WEBVTT
Language: en-US

1
00:00:00.180 --> 00:00:04.980
<v Kalim Saliba>I think one of the biggest data quality
problems the industry has right now is one

2
00:00:04.980 --> 00:00:07.739
<v Kalim Saliba>of context and missing contexts, right?

3
00:00:07.849 --> 00:00:10.939
<v Kalim Saliba>Because even if you have data and you're
sure it's the right data and it's

4
00:00:10.939 --> 00:00:16.500
<v Kalim Saliba>represented correctly, if it's not
contextualized within the appropriate surrounding

5
00:00:16.500 --> 00:00:19.459
<v Kalim Saliba>data, it's very hard to really understand
what you're supposed to do with it.

6
00:00:29.049 --> 00:00:33.939
<v Speaker 2>Welcome to a brand new episode of our
podcast, Human and AI Mind Machines in a crate

7
00:00:33.939 --> 00:00:34.700
<v Speaker 2>in the sand.

8
00:00:35.060 --> 00:00:39.459
<v Speaker 3>Thanks for tuning in to our geeky podcast
where we discuss the fascinating field of

9
00:00:39.459 --> 00:00:43.340
<v Speaker 3>AI and machine learning, corporate
craziness, passion for technology and the role of

10
00:00:43.340 --> 00:00:43.849
<v Speaker 3>humans.

11
00:00:44.099 --> 00:00:48.419
<v Speaker 3>We're your hosts, Oli and Abuy, and it
means the world to us that you tune in to

12
00:00:48.419 --> 00:00:50.259
<v Speaker 3>listen, learn and grow with us.

13
00:00:50.500 --> 00:00:52.860
<v Speaker 2>And today we're heading into the lab.

14
00:00:53.020 --> 00:00:56.400
<v Speaker 2>AI has already changed how we predict what
a protein looks like.

15
00:00:56.540 --> 00:00:56.610
<v Speaker 2>Now.

16
00:00:56.793 --> 00:01:01.993
<v Speaker 2>The races on to design and make molecules
that actually work as drugs with AI agents

17
00:01:01.993 --> 00:01:05.903
<v Speaker 2>writing their own scripts and running
their own analysis along the way.

18
00:01:06.113 --> 00:01:08.893
<v Speaker 3>And that creates quite an interesting
tension.

19
00:01:09.113 --> 00:01:13.583
<v Speaker 3>So in software, you can copy a project
with one command and get the same results

20
00:01:13.583 --> 00:01:14.814
<v Speaker 3>over and over again.

21
00:01:14.954 --> 00:01:17.503
<v Speaker 3>And in a lab, that's still kind of a
dream.

22
00:01:17.753 --> 00:01:23.793
<v Speaker 3>Recently, AI agent designed a protein
starting from a 16,000 word prompt, writing

23
00:01:23.793 --> 00:01:24.634
<v Speaker 3>its own script.

24
00:01:25.017 --> 00:01:29.367
<v Speaker 3>As it went, impressive, but nearly
impossible to repeat exactly.

25
00:01:29.648 --> 00:01:31.477
<v Speaker 3>The 1st run can be creative.

26
00:01:31.648 --> 00:01:34.708
<v Speaker 3>By the 100th run, it has to be a reliable
product.

27
00:01:35.248 --> 00:01:40.127
<v Speaker 2>How do you let AI build freely while
scientists can still inspect it, trust it, and

28
00:01:40.127 --> 00:01:41.227
<v Speaker 2>sign off on it?

29
00:01:41.447 --> 00:01:46.157
<v Speaker 2>To help us detangle that, we have someone
who sits right at the intersection.

30
00:01:46.407 --> 00:01:50.828
<v Speaker 2>Uh, he has spent 17 years growing product
in engineering teams and scaling SUSS

31
00:01:50.828 --> 00:01:51.557
<v Speaker 2>businesses.

32
00:01:51.928 --> 00:01:52.648
<v Speaker 2>Today he's the chief.

33
00:01:52.772 --> 00:01:57.222
<v Speaker 2>Product officer, that's a cool, cool name
anyway, at Dot Mattox, leading the Duma

34
00:01:57.222 --> 00:01:57.601
<v Speaker 2>platform.

35
00:01:57.781 --> 00:02:03.061
<v Speaker 2>He's also a writer, a public speaker,
Hauldas, a photographer, I love this, and a

36
00:02:03.061 --> 00:02:06.622
<v Speaker 2>self-declared follower of the Ted Lasso
School of Coaching.

37
00:02:06.802 --> 00:02:08.901
<v Speaker 2>Please welcome Kalim Saliba.

38
00:02:09.171 --> 00:02:09.742
<v Kalim Saliba>Thank you guys.

39
00:02:09.861 --> 00:02:10.722
<v Kalim Saliba>Really great to be here.

40
00:02:10.891 --> 00:02:13.822
<v Kalim Saliba>Appreciate your time and the opportunity
to talk about these really important and

41
00:02:13.822 --> 00:02:16.492
<v Kalim Saliba>exciting subjects that are impacting our
industry and our customers.

42
00:02:16.701 --> 00:02:18.212
<v Speaker 3>Yeah, we can't wait to dive in.

43
00:02:18.421 --> 00:02:21.916
<v Speaker 3>And before we do so, we have a little
Challenge for you, Kalim.

44
00:02:22.296 --> 00:02:26.876
<v Speaker 3>Since the first run can be creative, as we
heard in the intro, why don't you ask

45
00:02:26.876 --> 00:02:28.035
<v Speaker 3>your first run answer?

46
00:02:28.136 --> 00:02:30.735
<v Speaker 3>So maybe you could introduce yourself real
quick.

47
00:02:30.916 --> 00:02:31.835
<v Speaker 3>Who is Kalim?

48
00:02:31.916 --> 00:02:33.595
<v Speaker 3>What is it that really drives you?

49
00:02:33.796 --> 00:02:34.805
<v Speaker 3>Why don't you go ahead?

50
00:02:35.115 --> 00:02:38.916
<v Kalim Saliba>So, Kaliem Saliba, I'm the chief product
officer at Dotmatt, as you mentioned, I'm

51
00:02:38.916 --> 00:02:43.305
<v Kalim Saliba>leading product and engineering, heavily
focused on our lab informatics platform

52
00:02:43.305 --> 00:02:44.016
<v Kalim Saliba>called Luma.

53
00:02:44.156 --> 00:02:47.826
<v Kalim Saliba>I've been with the company for about 4
years and prior to that, I worked at a

54
00:02:47.826 --> 00:02:52.529
<v Kalim Saliba>company called uh Peas Systems for about
15 years, which is a leader in low code

55
00:02:52.529 --> 00:02:55.169
<v Kalim Saliba>data management and, uh, workflow
orchestration.

56
00:02:55.440 --> 00:03:00.089
<v Kalim Saliba>A lot of important principles and and
concepts I've kind of been a part of over my

57
00:03:00.089 --> 00:03:03.970
<v Kalim Saliba>career that have a lot of relevance to
life sciences, and I think what drives me and

58
00:03:03.970 --> 00:03:08.679
<v Kalim Saliba>what I'm the most passionate about is kind
of using the knowledge that I've learned,

59
00:03:08.889 --> 00:03:13.009
<v Kalim Saliba>the leadership skills that I've gained to
work with teams and work with software and

60
00:03:13.009 --> 00:03:16.570
<v Kalim Saliba>technology to really do that, which has
been very difficult.

61
00:03:16.813 --> 00:03:21.544
<v Kalim Saliba>You know, in some cases seemingly
impossible, and especially now with the advent of

62
00:03:21.544 --> 00:03:26.664
<v Kalim Saliba>AI and all of these amazing technologies,
like the ability to unlock value if

63
00:03:26.664 --> 00:03:30.923
<v Kalim Saliba>approached in just the right way is
incredible, and I think that's very motivating

64
00:03:31.223 --> 00:03:36.464
<v Kalim Saliba>to me and the, the teams that I work with,
just really trying to do whatever we can

65
00:03:36.464 --> 00:03:40.824
<v Kalim Saliba>to bring together software for the benefit
of our customers and patients and others

66
00:03:40.824 --> 00:03:44.544
<v Kalim Saliba>that are really stand to benefit
significantly from the work that we are doing.

67
00:03:45.050 --> 00:03:47.970
<v Speaker 2>Yeah, that was almost 60 seconds, I would
say, right?

68
00:03:48.210 --> 00:03:53.210
<v Speaker 2>Hey Kalim, so you, you know, you, you
mentioned something on Ted Lasso school of

69
00:03:53.210 --> 00:03:54.240
<v Speaker 2>coaching, right?

70
00:03:54.490 --> 00:03:58.570
<v Speaker 2>So obviously maybe for the audience
worldwide, you know, give us, give us a bit of

71
00:03:58.570 --> 00:03:59.529
<v Speaker 2>your perspective.

72
00:03:59.850 --> 00:04:02.250
<v Speaker 2>We haven't seen maybe also the show, not
yet, right?

73
00:04:02.330 --> 00:04:06.210
<v Speaker 2>So what actually means for you, Ted
Lasso's school of coaching?

74
00:04:06.250 --> 00:04:09.029
<v Speaker 2>Why this matters for this conversation
here with us?

75
00:04:09.440 --> 00:04:13.399
<v Kalim Saliba>Yeah, that's a great question, and it's an
interesting question because I am not at

76
00:04:13.399 --> 00:04:14.759
<v Kalim Saliba>all a sports fan.

77
00:04:14.880 --> 00:04:18.200
<v Kalim Saliba>I, you know, I think, um, you know,
there's, there's always a running joke with my

78
00:04:18.200 --> 00:04:22.200
<v Kalim Saliba>friends that when an important game was on
and they were all, you know, we're all

79
00:04:22.200 --> 00:04:26.480
<v Kalim Saliba>talking about which team do you prefer,
who do you like, my go to was just, yeah, go

80
00:04:26.480 --> 00:04:30.260
<v Kalim Saliba>sports, like very generic, like, I don't
even know who's playing in this game, but

81
00:04:30.440 --> 00:04:31.829
<v Kalim Saliba>happy to be a part of it with you guys.

82
00:04:32.079 --> 00:04:37.079
<v Kalim Saliba>I think for me, Ted Lasso and really like
the coaching style that's embodied by the

83
00:04:37.079 --> 00:04:37.440
<v Kalim Saliba>main character.

84
00:04:37.967 --> 00:04:43.007
<v Kalim Saliba>Ted, it's an exaggeration of what good
coaching looks like, I think.

85
00:04:43.208 --> 00:04:48.447
<v Kalim Saliba>And by being an exaggeration, it
highlights the things that I think a lot of leaders

86
00:04:48.447 --> 00:04:53.208
<v Kalim Saliba>try and strive to be really empowering,
being able to connect with people at a very,

87
00:04:53.398 --> 00:04:56.928
<v Kalim Saliba>you know, personal level to understand
their specific struggles and then their

88
00:04:56.928 --> 00:05:00.407
<v Kalim Saliba>broader struggles within the team, but
also knowing when to be decisive, right?

89
00:05:00.497 --> 00:05:05.088
<v Kalim Saliba>Knowing how to balance distinction between
decisive and empowering leadership is

90
00:05:05.088 --> 00:05:07.365
<v Kalim Saliba>something that I'm not perfect at it.

91
00:05:07.496 --> 00:05:11.976
<v Kalim Saliba>I think it's, it's always going to be a
lifelong, uh, piece of learning for me, and

92
00:05:12.216 --> 00:05:17.175
<v Kalim Saliba>I think Ted Lasso represents that
objective for me, and I believe others, which is

93
00:05:17.175 --> 00:05:18.355
<v Kalim Saliba>why it's such a popular show.

94
00:05:18.575 --> 00:05:20.156
<v Kalim Saliba>I will say though, I've watched it way too
many times.

95
00:05:20.256 --> 00:05:25.635
<v Kalim Saliba>I think, I think I'm on like, I think I'm
on like the 5th run of the, of the earlier

96
00:05:25.635 --> 00:05:28.656
<v Kalim Saliba>seasons, and now they, of course, they
just released a new, a new season that's

97
00:05:28.656 --> 00:05:31.996
<v Kalim Saliba>about, um, the women's team, so we're
watching that with my wife as well.

98
00:05:32.175 --> 00:05:33.855
<v Kalim Saliba>So yeah, it's just been a nice kind of
thing.

99
00:05:33.984 --> 00:05:36.824
<v Kalim Saliba>It's been in the background that's just
fun to watch, but also represents, I think,

100
00:05:36.984 --> 00:05:40.023
<v Kalim Saliba>you know, what I strive towards from a
leadership perspective.

101
00:05:40.264 --> 00:05:42.683
<v Speaker 3>Yeah, we can see that you're really
radiating that.

