WEBVTT - How open source can democratize AI

0:00:04.280 --> 0:00:07.880
<v Speaker 1>Hello, Hello, Welcome to Smart Talks with IBM, a podcast

0:00:07.960 --> 0:00:13.560
<v Speaker 1>from Pushkin Industries, iHeartRadio and IBM. I'm Malcolm Gladwell. This season,

0:00:13.600 --> 0:00:16.919
<v Speaker 1>we're diving back into the world of artificial intelligence, but

0:00:17.040 --> 0:00:22.919
<v Speaker 1>with a focus on the powerful concept of open its possibilities, implications,

0:00:23.079 --> 0:00:26.400
<v Speaker 1>and misconceptions. We'll look at openness from a variety of

0:00:26.440 --> 0:00:30.560
<v Speaker 1>angles and explore how the concept is already reshaping industries,

0:00:30.600 --> 0:00:34.199
<v Speaker 1>ways of doing business and a very notion of what's possible.

0:00:34.560 --> 0:00:38.040
<v Speaker 1>In today's episode, I sat down with mo Duffy, software

0:00:38.080 --> 0:00:41.840
<v Speaker 1>engineering manager at red Hat, who works on instruct Lab,

0:00:42.080 --> 0:00:46.040
<v Speaker 1>a project co developed by red Hat and IBM. Most

0:00:46.040 --> 0:00:50.800
<v Speaker 1>shared with me how this new initiative is revolutionizing AI training,

0:00:51.120 --> 0:00:54.400
<v Speaker 1>making it not only more accessible, but also more inclusive.

0:00:55.280 --> 0:00:59.280
<v Speaker 1>This project, unique in the industry, allows developers to submit

0:00:59.440 --> 0:01:04.800
<v Speaker 1>incremental contributions to one base AI model, creating a continuous

0:01:04.840 --> 0:01:09.200
<v Speaker 1>loop of development, much like normal open source software. By

0:01:09.319 --> 0:01:14.280
<v Speaker 1>leveraging community contributions and IBM's cutting edge granite models, Mo

0:01:14.520 --> 0:01:17.520
<v Speaker 1>in the team of ibmrs and red hatters are paving

0:01:17.560 --> 0:01:20.920
<v Speaker 1>the way for a future where AI development is a

0:01:20.959 --> 0:01:25.920
<v Speaker 1>communal endeavor. Our insights into open source software extend beyond

0:01:26.040 --> 0:01:31.520
<v Speaker 1>technical proficiency to the profound impact of collaborative effort. At

0:01:31.520 --> 0:01:35.440
<v Speaker 1>the heart of Moe's work is a belief in democratizing technology,

0:01:36.000 --> 0:01:40.440
<v Speaker 1>ensuring that AI becomes a tool accessible to all. So

0:01:40.560 --> 0:01:44.360
<v Speaker 1>let's explore how Mo, red Hat and IBM are empowering

0:01:44.360 --> 0:01:48.560
<v Speaker 1>individuals and businesses alike to reshape the future of technology

0:01:48.760 --> 0:01:57.640
<v Speaker 1>through collaboration and innovation. Mo, thank you for joining me

0:01:57.640 --> 0:02:01.000
<v Speaker 1>today so much, for you have just about the most

0:02:01.040 --> 0:02:06.760
<v Speaker 1>Irish name ever. I do very proud you weren't born

0:02:06.760 --> 0:02:07.360
<v Speaker 1>in Ireland.

0:02:07.560 --> 0:02:09.320
<v Speaker 2>No, my grandparents or your grandparents.

0:02:09.480 --> 0:02:10.560
<v Speaker 1>I see, where did you grow up?

0:02:10.960 --> 0:02:11.960
<v Speaker 2>New York Queens?

0:02:12.280 --> 0:02:15.520
<v Speaker 1>Oh you're a see. So tell me a little bit

0:02:15.560 --> 0:02:18.280
<v Speaker 1>about how how you got to red hat? What was

0:02:18.320 --> 0:02:18.800
<v Speaker 1>your path?

