1
00:00:00,127 --> 00:00:03,807
I'm going to introduce Andrew and he's going to get started because he's got a lot to show you.

2
00:00:04,127 --> 00:00:06,767
Hopefully we have a lot of great back and forth today.

3
00:00:07,487 --> 00:00:10,047
Questions, Q&A, that sort of thing.

4
00:00:10,127 --> 00:00:11,087
You ready for that, Andrew?

5
00:00:11,087 --> 00:00:12,127
Yeah, we're good to go.

6
00:00:12,207 --> 00:00:12,767
All right.

7
00:00:13,087 --> 00:00:22,047
So Andrew Severin is the director of Genome Informatics Facility at Iowa State University, where he develops documentation

8
00:00:22,487 --> 00:00:30,047
computation-centered computational workflows that help researchers navigate complex data and analysis.

9
00:00:30,287 --> 00:00:34,527
You should not do it in a complete lap as fast as you can before you come in here and run out of breath.

10
00:00:35,087 --> 00:00:35,727
Oh my goodness.

11
00:00:36,367 --> 00:00:46,447
With more than 20 years of experience in genomics and bioinformatics, his work now focuses on how artificial intelligence is transforming the way we interact with information.

12
00:00:47,727 --> 00:00:49,167
Shifting systems from

13
00:00:49,647 --> 00:00:52,407
passive tools into active collaborators.

14
00:00:53,967 --> 00:01:04,927
In this session, Andrew will move beyond the idea of AI as a chatbot and demonstrate what it looks like to embed it, to embed AI directly into real projects and workflows.

15
00:01:06,287 --> 00:01:16,207
Drawing from his work on the 100 Days of Bioinformatics project, he will show how tools like ChatGPT and Claude can be used inside development environments

16
00:01:17,807 --> 00:01:21,727
reduce friction, accelerate work, and support practical day-to-day problem solving.

17
00:01:22,527 --> 00:01:24,207
Please join me in welcoming Andrew Zeperant.

18
00:01:28,607 --> 00:01:29,407
Hello, everybody.

19
00:01:30,127 --> 00:01:35,727
I'm listening to this, and I'm just like, that's the most boring thing that you can possibly describe what I do.

20
00:01:36,487 --> 00:01:43,087
In all honesty, I think of myself as a problem solver, as a troubleshooter, and I think pretty much everybody in this room

21
00:01:43,807 --> 00:01:45,167
Can relate with that.

22
00:01:45,327 --> 00:01:52,127
We're here to try to solve problems, trying to figure out how to use this technology, and this technology it keeps advancing.

23
00:01:52,127 --> 00:01:55,647
In fact, when I was writing this, I had originally told them, Paul, there's nothing new.

24
00:01:55,647 --> 00:01:57,007
I'm not doing anything.

25
00:01:57,007 --> 00:01:59,847
I'm just applying the chatbot to what I do in my...

26
00:01:59,927 --> 00:02:00,767
my daily work.

27
00:02:01,167 --> 00:02:02,687
And he said, that could be really useful.

28
00:02:02,767 --> 00:02:04,407
There are people that want to hear about that.

29
00:02:04,407 --> 00:02:05,487
I'm like, really?

30
00:02:06,047 --> 00:02:07,927
Like, no.

31
00:02:08,127 --> 00:02:15,087
I mean, and then I, heard about, I don't know how many people have heard about OpenClaw and agents and stuff like that.

32
00:02:15,807 --> 00:02:23,567
When it came out in November, I was like, there's no way am I ever going to let something like that use my computer on its own.

33
00:02:23,807 --> 00:02:27,087
That's like the most dangerous thing that I could possibly think of.

34
00:02:27,327 --> 00:02:28,767
I ignored it for three months.

35
00:02:29,127 --> 00:02:29,727
four months.

36
00:02:30,327 --> 00:02:33,927
And then basically all social media exploded.

37
00:02:33,927 --> 00:02:37,327
And I was like, holy, I cannot believe I'm this far behind.

38
00:02:38,047 --> 00:02:39,727
And I was starting looking at it.

39
00:02:39,807 --> 00:02:55,327
And there are other bioinformaticians, people that help faculty analyze their high throughput DNA sequencing, RNA sequencing data to try to solve these biological questions to help improve agriculture or make drugs or whatever else.

40
00:02:56,367 --> 00:02:57,327
They basically

41
00:02:57,807 --> 00:03:04,047
have agents or skills now, which I'll explain in a minute, that can do pretty much everything I do.

42
00:03:04,367 --> 00:03:08,727
And I had this moment like, I'm going to lose my job.

43
00:03:09,567 --> 00:03:17,247
You know, this feeling that AI is going to replace me, which if you asked me six months ago, like, no way.

44
00:03:17,247 --> 00:03:18,207
There's no way.

45
00:03:19,087 --> 00:03:24,447
I have a knowledge base and I know enough that there's no way that AI is going to replace me.

46
00:03:24,767 --> 00:03:26,487
And then this happened and I was just,

47
00:03:27,207 --> 00:03:29,167
It was like an existential crisis.

48
00:03:29,167 --> 00:03:34,127
I don't know how many people have had that kind of feeling with all this AI.

49
00:03:34,687 --> 00:03:38,927
So really this talk is about my, it's like multiple things.

50
00:03:38,927 --> 00:03:46,847
It's partly my journey, my implementation, and how I'm using AI in what I do on a daily basis, which I think is generally useful.

51
00:03:48,007 --> 00:03:56,607
It's also, I guess, an introduction of where do you start, where is the second step, and where is it going?

52
00:03:57,087 --> 00:04:00,287
So all of that will kind of be part of that.

53
00:04:01,887 --> 00:04:03,967
What I do, I've already pretty much explained.

54
00:04:03,967 --> 00:04:07,847
I help the faculty transform all of this large data.

55
00:04:07,847 --> 00:04:13,327
Back in 2011, when they hired me and I founded this facility, they would send data to...

56
00:04:13,807 --> 00:04:17,607
the genome core, and they would sequence it, and they would give this data back.

57
00:04:17,887 --> 00:04:23,447
The faculty would open that data file in Word or Excel, and it would crash their computers.

58
00:04:23,447 --> 00:04:27,407
And the university is like, well, we should probably hire somebody to help them figure out how to analyze the data.

59
00:04:28,127 --> 00:04:30,367
And that's how this facility came about.

60
00:04:30,367 --> 00:04:31,567
That's how my job came about.

61
00:04:31,967 --> 00:04:38,607
Since then, I've hired five more bioinformaticians that help analyze data across the university and at the USDA.

62
00:04:40,767 --> 00:04:59,807
But with the recent advances in AI, I've realized that that's not going to be sufficient to be able to maintain salaries for my people, especially with the current funding environment, and that we're going to start needing to pivot to help faculty and staff use AI and these tools to help advance scientific research.

63
00:05:00,847 --> 00:05:03,167
It's no different than what I've been doing before.

64
00:05:03,247 --> 00:05:05,887
Basically, I download a program off the internet.

65
00:05:06,567 --> 00:05:11,167
I figure out how to use it, and then I can explain it to the researchers or the students.

66
00:05:12,047 --> 00:05:23,167
I create pipelines, like raw data in, transform it, make that transform data the right input for the next program, boom, boom, boom, and then they get multiple outputs.

67
00:05:23,167 --> 00:05:25,647
I make it look pretty, beautiful documentation.

68
00:05:40,927 --> 00:05:41,887
can be okay.

69
00:05:42,127 --> 00:05:44,927
The quality of the documentation, maybe less so.

70
00:05:46,127 --> 00:05:59,007
And so I'm in charge of finding the best tool to help my researchers analyze their data to get, I guess, closer approximation to truth for the scientific question.

71
00:05:59,887 --> 00:06:04,927
You typically with software that has no tutorial or very bad documentation.

72
00:06:05,327 --> 00:06:12,047
Well, we're currently living in an era where we have these amazing tools, but nobody knows how to use it.

