICK ON VIDEOS

3083
01:57:25,338 --> 01:57:27,140
AND YOU GET THEM, SO THAT'S MY

3084
01:57:27,206 --> 01:57:29,075
QUESTION, HOW DO WE GAIN MORE

3085
01:57:29,142 --> 01:57:34,680
WITH THIS.

3086
01:57:34,747 --> 01:57:34,981
>> RIGHT.

3087
01:57:35,047 --> 01:57:37,250
SO, THE KIND OF YOU KNOW

3088
01:57:37,316 --> 01:57:38,518
SEVERALLA ASPECTS, SOME PEOPLE

3089
01:57:38,584 --> 01:57:43,122
ARE MORE VISUAL AND SOME PEOPLE

3090
01:57:43,189 --> 01:57:44,757
ARE MORE SAVVY, SO SOME PEOPLE

3091
01:57:44,824 --> 01:57:46,459
WHO ARE MORE VISUAL, THEY WILL

3092
01:57:46,526 --> 01:57:52,632
WANT TO SEE A VISUAL ANSWER A

3093
01:57:52,698 --> 01:57:56,035
PICTURE, I MYSELF AM A TEXT

3094
01:57:56,102 --> 01:57:57,737
PERSON, I'M FINE READING A

3095
01:57:57,804 --> 01:58:01,040
DESCRIPTION OF SOMETHING, BUT I

3096
01:58:01,107 --> 01:58:02,341
UNDERSTAND THAT DIFFERENT

3097
01:58:02,408 --> 01:58:03,376
VIEWERS WILL WANT DIFFERENT OPGS

3098
01:58:03,443 --> 01:58:07,013
AND THAT'S KIND OF LIKE TOWARD

3099
01:58:07,079 --> 01:58:08,214
THAT NLM 1 ENTRY POINT WHERE YOU

3100
01:58:08,281 --> 01:58:10,716
CAN GET WHATEVER YOU WANT FROM

3101
01:58:10,783 --> 01:58:11,150
THAT ENTRY POINT.

3102
01:58:11,217 --> 01:58:13,286
YOU DON'T HAVE TO BE AWARE OF

3103
01:58:13,352 --> 01:58:15,254
LIKE, I NEED TO GO OVER THERE

3104
01:58:15,321 --> 01:58:16,289
AND LOOK FOR SOMETHING OVER

3105
01:58:16,355 --> 01:58:21,661
THERE IF I WANT A--AN IMAGE

3106
01:58:21,727 --> 01:58:22,128
BASED ANSWER.

3107
01:58:22,195 --> 01:58:29,035
THE OTHER PART OF IT IS THIS

3108
01:58:29,101 --> 01:58:30,169
IS--PARTICULARLY FOR THE MEDICAL

3109
01:58:30,236 --> 01:58:34,941
DOMAIN, SO CAN YOU SEE, YOU KNOW

3110
01:58:35,007 --> 01:58:36,609
THIS STEP CAPTURING FULLY OPEN

3111
01:58:36,676 --> 01:58:37,844
DOMAIN ALREADY EXISTS AND THERE

3112
01:58:37,910 --> 01:58:39,879
ARE EVALUATION ON THAT, SO IT'S

3113
01:58:39,946 --> 01:58:41,380
NOT A SOLVED PROBLEM EVEN FOR

3114
01:58:41,447 --> 01:58:44,584
THE OPEN DOMAIN BUT WE CAN

3115
01:58:44,650 --> 01:58:46,252
ALREADY SEE THE RESULTS ON

3116
01:58:46,319 --> 01:58:49,722
GOOGLE SEARCHES WHERE THERE IS

3117
01:58:49,789 --> 01:58:50,790
THE STEP DESCRIPTION FOR THE

3118
01:58:50,857 --> 01:58:52,959
ITEM OR WHARF YOU WERE LOOKING

3119
01:58:53,025 --> 01:58:54,393
FOR, BUT FOR THE MEDICAL DOMAIN,

3120
01:58:54,460 --> 01:58:58,364
NOTHING LIKE THAT EXISTS, YET.