102
00:05:42.783 --> 00:05:45.063
<v Speaker 3>And I also, I I'm such a big fan of that
show.

103
00:05:45.403 --> 00:05:50.584
<v Speaker 3>So we touched upon Ted Lasso now, but um
actually your your day to day role um is

104
00:05:50.584 --> 00:05:54.063
<v Speaker 3>like the CPO at Dotmatt for the Luma
platform.

105
00:05:54.343 --> 00:05:59.463
<v Speaker 3>Maybe for those in the audience who don't
really know what Dotmatt does and what

106
00:05:59.463 --> 00:06:02.063
<v Speaker 3>Luma is about, can you maybe share a bit?

107
00:06:02.381 --> 00:06:06.052
<v Speaker 3>About like, how did you get in that space
and what do you guys actually even do?

108
00:06:06.432 --> 00:06:06.652
<v Kalim Saliba>Sure.

109
00:06:06.832 --> 00:06:12.052
<v Kalim Saliba>So Dotmattox is essentially a life
sciences and material sciences software company.

110
00:06:12.191 --> 00:06:16.832
<v Kalim Saliba>We have about over 800 people, many of
them with advanced degrees and PhDs.

111
00:06:17.071 --> 00:06:21.421
<v Kalim Saliba>Over 2 million scientists use our software
and we have, you know, around 10,000

112
00:06:21.421 --> 00:06:26.671
<v Kalim Saliba>customers and what we do is really bring
together a combination of highly specific,

113
00:06:26.832 --> 00:06:31.980
<v Kalim Saliba>precise scientific software within
specific Verticals and domains such as

114
00:06:32.209 --> 00:06:37.829
<v Kalim Saliba>proteomics, chem informatics, biologics,
cloning, but we also are working on

115
00:06:37.829 --> 00:06:43.559
<v Kalim Saliba>Dotmattic Luma, which is this cohesive lab
informatics platform that drives

116
00:06:43.559 --> 00:06:46.839
<v Kalim Saliba>precision in how materials are represented
and in how

117
00:06:46.839 --> 00:06:47.899
<v Kalim Saliba>processes are orchestrated.

118
00:06:48.079 --> 00:06:51.519
<v Kalim Saliba>Because if you think about the scientific
landscape, there's so much software,

119
00:06:51.679 --> 00:06:51.799
<v Kalim Saliba>right?

120
00:06:51.880 --> 00:06:57.079
<v Kalim Saliba>There's so much software, and it's so
precise in what it does, and the number of

121
00:06:57.079 --> 00:07:00.638
<v Kalim Saliba>software is growing, of course, with the
advent of AI, lots of our customers are

122
00:07:00.638 --> 00:07:01.708
<v Kalim Saliba>writing their own software.

123
00:07:01.967 --> 00:07:07.167
<v Kalim Saliba>And so how do you actually stitch together
work that's being done in these different

124
00:07:07.167 --> 00:07:11.967
<v Kalim Saliba>pieces of software across scientific
functions in a way that allows our customers to

125
00:07:11.967 --> 00:07:16.328
<v Kalim Saliba>extract higher order insights around what
it all means, how is it all connected

126
00:07:16.328 --> 00:07:16.627
<v Kalim Saliba>together.

127
00:07:16.687 --> 00:07:20.928
<v Kalim Saliba>So, that's really what we're driving
forward as a company is like excellence in

128
00:07:20.928 --> 00:07:24.847
<v Kalim Saliba>science, working with our champion and
challenger brands that we've acquired over

129
00:07:24.847 --> 00:07:26.588
<v Kalim Saliba>the history of the company, plus.

130
00:07:26.816 --> 00:07:32.756
<v Kalim Saliba>Excellence in scientific orchestration and
lab informatics, really driving autonomy

131
00:07:32.925 --> 00:07:38.855
<v Kalim Saliba>and effective use of AI in end to end
capability sets that really bring the context

132
00:07:38.855 --> 00:07:41.566
<v Kalim Saliba>cohesion that's needed to apply AI
effectively.

133
00:07:41.895 --> 00:07:44.895
<v Kalim Saliba>How did I get into this space
specifically?

134
00:07:44.976 --> 00:07:48.135
<v Kalim Saliba>I, you know, I was always interested in
life sciences.

135
00:07:48.216 --> 00:07:52.055
<v Kalim Saliba>After my engineering degree, I actually
almost went into med school until my father,

136
00:07:52.135 --> 00:07:55.036
<v Kalim Saliba>who was a physician, lovingly talked me
out of it.

137
00:07:55.696 --> 00:07:56.615
<v Kalim Saliba>Let's just say.

138
00:07:56.950 --> 00:08:00.519
<v Kalim Saliba>And encouraged me to go use my my degree
in in in other pursuits.

139
00:08:00.679 --> 00:08:05.380
<v Kalim Saliba>But, you know, hearing about the
opportunity at Dot Mattox to work on something as

140
00:08:05.760 --> 00:08:10.760
<v Kalim Saliba>really groundbreaking and comprehensive as
what we're doing here was an amazing

141
00:08:10.760 --> 00:08:11.880
<v Kalim Saliba>opportunity that I just wanted to jump
out.

142
00:08:11.970 --> 00:08:15.000
<v Kalim Saliba>And I know a lot of people on the team
are, are there for a lot of the same reasons.

143
00:08:15.160 --> 00:08:18.100
<v Kalim Saliba>So it's a really good group of people
working on a really important mission.

144
00:08:18.399 --> 00:08:19.579
<v Speaker 2>Hi, Colleen, thanks so much.

145
00:08:19.880 --> 00:08:25.160
<v Speaker 2>So, drug discovery in an area and then
protein, right, and documenting about the lab

146
00:08:25.160 --> 00:08:26.440
<v Speaker 2>and moving from lab.

147
00:08:26.579 --> 00:08:31.510
<v Speaker 2>To let's say the more the manufacturing
space, but if you engage with the community

148
00:08:31.510 --> 00:08:36.260
<v Speaker 2>on on drug discovery from early innovation
to scientific aspects, right?

149
00:08:36.700 --> 00:08:41.400
<v Speaker 2>One of the key themes over the last, let's
say, a couple of 4 or 5 years is now

150
00:08:41.400 --> 00:08:43.940
<v Speaker 2>radiate around alpha fold, right?

151
00:08:43.950 --> 00:08:48.820
<v Speaker 2>When somehow the jet GPT momentum was
right in the consumer world, we have the

152
00:08:48.820 --> 00:08:51.859
<v Speaker 2>feeling the alpha-fold did a similar kind
of aspect somehow, right?

153
00:08:51.900 --> 00:08:55.900
<v Speaker 2>It showed us what proteins maybe looked
like, it earned, you know, creators a Nobel

154
00:08:55.900 --> 00:08:56.059
<v Speaker 2>Prize.

155
00:08:56.599 --> 00:09:01.640
<v Speaker 2>And then independent analysts found out
that researchers using it submitted over 40%

156
00:09:01.640 --> 00:09:07.239
<v Speaker 2>more new protein structure, but it hasn't
led yet to a new wave of new drugs, has

157
00:09:07.239 --> 00:09:07.750
<v Speaker 2>it, right?

158
00:09:07.880 --> 00:09:13.039
<v Speaker 2>Knowing the shape is only half of the
story probably or how is AI now helping teams

159
00:09:13.039 --> 00:09:18.400
<v Speaker 2>getting from predicted structure,
probabilistic stuff, right, to molecule that, you

160
00:09:18.400 --> 00:09:22.960
<v Speaker 2>know, works as a drug and brings this new
innovation, or are we at the pivotal stage

161
00:09:22.960 --> 00:09:23.119
<v Speaker 2>here?

162
00:09:23.200 --> 00:09:25.679
<v Speaker 2>What's your view on this kind of, let's
say momentum?

163
00:09:26.219 --> 00:09:30.929
<v Kalim Saliba>I think there's a lot that has to come
together in order to harness the advances in

164
00:09:30.929 --> 00:09:34.929
<v Kalim Saliba>AI in an operational way, in a drug
discovery pipeline.

165
00:09:35.140 --> 00:09:40.280
<v Kalim Saliba>I think typically these days, the advances
in AI and the application of AI within

166
00:09:40.619 --> 00:09:42.619
<v Kalim Saliba>the pharma space, it exists, right?

167
00:09:42.659 --> 00:09:46.900
<v Kalim Saliba>It exists in spades, but it's not
necessarily as connected to the operational way

168
00:09:46.900 --> 00:09:50.820
<v Kalim Saliba>that scientists and different functions in
the company do work, and I think that has

169
00:09:50.820 --> 00:09:52.859
<v Kalim Saliba>to really change in order for AI to be
effective.

170
00:09:52.940 --> 00:09:54.315
<v Kalim Saliba>But I think there's Combination of things.

171
00:09:54.395 --> 00:09:56.265
<v Kalim Saliba>So, let's start with the AI models.

172
00:09:56.515 --> 00:09:58.036
<v Kalim Saliba>There've been massive, massive
advancements.

173
00:09:58.075 --> 00:09:59.306
<v Kalim Saliba>You you mentioned alpha fold.

174
00:09:59.356 --> 00:10:01.075
<v Kalim Saliba>There's also bolts, RF diffusion.

175
00:10:01.315 --> 00:10:04.895
<v Kalim Saliba>There, there's a growing number of
increasingly capable models.

176
00:10:05.106 --> 00:10:08.226
<v Kalim Saliba>And I think the first thing that has to
happen is the accuracy of these models and

177
00:10:08.226 --> 00:10:10.466
<v Kalim Saliba>the capability of these models has to
improve.

178
00:10:10.515 --> 00:10:14.046
<v Kalim Saliba>And I think as you were alluding to, and
this is already starting to happen, moving

179
00:10:14.046 --> 00:10:18.116
<v Kalim Saliba>beyondtru prediction into things that are
more tantamount to simulating aspects of

180
00:10:18.116 --> 00:10:20.716
<v Kalim Saliba>the biological system, things like binding
affinity.

181
00:10:20.882 --> 00:10:25.452
<v Kalim Saliba>selectivity, really understanding like
where does the liability, which would be the

182
00:10:25.452 --> 00:10:29.731
<v Kalim Saliba>thing within a protein or therapeutic that
might cause off-target effects or not be

183
00:10:29.731 --> 00:10:34.411
<v Kalim Saliba>as selective as as you want, how do you
actually have a model predict what that

184
00:10:34.411 --> 00:10:34.831
<v Kalim Saliba>looks like?

185
00:10:34.932 --> 00:10:39.711
<v Kalim Saliba>How do you have a model predict the
ability of a of a molecule to bind to your

186
00:10:39.711 --> 00:10:40.432
<v Kalim Saliba>target of interest?

187
00:10:40.531 --> 00:10:45.372
<v Kalim Saliba>How do you use models to look for targets
and expand on our understanding of

188
00:10:45.372 --> 00:10:49.607
<v Kalim Saliba>biological pathways, which is a
fundamental part of Really any drug discovery

189
00:10:49.607 --> 00:10:50.098
<v Kalim Saliba>pipeline.

190
00:10:50.148 --> 00:10:54.918
<v Kalim Saliba>So I think that's number one, like
continuous advancement of the models, and I do

191
00:10:54.918 --> 00:10:58.867
<v Kalim Saliba>truly believe we're going to see
advancements as a result of a lot of the, you know,

192
00:10:59.067 --> 00:11:03.427
<v Kalim Saliba>change and rapid improvement in these
frontier models that are going to continue to

193
00:11:03.427 --> 00:11:05.447
<v Kalim Saliba>surprise us in terms of how capable they
are.

194
00:11:05.668 --> 00:11:08.947
<v Kalim Saliba>But then I think the other big thing
that's needed and that frankly is, is kind of

195
00:11:08.947 --> 00:11:12.507
<v Kalim Saliba>missing and that we're trying to resolve
is that the software systems have to

196
00:11:12.507 --> 00:11:16.914
<v Kalim Saliba>support the follow through, because if you
think about how AI is used in these

197
00:11:16.914 --> 00:11:20.564
<v Kalim Saliba>pipelines, it's at discrete points in the
process.

198
00:11:20.883 --> 00:11:27.083
<v Kalim Saliba>You're not really sending AI a prompt and
saying, hey, go discover, make tests,

199
00:11:27.283 --> 00:11:32.463
<v Kalim Saliba>analyze, and do a tech transfer and then
manufacture the drug.

200
00:11:32.804 --> 00:11:36.484
<v Kalim Saliba>That'd be lovely if you could have one
prompt that did all of that, but that's just

201
00:11:36.484 --> 00:11:37.314
<v Kalim Saliba>not how it works.

202
00:11:37.604 --> 00:11:41.763
<v Kalim Saliba>So really, I think that the market and our
customers are thinking about how do we

203
00:11:41.763 --> 00:11:42.723
<v Kalim Saliba>insert AI.