73
00:06:12,767 --> 00:06:21,887
So to some degree, I feel like I'm pretty well qualified to be able to learn how to use a very under-documented tool to help solve problems.

74
00:06:25,007 --> 00:06:28,927
The first thing back in 2023 were the chatbots.

75
00:06:29,407 --> 00:06:32,687
GPT-4 came out and everybody was talking about it.

76
00:06:33,687 --> 00:06:35,327
And the things they could do were amazing.

77
00:06:35,487 --> 00:06:41,167
If you have not learned how to create your own prompts, do prompt engineering, this is where you start.

78
00:06:41,567 --> 00:06:42,847
This has not gone away.

79
00:06:43,327 --> 00:06:47,407
This is the foundation of skills and agents and everything downstream.

80
00:06:47,967 --> 00:06:51,407
Because in order to write a skill, you have to be able to prompt it well.

81
00:06:51,927 --> 00:06:56,087
You have to make sure it's executing your instructions the way you want it to.

82
00:06:58,047 --> 00:07:00,927
And a lot of this is about loops.

83
00:07:01,967 --> 00:07:03,807
You ask a question, you get an output.

84
00:07:03,807 --> 00:07:04,607
Do you like the output?

85
00:07:04,607 --> 00:07:08,687
No, that's not what I meant, and usually that's because you didn't provide enough context.

86
00:07:09,407 --> 00:07:22,527
I have researchers come into my office and they start talking about their problem that they told me like 8 months ago, and I have no clue what they're talking about, and I'm like, Let's step back and tell me the biological question you're trying to answer.

87
00:07:23,127 --> 00:07:27,727
there's lots of technology, a lot of stacks, lots of sequencing, whatever that you could apply to a problem.

88
00:07:27,887 --> 00:07:30,767
But I want to know what is the thing you're trying to solve?

89
00:07:30,767 --> 00:07:32,927
What is the question you're trying to answer?

90
00:07:33,487 --> 00:07:35,167
And I think that's also true here.

91
00:07:35,727 --> 00:07:39,967
And you have to tell me more about what you are, what you know.

92
00:07:40,127 --> 00:07:43,887
I can teach people how to use and how to do bioinformatics.

93
00:07:44,127 --> 00:07:47,007
But the first thing I ask is, do you know how to use Unix?

94
00:07:47,047 --> 00:07:48,207
Do you know what a terminal is?

95
00:07:49,887 --> 00:07:51,247
Have you ever done any programming before?

96
00:07:51,687 --> 00:08:02,047
You have to tell these kinds of things also to the agent, because they don't, the program, the ChatGPT or whatever you're using, the large language model, doesn't know what you know.

97
00:08:02,687 --> 00:08:10,287
And so now people are actually creating identity files in their Claude folder that says, this is who I am.

98
00:08:11,487 --> 00:08:12,927
These are all the things I know.

99
00:08:13,127 --> 00:08:19,807
And then every time you start a new session, some part of that can be uploaded, and then you don't have to tell it every single time.

100
00:08:20,687 --> 00:08:24,287
And some people have things like, these are my current objectives.

101
00:08:24,767 --> 00:08:30,607
I'm trying to make this program, or I'm trying to make my company more profitable, or something like that.

102
00:08:30,847 --> 00:08:33,807
And that is something that could be updated on a more regular basis.

103
00:08:35,727 --> 00:08:45,567
So this kind of loop structure where you keep asking it to refine to get a better answer until the final output, you're going to see that over and over again as we get to skills and agents.

104
00:08:48,287 --> 00:08:49,087
After that,

105
00:08:49,927 --> 00:08:55,487
The thing that I was going to talk to you about, but I've kind of tried to expand upon that, is talking with your documents.

106
00:08:55,807 --> 00:08:58,287
So I use a program called VS Code.

107
00:08:58,527 --> 00:09:03,327
It's basically a text editor that uses markdown.

108
00:09:03,327 --> 00:09:05,247
It doesn't really matter.

109
00:09:05,247 --> 00:09:11,247
Basically, a pound is title, pound pound is subtitle, a bullet point is a star, you know, stuff like that.

110
00:09:11,247 --> 00:09:15,327
It's a very, very simple way to mark up your text documents.

111
00:09:15,647 --> 00:09:17,087
It also happens to be

112
00:09:17,607 --> 00:09:21,487
what is a lot of the AI agents find easy to read.

113
00:09:21,967 --> 00:09:22,927
So a lot of this stuff.

114
00:09:22,927 --> 00:09:26,847
So if you have things in text, it prefers that over a PDF or something like that.

115
00:09:26,847 --> 00:09:28,847
So you have to convert that to some kind of a text file.

116
00:09:31,567 --> 00:09:33,967
And you can convert any folder.

117
00:09:33,967 --> 00:09:37,527
We all have folders with all of our documentation, all of our information.

118
00:09:37,527 --> 00:09:45,487
You can convert any folder into basically a retrieval augmented generation system by plugins that you can just chat with the folder.

119
00:09:45,887 --> 00:09:51,247
And that's advantageous because there are times where I'm working on a very large website.

120
00:09:51,647 --> 00:10:02,607
It's Jekyll-based, minimal mistakes, basically lots of JavaScript and Jekyll code and my own information, my tutorials I'm putting in there.

121
00:10:03,007 --> 00:10:06,047
And they're all asking questions like, where do I...

122
00:10:06,567 --> 00:10:11,007
where's the file that allows me to change the color of this thing?

123
00:10:11,007 --> 00:10:15,807
It's a huge database, or not a huge website, and I might not know where that file is.

124
00:10:16,047 --> 00:10:19,487
I might have known at one point, but now I can find it immediately.

125
00:10:20,607 --> 00:10:22,047
And I can ask it to do tasks.

126
00:10:22,047 --> 00:10:25,087
Like, oh, can you add this information to this file?

127
00:10:25,407 --> 00:10:27,887
And I can keep staying within focus.

128
00:10:27,967 --> 00:10:29,087
And I'll talk about that more.

129
00:10:31,327 --> 00:10:32,367
So this is what it looks like.

130
00:10:32,527 --> 00:10:33,247
Let's see if I can...

131
00:10:34,207 --> 00:10:35,727
You've got your folders on the left.

132
00:10:37,487 --> 00:10:40,287
The markdown text here, and this is what the markdown would look like.

133
00:10:40,287 --> 00:10:43,967
Again, a single pound gave it a big title with an underline, double pound.

134
00:10:44,207 --> 00:10:46,927
You got the bullet points here with dashes, et cetera.

135
00:10:47,727 --> 00:10:50,127
Not critical, but you have something like that.

136
00:10:50,607 --> 00:10:55,087
You can have a plug-in that basically gives you a chat window on the left-hand side here.

137
00:10:56,527 --> 00:11:01,287
And you can ask, in this case, I was writing a tutorial about GitHub.

138
00:11:01,287 --> 00:11:04,127
And I'm like, oh, I should add GitHub to the glossary.

139
00:11:04,447 --> 00:11:06,127
Glossary is a completely different file.

140
00:11:06,607 --> 00:11:08,527
And it's alphabetically sorted.

141
00:11:08,807 --> 00:11:16,287
And so in order to add GitHub as a term, I would have to think about how do I want to describe GitHub.

142
00:11:16,367 --> 00:11:21,167
It's a repository that you can use for software and text and things like that.

143
00:11:21,167 --> 00:11:23,367
It's version controlled, all these other things.

144
00:11:23,367 --> 00:11:25,407
And then I have to code to the glossary.

145
00:11:25,647 --> 00:11:27,247
I'd have to find the correct

146
00:11:28,607 --> 00:11:31,007
alphabetical order, insert it, paste it.

147
00:11:31,167 --> 00:11:34,847
That completely breaks me out of my flow that I was in for my tutorial.

148
00:11:35,327 --> 00:11:39,647
So instead, I had this thought that was going to distract me from my tutorial.

149
00:11:39,887 --> 00:11:48,447
I went over to the chat window and I said, hey, can you check if I was in my glossary?

150
00:11:49,047 --> 00:11:50,367
That's something I didn't know at the time.