0:02:19.280 --> 0:02:21.800
<v Speaker 2>When I was in high school, it was a chatty girl,

0:02:22.160 --> 0:02:24.640
<v Speaker 2>teenage girl on the phone. We had one phone line.

0:02:24.960 --> 0:02:27.760
<v Speaker 2>My older brother was studying at the local state college

0:02:27.800 --> 0:02:30.360
<v Speaker 2>computer science, and he had to tell that end to

0:02:30.400 --> 0:02:33.240
<v Speaker 2>compile his homework. One phone line and I'm on it

0:02:33.280 --> 0:02:36.239
<v Speaker 2>all the time. He got very frustrated and he needed

0:02:36.240 --> 0:02:38.799
<v Speaker 2>a compiler to do his homework. So he bought red

0:02:38.840 --> 0:02:43.240
<v Speaker 2>hat Linux from a CompUSA, brought it home and that

0:02:43.360 --> 0:02:45.320
<v Speaker 2>was on the family computer. So I learned Linux and

0:02:45.320 --> 0:02:48.040
<v Speaker 2>I started playing around with it. I really liked it

0:02:48.120 --> 0:02:52.000
<v Speaker 2>because you could customize everything, like the entire user interface.

0:02:52.360 --> 0:02:54.560
<v Speaker 2>You could actually modify the code of the programs you

0:02:54.560 --> 0:02:56.840
<v Speaker 2>were using to do what you wanted. And for me,

0:02:56.880 --> 0:02:59.079
<v Speaker 2>it was really cool because, especially when you're a kid

0:02:59.120 --> 0:03:00.880
<v Speaker 2>and like people tell you this is the way things

0:03:00.880 --> 0:03:02.639
<v Speaker 2>are and you just have to deal with it, it's

0:03:02.720 --> 0:03:04.280
<v Speaker 2>nice to be like, I'm going to make things the

0:03:04.320 --> 0:03:07.600
<v Speaker 2>way I want, modify the code and playing. Yeah, it

0:03:07.680 --> 0:03:09.520
<v Speaker 2>was amazing and it was just such a time and

0:03:09.560 --> 0:03:12.600
<v Speaker 2>like before it was cool, I was doing it and

0:03:13.360 --> 0:03:15.320
<v Speaker 2>what I saw on that is sort of the potential

0:03:15.360 --> 0:03:18.400
<v Speaker 2>like number one of like a community of people working together.

0:03:18.440 --> 0:03:21.480
<v Speaker 2>And like the Internet existed, it was slow, it involved modems,

0:03:21.880 --> 0:03:23.840
<v Speaker 2>but there were people that you could talk to who

0:03:23.840 --> 0:03:27.600
<v Speaker 2>would give you tips and you'd share information, and this

0:03:27.720 --> 0:03:31.400
<v Speaker 2>collaborative building something together is really something special. Right. I

0:03:31.440 --> 0:03:35.240
<v Speaker 2>could file a complaint to whatever large software company made

0:03:35.240 --> 0:03:37.440
<v Speaker 2>whatever software I was into, or I could go to

0:03:37.480 --> 0:03:39.640
<v Speaker 2>an open source software community and be like, hey, guys,

0:03:39.680 --> 0:03:42.000
<v Speaker 2>I think we should do this. I'm like, yeah, okay,

0:03:42.080 --> 0:03:44.200
<v Speaker 2>I'll help. I'll pitch in so you don't feel powerless.