204
00:11:43.359 --> 00:11:49.640
<v Kalim Saliba>Within software systems that allow us to
orchestrate its use in a way where we can

205
00:11:49.640 --> 00:11:54.359
<v Kalim Saliba>harness what it's good at, but make more
predictable that which it's not.

206
00:11:54.489 --> 00:11:58.119
<v Kalim Saliba>And I think all of those things really
have to come to play in order for the space

207
00:11:58.119 --> 00:12:00.380
<v Kalim Saliba>to see some more of the transformative.

208
00:12:00.599 --> 00:12:02.239
<v Kalim Saliba>I'll say transformative end to end.

209
00:12:02.719 --> 00:12:03.969
<v Kalim Saliba>Capabilities that we're looking for.

210
00:12:04.010 --> 00:12:08.849
<v Kalim Saliba>There are plenty of transformative
examples of AI being used to predict structure

211
00:12:08.849 --> 00:12:13.630
<v Kalim Saliba>better or with significantly higher
performance or to do binding infinity

212
00:12:13.849 --> 00:12:18.960
<v Kalim Saliba>calculations or to even do de novo protein
design like Bolt'sGen, for example, uses

213
00:12:18.960 --> 00:12:21.450
<v Kalim Saliba>generative model to actually predict the
protein.

214
00:12:21.989 --> 00:12:26.549
<v Kalim Saliba>That will uh feasibly be able to address a
particular target, but again, that's just

215
00:12:26.549 --> 00:12:30.989
<v Kalim Saliba>one part of many, many, many, many steps
that have to happen to validate and

216
00:12:30.989 --> 00:12:35.270
<v Kalim Saliba>understand whether or not that thing is
therapeutically significant and viable or

217
00:12:35.270 --> 00:12:35.590
<v Kalim Saliba>not.

218
00:12:35.950 --> 00:12:40.140
<v Speaker 3>Very well put and also interesting how you
described like that end to end process

219
00:12:40.140 --> 00:12:44.669
<v Speaker 3>and how AI can can accelerate that and
where it really can unfold a lot of value.

220
00:12:44.900 --> 00:12:47.150
<v Speaker 3>Do you feel like there's certain
organizational structure?

221
00:12:47.674 --> 00:12:53.335
<v Speaker 3>um or setups that are have like greater
benefits in that space like to win that game

222
00:12:53.335 --> 00:12:57.075
<v Speaker 3>and I'm thinking about, I don't know, we
hear a lot about like the AI native

223
00:12:57.075 --> 00:13:01.914
<v Speaker 3>companies, a lot of startups out there, as
opposed to like maybe big pharma tech

224
00:13:01.914 --> 00:13:02.515
<v Speaker 3>companies.

225
00:13:02.864 --> 00:13:07.784
<v Speaker 3>Do you have a feel on like who will be
able to actually translate that AI

226
00:13:07.784 --> 00:13:12.594
<v Speaker 3>accelerated end to end pipeline in drug
discovery quicker or better?

227
00:13:13.049 --> 00:13:19.250
<v Kalim Saliba>Yeah, it's a good question, and I think um
it's hard to say for sure which style of

228
00:13:19.250 --> 00:13:24.250
<v Kalim Saliba>company is going to lead to the most
disruption, but I believe that for sure there

229
00:13:24.250 --> 00:13:25.330
<v Kalim Saliba>will be disruption, right?

230
00:13:25.489 --> 00:13:29.330
<v Kalim Saliba>It's a question of where is it going to
come from, and it may not come from one or

231
00:13:29.330 --> 00:13:29.650
<v Kalim Saliba>the other.

232
00:13:29.729 --> 00:13:34.169
<v Kalim Saliba>I think if you look at a lot of the AI
native or AI focused startups in the space,

233
00:13:34.289 --> 00:13:36.890
<v Kalim Saliba>there's typically heavy involvement from
pharma, right?

234
00:13:36.960 --> 00:13:38.010
<v Kalim Saliba>Either the pharma is providing.

235
00:13:38.578 --> 00:13:42.918
<v Kalim Saliba>information about the targets and pathways
or providing funding to receive

236
00:13:43.538 --> 00:13:47.807
<v Kalim Saliba>understanding or maps that increase their
understanding of the biological system or

237
00:13:47.807 --> 00:13:50.617
<v Kalim Saliba>in some cases, you know, out licensing
some of the molecules.

238
00:13:50.937 --> 00:13:56.257
<v Kalim Saliba>So there is a very strong relationship
between pharma and some of these AI native

239
00:13:56.257 --> 00:13:56.978
<v Kalim Saliba>startups that we're seeing.

240
00:13:57.018 --> 00:14:01.328
<v Kalim Saliba>Like for example, recursion does a lot of
work on disease biology, doing experiments

241
00:14:01.328 --> 00:14:03.218
<v Kalim Saliba>on cells and then imaging them to help.

242
00:14:03.366 --> 00:14:07.265
<v Kalim Saliba>Increase the understanding of how the
biology works, and they have tons and tons of

243
00:14:07.265 --> 00:14:07.846
<v Kalim Saliba>data on that.

244
00:14:07.986 --> 00:14:10.546
<v Kalim Saliba>They're working heavily with Roche,
Genentech, and Sanofi.

245
00:14:10.736 --> 00:14:14.466
<v Kalim Saliba>Isomorphic is interesting because you
mentioned Alphafold at the beginning of the

246
00:14:14.466 --> 00:14:14.726
<v Kalim Saliba>call.

247
00:14:14.786 --> 00:14:19.526
<v Kalim Saliba>It's a DeepMind spin-off from Google,
which is essentially they took Alpha Fold 3

248
00:14:19.745 --> 00:14:21.385
<v Kalim Saliba>and improved upon it significantly.

249
00:14:21.505 --> 00:14:25.385
<v Kalim Saliba>So they are the most computationally
focused, but they're relying on partnerships

250
00:14:25.385 --> 00:14:28.466
<v Kalim Saliba>with Lilly, Novartis, and most recently J
and J, you know.

251
00:14:28.554 --> 00:14:32.513
<v Kalim Saliba>One of their models is to source targets
from these customers to work with them to

252
00:14:32.513 --> 00:14:37.013
<v Kalim Saliba>help build out using their computational
pipeline, potential candidate molecules.

253
00:14:37.244 --> 00:14:38.914
<v Kalim Saliba>The most interesting one is probably in
silica.

254
00:14:39.093 --> 00:14:40.133
<v Kalim Saliba>They're they're the furthest along.

255
00:14:40.273 --> 00:14:44.664
<v Kalim Saliba>You may have heard about them as, as I
think in early September, they just entered

256
00:14:44.664 --> 00:14:49.223
<v Kalim Saliba>phase 3 clinical trials for one of their
pulmonary fibrosis drugs.

257
00:14:49.354 --> 00:14:53.593
<v Kalim Saliba>So this would be, if it's successful, the
first end to end AI discovered.

258
00:14:53.802 --> 00:14:55.262
<v Kalim Saliba>Drug that's brought to market.

259
00:14:55.562 --> 00:14:57.611
<v Kalim Saliba>So lots of exciting stuff is happening,
right?

260
00:14:57.692 --> 00:15:02.081
<v Kalim Saliba>And like different companies have
different approaches, but in all cases, there

261
00:15:02.081 --> 00:15:05.122
<v Kalim Saliba>appears to be heavy involvement with
pharma, and that's not all that surprising

262
00:15:05.122 --> 00:15:08.041
<v Kalim Saliba>because these pharma companies are the
ones that have all of the data.

263
00:15:08.322 --> 00:15:11.031
<v Kalim Saliba>They have all of the relationships
cataloged.

264
00:15:11.081 --> 00:15:16.041
<v Kalim Saliba>They have like decades of experience in
understanding disease biology and

265
00:15:16.041 --> 00:15:18.432
<v Kalim Saliba>information about targets and what has
worked and what hasn't worked.

266
00:15:18.442 --> 00:15:23.690
<v Kalim Saliba>And so, I don't know exactly which style
is best for the time being, I believe it's

267
00:15:23.690 --> 00:15:28.750
<v Kalim Saliba>more about making sure folks work together
to bring the best of what each one is

268
00:15:28.750 --> 00:15:30.909
<v Kalim Saliba>able to deliver, to bring something to
market.

269
00:15:31.229 --> 00:15:33.770
<v Speaker 2>Do you think they speak the same language?

270
00:15:33.869 --> 00:15:38.429
<v Speaker 2>So usually, right, traditionally you have
a little bit of the designs back, you have

271
00:15:38.429 --> 00:15:41.669
<v Speaker 2>the task, you have model, these are
different personas, these are different

272
00:15:41.669 --> 00:15:44.859
<v Speaker 2>scientific backgrounds or educational
backgrounds even, right?

273
00:15:44.950 --> 00:15:47.270
<v Speaker 2>So discovery has lifted, you know, the
chemist.

274
00:15:47.669 --> 00:15:51.270
<v Speaker 2>Design the components, the people running
the lessons and screens, and then you have

275
00:15:51.270 --> 00:15:55.859
<v Speaker 2>the modelers who are rather computational
work, let's say, and then, you know, you

276
00:15:55.859 --> 00:16:00.390
<v Speaker 2>have this interface between prediction and
the lab result, how to compare it and how

277
00:16:00.390 --> 00:16:01.700
<v Speaker 2>to speak actually the language.

278
00:16:01.830 --> 00:16:08.549
<v Speaker 2>Do you think AI will, you know, enable
this kinds of or is everyone moving now into

279
00:16:08.549 --> 00:16:13.179
<v Speaker 2>computational design, computational
modeling, computational this, so AI is the

280
00:16:13.179 --> 00:16:15.460
<v Speaker 2>world, Colleen, in these kinds of worlds.

281
00:16:16.679 --> 00:16:21.859
<v Kalim Saliba>I think we're probably a ways away from AI
doing everything, but I believe we're

282
00:16:21.859 --> 00:16:26.690
<v Kalim Saliba>closer than people realize to the market
trying to address higher degrees of

283
00:16:26.690 --> 00:16:30.000
<v Kalim Saliba>autonomy in the way agentic systems work,
right?

284
00:16:30.099 --> 00:16:33.960
<v Kalim Saliba>And I think that's in part due to the
massive advancements in all of these models

285
00:16:34.099 --> 00:16:39.179
<v Kalim Saliba>and the increasing focus on orchestration
and making sure like the Discovery

286
00:16:39.179 --> 00:16:44.020
<v Kalim Saliba>pipelines are well engineered systems that
are able to kind of employ.

287
00:16:44.234 --> 00:16:49.544
<v Kalim Saliba>AI and humans across the design make test
to decide cycle at the right opportune

288
00:16:49.544 --> 00:16:49.994
<v Kalim Saliba>points.

289
00:16:50.265 --> 00:16:52.625
<v Kalim Saliba>In terms of, you know, is it the same
language?

290
00:16:52.664 --> 00:16:58.385
<v Kalim Saliba>Like there couldn't be more variability in
how science is, is discussed and how it's

291
00:16:58.385 --> 00:17:02.655
<v Kalim Saliba>interpreted across these different phases
of research, development, and

292
00:17:02.655 --> 00:17:07.025
<v Kalim Saliba>manufacturing, and it very much is a
different language, and I'll, I'll give you a

293
00:17:07.025 --> 00:17:11.505
<v Kalim Saliba>small personal anecdote, you know, as, as
an engineer by training, I joined the

294
00:17:11.505 --> 00:17:11.785
<v Kalim Saliba>company.

295
00:17:12.030 --> 00:17:17.010
<v Kalim Saliba>And I think naively thought that learning
the science, working with my colleagues to

296
00:17:17.510 --> 00:17:22.260
<v Kalim Saliba>understand where are some of the greatest
problems would be like applying, you know,

297
00:17:22.310 --> 00:17:27.859
<v Kalim Saliba>an engineering mindset to understanding
the concepts, the language, the relationship

298
00:17:27.859 --> 00:17:28.670
<v Kalim Saliba>between those concepts.

299
00:17:28.739 --> 00:17:33.209
<v Kalim Saliba>But there are things that are very
fundamental things in the space that don't even

300
00:17:33.430 --> 00:17:36.550
<v Kalim Saliba>called the same thing by the market or our
customers.

301
00:17:36.790 --> 00:17:39.550
<v Kalim Saliba>And so one of the things that emerges as,
as a really important.

302
00:17:39.714 --> 00:17:44.344
<v Kalim Saliba>Characteristics of these systems and in
driving, maybe not necessarily commonality,

303
00:17:44.395 --> 00:17:50.114
<v Kalim Saliba>but interpretability across the full life
cycle are ontologies, right, that allow

304
00:17:50.114 --> 00:17:55.515
<v Kalim Saliba>our customers to specialize the science to
the way that they think about, talk

305
00:17:55.515 --> 00:17:58.015
<v Kalim Saliba>about, rationalize, conceptualize the
work.