151
00:11:50,847 --> 00:11:54,447
And if not, I added to the glossary at the appropriate location.

152
00:11:54,607 --> 00:11:56,527
Now granted, I sat there and I was

153
00:11:57,007 --> 00:12:06,047
looking, just staring at it, like, We'll do it, we'll do it, and of course it did it, but at this point, and basically, does this look good?

154
00:12:06,047 --> 00:12:25,727
I'm like, Yeah, that's good, or I could have corrected it, but I didn't have to be staring at it, but you could literally hand off that task that was keeping up mental energy and keeping me from breaking me out of my flow, but I didn't want to forget.

155
00:12:26,767 --> 00:12:32,207
and finish where I was in my tutorial and then go back and verify that I worked.

156
00:12:32,487 --> 00:12:40,527
And that's the kind of thing that we can now do with these basically local agents.

157
00:12:41,367 --> 00:12:51,327
In this case, it's a coding agent is what you're working with.

158
00:12:51,407 --> 00:12:54,927
So beyond these retrieval augmented systems,

159
00:12:55,407 --> 00:12:59,007
Or generation systems, which is basically you're taking a folder and you're making it so you can chat with it.

160
00:13:00,447 --> 00:13:04,527
This guy came up with, you may have heard of Andre Karpathy.

161
00:13:04,687 --> 00:13:06,647
He's really big in the AILM field.

162
00:13:07,167 --> 00:13:15,607
He basically just tweeted out or posted something that's like, you know, I feel like we're doing augmented generation grog.

163
00:13:16,447 --> 00:13:19,007
I feel like we have the source documents.

164
00:13:20,127 --> 00:13:25,327
I feel like we should just have the large language model extract the important information and create a wiki.

165
00:13:26,127 --> 00:13:27,887
And then we can just talk with the wiki.

166
00:13:27,967 --> 00:13:31,967
That way, every time I ask a question, it doesn't have to go through 1,000 documents.

167
00:13:32,127 --> 00:13:37,567
It can go through this much smaller curated data set based off of those documents.

168
00:13:37,567 --> 00:13:42,687
So you still have the ground source of truth, but you have a wiki that was generated from that.

169
00:13:42,967 --> 00:13:46,367
And then you can have the large language model or these agents curate the wiki.

170
00:13:46,367 --> 00:13:48,127
So you don't even have to manage it.

171
00:13:49,167 --> 00:13:50,367
It's like, you know, I think that might work.

172
00:13:50,367 --> 00:13:55,647
And so a day after that, there was like 20 people or 30 people that had already created, oh my God, I got it working.

173
00:13:56,247 --> 00:13:58,887
And honestly, nobody's got it working yet.

174
00:14:00,287 --> 00:14:01,967
It's more complicated.

175
00:14:03,487 --> 00:14:04,407
It's really cool.

176
00:14:04,407 --> 00:14:08,767
And I can show you later after this presentation, because it's not that long.

177
00:14:09,967 --> 00:14:15,167
It's a demo, so I'll hook up my computer and I'll show you.

178
00:14:15,567 --> 00:14:21,647
But what people are doing, it helps with linking entities.

179
00:14:22,127 --> 00:14:24,367
And what I did was I said,

180
00:14:24,887 --> 00:14:27,607
He kindly gave us all these nice JSON files.

181
00:14:27,607 --> 00:14:37,207
So I just said, hey, Claude, can you create an obsidian vault that has all the connections for people, the speakers, what they're talking about, and all those kinds of things.

182
00:14:37,207 --> 00:14:43,167
And you can ask questions like, since you know who I am, who should I connect with?

183
00:14:43,567 --> 00:14:52,847
Of the speakers, who would be most interested in working with me or I should connect with because they're doing cool stuff similar to what I'm doing?

184
00:14:54,047 --> 00:15:01,407
it completely changes how we do conferences and how we interact with networking.

185
00:15:01,887 --> 00:15:12,127
Because not only do you have that, you can ask those questions, but now you have little profiles of everybody that might be at the conference and a picture and stuff like that.

186
00:15:12,487 --> 00:15:13,967
And I can show you that later.

187
00:15:14,687 --> 00:15:22,127
But the main challenge that we have with these curated knowledge bases is you have to extract data.

188
00:15:23,007 --> 00:15:26,447
and the extraction process or entities, like what's important.

189
00:15:26,447 --> 00:15:33,567
So if you were looking at a series of books, say like Harry Potter, you might chunk the book into different scenes.

190
00:15:34,127 --> 00:15:40,447
And in that scene, you'd extract the characters, the items that he used, the places that they were at.

191
00:15:41,887 --> 00:15:44,927
And that, once you have that, you would have that in the Wiki.

192
00:15:44,927 --> 00:15:46,687
It'd be these pages based on scenes.

193
00:15:47,647 --> 00:15:48,767
You can't ask

194
00:15:50,447 --> 00:15:54,527
A rag system, something like, How many times was how many scenes was Harry Potter in?

195
00:15:56,127 --> 00:16:00,287
It would fail, but if you had pre-computed that...

196
00:16:01,087 --> 00:16:01,807
You can do that.

197
00:16:01,807 --> 00:16:09,807
You can actually compute over the Wiki, and that's where you'd want to extract the useful information and every data type you have in there.

198
00:16:10,047 --> 00:16:11,327
So that was a fantasy book.

199
00:16:11,327 --> 00:16:13,087
But what if you have a non-fiction book?

200
00:16:13,407 --> 00:16:16,607
That's going to have very different items that you might want to extract.

201
00:16:16,607 --> 00:16:17,967
Maybe it was a textbook.

202
00:16:18,287 --> 00:16:22,687
Well, each chapter might have useful information that's relevant to the topics that were covered.

203
00:16:22,927 --> 00:16:26,527
So chapter, topic, et cetera.

204
00:16:27,647 --> 00:16:30,847
But you can see the interactions of these entities and you can compute over them.

205
00:16:31,007 --> 00:16:32,367
How many places were there?

206
00:16:32,767 --> 00:16:37,407
What was the place that the majority of the characters interacted at?

207
00:16:37,407 --> 00:16:39,047
Those are the kinds of questions you'll be able to do.

208
00:16:39,047 --> 00:16:46,807
And you could do that with any of your documents and information that you might have in your companies.

209
00:16:46,807 --> 00:16:53,887
Some of the challenges are calibrating those extraction pipelines.

210
00:16:54,367 --> 00:16:59,887
Every time you get a new type, you're going to have to, or data type, you're going to have to have different extraction entities.

211
00:17:00,247 --> 00:17:01,247
What are those entities?

212
00:17:01,967 --> 00:17:03,007
Are they really useful?

213
00:17:04,767 --> 00:17:06,607
You can have it do it, and that's what people are doing.

214
00:17:06,607 --> 00:17:09,967
The quick and dirty, it's like, oh, Claude, just extract useful entities.

215
00:17:10,047 --> 00:17:13,727
Like, Claude doesn't know, or ChatGPT doesn't know what those entities are.

216
00:17:13,727 --> 00:17:18,247
You know, because you are a domain expert, and you have to be doing that manually.

217
00:17:18,247 --> 00:17:21,007
That's what's taking the longer time for this.

218
00:17:21,647 --> 00:17:38,527
Sure, these large language models, these agents are super powerful, but if you want to actually significantly increase your productivity, you have to take your knowledge and insert it into the agent so that it can do what you're doing and do it well.

219
00:17:40,687 --> 00:17:43,327
Error handling, like there's duplications.

220
00:17:43,567 --> 00:17:45,727
contradiction detection and resolution.

221
00:17:45,727 --> 00:17:56,047
So if I was to extract information from scientific literature, one paper might say gene X does Y, and another one would say, well, gene X actually does Z.

222
00:17:56,047 --> 00:17:57,007
Well, are they right?

223
00:17:57,647 --> 00:17:59,327
Maybe it goes up or down.

224
00:17:59,567 --> 00:18:01,647
How do I know which is the correct one?