3121
01:58:58,431 --> 01:59:00,433
SO WHEN YOU SEARCH THESE

3122
01:59:00,500 --> 01:59:02,502
YouTUBE OR GOOGLE VIEWS,

3123
01:59:02,568 --> 01:59:05,037
THEY'RE NOT, YOU ARE JUST

3124
01:59:05,104 --> 01:59:07,940
GETTING THE WHOLE VIDEO AND VERY

3125
01:59:08,007 --> 01:59:11,611
OFTEN IT'S NOT EVEN AN INTRUKSAL

3126
01:59:11,677 --> 01:59:14,213
VIDEO, IT'S TOPICAL RELEVANT,

3127
01:59:14,280 --> 01:59:16,549
BUT IT COULD BE SOMEONE SHOWING

3128
01:59:16,616 --> 01:59:17,083
THE SLIDES.

3129
01:59:17,149 --> 01:59:24,857
WE ALSO MAKE AN EFFORT TO SELECT

3130
01:59:24,924 --> 01:59:27,260
REPUTABLE SOURCES AND SELECT

3131
01:59:27,326 --> 01:59:28,027
PATIENT-LEVEL SOURCES SO IT

3132
01:59:28,094 --> 01:59:29,161
WON'T BE AT A PROFESSIONAL

3133
01:59:29,228 --> 01:59:32,331
LEVEL, SO ALL OF THAT IS STILL

3134
01:59:32,398 --> 01:59:35,935
SORT OF ACTIVE RESEARCH AND

3135
01:59:36,002 --> 01:59:37,103
CREATING THESE COLLECTIONS AND

3136
01:59:37,169 --> 01:59:39,305
HAVING THESE COLLECTIONS IN THE

3137
01:59:39,372 --> 01:59:46,379
PUBLIC DOMAIN IS USEFUL FOR THEM

3138
01:59:46,445 --> 01:59:49,181
INCLUDING GOOGLE.

3139
01:59:49,248 --> 01:59:51,417
NOKAY, THANKS.

3140
01:59:51,484 --> 01:59:56,656
>> DINA, WONDERFUL PRESENTATION

3141
01:59:56,722 --> 01:59:59,091
AND EXTREMELY USEFUL AREA OF

3142
01:59:59,158 --> 02:00:00,092
RESEARCH, NOT MY PARTICULAR AREA

3143
02:00:00,159 --> 02:00:01,794
OF RESEARCH BUT I HAVE A

3144
02:00:01,861 --> 02:00:03,462
QUESTION FOR YOU IN YOUR FUTURE

3145
02:00:03,529 --> 02:00:04,964
CHALLENGES OF RESEARCH BUT I

3146
02:00:05,031 --> 02:00:06,165
WANTED TO FIND OUT A LITTLE BIT

3147
02:00:06,232 --> 02:00:08,267
MORE AS A SCIENTIFIC, IF

3148
02:00:08,334 --> 02:00:10,102
SUDDENLY MY METRICS, I MEAN,

3149
02:00:10,169 --> 02:00:11,871
WHAT I'M TRYING TO DO IS

3150
02:00:11,938 --> 02:00:14,140
CHALLENGE, I CAN SEE IN YOUR

3151
02:00:14,206 --> 02:00:17,109
CASE, IT'S LIKE, BEHAVIORIAL AND

3152
02:00:17,176 --> 02:00:18,544
PSYCHOLOGICAL ISSUES, I MEAN, I

3153
02:00:18,611 --> 02:00:20,846
NEVER HAD TO WORRY ABOUT THAT.