306
00:17:58.155 --> 00:18:00.704
<v Kalim Saliba>And I think without that, it's very hard
because without that, sure, you have data,

307
00:18:00.755 --> 00:18:04.555
<v Kalim Saliba>you can put it all in one place, but if
you can't understand, you know, how does

308
00:18:04.555 --> 00:18:11.109
<v Kalim Saliba>design A actually relate to Protein panel
B relate to assay

309
00:18:11.109 --> 00:18:16.040
<v Kalim Saliba>results C and then the decisions that
derive from that and all the things that have

310
00:18:16.040 --> 00:18:19.510
<v Kalim Saliba>to be described in order to analyze and
understand that data.

311
00:18:19.640 --> 00:18:21.839
<v Kalim Saliba>It's very hard for people, even harder for
AI.

312
00:18:22.239 --> 00:18:25.920
<v Speaker 3>And it's interesting, right, because you
still have to put in so much groundwork to

313
00:18:25.920 --> 00:18:30.040
<v Speaker 3>actually establish that and make those
connections and then lay that foundation

314
00:18:30.040 --> 00:18:34.699
<v Speaker 3>before you can actually start to like reap
the benefits that AI can enable.

315
00:18:34.839 --> 00:18:39.564
<v Speaker 3>And I guess It's like another aspect that
um always comes up is like the question

316
00:18:39.564 --> 00:18:40.994
<v Speaker 3>around like data quality.

317
00:18:41.244 --> 00:18:46.285
<v Speaker 3>I assume that there's lots of context and
like data being generated um throughout

318
00:18:46.285 --> 00:18:49.435
<v Speaker 3>like all the testing that you have to do,
a lot of the screening.

319
00:18:49.685 --> 00:18:54.805
<v Speaker 3>How do you enforce that like the data
quality is always top notch and that we don't

320
00:18:54.805 --> 00:19:00.564
<v Speaker 3>like end up building on top of like poor
data quality and then deteriorating um the

321
00:19:00.564 --> 00:19:02.854
<v Speaker 3>whole system that we're trying to build.

322
00:19:03.329 --> 00:19:08.540
<v Kalim Saliba>I think one of the biggest data quality
problems the industry has right now is one

323
00:19:08.540 --> 00:19:11.380
<v Kalim Saliba>of context and missing contexts, right?

324
00:19:11.489 --> 00:19:15.180
<v Kalim Saliba>Cause even if you have data and you're
sure it's the right data and it's represented

325
00:19:15.180 --> 00:19:21.300
<v Kalim Saliba>correctly, if it's not contextualized
within the appropriate surrounding data, it's

326
00:19:21.300 --> 00:19:23.420
<v Kalim Saliba>very hard to really understand what you're
supposed to do with it.

327
00:19:23.500 --> 00:19:24.880
<v Kalim Saliba>And I'll, I'll give you a practical
example.

328
00:19:25.060 --> 00:19:29.060
<v Kalim Saliba>So the, the traditional buying Categories
in pharma, and there are many of them.

329
00:19:29.180 --> 00:19:32.619
<v Kalim Saliba>I'll I'll just reference a few are things
like registration systems, right, where

330
00:19:32.619 --> 00:19:36.300
<v Kalim Saliba>intellectual property is tracked and
registered where the molecules, knowledge of

331
00:19:36.300 --> 00:19:40.420
<v Kalim Saliba>the molecules themselves and their
uniqueness is actually recorded, very important

332
00:19:40.420 --> 00:19:40.900
<v Kalim Saliba>systems.

333
00:19:41.140 --> 00:19:45.219
<v Kalim Saliba>And then you have electronic lab notebooks
or ELNs where the experimental work

334
00:19:45.219 --> 00:19:48.900
<v Kalim Saliba>happens, the protocol is established, the
work is signed or countersigned.

335
00:19:49.219 --> 00:19:53.699
<v Kalim Saliba>And there's just this record of what was
done and what were the results.

336
00:19:53.989 --> 00:19:58.270
<v Kalim Saliba>And then you have limb systems, which are
all about, like, samples of the molecule,

337
00:19:58.390 --> 00:20:02.630
<v Kalim Saliba>and, you know, where, how are they being
tracked and routed and tested, and what

338
00:20:02.630 --> 00:20:04.410
<v Kalim Saliba>inventory systems are they coming from.

339
00:20:04.729 --> 00:20:07.810
<v Kalim Saliba>And all these things are traditionally
separate and disparate, and when you have

340
00:20:07.810 --> 00:20:11.599
<v Kalim Saliba>silos in software, you end up having silos
in data, and silos in the way that you

341
00:20:11.599 --> 00:20:14.589
<v Kalim Saliba>work, and missing context across those
silos.

342
00:20:14.849 --> 00:20:20.000
<v Kalim Saliba>And so, really, I think what needs to
happen, and what very much is happening, is

343
00:20:20.000 --> 00:20:22.609
<v Kalim Saliba>that the lines between these traditionally
discrete software.

344
00:20:22.694 --> 00:20:25.614
<v Kalim Saliba>Categories are starting to blur very
intentionally.

345
00:20:25.675 --> 00:20:29.775
<v Kalim Saliba>Like we're doing it very intentionally to
make sure that when you are orchestrating

346
00:20:29.775 --> 00:20:35.675
<v Kalim Saliba>a piece of work across a campaign for a
specific target that has a set of designs

347
00:20:35.935 --> 00:20:39.655
<v Kalim Saliba>that has instructions for how to make it
in the lab, that has specific assays that

348
00:20:39.655 --> 00:20:40.435
<v Kalim Saliba>you want to run.

349
00:20:41.099 --> 00:20:44.469
<v Kalim Saliba>You're not bouncing between systems to do
that, right?

350
00:20:44.579 --> 00:20:50.420
<v Kalim Saliba>There is one orchestrated view that blends
aspects of samples and experiments and

351
00:20:50.420 --> 00:20:55.660
<v Kalim Saliba>registered compounds within the same
orchestrated workflows, which gets you this

352
00:20:55.660 --> 00:20:56.099
<v Kalim Saliba>really clear.

353
00:20:56.189 --> 00:21:01.709
<v Kalim Saliba>Their lineage or digital thread, if you
will, from the inception of an idea all the

354
00:21:01.709 --> 00:21:05.670
<v Kalim Saliba>way through to its conclusion, whether
that is advancing it forward or revising the

355
00:21:05.670 --> 00:21:07.609
<v Kalim Saliba>hypothesis based on what was learned.

356
00:21:07.869 --> 00:21:12.030
<v Kalim Saliba>That lineage, lineage is another word for
context, right?

357
00:21:12.189 --> 00:21:17.030
<v Kalim Saliba>There's contextual linkages between these
data sets that ultimately drives the

358
00:21:17.030 --> 00:21:18.520
<v Kalim Saliba>understanding of the science and what
works.

359
00:21:18.619 --> 00:21:21.979
<v Kalim Saliba>So, data hygiene, data quality, very
important.

360
00:21:22.229 --> 00:21:26.069
<v Kalim Saliba>The most important dimension of that is
context, the glue.

361
00:21:26.530 --> 00:21:28.099
<v Speaker 2>Uh you mentioned software, right?

362
00:21:28.260 --> 00:21:33.339
<v Speaker 2>So I think it's one of the aspects I guess
is, is also in software, it's easy to

363
00:21:33.339 --> 00:21:38.130
<v Speaker 2>make, I don't know, a repro clone, right,
in terms of, you know, copycatting, right,

364
00:21:38.140 --> 00:21:39.689
<v Speaker 2>and and reproduce, right?

365
00:21:39.780 --> 00:21:41.650
<v Speaker 2>Running integration tests, right?

366
00:21:41.699 --> 00:21:46.699
<v Speaker 2>And Uh, copy a project, one command, you
know, and then run, run the integration

367
00:21:46.699 --> 00:21:48.400
<v Speaker 2>test is like, OK, it still works, right?

368
00:21:48.540 --> 00:21:52.859
<v Speaker 2>Still runs, I guess, you know, for a lab
experiment in your environment, that's,

369
00:21:53.099 --> 00:21:55.209
<v Speaker 2>that's still a learning journey to go
ahead.

370
00:21:55.420 --> 00:21:58.150
<v Speaker 2>You, you have a great sub stack, right,
for people out there, right?

371
00:21:58.219 --> 00:22:05.140
<v Speaker 2>I guess check out Kunal stack, right?

372
00:22:06.010 --> 00:22:09.310
<v Kalim Saliba>There are better ones out there, but check
it out if you want.

373
00:22:10.589 --> 00:22:15.510
<v Speaker 2>So, and you alluded a bit of, you know,
in, in terms of about Anthropic protein

374
00:22:15.510 --> 00:22:20.030
<v Speaker 2>design experiments and cloud and the role
of, you know, 16,000 prompts I guess

375
00:22:20.030 --> 00:22:25.150
<v Speaker 2>around this, right, where its own scripts
and you, you, your point was, right, it

376
00:22:25.150 --> 00:22:26.790
<v Speaker 2>can be super cool in creative stuff.

377
00:22:26.905 --> 00:22:28.185
<v Speaker 2>Maybe it could be, I'm not sure.

378
00:22:28.265 --> 00:22:32.224
<v Speaker 2>I'm happy to hear your thoughts, but it's,
how do we make a reliable product on a

379
00:22:32.224 --> 00:22:35.064
<v Speaker 2>different note, right on a on a on a
different scale.

380
00:22:35.464 --> 00:22:36.744
<v Speaker 2>So what's your view on that?

381
00:22:36.824 --> 00:22:40.864
<v Speaker 2>Can you a bit of allude to it and what
does it mean actually building let's say

382
00:22:40.864 --> 00:22:42.145
<v Speaker 2>agent workflows, right?

383
00:22:42.265 --> 00:22:47.625
<v Speaker 2>Making sure that, you know, these are
integrated in the rigid purpose of, you know,

384
00:22:47.785 --> 00:22:51.604
<v Speaker 2>science, trusted, right, experimentation
track.

385
00:22:52.079 --> 00:22:55.000
<v Kalim Saliba>Yeah, that's a good question, but
particularly because there's just so much

386
00:22:55.000 --> 00:22:57.819
<v Kalim Saliba>incredible innovation happening in the
space.

387
00:22:57.890 --> 00:23:00.880
<v Kalim Saliba>Like, it's like every week seemingly
something changes in the AI world, and

388
00:23:00.880 --> 00:23:05.459
<v Kalim Saliba>anthropic or open AI are doing something
incredible, or scary, or sometimes both.

389
00:23:07.219 --> 00:23:07.930
<v Kalim Saliba>Sometimes both.

390
00:23:08.260 --> 00:23:10.430
<v Kalim Saliba>It's like it's a constant debate between
me and some of my colleagues.

391
00:23:10.469 --> 00:23:13.900
<v Kalim Saliba>Are we excited today or are we scared
today about what's happening?

392
00:23:14.109 --> 00:23:15.119
<v Kalim Saliba>What's the feeling today?

393
00:23:15.270 --> 00:23:15.900
<v Kalim Saliba>Excited.

394
00:23:15.989 --> 00:23:20.229
<v Kalim Saliba>I'm, I'm, I tend to be more excited these
days, but who, who knows that that could

395
00:23:20.229 --> 00:23:20.630
<v Kalim Saliba>change.

396
00:23:20.839 --> 00:23:24.630
<v Kalim Saliba>I think on the topic of, of some of that
innovation, and I guess we can talk maybe

397
00:23:24.630 --> 00:23:28.270
<v Kalim Saliba>more specifically about what Anthropic
did, there's a lot of research going on to

398
00:23:28.270 --> 00:23:30.530
<v Kalim Saliba>really understand how capable are these
models.

399
00:23:30.709 --> 00:23:34.589
<v Kalim Saliba>It's starting to go beyond that, but I
think for the moment, the majority of the

400
00:23:34.589 --> 00:23:36.550
<v Kalim Saliba>research is focused on how capable.

401
00:23:36.816 --> 00:23:37.495
<v Kalim Saliba>is the model?

402
00:23:37.816 --> 00:23:38.765
<v Kalim Saliba>What can it do?

403
00:23:38.975 --> 00:23:40.046
<v Kalim Saliba>Where does it get stuck?

404
00:23:40.296 --> 00:23:41.186
<v Kalim Saliba>How accurate is it?

405
00:23:41.255 --> 00:23:47.056
<v Kalim Saliba>How much better is it than what humans can
do to really gauge and assess where are

406
00:23:47.056 --> 00:23:51.696
<v Kalim Saliba>its strengths and weak points for the
purposes of improving the model, right?