225
00:18:02,047 --> 00:18:06,047
Maybe they're both correct, and it depends on the scientific or experimental setup.

226
00:18:06,527 --> 00:18:09,127
So that resolution of those errors is going to be important.

227
00:18:09,487 --> 00:18:12,687
It's not necessarily just going to be, oh, well, the consensus is this, so it must be this.

228
00:18:12,847 --> 00:18:18,207
It might in some cases, but you'll have to figure that out depending on your particular domain.

229
00:18:19,207 --> 00:18:20,687
And then being able to scale that.

230
00:18:20,847 --> 00:18:34,447
Because at some point, I'm finding so far is unless you have a very strict template in your wikis, as you ask questions and then you add that to the wiki, can you add it in a way that it still fits that structure?

231
00:18:35,247 --> 00:18:37,007
It starts getting messy and kind of

232
00:18:37,527 --> 00:18:38,207
all over the place.

233
00:18:38,207 --> 00:18:41,167
And I don't think anybody has a really good solution for that yet.

234
00:18:44,607 --> 00:18:58,367
After using chatbots, the next level up, which people are just now figuring out, and I mean, I'm only like 3 weeks or two weeks into this because I was ignoring the field, is our skills.

235
00:18:59,007 --> 00:19:05,327
And I think of skills as prompts with computer access, okay?

236
00:19:06,127 --> 00:19:06,527
So

237
00:19:07,487 --> 00:19:09,847
You have prompting, so you have to know prompt engineering.

238
00:19:09,847 --> 00:19:12,447
You have to be able to create a good prompt that does something you want to do.

239
00:19:13,247 --> 00:19:16,327
Skills are prompts for computer access.

240
00:19:16,327 --> 00:19:23,247
Technically, you're using skills or agents as well.

241
00:19:23,287 --> 00:19:24,887
So ChatGPT is a chat agent.

242
00:19:25,007 --> 00:19:27,327
Those agents are executing skills on your behalf.

243
00:19:28,527 --> 00:19:33,247
A web search, file ingestion, image generation, those are all skills.

244
00:19:33,567 --> 00:19:37,167
that was developed that you're using already on a daily basis.

245
00:19:37,567 --> 00:19:38,687
You just didn't make them yet.

246
00:19:39,247 --> 00:19:43,407
It's kind of like, these 10 prompts will do everything you need to do.

247
00:19:43,487 --> 00:19:44,607
That was such a dumb idea.

248
00:19:44,607 --> 00:19:48,447
Now, if you know these 100 prompts, here's a prompt library.

249
00:19:48,847 --> 00:19:50,607
Same thing's happening now with skills.

250
00:19:51,567 --> 00:19:53,807
People say, I have all the skills you will need.

251
00:19:53,807 --> 00:19:55,807
No, ignore all of that.

252
00:19:55,967 --> 00:19:56,767
That's just the hype.

253
00:19:57,007 --> 00:19:58,527
Learn how to write your own skills.

254
00:19:59,007 --> 00:20:01,887
And in fact, this is getting easier because you can actually ask,

255
00:20:02,527 --> 00:20:07,887
Claude or ChatGPT, they have skill builder skills that will help you get that first draft.

256
00:20:10,207 --> 00:20:12,207
And then people are using codecs and code.

257
00:20:12,207 --> 00:20:19,327
Those are coding agents that are code execution, executing commands, editing files, and so forth.

258
00:20:19,647 --> 00:20:26,727
And as I was saying, these skill builders, that's another one of those feedback loops.

259
00:20:26,727 --> 00:20:28,527
You're going to describe the process,

260
00:20:28,887 --> 00:20:29,727
what you're doing.

261
00:20:30,047 --> 00:20:32,127
It will build you on a first draft of a skill.

262
00:20:32,447 --> 00:20:34,607
And then you say, let me look at this.

263
00:20:34,607 --> 00:20:35,887
You say, oh, I don't like this.

264
00:20:35,887 --> 00:20:36,687
I don't like this.

265
00:20:36,927 --> 00:20:38,527
Can you change the skill to make it better?

266
00:20:38,687 --> 00:20:39,727
And then it iterates.

267
00:20:39,807 --> 00:20:44,367
You keep iterating until you have something that gets to 95 plus of what you want to do.

268
00:20:47,207 --> 00:20:50,047
And my example is, I've been working on a project.

269
00:20:50,687 --> 00:20:52,207
It's this website.

270
00:20:52,567 --> 00:20:53,647
It's 100 days.

271
00:20:54,207 --> 00:21:06,687
So a lot of times we have these workshops, you may have gone to a bunch for AI or whatnot, that they promise you that you will learn everything you need to know in this four-hour workshop, or maybe it's 4 hours over 2 days.

272
00:21:07,167 --> 00:21:18,887
The truth is, it's just an information dump, and unless you're using it on a regular basis going forward, most people will leave that workshop feeling great, and then two weeks later have forgotten everything and it's just gone.

273
00:21:19,647 --> 00:21:21,967
The way you get around some of that, you see these

274
00:21:22,447 --> 00:21:25,487
Websites are like 100 days of Python.

275
00:21:25,927 --> 00:21:31,247
And basically every day you spend an hour learning a little bit of Python, and then it builds and builds and builds and builds.

276
00:21:31,327 --> 00:21:33,327
And after 100 days, it's like riding a bike.

277
00:21:34,527 --> 00:21:36,127
It never goes away.

278
00:21:36,527 --> 00:21:37,167
So I'm trying to do that.

279
00:21:37,167 --> 00:21:42,287
It's been taking me about a week to go to create a single day as I envision it.

280
00:21:42,287 --> 00:21:46,207
So I'm kind of this, I have the broad idea of what I want.

281
00:21:46,527 --> 00:21:48,687
So I feel like I'm an architect and I know how to do it.

282
00:21:48,927 --> 00:21:50,127
I know how I want to do it.

283
00:21:50,607 --> 00:21:51,807
You're like, well, why don't you just have

284
00:21:52,207 --> 00:21:53,967
ChatGPT, right, the whole thing.

285
00:21:54,367 --> 00:21:57,487
It's because it doesn't have my vision.

286
00:21:58,207 --> 00:22:05,007
It doesn't have how I, all the experience I have seen as I've trained students and taught them bioinformatics.

287
00:22:06,527 --> 00:22:08,127
And I want it a specific way.

288
00:22:08,487 --> 00:22:16,047
And so what we do is, or some of the challenges we have is like this content has YAML front matter.

289
00:22:16,287 --> 00:22:19,087
I have very specific requirements that I want.

290
00:22:19,087 --> 00:22:21,567
I don't want any page to be more than 2,000 words.

291
00:22:21,927 --> 00:22:24,527
And I want to make sure that all the content I have is updated.

292
00:22:25,087 --> 00:22:26,207
And so what does that look like?

293
00:22:26,207 --> 00:22:28,687
So I actually, so this is kind of fun.

294
00:22:29,647 --> 00:22:30,367
You may not know this.

295
00:22:30,367 --> 00:22:39,087
You can actually ask Claude or Codex to make a diagram of your process or of your website or of your documentation.

296
00:22:39,327 --> 00:22:45,367
And it will actually spit out something that looks like this and create a flow chart for you, which is extremely helpful.

297
00:22:46,607 --> 00:22:50,047
So in this case, I had my, I'm going to have a request.

298
00:22:50,367 --> 00:23:00,607
I gave it something like, I want you to provide a tutorial that helps a user get a GitHub account and create their first repo and put a few lines of text in it.

299
00:23:01,007 --> 00:23:01,807
Super simple.

300
00:23:02,207 --> 00:23:10,527
But I want it in a way that has roughly 4 subpages.

301
00:23:10,687 --> 00:23:11,727
It has an introduction.

302
00:23:11,727 --> 00:23:12,687
It has a wrap up.

303
00:23:12,927 --> 00:23:13,807
We have that here.

304
00:23:15,007 --> 00:23:18,687
And it may depend on the type of the day.

305
00:23:19,087 --> 00:23:20,847
This is where I'm talking about types of data.