3154
02:00:20,913 --> 02:00:22,682
I FOUND A MUTATION OR I DON'T

3155
02:00:22,748 --> 02:00:25,651
EMPLOY IN YOUR CASE, IT'S LIKE

3156
02:00:25,718 --> 02:00:26,819
IS THERE EMOTIONAL DISTRESS FROM

3157
02:00:26,886 --> 02:00:30,022
THE ANSWER OF THESE VERSUS, YOU

3158
02:00:30,089 --> 02:00:33,859
KNOW, IT'S USEFUL OR NOT, SO IS

3159
02:00:33,926 --> 02:00:36,162
IT SOMETHING THAT YOU WOULD

3160
02:00:36,228 --> 02:00:37,096
REQUIRE--I CAN IMAGINE THAT THAT

3161
02:00:37,163 --> 02:00:39,398
WILL BE PART OF YOUR CHALLENGES

3162
02:00:39,465 --> 02:00:41,500
AND I ALSO IMAGINE THAT THE

3163
02:00:41,567 --> 02:00:46,005
RESOURCES THAT HAVE YOU AROUND

3164
02:00:46,072 --> 02:00:47,239
AT NNLM MIGHT NOT BE PROVIDING

3165
02:00:47,306 --> 02:00:50,876
SOME OF THOSE, YOU KNOW, SOME

3166
02:00:50,943 --> 02:00:52,378
BETTER WAYS TO WORK ON THOSE

3167
02:00:52,445 --> 02:00:53,446
CHANCES, SO TOMORROW ON THE

3168
02:00:53,512 --> 02:00:55,982
QUESTION, WHAT ARE YOUR

3169
02:00:56,048 --> 02:00:56,849
CHALLENGES, WHAT ARE THE

3170
02:00:56,916 --> 02:00:58,250
RESOURCES THAT YOU HAVE OR

3171
02:00:58,317 --> 02:01:00,519
MOSTLY YOU DON'T HAVE TO FACE

3172
02:01:00,586 --> 02:01:03,656
THOSE CHALLENGES, ESPECIALLY

3173
02:01:03,723 --> 02:01:05,191
CONSIDERING THAT YOU ARE NOW IN

3174
02:01:05,257 --> 02:01:06,692
THIS AREA WHERE THINGS LIKE

3175
02:01:06,759 --> 02:01:07,994
THAT, BECOME A PART OF THE

3176
02:01:08,060 --> 02:01:12,398
METRIC OF YOUR WORK.