407
00:23:51.776 --> 00:23:54.406
<v Kalim Saliba>Because if you think about what these
frontier companies are doing, they're doing a

408
00:23:54.406 --> 00:23:58.255
<v Kalim Saliba>lot, but the most important existential
thing that they need to be focused on is

409
00:23:58.255 --> 00:23:59.316
<v Kalim Saliba>having the best model.

410
00:23:59.576 --> 00:24:06.456
<v Kalim Saliba>And one of the greatest areas of untapped
human potential is in human biology and

411
00:24:06.456 --> 00:24:11.962
<v Kalim Saliba>being Able to design, you know, very like
precise or personalized medicines at an

412
00:24:11.962 --> 00:24:16.082
<v Kalim Saliba>effective cost that, you know, is just
unfathomable today with how expensive it is

413
00:24:16.082 --> 00:24:20.881
<v Kalim Saliba>to do these research projects or being
able to, you know, identify how to drug, you

414
00:24:20.881 --> 00:24:25.232
<v Kalim Saliba>know, traditionally undruggable targets
based on insights that AI is able to

415
00:24:25.232 --> 00:24:25.562
<v Kalim Saliba>discover.

416
00:24:25.682 --> 00:24:29.202
<v Kalim Saliba>So there's the reason I say all that is
because there's lots of opportunity and

417
00:24:29.202 --> 00:24:33.222
<v Kalim Saliba>potential, and it's not just opportunity
and potential that folks in life sciences

418
00:24:33.222 --> 00:24:34.942
<v Kalim Saliba>and material sciences see.

419
00:24:35.121 --> 00:24:39.547
<v Kalim Saliba>It's potential that I think is Very
relatable at a human level because we all have

420
00:24:39.547 --> 00:24:44.218
<v Kalim Saliba>experience in one form or another with
illness and hardship and like, you know,

421
00:24:44.307 --> 00:24:48.167
<v Kalim Saliba>people in our lives either that we know
directly or through friends or family that

422
00:24:48.167 --> 00:24:50.647
<v Kalim Saliba>have been impacted in some way by disease.

423
00:24:50.708 --> 00:24:53.537
<v Kalim Saliba>And so it's not surprising to me that
there's so much focus on this.

424
00:24:53.748 --> 00:24:58.547
<v Kalim Saliba>I think what it means though is looking at
how to extract the results from these

425
00:24:58.547 --> 00:25:02.647
<v Kalim Saliba>types of research projects into more
operable pipelines.

426
00:25:02.907 --> 00:25:06.387
<v Kalim Saliba>And so what Anthropic did is they had this
study on um called binder design.

427
00:25:06.494 --> 00:25:11.014
<v Kalim Saliba>Essentially they're they're designing an
aspect of a protein to bind to a specific

428
00:25:11.014 --> 00:25:11.394
<v Kalim Saliba>target.

429
00:25:11.654 --> 00:25:16.173
<v Kalim Saliba>These were known targets, and they were
trying to assess how good of a binder could

430
00:25:16.173 --> 00:25:18.644
<v Kalim Saliba>it design for these known targets.

431
00:25:18.843 --> 00:25:24.734
<v Kalim Saliba>And in about 14 of the 16 targets, it was
able to design binders around double the

432
00:25:24.734 --> 00:25:27.693
<v Kalim Saliba>typical success rate that's seen from
human campaigns.

433
00:25:27.894 --> 00:25:30.813
<v Kalim Saliba>But there are some big caveats to that,
right?

434
00:25:31.014 --> 00:25:32.734
<v Kalim Saliba>And I highly encourage folks to go read
it.

435
00:25:32.813 --> 00:25:33.413
<v Kalim Saliba>They weren't kidding.

436
00:25:33.453 --> 00:25:36.293
<v Kalim Saliba>There is a 16,000 word prompt that you can
go.

437
00:25:36.770 --> 00:25:41.469
<v Kalim Saliba>That encodes everything from the
scientific guidance that was expected for this

438
00:25:41.469 --> 00:25:46.239
<v Kalim Saliba>experiment to the operational
characteristics, to the tools that it could use to

439
00:25:46.239 --> 00:25:49.079
<v Kalim Saliba>scoring criteria, and it could use tools,
right?

440
00:25:49.119 --> 00:25:53.800
<v Kalim Saliba>If you call models, it could use open
source techniques and technologies, but

441
00:25:53.800 --> 00:25:55.250
<v Kalim Saliba>everything else it had to build, right?

442
00:25:55.500 --> 00:25:57.640
<v Kalim Saliba>It needed to build its own orchestration
pipeline.

443
00:25:57.680 --> 00:26:00.920
<v Kalim Saliba>It needed to build its own scoring parts
of the software.

444
00:26:00.959 --> 00:26:05.780
<v Kalim Saliba>It needed to do all of the wiring between
the open source tooling and the workflow

445
00:26:05.780 --> 00:26:06.280
<v Kalim Saliba>pipeline.

446
00:26:06.595 --> 00:26:10.265
<v Kalim Saliba>And that's interesting on the one hand
because it shows just how capable the thing

447
00:26:10.265 --> 00:26:14.225
<v Kalim Saliba>is, but then if you think about how
science works, right, and how the scientific

448
00:26:14.225 --> 00:26:19.505
<v Kalim Saliba>method itself, like what it's actually
founded upon, which is iteration and trial

449
00:26:19.505 --> 00:26:24.505
<v Kalim Saliba>and error, and being able to very clearly
distinguish your dependent and independent

450
00:26:24.505 --> 00:26:30.365
<v Kalim Saliba>variables from each run of an experiment,
there isn't really any room for inference

451
00:26:30.576 --> 00:26:31.385
<v Kalim Saliba>in the pipeline.

452
00:26:31.666 --> 00:26:36.225
<v Kalim Saliba>So, if you want to say, for example,
change one aspect of the test.

453
00:26:36.332 --> 00:26:40.862
<v Kalim Saliba>That was run with um with what anthropic
did, the way to do that would be to go find

454
00:26:40.862 --> 00:26:45.171
<v Kalim Saliba>the words in the prompt and change the
words and then hope that it infers the

455
00:26:45.171 --> 00:26:48.452
<v Kalim Saliba>correct update to the, and it's a big
challenge because it's not predictable, it's

456
00:26:48.452 --> 00:26:54.171
<v Kalim Saliba>not deterministic, and it makes even those
bits that don't need to be inferred, it

457
00:26:54.171 --> 00:26:55.052
<v Kalim Saliba>makes them inferred.

458
00:26:55.241 --> 00:26:58.771
<v Kalim Saliba>By the way, I don't think they were trying
to build a like a deterministic.

459
00:26:58.901 --> 00:27:04.031
<v Kalim Saliba>I don't think that was the objective of
that experiment, but it's important for

460
00:27:04.291 --> 00:27:06.171
<v Kalim Saliba>those listening in the industry and
customers to.

461
00:27:06.317 --> 00:27:08.767
<v Kalim Saliba>Recognize because it's easy to look at
that and say that's the future.

462
00:27:09.078 --> 00:27:10.248
<v Kalim Saliba>That's part of the future.

463
00:27:10.598 --> 00:27:15.657
<v Kalim Saliba>But really to harness that part of the
future in a predictable way, you need to know

464
00:27:16.037 --> 00:27:21.478
<v Kalim Saliba>how to insert that tremendously capable
judgment and execution capability within a

465
00:27:21.478 --> 00:27:27.998
<v Kalim Saliba>durable harness that allows the AI to make
and or execute within very,

466
00:27:28.078 --> 00:27:32.098
<v Kalim Saliba>very specific changes to deterministic
pipelines.

467
00:27:32.358 --> 00:27:34.437
<v Kalim Saliba>Otherwise, you don't know what's changing.

468
00:27:34.508 --> 00:27:36.118
<v Kalim Saliba>You, you, you kind of will lose track.

469
00:27:36.343 --> 00:27:41.453
<v Kalim Saliba>About what's changing and what's staying
fixed on each subsequent run, and what that

470
00:27:41.463 --> 00:27:45.183
<v Kalim Saliba>means at a fundamental level is that
there's very little you can conclude about the

471
00:27:45.183 --> 00:27:49.223
<v Kalim Saliba>experiment at that point regarding what
did or did not work, unless it's a fully

472
00:27:49.223 --> 00:27:53.303
<v Kalim Saliba>closed loop and you walk away and you say,
hey, like go do this thing in its

473
00:27:53.303 --> 00:27:53.734
<v Kalim Saliba>entirety.

474
00:27:53.973 --> 00:27:56.504
<v Kalim Saliba>The reality is there are still scientists,
right?

475
00:27:56.543 --> 00:28:00.624
<v Kalim Saliba>And those scientists still add tremendous
value and they have to be inserted at the

476
00:28:00.624 --> 00:28:04.784
<v Kalim Saliba>correct parts of these pipelines in order
to kind of drive these increasingly

477
00:28:04.784 --> 00:28:06.104
<v Kalim Saliba>autonomous labs.

478
00:28:06.449 --> 00:28:12.699
<v Speaker 2>But isn't that that to check, so it's now
then end up in a complexity in the

479
00:28:12.699 --> 00:28:16.540
<v Speaker 2>software engineering and modeling the
engineering process getting right because it's

480
00:28:16.540 --> 00:28:21.939
<v Speaker 2>a one-off in terms of, you know, fully
loaded prompt aspects, right, which we now

481
00:28:21.939 --> 00:28:24.560
<v Speaker 2>call in in AI harness and let's say.

482
00:28:24.780 --> 00:28:29.890
<v Speaker 2>Workflow, orchestration, orchestration of
different agents, you know, promoting,

483
00:28:30.060 --> 00:28:31.660
<v Speaker 2>subtasking it, right?

484
00:28:31.739 --> 00:28:34.319
<v Speaker 2>And, and slicing down the capabilities.

485
00:28:34.979 --> 00:28:39.260
<v Speaker 2>Isn't that maybe the precursors like, you
know, OK, now we need to computational

486
00:28:39.260 --> 00:28:43.900
<v Speaker 2>engineering basically or engineering with
computational logic behind it, that moves

487
00:28:43.900 --> 00:28:46.780
<v Speaker 2>them further or farther into the stream.

488
00:28:46.859 --> 00:28:51.290
<v Speaker 2>Would you anticipate that this is most
likely to pivot in this direction?

489
00:28:51.689 --> 00:28:53.439
<v Speaker 2>Smaller agents, smaller things.

490
00:28:54.170 --> 00:28:56.449
<v Kalim Saliba>I think it isn't just one thing, right?

491
00:28:56.479 --> 00:29:01.329
<v Kalim Saliba>I think what we'll see is an increasing
focus, and when I say harness, I don't

492
00:29:01.329 --> 00:29:04.550
<v Kalim Saliba>necessarily mean like the AI, the
underlying AI harness.

493
00:29:04.609 --> 00:29:08.449
<v Kalim Saliba>I mean more platform harness in which AI
runs, right?

494
00:29:08.569 --> 00:29:14.770
<v Kalim Saliba>So, like, if I have an orchestration
pipeline that does design of uh small molecules

495
00:29:14.770 --> 00:29:18.650
<v Kalim Saliba>or or biologic molecules, and then there's
a whole pipeline for how I go about

496
00:29:18.650 --> 00:29:19.709
<v Kalim Saliba>making those molecules.

497
00:29:20.472 --> 00:29:25.352
<v Kalim Saliba>assaying them, analyzing results from
them, that's a workflow, that's a process

498
00:29:25.352 --> 00:29:25.871
<v Kalim Saliba>model, right?

499
00:29:25.942 --> 00:29:31.212
<v Kalim Saliba>That has to actually be defined in order
to be reasoned about in a consistent way.

500
00:29:31.432 --> 00:29:33.592
<v Kalim Saliba>And AI can do a few things.

501
00:29:33.671 --> 00:29:39.312
<v Kalim Saliba>It can help build that model, it can
execute within that model, and it can help

502
00:29:39.312 --> 00:29:44.732
<v Kalim Saliba>interpret the data both at the start of
the pipeline, right, in terms of like

503
00:29:44.871 --> 00:29:45.712
<v Kalim Saliba>generating predictions.

504
00:29:46.104 --> 00:29:50.654
<v Kalim Saliba>And at the end of the pipeline, in the
case of analyzing results and determining

505
00:29:50.654 --> 00:29:53.413
<v Kalim Saliba>based on success or failure how to refine
those predictions.

506
00:29:53.614 --> 00:29:59.053
<v Kalim Saliba>The important part is to ensure that
whatever the orchestrator is running is that it

507
00:29:59.053 --> 00:30:04.024
<v Kalim Saliba>is not purely inference based, meaning if
your methodology for running a drug

508
00:30:04.024 --> 00:30:09.734
<v Kalim Saliba>discovery pipeline is a prompt and is not
an introspectable discrete workflow that a

509
00:30:09.734 --> 00:30:11.573
<v Kalim Saliba>human or another AI can go and.