306
00:23:21,647 --> 00:23:24,847
It might be, in this case, it's basic.

307
00:23:26,767 --> 00:23:28,847
I needed them to know about GitHub tools.

308
00:23:29,007 --> 00:23:30,047
So there are different tools.

309
00:23:30,207 --> 00:23:32,527
But it might also be, how do you use a tool?

310
00:23:32,927 --> 00:23:35,487
Or it might be biology in the background.

311
00:23:35,647 --> 00:23:41,087
But that type of data that I'm introducing will have different specifications of how I want it to be done.

312
00:23:41,847 --> 00:23:42,367
And so

313
00:23:42,767 --> 00:23:55,487
It has parts of the skill, which is just a giant text file at this point, that describes all of these steps for me, and then it has a self-check at the end, and finally I'll write it to the repo.

314
00:23:55,487 --> 00:23:59,847
So, what you'll see is what it did was it actually...

315
00:23:59,927 --> 00:24:03,487
It created day nine and day 10, which is what I asked it to.

316
00:24:03,567 --> 00:24:06,847
It had the intro, it had the wrap up, it had the read me file.

317
00:24:07,327 --> 00:24:09,647
All of that was generated in a few minutes.

318
00:24:10,207 --> 00:24:13,647
And then I went through it, and I had about 11 changes I wanted to do.

319
00:24:13,887 --> 00:24:18,447
It auto-generated questions in the front matter, which is what I asked it to do.

320
00:24:18,447 --> 00:24:21,527
But it only had three options instead of four.

321
00:24:21,527 --> 00:24:22,607
I'm like, I want it to be 4.

322
00:24:24,527 --> 00:24:25,727
And that could be variable.

323
00:24:25,727 --> 00:24:30,447
If you don't specify this in these skill files, you might get two, you might get 4.

324
00:24:33,407 --> 00:24:35,007
And it was incredible.

325
00:24:35,007 --> 00:24:43,487
So something that would have taken me two weeks was done in about 10 minutes, which left me as the expert, as in the domain, just having to review it.

326
00:24:44,287 --> 00:24:46,447
And I'm not saying that, okay, now just create the next two days.

327
00:24:46,447 --> 00:24:53,087
No, I would still need to sit down and plan out what those next two days should be because I'm architecting this 100 days

328
00:24:53,727 --> 00:24:55,727
And I need to know what I want to do.

329
00:24:56,127 --> 00:25:03,247
And it won't be able to do as good of a job compared to what I am envisioning.

330
00:25:03,407 --> 00:25:07,487
Like if you want it to do what you envision, you have to sit down and outline it.

331
00:25:07,967 --> 00:25:14,767
And then if you create a skill, it can go out, do the searches, and actually give you a first draft, which then you can polish.

332
00:25:14,967 --> 00:25:16,847
And that's where you're going to get that productivity.

333
00:25:17,567 --> 00:25:20,407
So the first prompt is not do my job?

334
00:25:21,167 --> 00:25:21,647
Yes.

335
00:25:22,047 --> 00:25:22,407
OK.

336
00:25:23,247 --> 00:25:27,247
Could you quickly define front manner?

337
00:25:27,887 --> 00:25:31,047
And there is YAML that you mentioned.

338
00:25:31,047 --> 00:25:33,087
And some people might not know YAML.

339
00:25:33,967 --> 00:25:35,887
I don't even know if I know how to describe YAML.

340
00:25:35,887 --> 00:25:50,607
Basically, at the top of a website page, there are variables that are not shown when you show the page that you can specify that can be

341
00:25:51,407 --> 00:25:53,407
used as content in the page.

342
00:25:54,447 --> 00:26:00,207
For example, I put my questions in the YAML, so it doesn't display on the page, but it's tied to the page.

343
00:26:00,767 --> 00:26:15,327
And then later on, the Jekyll website will go through all the pages for that particular day and pull all those questions from the front matter and then create a quiz page for me, for example.

344
00:26:15,727 --> 00:26:21,567
So I needed that in that front matter in order for the little script to run later on when they're taking the quiz.

345
00:26:23,487 --> 00:26:24,767
I think that explains most of it.

346
00:26:25,887 --> 00:26:26,287
Oh, sorry.

347
00:26:27,007 --> 00:26:27,727
Agents.

348
00:26:27,887 --> 00:26:31,327
So chat bot, skills, agents.

349
00:26:32,847 --> 00:26:36,887
Took me a while to wrap my head around the difference between a skill and agent.

350
00:26:36,887 --> 00:26:39,327
I don't know if anybody else is struggling with that.

351
00:26:39,807 --> 00:26:41,647
But the way I think of it is

352
00:26:42,607 --> 00:26:49,007
at the highest level is an agent is a skill with a separate context window.

353
00:26:49,487 --> 00:26:54,127
Has anybody used Claude and hit the 200,000 tokens limit?

354
00:26:54,967 --> 00:26:56,207
So I know people have done that.

355
00:26:56,527 --> 00:27:07,327
So first off, try never to hit that limit because apparently keeping all of that context in your window multiplies the number of tokens you use.

356
00:27:07,727 --> 00:27:08,207
So

357
00:27:08,847 --> 00:27:17,007
As soon as you get to the next feature that you're trying to development or the next question and you don't need any previous context, start a new window or clear the window.

358
00:27:17,927 --> 00:27:19,967
OK, because you will use fewer tokens.

359
00:27:23,847 --> 00:27:33,727
When you want an agent is when you want to minimize how much of that context you're generating from polluting your main context window.

360
00:27:33,887 --> 00:27:37,087
So when you're chatting with a chat bot, it has a history.

361
00:27:37,087 --> 00:27:38,367
It has the context window.

362
00:27:39,167 --> 00:27:48,047
You might say, hey, can you go out and search kittens and puppies for me, as an example, or market strategies.

363
00:27:48,527 --> 00:27:52,287
But you only want the final output from that search.

364
00:27:52,367 --> 00:27:56,607
You don't care about all the websites it did and all the different texts it brought in to figure that out.

365
00:27:57,007 --> 00:28:03,087
That's just polluting your main context window with information you don't care about.

366
00:28:04,367 --> 00:28:13,807
So you launch a sub-agent with its own context window that can then store all of that context, produce a final output, and then bring that into your main context window.

367
00:28:13,807 --> 00:28:18,447
So you're only grabbing what you need for the next step.

368
00:28:18,447 --> 00:28:21,647
And that's what agents and sub-agents are useful for.

369
00:28:27,487 --> 00:28:31,487
Now that we've talked about chatbots, skills, and agents,

370
00:28:32,047 --> 00:28:34,767
How am I using it in my daily work?

371
00:28:34,767 --> 00:28:43,807
Well, one of the things that I've realized is that one-off software development is now basically the norm.

372
00:28:43,807 --> 00:28:47,407
Like, it shouldn't be something that is restricting us from accomplishing it.

373
00:28:47,487 --> 00:28:49,647
used to be that you had to have...

374
00:28:50,687 --> 00:28:57,807
somebody that knew everything about coding and programming to develop a software for whatever your need is.

375
00:28:57,967 --> 00:29:04,287
But now, if you just need something that will get you through and maybe it doesn't have to be super finely polished, you can vibe code it.

376
00:29:05,727 --> 00:29:10,767
If you want something that's more permanent, it's better to have somebody that has programming experience because they will do a better job.

377
00:29:11,087 --> 00:29:14,847
Because they have that domain expertise and they can prompt it

378
00:29:15,407 --> 00:29:17,447
in a way that will give you a better result.

379
00:29:17,487 --> 00:29:19,167
They know what libraries to use.

380
00:29:19,487 --> 00:29:23,167
They know what programming language to use that makes it most efficient.

381
00:29:23,407 --> 00:29:27,087
So I don't think they should be getting rid of programmers.

382
00:29:27,447 --> 00:29:35,407
Yeah, there may be someday that you have an automatic programmer, but basically they become architects with a broader vision for what you want.