3177
02:01:12,465 --> 02:01:14,200
>> YEAH, SO WE WERE VERY

3178
02:01:14,266 --> 02:01:17,903
FORTUNATE AT LAST YEAR, YES, NLM

3179
02:01:17,970 --> 02:01:19,772
GAVE US RESOURCES TO HIRE THAT

3180
02:01:19,839 --> 02:01:23,609
ICF COMPANY TO DO THESE LINE BY

3181
02:01:23,676 --> 02:01:25,811
LINE TRANSLATIONS, SO THEN, I

3182
02:01:25,878 --> 02:01:29,782
THINK IT WAS 30 K, WHICH IS YOU

3183
02:01:29,849 --> 02:01:32,385
KNOW, THE BENEFIT IS TREMENDOUS

3184
02:01:32,451 --> 02:01:35,421
BECAUSE THESE ARE PEOPLE WHO ARE

3185
02:01:35,488 --> 02:01:38,357
TRAINED, THEY ARE SUPPORTING MED

3186
02:01:38,424 --> 02:01:43,496
LINE PLUS, AND THEY WERE TRAINED

3187
02:01:43,562 --> 02:01:44,897
TO SUPPORT EHR, SO THEY DO THAT

3188
02:01:44,964 --> 02:01:46,799
FOR A LIVING, THEY KNOW HOW TO

3189
02:01:46,866 --> 02:01:50,069
EXPLAIN THAT RESEARCH IN PLAIN

3190
02:01:50,136 --> 02:01:52,171
LANGUAGE SOPHISTICATED THEIR

3191
02:01:52,238 --> 02:01:52,571
TRANSLATIONS WERE

3192
02:01:52,638 --> 02:01:55,608
TRIEWMENTAL--AND THEY ALSO

3193
02:01:55,675 --> 02:01:57,476
HELPED EVALUATING THESE LARGE

3194
02:01:57,543 --> 02:01:58,944
LANGUAGE MODELS, UPWARDS IN THE

3195
02:01:59,011 --> 02:01:59,912
CHALLENGE, SO AGAIN THIS, IS

3196
02:01:59,979 --> 02:02:01,414
LIKE A SERVICE TO THE COMMUNITY

3197
02:02:01,480 --> 02:02:03,716
AS WELL, BECAUSE NOW THE

3198
02:02:03,783 --> 02:02:05,184
COMMUNITY GETS THEIR WORK

3199
02:02:05,251 --> 02:02:09,355
EVALUATED IN THE EXACT SAME

3200
02:02:09,422 --> 02:02:11,357
ENVIRONMENT SO NOW YOU HAVE, YOU

3201
02:02:11,424 --> 02:02:13,225
CAN AT LEAST COMPARE THESE TEAMS

3202
02:02:13,292 --> 02:02:14,126
THAT PARTICIPATED IN THESE

3203
02:02:14,193 --> 02:02:17,563
APPROACHES THAT WE USED IN THAT

3204
02:02:17,630 --> 02:02:19,799
CHALLENGE TO SEE WHICH 1 IS MORE

3205
02:02:19,865 --> 02:02:24,704
PROMISING AND WHICH 1 SHOULD BE

3206
02:02:24,770 --> 02:02:29,842
DEVELOPED MORE.

3207
02:02:29,909 --> 02:02:32,178
AND YES, WE HAD--IS THE PURPOSE

3208
02:02:32,244 --> 02:02:35,414
WHO HAS BEEN THE PERSON

3209
02:02:35,481 --> 02:02:38,117
EVALUATING THESE APPROACHES AND

3210
02:02:38,184 --> 02:02:39,051
THE TEXT RETRIEVAL CONFERENCE

3211
02:02:39,118 --> 02:02:40,586
FOR THE LAST 3 OR 4 YEARS AND

3212
02:02:40,653 --> 02:02:43,989
SHE WAS VERY CONCERNED IN THESE

3213
02:02:44,056 --> 02:02:46,125
LAST TRACK MEETING THAT WE DO

3214
02:02:46,192 --> 02:02:49,628
NEED TO LIKE HAVE A COMMUNITY

3215
02:02:49,695 --> 02:02:53,532
EFFORT ON FINDING THESE, BECAUSE

3216
02:02:53,599 --> 02:02:57,903
YOU HAVE SEEN THE ANSWERS, THE

3217
02:02:57,970 --> 02:03:00,172
ANSWERS ARE BEAUTIFUL SO IT'S NO

3218
02:03:00,239 --> 02:03:02,875
LONGER THE PROBLEM OF

3219
02:03:02,942 --> 02:03:06,078
[INDISCERNIBLE] AND FLUENCY, THE

3220
02:03:06,145 --> 02:03:08,114
PROBLEM NOW IS THAT THE ANSWERS

3221
02:03:08,180 --> 02:03:11,684
ARE SO PLAUSIBLE AND THEY ARE

3222
02:03:11,751 --> 02:03:14,720
ALL SO RELATIVELY RARELY THAT IS

3223
02:03:14,787 --> 02:03:20,893
NOW TURNING INTO LIKE DETECTING

3224
02:03:20,960 --> 02:03:21,727
THAT OUTLIER, EVALUATING LIKE

3225
02:03:21,794 --> 02:03:26,031
WHAT WE USE TO DO SO IT'S A

3226
02:03:26,098 --> 02:03:27,299
CHALLENGE BUT THAT'S A GOOD

3227
02:03:27,366 --> 02:03:27,933
THING, RIGHT?

3228
02:03:28,000 --> 02:03:38,377
WE ARE MOVING FORWARD.

3229
02:03:44,083 --> 02:03:45,918
ALL RIGHT, WE HAVE TIME FOR 1

3230
02:03:45,985 --> 02:03:47,119
MORE QUESTION BEFORE WE MOVE TO

3231
02:03:47,186 --> 02:03:55,995
CLOSED SESSION WITH THE POST

3232
02:03:56,061 --> 02:03:56,262
DOCS.

3233
02:03:56,328 --> 02:03:58,464
ALL RIGHT, SO WITH THAT, IF WE

3234
02:03:58,531 --> 02:03:59,665
CAN MOVE ALL OF THE BOARD AND

3235
02:03:59,732 --> 02:04:02,968
THE POST DOCS TO THE CLOSED

3236
02:04:03,035 --> 02:04:04,403
SESSION BREAK OUT ROOM, THAT

3237
02:04:04,470
WOULD BE GREAT, THANK YOU