510
00:30:11.676 --> 00:30:16.115
<v Kalim Saliba>Respect and modify in a targeted way, you
do not have a repeatable or durable

511
00:30:16.115 --> 00:30:16.796
<v Kalim Saliba>process, right?

512
00:30:16.836 --> 00:30:22.715
<v Kalim Saliba>And so it's about figuring out how to
unfetter AI within these really, really well

513
00:30:22.715 --> 00:30:27.556
<v Kalim Saliba>defined areas of science where you know
that they're going to be able to deliver

514
00:30:27.556 --> 00:30:33.156
<v Kalim Saliba>something at a scale and capability level
that is a massive accelerant, but still

515
00:30:33.156 --> 00:30:36.676
<v Kalim Saliba>governable and still executable within a
predictable pipeline.

516
00:30:36.946 --> 00:30:37.436
<v Kalim Saliba>Does that make sense?

517
00:30:37.848 --> 00:30:40.057
<v Speaker 3>It does an interesting view on that.

518
00:30:40.417 --> 00:30:44.738
<v Speaker 3>What you also just touched upon was like
that closed loop tipping point a little

519
00:30:44.738 --> 00:30:48.618
<v Speaker 3>bit, like, do you first of all think that
we're like close to getting there?

520
00:30:48.657 --> 00:30:51.248
<v Speaker 3>Is that even something like that is
desirable?

521
00:30:51.267 --> 00:30:54.858
<v Speaker 3>Because you also mentioned like we could
lose like track on like all the experiments

522
00:30:54.858 --> 00:30:59.537
<v Speaker 3>if, if there's no human in the loop
anymore and the AI just like self recursively

523
00:30:59.537 --> 00:31:04.739
<v Speaker 3>keeps improving, keeps on bringing up new
results, new tests, and Maybe lead, that

524
00:31:04.739 --> 00:31:08.300
<v Speaker 3>could lead to great discoveries, but on
the other hand, we might also lose control.

525
00:31:08.400 --> 00:31:09.530
<v Speaker 3>So what's your view on that?

526
00:31:09.959 --> 00:31:13.619
<v Kalim Saliba>I think, you know, I mentioned it earlier,
I think we're already seeing an increase

527
00:31:13.619 --> 00:31:18.119
<v Kalim Saliba>in the focus on greater autonomy, but
reaching a level of autonomy where the loop is

528
00:31:18.119 --> 00:31:20.400
<v Kalim Saliba>fully closed, I think we're a little ways
away.

529
00:31:20.520 --> 00:31:24.760
<v Kalim Saliba>And I think that's for a number of
reasons, but mostly because it's not really a

530
00:31:24.760 --> 00:31:26.040
<v Kalim Saliba>binary question, right?

531
00:31:26.079 --> 00:31:29.839
<v Kalim Saliba>It's not like one day we're going to turn
the lights off and only agents are going

532
00:31:29.839 --> 00:31:30.989
<v Kalim Saliba>to be running the labs.

533
00:31:32.949 --> 00:31:34.280
<v Kalim Saliba>I think we're going to learn by from
what's working.

534
00:31:34.401 --> 00:31:38.041
<v Kalim Saliba>What's not working as we increase the
autonomy, and scientists are going to have to

535
00:31:38.041 --> 00:31:39.401
<v Kalim Saliba>be involved in that process, right?

536
00:31:39.521 --> 00:31:43.472
<v Kalim Saliba>Scientists are very much needed for things
like defining the success criteria, like

537
00:31:43.661 --> 00:31:47.482
<v Kalim Saliba>what is successful efficacy, toxicity,
off-target effects look like?

538
00:31:47.602 --> 00:31:49.592
<v Kalim Saliba>Are we just going to rely on AI to tell us
that?

539
00:31:49.761 --> 00:31:52.021
<v Kalim Saliba>I hope not, because as a human, I hope
not.

540
00:31:52.151 --> 00:31:55.362
<v Kalim Saliba>And like for validating the foundational
science ranging from like whether we're

541
00:31:55.362 --> 00:31:58.761
<v Kalim Saliba>talking about pathways or targets or, you
know, making sure.

542
00:31:58.884 --> 00:32:01.364
<v Kalim Saliba>That there are appropriate sign-offs on
the conclusions.

543
00:32:01.443 --> 00:32:05.473
<v Kalim Saliba>I think there's very much right now a need
for not just scientists.

544
00:32:05.524 --> 00:32:09.083
<v Kalim Saliba>I think scientists are needed for a lot of
what I just mentioned, but also like

545
00:32:09.083 --> 00:32:14.163
<v Kalim Saliba>engineers, like people who are putting the
systems in place that will support this

546
00:32:14.163 --> 00:32:17.144
<v Kalim Saliba>way of working, because it's a
transformation to the way of working.

547
00:32:17.223 --> 00:32:20.843
<v Kalim Saliba>It's not like you can say, hey, we're
going to do everything we did before, but now

548
00:32:20.843 --> 00:32:22.323
<v Kalim Saliba>we're going to do it more autonomously.

549
00:32:22.484 --> 00:32:27.485
<v Kalim Saliba>It's much more of a Mindset shift, where
you have to think about to develop and run

550
00:32:27.485 --> 00:32:31.956
<v Kalim Saliba>and analyze this assay in a fully
automated way, what does that actually mean?

551
00:32:32.086 --> 00:32:33.086
<v Kalim Saliba>Like, what systems do I need?

552
00:32:33.135 --> 00:32:34.086
<v Kalim Saliba>What software do I need?

553
00:32:34.166 --> 00:32:36.005
<v Kalim Saliba>Where's the human in the in this loop?

554
00:32:36.166 --> 00:32:38.105
<v Kalim Saliba>Where should the human not be in this
loop?

555
00:32:38.326 --> 00:32:39.485
<v Kalim Saliba>Where am I calling an agent?

556
00:32:39.605 --> 00:32:42.145
<v Kalim Saliba>Where am I calling a more automated piece
of processing?

557
00:32:42.406 --> 00:32:46.796
<v Kalim Saliba>There's lots of functionality to consider,
and wiring it together in a in a coherent

558
00:32:46.796 --> 00:32:47.686
<v Kalim Saliba>way is very important.

559
00:32:48.199 --> 00:32:51.329
<v Kalim Saliba>But long answer to your question, the
short version is, I don't think we'll see

560
00:32:51.329 --> 00:32:56.369
<v Kalim Saliba>fully closed loop labs for a little while,
but we will see labs of higher and higher

561
00:32:56.369 --> 00:32:58.890
<v Kalim Saliba>autonomy getting developed and deployed.

562
00:32:59.709 --> 00:33:04.349
<v Speaker 2>One, you know, prediction which I hope,
let's say is, you know, coming out of the

563
00:33:04.349 --> 00:33:07.829
<v Speaker 2>lab in terms of manufacture and
production, right?

564
00:33:08.290 --> 00:33:14.469
<v Speaker 2>Um, so making small things that are that
small in a sizable scale of performance,

565
00:33:14.550 --> 00:33:14.760
<v Speaker 2>right?

566
00:33:14.829 --> 00:33:19.930
<v Speaker 2>And this moving this transfer from, from
the lab into the Into manufacturing, right?

567
00:33:20.089 --> 00:33:21.729
<v Speaker 2>Uh, we call tech transfer, right?

568
00:33:21.849 --> 00:33:26.329
<v Speaker 2>So, and there there is a bit of, you know,
somehow it's again a silo thing, right?

569
00:33:26.449 --> 00:33:30.969
<v Speaker 2>Tech, you have the R&amp;D discovery and
then you have the GMP, a good manufacturing

570
00:33:30.969 --> 00:33:34.890
<v Speaker 2>practice manufacturing, you know, has been
fragmented, the bottleneck, you know,

571
00:33:35.130 --> 00:33:37.430
<v Speaker 2>they are data rich, they are not data
rich, right?

572
00:33:37.925 --> 00:33:42.405
<v Speaker 2>No, um, you know, everyone says like, hey,
how about a one click tech transfer.

573
00:33:42.484 --> 00:33:45.954
<v Speaker 2>So you with a click on a button, right,
you can scale the things up.

574
00:33:46.204 --> 00:33:48.564
<v Speaker 2>What's your view on on tech transfer?

575
00:33:52.324 --> 00:33:55.864
<v Kalim Saliba>If I had, if I had the button, I would
have pushed it by now.

576
00:33:58.229 --> 00:33:59.750
<v Kalim Saliba>I have a pretty specific opinion on this.

577
00:33:59.829 --> 00:34:02.729
<v Kalim Saliba>I think one click is a lovely story.

578
00:34:02.829 --> 00:34:06.989
<v Kalim Saliba>It's a lovely aspiration, but it's not
particularly practical.

579
00:34:07.189 --> 00:34:11.510
<v Kalim Saliba>And I think the reason for that is because
you're dealing with parts of our

580
00:34:11.510 --> 00:34:15.909
<v Kalim Saliba>customers that work, think, and optimize
for completely different things.

581
00:34:16.110 --> 00:34:19.629
<v Kalim Saliba>On the research end of the spectrum,
you're optimizing for what works, you know,

582
00:34:19.909 --> 00:34:24.510
<v Kalim Saliba>with within which acceptable parameters
and trying to understand the biology of how

583
00:34:24.510 --> 00:34:27.169
<v Kalim Saliba>it's working or not working, whereas as
you go.

584
00:34:27.354 --> 00:34:31.554
<v Kalim Saliba>Into the development and then ultimately
to the manufacturing space, you're focused

585
00:34:31.554 --> 00:34:36.915
<v Kalim Saliba>on optimizing for repeatability,
predictability, and scale up, right?

586
00:34:36.955 --> 00:34:40.254
<v Kalim Saliba>And those are just very different things,
and the data sets don't necessarily

587
00:34:40.834 --> 00:34:45.905
<v Kalim Saliba>support this idea of one click because
they have to be almost translated, right?

588
00:34:45.955 --> 00:34:50.715
<v Kalim Saliba>We talked a little bit about the different
ways of talking about data across these

589
00:34:50.715 --> 00:34:52.344
<v Kalim Saliba>cycles, and is there a common language?

590
00:34:52.364 --> 00:34:56.235
<v Kalim Saliba>I think the interface point between
research and development and then manufacturing.

591
00:34:56.610 --> 00:35:00.649
<v Kalim Saliba>is the interface at which there is perhaps
the greatest disparity and difference

592
00:35:00.649 --> 00:35:03.010
<v Kalim Saliba>between language that's used.

593
00:35:03.280 --> 00:35:07.370
<v Kalim Saliba>Now, one of the things that I think is
helping is capturing more context and

594
00:35:07.370 --> 00:35:08.770
<v Kalim Saliba>structure during the research process.

595
00:35:08.889 --> 00:35:12.770
<v Kalim Saliba>So I talked a lot about orchestration and
digital threads and like blurring the line

596
00:35:12.770 --> 00:35:16.290
<v Kalim Saliba>between discrete software categories to
have lots of functionality kind of

597
00:35:16.290 --> 00:35:19.770
<v Kalim Saliba>interspersed so that you can naturally
draw those contextual connections.

598
00:35:20.676 --> 00:35:24.926
<v Kalim Saliba>If research is increasingly done that way,
and if you increase the autonomy of

599
00:35:24.926 --> 00:35:28.676
<v Kalim Saliba>research, which you can't do autonomous
research if that that autonomy isn't

600
00:35:28.676 --> 00:35:33.085
<v Kalim Saliba>organically building those connections
anyway, you end up with, you know, your

601
00:35:33.085 --> 00:35:38.166
<v Kalim Saliba>research data not just described, but
organized and structured in such a way, the

602
00:35:38.166 --> 00:35:40.686
<v Kalim Saliba>tech transfer translation becomes easier.

603
00:35:40.926 --> 00:35:47.122
<v Kalim Saliba>So not necessarily push button, but Not as
hard as spending a year plus trying to go

604
00:35:47.122 --> 00:35:51.562
<v Kalim Saliba>find the research data from across silos,
trying to figure out how to reproduce

605
00:35:51.562 --> 00:35:56.642
<v Kalim Saliba>elements of it within a uh scale
requirements that weren't true, you know, in the

606
00:35:56.642 --> 00:35:57.941
<v Kalim Saliba>earlier stage research.

607
00:35:58.122 --> 00:36:00.471
<v Kalim Saliba>I think the aspiration, one click is
wonderful.

608
00:36:00.602 --> 00:36:05.711
<v Kalim Saliba>I think practically speaking, what we will
see more of is better interfaces for

609
00:36:05.711 --> 00:36:06.521
<v Kalim Saliba>translation.