383
00:29:36,607 --> 00:29:42,927
I've tried to write apps or have it just write an app that I've envisioned.

384
00:29:43,407 --> 00:29:44,927
And it gets

385
00:29:45,247 --> 00:29:57,487
Unwieldy very quickly, and I think you almost have to break that bigger idea up into smaller apps first, so you can get like an app example library, and then from there I think you could take that and say, OK, now make this bigger app.

386
00:30:01,007 --> 00:30:01,967
Oops, sorry.

387
00:30:02,367 --> 00:30:05,327
So the problem I was facing was, anybody uses ZenHub?

388
00:30:05,807 --> 00:30:07,087
It's like an agile.

389
00:30:10,207 --> 00:30:11,967
It's built on top of GitHub, which

390
00:30:13,327 --> 00:30:19,567
is a system or a place where most programmers will put their code for version control and things like that.

391
00:30:19,567 --> 00:30:23,247
I use it as an online notebook for my documentation.

392
00:30:23,327 --> 00:30:24,367
You can make them private.

393
00:30:24,367 --> 00:30:25,007
It's great.

394
00:30:25,647 --> 00:30:26,607
It's all a markdown.

395
00:30:29,167 --> 00:30:34,847
I was using this program, it was actually a Chrome plug-in, to create a Kanban board, which I'll explain in a minute.

396
00:30:35,487 --> 00:30:39,087
And the company was like, hey, I've got a lot of users.

397
00:30:39,087 --> 00:30:41,487
I'm going to pull all that into my local servers.

398
00:30:41,847 --> 00:30:45,087
get rid of the Chrome plugin and then start charging 6 bucks per person.

399
00:30:45,087 --> 00:30:47,087
And I'm like, I don't use your full suite.

400
00:30:47,087 --> 00:30:48,927
I just need this one tool.

401
00:30:49,647 --> 00:30:54,687
And I'm like, you know, so I spent, this was 2023, and we only had the chat bots, no skills, no agents.

402
00:30:55,007 --> 00:30:58,167
I asked, hey, I want to make a Chrome plugin.

403
00:30:58,167 --> 00:31:00,127
I didn't know JavaScript.

404
00:31:00,287 --> 00:31:01,127
I know scripting.

405
00:31:01,127 --> 00:31:03,207
I know programming, but not those particular languages.

406
00:31:03,207 --> 00:31:04,607
And nothing about Chrome plugins.

407
00:31:05,007 --> 00:31:07,167
This would have taken me months to figure out.

408
00:31:07,647 --> 00:31:10,287
And within like 2 days, I had a functional

409
00:31:11,807 --> 00:31:12,487
Kanban board.

410
00:31:12,487 --> 00:31:17,247
So A Kanban board, you visualize a white board, and you divide it into sections.

411
00:31:17,407 --> 00:31:21,847
You have your sticky notes, your post-its, and you have, this is the things that are the tasks.

412
00:31:21,847 --> 00:31:25,647
I'm going to do these tasks today, and these are the ones, and once you finish it, you can move it over.

413
00:31:25,887 --> 00:31:32,927
So a digital version of this, you can create, and it's basically, I like to think of it as like a 2D to-do list, okay?

414
00:31:35,007 --> 00:31:37,727
And I wanted to make a digital version that I didn't have to pay for.

415
00:31:39,807 --> 00:31:40,527
And so I did.

416
00:31:40,887 --> 00:31:52,647
And in fact, if you guys want to play around with it, you can go to this GitHub repo, and you can download the Chrome plugin, follow the directions, and you can load it up.

417
00:31:53,167 --> 00:31:57,727
Now you do have to have a GitHub account, and you'll have to figure out how to add the token.

418
00:31:57,807 --> 00:32:01,327
But anybody that uses GitHub, this shouldn't be too big of a deal.

419
00:32:01,567 --> 00:32:07,407
And then once you have issues, you can literally add issues, and the labels correspond to the columns.

420
00:32:10,767 --> 00:32:11,807
It looks like this.

421
00:32:11,807 --> 00:32:13,247
Wow, these are terrible slides.

422
00:32:13,247 --> 00:32:14,607
I apologize.

423
00:32:14,927 --> 00:32:17,007
But basically, you've got your columns here.

424
00:32:18,207 --> 00:32:19,967
And you can move around the columns.

425
00:32:21,167 --> 00:32:24,447
And you can, I don't know if you saw that very well.

426
00:32:24,527 --> 00:32:26,927
So you can see that the left column became the right column.

427
00:32:27,327 --> 00:32:29,967
And you can move tasks or issues around.

428
00:32:30,367 --> 00:32:31,687
You see how that one got shorter?

429
00:32:34,847 --> 00:32:38,127
And you can do this, you can actually pull this up on any GitHub repo.

430
00:32:38,367 --> 00:32:42,367
Now, if the repo has a lot of issues, I haven't tested it since it failed the first time.

431
00:32:43,167 --> 00:32:48,447
It might say, you've requested too many calls, but I think I made a fix for that.

432
00:32:49,087 --> 00:32:54,927
But the point is that you can have your own private repo and put your own issues and have your own tasks and you can play around with that.

433
00:32:55,647 --> 00:32:56,607
And this solved my problem.

434
00:32:56,607 --> 00:33:01,727
So this one-off software solutions where you just need something to do the thing that you're trying to do,

435
00:33:02,287 --> 00:33:11,087
You can probably do that in-house with somebody that has some basic coding skills without having to be a full-blown programmer.

436
00:33:11,647 --> 00:33:21,007
But what this also means is that any digital product is now basically approaching 0 in its worth.

437
00:33:22,527 --> 00:33:26,607
Because you can say, hey, I really like what this program can do.

438
00:33:26,767 --> 00:33:28,527
Can you reimburse engineer it?

439
00:33:29,087 --> 00:33:31,167
And chances are, within a few days, you could do that.

440
00:33:33,087 --> 00:33:44,127
In fact, it's becoming a problem in my field because before, graduate students and postdocs would create a program to solve a problem, and then they would publish a paper.

441
00:33:44,607 --> 00:33:45,767
And that was great.

442
00:33:46,607 --> 00:33:48,727
They get credit for it, academic, all that kind of stuff.

443
00:33:49,167 --> 00:33:54,447
Now people are taking that and saying, hey, I like this, but I'm going to make it faster and then publish on it.

444
00:33:54,847 --> 00:34:12,687
Instead of having 20 or 30 programs that claim to be the best assembly program in the world, now we might have 100 because everybody's just rewriting it over and over again, which makes determining which software is the best one to use exponentially harder to figure out.

445
00:34:13,007 --> 00:34:16,447
And those people that are rewriting it, are they doing it well?

446
00:34:17,247 --> 00:34:19,087
Did they test every edge case?

447
00:34:19,647 --> 00:34:29,087
So you have to be careful to some degree as to what you're downloading, because people are just making digital products, programs, and you don't know where that came from.

448
00:34:29,087 --> 00:34:36,927
You don't know if they inserted some kind of prompt injection that says, oh, send me all of your Bitcoin tokens or your opening AI keys.

449
00:34:37,327 --> 00:34:41,727
You know, these are serious questions and concerns that you have to have as you go forward.

450
00:34:46,287 --> 00:34:53,327
Some of the other features include, you can open up the task and look at it.

451
00:34:53,807 --> 00:34:59,767
I then took it a step farther, and instead of a Chrome plugin, I said, Hey, I want you to rewrite this Chrome plugin as a...

452
00:34:59,927 --> 00:35:02,367
native Mac app using the Swift language.

453
00:35:02,767 --> 00:35:04,367
And in a one shot, it did it.

454
00:35:05,727 --> 00:35:13,807
I did some programming for fun as a hobby on apps when my kids were a lot younger, because I was looking at the apps, like, these apps are terrible.

455
00:35:13,967 --> 00:35:15,047
I could do a better job.

456
00:35:15,567 --> 00:35:19,647
I never got around to doing it fully, but I learned a lot about app development.

457
00:35:19,647 --> 00:35:23,247
So once I had it in here, I was like, oh, I want this feature and I want this feature.