610
00:36:06.870 --> 00:36:11.669
<v Kalim Saliba>Between these functions driven by growing
contextual awareness, you know, how all of

611
00:36:11.669 --> 00:36:15.989
<v Kalim Saliba>the steps in the research worked and how
they were connected and therefore how they

612
00:36:15.989 --> 00:36:19.020
<v Kalim Saliba>should materialize within development and
tech transfer.

613
00:36:19.459 --> 00:36:23.669
<v Speaker 3>I mean, like that would alleviate so much
already, even though maybe the the one

614
00:36:23.669 --> 00:36:28.429
<v Speaker 3>click button thing is a stretch target,
stretch goal, but even taking like some

615
00:36:28.429 --> 00:36:33.110
<v Speaker 3>smaller steps in between and not having
like, finding the data being such a big

616
00:36:33.110 --> 00:36:37.060
<v Speaker 3>challenge um across silos, feeling
sometimes feeling like a scavenger hunt.

617
00:36:37.189 --> 00:36:40.949
<v Speaker 3>I think that would already tremendously
accelerate the entire process.

618
00:36:41.229 --> 00:36:42.389
<v Kalim Saliba>We don't want scavenger hunts.

619
00:36:42.629 --> 00:36:43.429
<v Kalim Saliba>That's exactly right.

620
00:36:43.540 --> 00:36:44.350
<v Kalim Saliba>That's a great way to put it.

621
00:36:44.379 --> 00:36:45.709
<v Speaker 3>At least not like in the work context.

622
00:36:45.979 --> 00:36:46.330
<v Speaker 3>Right?

623
00:36:46.510 --> 00:36:47.219
<v Speaker 3>Outside.

624
00:36:47.500 --> 00:36:48.379
<v Speaker 3>I, I don't mind.

625
00:36:48.879 --> 00:36:49.090
<v Kalim Saliba>Halloween.

626
00:36:49.820 --> 00:36:50.699
<v Speaker 3>Yeah, for Halloween.

627
00:36:50.780 --> 00:36:52.600
<v Speaker 3>Yeah, um, it's Halloween soon.

628
00:36:52.939 --> 00:36:58.199
<v Speaker 3>So, Kelly, you mentioned earlier that you
still see people and scientists um taking

629
00:36:58.659 --> 00:37:02.679
<v Speaker 3>over very important roles and a lot of
responsibility.

630
00:37:02.939 --> 00:37:07.020
<v Speaker 3>Let's maybe talk a little bit about like
how we train like these people, maybe in

631
00:37:07.020 --> 00:37:11.860
<v Speaker 3>the next generation in the future, because
today if we go to campuses, we see that

632
00:37:11.860 --> 00:37:16.449
<v Speaker 3>at most Universities, there's like the
biology chair and it's in one building and

633
00:37:16.449 --> 00:37:20.169
<v Speaker 3>then maybe we have the chemistry chair and
in another building and then computer

634
00:37:20.169 --> 00:37:23.010
<v Speaker 3>science is like in a different part of the
city somewhere.

635
00:37:23.449 --> 00:37:24.760
<v Speaker 3>So even like underground underground,

636
00:37:26.810 --> 00:37:27.489
<v Speaker 2>not anymore.

637
00:37:27.810 --> 00:37:31.090
<v Speaker 2>I think they're on the rooftop, you know,
on the surfing, surfing side,

638
00:37:31.169 --> 00:37:31.540
<v Kalim Saliba>right?

639
00:37:32.010 --> 00:37:36.139
<v Kalim Saliba>To be fair, I always had a nice computer
science building when I went to school.

640
00:37:36.209 --> 00:37:38.010
<v Kalim Saliba>So they didn't, they didn't stick us
underground.

641
00:37:39.790 --> 00:37:42.989
<v Speaker 3>I was at university in Kai school that was
tough.

642
00:37:43.189 --> 00:37:46.830
<v Speaker 3>So to all the KT folks out there, if you
can make it there, you can make it

643
00:37:46.830 --> 00:37:47.110
<v Speaker 3>anywhere.

644
00:37:47.270 --> 00:37:51.590
<v Speaker 3>But like that setup um already um creates
quite some barriers, so it's like hard for

645
00:37:51.590 --> 00:37:53.649
<v Speaker 3>people to even meet each other to
collaborate.

646
00:37:53.790 --> 00:37:58.360
<v Speaker 3>So, do you feel like, like that way of
like teaching or like that curricular should

647
00:37:58.370 --> 00:38:04.510
<v Speaker 3>should somehow like change and so students
learn more interdisciplinary skills like

648
00:38:04.510 --> 00:38:05.909
<v Speaker 3>at an early age?

649
00:38:06.320 --> 00:38:08.120
<v Kalim Saliba>Yes, this is a short answer to your
question.

650
00:38:08.199 --> 00:38:13.919
<v Kalim Saliba>I, I, I do think the way in which we
educate and also learn within our fields needs

651
00:38:13.919 --> 00:38:18.229
<v Kalim Saliba>to fundamentally change, and in some areas
more than in others, like I'll give you

652
00:38:18.229 --> 00:38:22.879
<v Kalim Saliba>an example, in in in software engineering,
it's fundamentally changed the way

653
00:38:22.879 --> 00:38:24.260
<v Kalim Saliba>engineers work today.

654
00:38:24.550 --> 00:38:25.260
<v Kalim Saliba>It's on a spectrum.

655
00:38:25.360 --> 00:38:26.719
<v Kalim Saliba>Again, none of these things are binary.

656
00:38:26.820 --> 00:38:31.739
<v Kalim Saliba>There are different levels and ways of
working with AI, but the net is computer

657
00:38:31.739 --> 00:38:34.830
<v Kalim Saliba>engineering and computer science will
never be the same.

658
00:38:35.000 --> 00:38:35.560
<v Kalim Saliba>It's very interesting.

659
00:38:35.864 --> 00:38:41.524
<v Kalim Saliba>If you go look at the stack overflow site
visits on a monthly basis, you'll see that

660
00:38:41.524 --> 00:38:45.705
<v Kalim Saliba>right around the point that AI starts to
get good at doing software engineering, it

661
00:38:45.705 --> 00:38:50.625
<v Kalim Saliba>kind of precipitously drops, meaning this
forum, which was rich and vibrant and the

662
00:38:50.625 --> 00:38:54.014
<v Kalim Saliba>place where everyone went to discuss
software engineering concepts and problems, get

663
00:38:54.014 --> 00:38:56.104
<v Kalim Saliba>solutions, changed almost overnight.

664
00:38:56.225 --> 00:39:01.425
<v Kalim Saliba>And I think we're seeing lots of similar
such changes, changes happening already

665
00:39:01.425 --> 00:39:04.495
<v Kalim Saliba>happening or will be happening in other
industries and life sciences is no

666
00:39:04.495 --> 00:39:05.024
<v Kalim Saliba>exception.

667
00:39:05.320 --> 00:39:08.209
<v Kalim Saliba>As we've been talking about this whole
call, like all of the frontier model

668
00:39:08.209 --> 00:39:12.969
<v Kalim Saliba>providers and other companies are taking
direct aim at making massive gains in the

669
00:39:12.969 --> 00:39:13.389
<v Kalim Saliba>space.

670
00:39:13.500 --> 00:39:17.429
<v Kalim Saliba>So the expectation should be that it's
going to fundamentally change.

671
00:39:18.250 --> 00:39:21.290
<v Kalim Saliba>We don't know exactly how fast, we don't
know exactly when, but it's going to

672
00:39:21.290 --> 00:39:21.750
<v Kalim Saliba>change.

673
00:39:22.010 --> 00:39:27.010
<v Kalim Saliba>And I think to help prepare folks going
through the education process to get ready

674
00:39:27.010 --> 00:39:27.949
<v Kalim Saliba>for this change.

675
00:39:28.162 --> 00:39:29.602
<v Kalim Saliba>The first thing is to accept it, right?

676
00:39:29.642 --> 00:39:32.751
<v Kalim Saliba>The first thing is to look it head on and
like have a really honest conversation

677
00:39:32.751 --> 00:39:34.052
<v Kalim Saliba>with yourself about what does this mean?

678
00:39:34.332 --> 00:39:37.562
<v Kalim Saliba>What does this mean for me personally and
like what I believe, and then what does

679
00:39:37.562 --> 00:39:38.711
<v Kalim Saliba>this mean for my career?

680
00:39:38.922 --> 00:39:42.402
<v Kalim Saliba>What am I seeing out there in the industry
about what people are prioritizing and

681
00:39:42.402 --> 00:39:46.122
<v Kalim Saliba>how are they putting campaigns together,
and then how do I need to orient myself to

682
00:39:46.122 --> 00:39:47.441
<v Kalim Saliba>be effective in that world.

683
00:39:47.681 --> 00:39:52.813
<v Kalim Saliba>And there are other examples that I think
illustrate Professions that tend to have

684
00:39:52.813 --> 00:39:54.023
<v Kalim Saliba>to be highly collaborative.

685
00:39:54.233 --> 00:39:58.624
<v Kalim Saliba>So I often look to those as kind of like
my inspiration for what I think is needed

686
00:39:58.624 --> 00:40:00.414
<v Kalim Saliba>and, you know, the advice I give people.

687
00:40:00.724 --> 00:40:03.793
<v Kalim Saliba>So maybe I can say it a little bit about
that, but I'll start, I guess I'll qualify

688
00:40:03.793 --> 00:40:08.793
<v Kalim Saliba>it by acknowledging that I'm biased
because part of my background is in systems

689
00:40:08.793 --> 00:40:13.233
<v Kalim Saliba>engineering, and as a systems engineer,
the whole job is how you bring different

690
00:40:13.233 --> 00:40:13.993
<v Kalim Saliba>functions together.

691
00:40:14.065 --> 00:40:14.256
<v Kalim Saliba>Right?

692
00:40:14.305 --> 00:40:18.385
<v Kalim Saliba>So you think about building a plane or a
ship or a building where you have to

693
00:40:18.385 --> 00:40:22.335
<v Kalim Saliba>involve electrical, mechanical, plumbing,
structural software systems.

694
00:40:22.666 --> 00:40:27.466
<v Kalim Saliba>That thing, that system does not work
unless every function comes together in

695
00:40:27.466 --> 00:40:28.496
<v Kalim Saliba>exactly the right way.

696
00:40:28.706 --> 00:40:32.416
<v Kalim Saliba>And I think we're moving very quickly to a
world where this type of systems thinking

697
00:40:32.416 --> 00:40:36.496
<v Kalim Saliba>and systems engineering of these systems,
systems engineering of these systems,

698
00:40:36.625 --> 00:40:36.885
<v Kalim Saliba>yeah.

699
00:40:38.610 --> 00:40:39.080
<v Speaker 3>systematic,

700
00:40:39.429 --> 00:40:44.350
<v Kalim Saliba>very systematic is kind of where these
industries need to go because you mentioned

701
00:40:44.350 --> 00:40:48.189
<v Kalim Saliba>biology, chemistry, and then you have
assay scientists, molecular biologists, have,

702
00:40:48.389 --> 00:40:52.189
<v Kalim Saliba>there's so many data scientists, there's
so many disciplines that end up having to

703
00:40:52.189 --> 00:40:57.489
<v Kalim Saliba>get involved in a therapeutic or material
science campaign that if you don't know

704
00:40:57.790 --> 00:41:02.030
<v Kalim Saliba>what that overall end to end system looks
like, you'll be able to do your, your

705
00:41:02.030 --> 00:41:02.310
<v Kalim Saliba>specific.

706
00:41:02.419 --> 00:41:07.379
<v Kalim Saliba>Discipline really well, but in terms of
being able to understand how it's changing

707
00:41:07.379 --> 00:41:11.419
<v Kalim Saliba>relative to how these companies are
changing the systems around that discipline,

708
00:41:11.659 --> 00:41:13.010
<v Kalim Saliba>it's much, much harder, right?

709
00:41:13.060 --> 00:41:16.120
<v Kalim Saliba>And I think it's kind of a wake-up call
for a lot of people where they have to now

710
00:41:16.379 --> 00:41:21.449
<v Kalim Saliba>not only be molecular biologists, they
have to understand, you know, data aspects of

711
00:41:21.449 --> 00:41:26.699
<v Kalim Saliba>data science and how the frontier models
are advancing in ways that impact their

712
00:41:26.699 --> 00:41:31.489
<v Kalim Saliba>space and like What systems are are pharma
and biotech putting in place to drive

713
00:41:31.489 --> 00:41:35.090
<v Kalim Saliba>greater autonomy, like, it's just all
interlinked is another way to say it.

714
00:41:35.209 --> 00:41:40.810
<v Kalim Saliba>And the more people can think about,
learn, and hopefully be taught more about

715
00:41:40.810 --> 00:41:45.649
<v Kalim Saliba>systems thinking, but in a way that's
specifically applicable to science, the

716
00:41:45.649 --> 00:41:48.209
<v Kalim Saliba>better, because that's, I think, what's
ultimately going to make or break the

717
00:41:48.209 --> 00:41:51.879
<v Kalim Saliba>difference in effectiveness in in a lot of
these uh education programs.

718
00:41:52.260 --> 00:41:52.570
<v Speaker 2>Kali.

719
00:41:52.719 --> 00:41:58.530
<v Speaker 2>It's a great reflection and you, you know,
you resonate a bit of anticipating the

720
00:41:58.530 --> 00:41:59.129
<v Speaker 2>change, right?

721
00:41:59.239 --> 00:42:02.199
<v Speaker 2>Based on stack overflow and um, you know,
and the impact they had.

722
00:42:02.479 --> 00:42:07.070
<v Speaker 2>If you look now a bit of, you know, if
your own boldness and and forward looking

723
00:42:07.070 --> 00:42:11.030
<v Speaker 2>perspective and you look, I don't know, is
the 3 year horizon, is it a 1 year

724
00:42:11.030 --> 00:42:12.439
<v Speaker 2>horizon or is it a 5 year?

725
00:42:12.639 --> 00:42:17.760
<v Speaker 2>Usually we, we would say in the past, past
all the AI hype, it was a 10 years.

726
00:42:17.870 --> 00:42:20.699
<v Speaker 2>You know, depiction what the future will
bring, right?

727
00:42:21.030 --> 00:42:24.179
<v Speaker 2>I felt like we are, we're talking 3 months
or sometimes, right?

728
00:42:24.479 --> 00:42:27.709
<v Speaker 2>So but let's, if you consider a bit of a 3
year horizon, right?

729
00:42:27.790 --> 00:42:31.830
<v Speaker 2>Um, so what is one bold contrarian maybe
prediction?

730
00:42:31.949 --> 00:42:36.820
<v Speaker 2>How would you imagine, you know, AI maybe
in drug discovery or others, right?

731
00:42:36.909 --> 00:42:39.629
<v Speaker 2>Um, actually it will be done in in 2029.

732
00:42:39.679 --> 00:42:42.810
<v Speaker 2>Is there some bold hypothesis you have or
a wish list?

733
00:42:43.350 --> 00:42:44.989
<v Kalim Saliba>I wish I had a better crystal ball.

734
00:42:46.169 --> 00:42:47.110
<v Kalim Saliba>I'll tell you that.

735
00:42:47.729 --> 00:42:48.649
<v Kalim Saliba>That's the first thing I

736
00:42:48.649 --> 00:42:49.209
<v Speaker 3>wish for.

737
00:42:49.489 --> 00:42:51.830
<v Speaker 3>obsessed with crystal balls lately.

738
00:42:52.639 --> 00:42:52.649
<v Kalim Saliba>Yeah.

739
00:42:53.850 --> 00:42:57.729
<v Kalim Saliba>No, but I guess if, if I really think
about it, it could go in so many different

740
00:42:57.729 --> 00:42:58.989
<v Kalim Saliba>directions, and I think it will.

741
00:42:59.129 --> 00:43:03.689
<v Kalim Saliba>But there will be a few common threads
that I believe will probably be true, but I

742
00:43:03.689 --> 00:43:08.209
<v Kalim Saliba>think we will probably still see the
frontier models keep improving significantly,

743
00:43:08.409 --> 00:43:12.120
<v Kalim Saliba>like to a point where having the ability
to use them will become.

744
00:43:12.176 --> 00:43:14.216
<v Kalim Saliba>More and more of a competitive
differentiator.

745
00:43:14.375 --> 00:43:18.256
<v Kalim Saliba>Right now it's important, don't get me
wrong, but it hasn't reached the level of

746
00:43:18.256 --> 00:43:24.295
<v Kalim Saliba>importance where having a model makes or
breaks the materials that are being brought

747
00:43:24.295 --> 00:43:24.825
<v Kalim Saliba>to market.

748
00:43:25.055 --> 00:43:27.615
<v Kalim Saliba>We're not there yet, so I, but I think
they're going to keep improving, and in 3

749
00:43:27.615 --> 00:43:31.895
<v Kalim Saliba>years' time, no one knows, but I think the
improvement will be exponential, not

750
00:43:31.895 --> 00:43:32.736
<v Kalim Saliba>linear, right?

751
00:43:32.815 --> 00:43:34.676
<v Kalim Saliba>So I think that that's, that's the first
prediction.

752
00:43:34.885 --> 00:43:40.562
<v Kalim Saliba>2 is, I think the base models will get
good enough that Harnessing them will become

753
00:43:40.862 --> 00:43:45.382
<v Kalim Saliba>a critical part of any drug discovery or
material development pipeline, which puts a

754
00:43:45.382 --> 00:43:47.221
<v Kalim Saliba>lot more focus on these lab in a loops.

755
00:43:47.382 --> 00:43:52.731
<v Kalim Saliba>How do you deliver and develop an
autonomous lab in a loop pipeline that runs these

756
00:43:52.731 --> 00:43:53.221
<v Kalim Saliba>models?

757
00:43:53.302 --> 00:43:55.181
<v Kalim Saliba>It's not, it's not a human running the
model, right?

758
00:43:55.332 --> 00:44:00.622
<v Kalim Saliba>It's a pipeline running the model at the
correct points in that workflow to really

759
00:44:00.622 --> 00:44:01.941
<v Kalim Saliba>extract the most value out of it.

760
00:44:01.981 --> 00:44:05.302
<v Kalim Saliba>And so, really, the loop will become the
most important thing in science.

761
00:44:05.498 --> 00:44:06.197
<v Kalim Saliba>In 3 years.

762
00:44:06.417 --> 00:44:10.667
<v Kalim Saliba>The loop, the system, the system that's
put in place to run the AI in my opinion,

763
00:44:10.907 --> 00:44:15.447
<v Kalim Saliba>will become the most important thing, and
that system, if it leverages the

764
00:44:15.508 --> 00:44:20.508
<v Kalim Saliba>upfrontier models, will be incredibly
powerful, and I'm very optimistic that we'll

765
00:44:20.508 --> 00:44:24.058
<v Kalim Saliba>start to see some of the first end to end
AI designed drugs, not just hit the

766
00:44:24.058 --> 00:44:27.708
<v Kalim Saliba>market, but maybe even be the default way
they are brought to market.

767
00:44:28.107 --> 00:44:31.937
<v Speaker 3>Wow, Kalim, that makes a lot of sense and
I'm very curious about.

768
00:44:32.114 --> 00:44:33.333
<v Speaker 3>How that will unfold.

769
00:44:33.394 --> 00:44:37.303
<v Speaker 3>So we'll we'll make sure to give you a
ring in 3 years latest, latest, but let's

770
00:44:37.303 --> 00:44:39.424
<v Speaker 3>make sure we stay in touch in the meantime
as well.

771
00:44:39.674 --> 00:44:44.023
<v Speaker 3>And speaking of time, the time for this
podcast is almost over.

772
00:44:44.354 --> 00:44:48.753
<v Speaker 3>But before we finish up, we would like to
play our signature game.

773
00:44:48.904 --> 00:44:50.634
<v Speaker 3>It's called Authentic Autocomplete.

774
00:44:50.993 --> 00:44:55.954
<v Speaker 3>You already started looking a little bit
into the future, and now you can anticipate

775
00:44:55.954 --> 00:44:58.374
<v Speaker 3>how um our sentence starters might end.

776
00:44:58.820 --> 00:45:00.959
<v Speaker 3>And ideally in the most authentic way.

777
00:45:01.139 --> 00:45:03.189
<v Speaker 3>Are you ready for some, some fun?

778
00:45:03.280 --> 00:45:03.889
<v Speaker 3>I don't know, but let's

779
00:45:03.889 --> 00:45:04.370
<v Kalim Saliba>give it a shot.

780
00:45:04.899 --> 00:45:05.489
<v Kalim Saliba>Let's give it a shot.

781
00:45:06.179 --> 00:45:10.479
<v Speaker 3>Let's believe if I weren't in tech, I'd be
a full-time

782
00:45:11.260 --> 00:45:11.860
<v Kalim Saliba>philosopher.

783
00:45:13.219 --> 00:45:14.310
<v Speaker 2>All right, that makes sense.

784
00:45:14.620 --> 00:45:18.500
<v Speaker 2>Uh, so the last time an AI agent surprised
me was.

785
00:45:19.600 --> 00:45:25.520
<v Kalim Saliba>When it turned two dimensional plans of
the house I'm building into a fully accurate

786
00:45:25.520 --> 00:45:27.120
<v Kalim Saliba>3 dimensional model.

787
00:45:27.520 --> 00:45:28.739
<v Speaker 3>Wow, here we go.

788
00:45:29.040 --> 00:45:31.419
<v Speaker 3>AI for drug discovery is

789
00:45:31.760 --> 00:45:32.239
<v Kalim Saliba>life changing.

790
00:45:32.320 --> 00:45:33.350
<v Kalim Saliba>It's going to change the world.

791
00:45:33.560 --> 00:45:36.280
<v Kalim Saliba>Maybe not this year, maybe not next year,
but eventually in 3

792
00:45:36.280 --> 00:45:37.669
<v Speaker 2>years, in 3 years' time.

793
00:45:37.919 --> 00:45:38.429
<v Speaker 2>So here we

794
00:45:38.429 --> 00:45:38.810
<v Kalim Saliba>go.

795
00:45:39.159 --> 00:45:40.030
<v Kalim Saliba>In 3 years' time.

796
00:45:40.159 --> 00:45:41.500
<v Kalim Saliba>That's what my crystal ball says.

797
00:45:42.969 --> 00:45:44.399
<v Speaker 2>So last, last one, easy one.

798
00:45:44.489 --> 00:45:48.010
<v Speaker 2>If I could invent one universal rule,
everyone must follow, it

799
00:45:48.010 --> 00:45:51.919
<v Kalim Saliba>would be the first thing that comes to
mind, yeah, be be being open minded,

800
00:45:51.929 --> 00:45:53.850
<v Kalim Saliba>especially these days, just really
important.

801
00:45:53.929 --> 00:45:55.169
<v Kalim Saliba>It's very easy to miscommunicate.

802
00:45:55.330 --> 00:45:57.610
<v Kalim Saliba>Open-mindedness is the best way to
understand each other.

803
00:45:58.169 --> 00:46:02.100
<v Speaker 2>Kalem, thanks so much for your time, for
sharing your passion, the sword leadership.

804
00:46:02.389 --> 00:46:05.439
<v Speaker 2>I think the clarity in addition is
amazing.

805
00:46:05.560 --> 00:46:10.120
<v Speaker 2>You're, you're, it's so nice to listen to
you and about your, your world model,

806
00:46:10.229 --> 00:46:15.399
<v Speaker 2>let's say, following up your world model
in a nice complex yet understandable

807
00:46:15.399 --> 00:46:15.679
<v Speaker 2>format.

808
00:46:15.780 --> 00:46:17.669
<v Speaker 2>So I appreciated that you shared it with
us.

809
00:46:18.030 --> 00:46:19.040
<v Kalim Saliba>Yeah, of course, anytime.

810
00:46:19.120 --> 00:46:20.679
<v Kalim Saliba>Thank you both for inviting me to this.

811
00:46:20.760 --> 00:46:23.500
<v Kalim Saliba>I hope uh hope listeners found it valuable
perspective.

812
00:46:23.959 --> 00:46:27.399
<v Speaker 2>Yeah, folks out there, um, you know, you
want to connect first of all to your

813
00:46:27.399 --> 00:46:30.810
<v Speaker 2>Netflix or any, any streamer, you know,
Ted Lasso for coaching, right?

814
00:46:30.879 --> 00:46:32.020
<v Speaker 2>You get to get the aspects.

815
00:46:32.159 --> 00:46:35.919
<v Speaker 2>Second, then you end up at sub stack and
reading some stuff of Kalim, right?

816
00:46:36.159 --> 00:46:39.959
<v Speaker 2>And third one, you connect, I guess with
him on LinkedIn, right, to let, to let and

817
00:46:39.959 --> 00:46:44.879
<v Speaker 2>also us know about what do you find good,
what we can do better, or what's your view

818
00:46:44.879 --> 00:46:46.870
<v Speaker 2>on AI and the role of drug discovery.

819
00:46:47.080 --> 00:46:50.959
<v Speaker 2>But definitely stay bold, committed and
open minded, and we hear us at the next

820
00:46:50.959 --> 00:46:52.179
<v Speaker 2>Siemens AI podcast.

821
00:46:52.360 --> 00:46:52.800
<v Speaker 2>Cheers.