458
00:35:24,287 --> 00:35:25,567
So I added things like...

459
00:35:26,287 --> 00:35:38,127
retaining the GitHub URL, retaining the access token, dragging cards between columns, dragging columns around so they have different places, color coding them.

460
00:35:38,527 --> 00:35:41,967
I found that once I opened the issue, I wanted to scroll through the issues.

461
00:35:42,207 --> 00:35:51,447
Well, it turns out that all those keyboard shortcuts, like shift, enter and stuff like that, you have to program them in if you want them, right?

462
00:35:51,447 --> 00:35:52,447
I mean, that makes sense.

463
00:35:53,167 --> 00:35:54,287
Seems obvious now.

464
00:35:54,607 --> 00:35:57,087
But so I was able to do that really, really quickly.

465
00:35:57,287 --> 00:36:03,927
And so once you have the initial state, it's fairly straightforward to add small features, and you just keep building on that.

466
00:36:04,207 --> 00:36:11,647
Now, again, ideally, you have somebody that has some coding experience to try to keep away from some of the bloat or the inefficiencies and so forth.

467
00:36:11,967 --> 00:36:16,607
But it's a super powerful tool that you should have in your toolkit going forward.

468
00:36:16,927 --> 00:36:21,487
You don't necessarily need a full-blown coder to solve or create these

469
00:36:22,127 --> 00:36:28,127
user interfaces to help you access your data or to analyze certain things and going forward.

470
00:36:30,207 --> 00:36:31,327
Here's another example.

471
00:36:31,487 --> 00:36:34,527
Actually, it kind of ties back to the graduate student problem.

472
00:36:35,807 --> 00:36:44,127
This guy, Phil Ewells, he's part of a company that creates workflows.

473
00:36:44,367 --> 00:36:49,887
So I run a script, it gives me a result, and then I have another script that gives me a result, and it goes, well,

474
00:36:51,167 --> 00:36:55,407
There are these people that now have workflows that do all those steps.

475
00:36:55,407 --> 00:36:56,527
You can put them all in there.

476
00:36:56,687 --> 00:37:04,567
It distributes all of that compute across supercomputers and then brings it all back automatically for you.

477
00:37:05,007 --> 00:37:05,647
Super awesome.

478
00:37:05,887 --> 00:37:10,887
He looked at one of these and he's like, man, this particular workflow was taking 15 hours.

479
00:37:10,887 --> 00:37:17,207
I was like, I wonder if I could rewrite that in Rust, which is a super fast programming language close to C, C.

480
00:37:18,687 --> 00:37:21,967
but more often used in the scientific community.

481
00:37:22,847 --> 00:37:28,527
And after a week, he had tested, validated, all that kind of stuff, what you're supposed to do.

482
00:37:29,487 --> 00:37:37,247
He went from 15 hours to 15 minutes, a 60-fold improvement in speed.

483
00:37:37,247 --> 00:37:40,287
So if you have any data processing bottlenecks,

484
00:37:40,727 --> 00:37:56,527
and you have existing software that's doing that for you, chances are you can find ways to improve that bottleneck, reduce that time to do the data analysis or generation of documentation or plots or figures and stuff like that.

485
00:37:56,767 --> 00:37:58,527
So that's something else to consider.

486
00:38:01,047 --> 00:38:01,967
I think I'm way ahead of it.

487
00:38:01,967 --> 00:38:03,327
No, I'm actually quite on time.

488
00:38:03,807 --> 00:38:08,687
So in summary, we're shifting from tools to more collaborators.

489
00:38:09,647 --> 00:38:11,247
especially as we get to more agents.

490
00:38:11,887 --> 00:38:26,527
Knowledge is going to be the most important asset that you have in your company, not just the data you have locally, but the people in your company that have that knowledge, that have the skills that you're trying to make skills and agents for to automate.

491
00:38:27,567 --> 00:38:31,767
Software is now disposable, which is crazy to think about.

492
00:38:32,367 --> 00:38:36,607
Skills and agents are being used for repeatable automation.

493
00:38:36,927 --> 00:38:39,167
I think that's where you're going to get the most efficiency gains.

494
00:38:41,407 --> 00:38:56,527
And what the challenges or the constraints we're going to have are reliability, validation, security, governance, like where, how exactly did it do the thing?

495
00:38:56,847 --> 00:38:59,487
You want to keep track of how things were moving forward.

496
00:39:00,567 --> 00:39:02,727
And then I have one final thought.

497
00:39:04,687 --> 00:39:06,447
As you are using implementing AI,

498
00:39:08,207 --> 00:39:11,727
I don't want, I don't think we should be thinking about how can I get rid of employees?

499
00:39:11,727 --> 00:39:15,487
Because I think that is, that's the wrong way to think about it.

500
00:39:15,727 --> 00:39:22,527
You should be thinking about how can I use my existing employees to multiply my production and multiply my profit?

501
00:39:23,327 --> 00:39:32,527
Because your employees are the ones that have the knowledge and have those skills that can utilize these agents to significantly increase your productivity.

502
00:39:33,007 --> 00:39:39,967
If you're using it just to maintain your same level of productivity, you're going to be out-competed going forward.

503
00:39:39,967 --> 00:39:46,767
You want to figure out a way to get your workforce to be able to do, instead of five people, being able to do 50 people's worth of work.

504
00:39:46,767 --> 00:39:47,887
Does that make sense?

505
00:39:49,567 --> 00:39:49,727
Yeah.

506
00:39:49,727 --> 00:39:50,927
Do you guys have questions?

507
00:39:52,367 --> 00:39:53,087
I don't know what I have.

508
00:39:53,487 --> 00:39:54,447
Who has a question?

509
00:39:54,607 --> 00:39:57,007
I'd like to bring the microphone to you so we can capture it.

510
00:39:57,487 --> 00:39:58,047
All righty.

511
00:40:02,047 --> 00:40:05,727
You say the value of a digital product is rapidly approaching 0.

512
00:40:05,807 --> 00:40:09,327
We're looking at new ERP software to run our whole entire business.

513
00:40:09,887 --> 00:40:13,007
Can I really build my own ERP with AI tools?

514
00:40:13,887 --> 00:40:22,927
Maybe not, but if you have access, if you get access to that code base or if you can do it with open source stuff that you can then add your own features to.

515
00:40:23,047 --> 00:40:25,327
So download the trial version and stick it in AI?

516
00:40:25,647 --> 00:40:26,367
I don't know.

517
00:40:26,727 --> 00:40:29,967
At some point it's going to be harder, but my point is that

518
00:40:30,447 --> 00:40:35,327
For the smaller tasks you have, there's a lot of things you can do with one-off software development.

519
00:40:35,727 --> 00:40:45,847
One thing I'm going to tell everybody here, and I probably need to repeat it several times, if you're talking and you're from Coimag or anything like that, and you're having that sort of question, call CIRAS.

520
00:40:46,607 --> 00:40:47,287
End of story.

521
00:40:47,287 --> 00:40:49,967
I don't want to sound cocky or anything.

522
00:40:50,367 --> 00:40:53,327
We know how to help you with those kinds of decisions.

523
00:40:54,007 --> 00:41:00,367
And I see people making those decisions all the time and then coming to me and saying, those idiots don't know what they're doing.

524
00:41:01,447 --> 00:41:03,967
And usually what I say is, can you lift your thumb for me?

525
00:41:06,047 --> 00:41:09,967
Because they don't know what they're asking for when they go to the ARP people.

526
00:41:10,447 --> 00:41:11,647
Is SAS going away?

527
00:41:11,727 --> 00:41:12,767
My belief it is.

528
00:41:13,327 --> 00:41:14,527
So be careful with that.

529
00:41:14,767 --> 00:41:17,567
Don't get involved with that at this point in time.

530
00:41:18,607 --> 00:41:23,247
So for those that might be curious, this is what this Obsidian Wiki Vault might look like.

531
00:41:24,207 --> 00:41:39,567
So these are all people and sessions and information in this vault where you can see who is connected to whom, and you can actually see how they're connected and play around with it.

532
00:41:39,887 --> 00:41:44,207
It's not the most useful interface, but it's still a powerful way to view things.

533
00:41:44,487 --> 00:41:47,567
And then I asked Claude, hey, based on

534
00:41:47,887 --> 00:41:56,007
everybody, all the speakers and people that are speaking, who do you think, based on how you know me, I should interact with?

535
00:41:56,127 --> 00:41:58,527
Who are the main people that I should be networking with?

536
00:41:58,607 --> 00:42:02,127
Who are potential people that might find my services valuable?

537
00:42:05,007 --> 00:42:13,087
So you can take, like I said, going to conferences, networking, it's changed how we do that.

538
00:42:13,567 --> 00:42:19,167
I might have done some of this before, but it took me hours of going through and looking up individuals.

539
00:42:19,167 --> 00:42:21,127
Oh, maybe this person, maybe this.

540
00:42:21,127 --> 00:42:22,527
gives you a jump start on all of that.

541
00:42:23,647 --> 00:42:30,847
Yeah, if you scroll up a little bit, just a little bit, the people you can help, that's me right there, Hans Coink.

542
00:42:31,167 --> 00:42:31,607
There you go.

543
00:42:31,607 --> 00:42:39,327
Yeah, so you nailed it, because we're a vaccine company, and so we use bioinformatics, that sort of thing.

544
00:42:39,647 --> 00:42:40,047
But

545
00:42:40,927 --> 00:42:43,087
I had just kind of more of a general question for you.

546
00:42:44,207 --> 00:42:47,007
showed some windows and I'm a Gemini guy.

547
00:42:47,007 --> 00:42:49,487
So, it's like Ford versus Chevy anymore.

548
00:42:49,487 --> 00:42:50,647
People are talking about this stuff.

549
00:42:50,647 --> 00:43:00,207
But how do you decide which LLM or which, you know, you talked about anthropic, you talked about OpenAI.

550
00:43:00,447 --> 00:43:03,807
I mean, I can sense that you're trying to avoid like paying for anything.

551
00:43:03,807 --> 00:43:04,527
You know what I mean?

552
00:43:04,527 --> 00:43:07,247
So, which, yeah, that's a good thing.

553
00:43:07,247 --> 00:43:09,727
So, but how do you make that decision?

554
00:43:10,847 --> 00:43:23,447
Right now, each of the big contenders, Gemini, Claude, and OpenAI, are all jump leapfrogging over each other for what is the best large language model.

555
00:43:24,127 --> 00:43:31,967
And some would argue, oh, you just, once that happens, you just switch, you know, and you can carry that over in an identity file with all the useful information.

556
00:43:31,967 --> 00:43:35,087
That's actually how Anthropic

557
00:43:35,567 --> 00:43:47,927
got that big boost at the end, if you saw that, why they're making so much money, because Chat or Open AI was like, basically Anthropic said, we're not going to let the government make killer robots that will come and like kill people.

558
00:43:47,927 --> 00:43:48,287
Like that.

559
00:43:49,407 --> 00:43:53,887
We don't think that's like, and then the next day Open AI is like, we just made a contract with the government, but we're not doing it.

560
00:43:54,167 --> 00:44:02,687
And like people ditched them, and then they said, just write in the thing saying, tell me everything you know about me, and then place that into a claw.

561
00:44:02,767 --> 00:44:03,567
And that's how they.

562
00:44:04,927 --> 00:44:06,607
not a big influx of users.

563
00:44:08,127 --> 00:44:13,407
I don't know if it makes a big difference from the foundational model.

564
00:44:13,647 --> 00:44:16,287
Right now, they're fighting over harnesses.

565
00:44:16,767 --> 00:44:25,247
Claude has a harness, and this is how the foundation model interacts and the skills that are in there.

566
00:44:25,487 --> 00:44:32,527
And I've heard arguments that right now it's really cheap and it's getting everybody hooked.

567
00:44:33,007 --> 00:44:37,807
And then at some point, they're just going to jack up the rates, and then we'll be like, well, shit, I don't know what else we can.

568
00:44:38,927 --> 00:44:41,167
I guess I'll have to pay $2,000 a month now.

569
00:44:41,967 --> 00:44:57,127
And so I'm planning on taking advantage of what we can now and use the open source models and all the open source environment and say, OK, I know they're not as good, but how close can I get it to the

570
00:44:58,607 --> 00:45:03,247
the OpenAI or Claude, and can I iterate it?

571
00:45:03,727 --> 00:45:23,247
Is there a way, if I create a skill, say, with these top models first, and then I switch the model to these open source that are like, I don't know, like what, six months, a year behind, which they're exceeding GPT-4, which we thought was the bee's knees back in 2023, can you iterate that?

572
00:45:23,247 --> 00:45:26,767
Is there a way that you can have a test

573
00:45:27,207 --> 00:45:28,287
What is Claude?

574
00:45:28,287 --> 00:45:34,927
Claude calls it the Ralph Wiggum approach, where you just, you keep doing it until it gets you the best result.

575
00:45:35,967 --> 00:45:39,167
I think that's where I'm going to spend my time.

576
00:45:39,327 --> 00:45:40,487
Can I do it all locally?

577
00:45:40,487 --> 00:45:41,727
Because then I'm not spending tokens.

578
00:45:41,727 --> 00:45:44,927
I don't hit my 200,000 token limit or my weekly limit.

579
00:45:45,247 --> 00:45:48,047
If I can get it working, so I have a spare laptop.

580
00:45:48,127 --> 00:45:54,207
I'm going to install either OpenCloud or whatever else I can find, Hermes, and see if I can

581
00:45:54,687 --> 00:45:59,847
give it an e-mail account, give it a Slack, invite it to Slack and just see what I can do with it, and...

582
00:46:00,447 --> 00:46:01,167
experiment.

583
00:46:01,567 --> 00:46:07,887
By no means am I going to put it on my main laptop that has all my stuff on it because that's just that's not a good idea.

584
00:46:09,007 --> 00:46:17,727
All right so we are at the time limit so I would like you to thank Andrew with me for his time and discussion.

585
00:46:20,207 --> 00:46:31,407
I also want to point out that what Andrew was showing you with his connections and everything was one way of taking the structured data that is on the website that you all have access to.

586
00:46:31,927 --> 00:46:35,887
And you can do it in really basic format with some of the prompts and things that we have.

587
00:46:36,447 --> 00:46:41,167
But that structured data that is put in there is for you to be able to use.

588
00:46:41,487 --> 00:46:45,247
Andrew just said, I'm going to use my tool and I'm going to pull the structured data in.

589
00:46:45,407 --> 00:46:49,367
My artificial intelligence is going to figure out how to organize it.

590
00:46:49,767 --> 00:46:51,647
and now I can use it in any way I want.

591
00:46:52,527 --> 00:46:56,847
We've given you some ways to do it online, but you can use any tool you'd like.

592
00:46:57,727 --> 00:46:58,207
Okay?

593
00:46:58,767 --> 00:47:01,327
If you have questions, you can check with the SIRIS desk.

594
00:47:01,327 --> 00:47:02,687
You can ask some questions there.

595
00:47:03,087 --> 00:47:05,167
We will be providing a lot more data.

596
00:47:05,167 --> 00:47:11,247
For instance, everything Andrew said here will be in a part of the structured data.

597
00:47:11,407 --> 00:47:12,847
It'll be a transcript.

598
00:47:13,007 --> 00:47:14,207
You'll be able to put that in.

599
00:47:14,207 --> 00:47:19,407
We'll have what he's shown us on the screen, et cetera, hopefully knocked down to some

600
00:47:20,127 --> 00:47:22,367
chunked information that he was talking about.

601
00:47:22,367 --> 00:47:25,407
So you're going to be able to talk with the entire conference.

602
00:47:26,207 --> 00:47:28,047
Now that's going to be over the next few weeks.

603
00:47:28,567 --> 00:47:28,927
Okay.