0:03:44.200 --> 0:03:45.600
<v Speaker 2>You feel like you can have an impact, and that

0:03:45.640 --> 0:03:49.560
<v Speaker 2>was really exciting to me. However, open source software has

0:03:49.680 --> 0:03:52.920
<v Speaker 2>a reputation for not having the best user interface, not

0:03:52.960 --> 0:03:57.480
<v Speaker 2>the best user experience. So I ended up studying computer

0:03:57.560 --> 0:04:01.760
<v Speaker 2>science and electronic media, and then I did human computeraction

0:04:01.920 --> 0:04:04.920
<v Speaker 2>as my master's And my thought was, wouldn't it be

0:04:05.000 --> 0:04:08.960
<v Speaker 2>nice if this free software accessible to anybody, if it

0:04:09.000 --> 0:04:11.120
<v Speaker 2>was easier to use, some more people could use it

0:04:11.160 --> 0:04:14.440
<v Speaker 2>and take advantage of it. And so, long story short,

0:04:14.600 --> 0:04:16.919
<v Speaker 2>I ended up going to Red Hat saying, Hey, I

0:04:16.920 --> 0:04:18.520
<v Speaker 2>want to learn how you guys work. Let me embed

0:04:18.560 --> 0:04:21.159
<v Speaker 2>in your team draft out of my graduate program, and

0:04:21.200 --> 0:04:22.880
<v Speaker 2>I'm like, I want to do this for a living.

0:04:23.000 --> 0:04:25.560
<v Speaker 2>This is cooler. So I thought this is the way

0:04:25.600 --> 0:04:27.240
<v Speaker 2>to go, and I've been there ever since. They haven't

0:04:27.240 --> 0:04:28.080
<v Speaker 2>been able to get rid of me.

0:04:30.160 --> 0:04:32.880
<v Speaker 1>To backtrack just a little bit, you were talking about

0:04:32.920 --> 0:04:36.640
<v Speaker 1>the sense of community that surrounds this way of thinking

0:04:36.680 --> 0:04:40.400
<v Speaker 1>about software. Talk a little bit more about what that

0:04:40.400 --> 0:04:43.240
<v Speaker 1>community is like, the benefits of that community, why it

0:04:43.240 --> 0:04:44.080
<v Speaker 1>appeals to you.

0:04:44.560 --> 0:04:46.880
<v Speaker 2>Sure, well, you know part of the reason I actually

0:04:46.960 --> 0:04:50.040
<v Speaker 2>ended up going to the graduate school track. Suddenly you're

0:04:50.080 --> 0:04:53.240
<v Speaker 2>a peer of your professors and you're working side by

0:04:53.320 --> 0:04:56.360
<v Speaker 2>side with them. At some point they retire and you're

0:04:56.400 --> 0:04:59.839
<v Speaker 2>in the next generation. So it's sharing information, building on

0:04:59.839 --> 0:05:02.640
<v Speaker 2>the work of others in sort of this cycle that

0:05:02.839 --> 0:05:07.240
<v Speaker 2>extends past human lifespan and in the same way, like

0:05:07.279 --> 0:05:10.719
<v Speaker 2>the open source model is very similar, but you're actually

0:05:10.760 --> 0:05:13.400
<v Speaker 2>you're building something, and it's something in me. I'm just

0:05:13.800 --> 0:05:16.200
<v Speaker 2>really attracted, Like I don't like talking about stuff. I

0:05:16.320 --> 0:05:20.280
<v Speaker 2>like doing stuff with open source software. The software doesn't

0:05:20.279 --> 0:05:23.919
<v Speaker 2>cost anything, the code is out there, generally uses open

0:05:23.960 --> 0:05:27.359
<v Speaker 2>standards for the file formats. I can open up files

0:05:27.400 --> 0:05:30.039
<v Speaker 2>that I created and open source tools as a high

0:05:30.040 --> 0:05:33.599
<v Speaker 2>school student today because they were using open formats and

0:05:33.640 --> 0:05:36.120
<v Speaker 2>that software still exists, I can still compile the code

0:05:36.240 --> 0:05:39.400
<v Speaker 2>and it's an active community project. Like these things can

0:05:39.440 --> 0:05:42.120
<v Speaker 2>outlast any single company in the same way that the

0:05:42.160 --> 0:05:44.760
<v Speaker 2>academic community has been going on for so many years,

0:05:44.760 --> 0:05:47.320
<v Speaker 2>and hopefully we'll continue moving on. So it's sort of

0:05:47.360 --> 0:05:50.240
<v Speaker 2>like not just the community around it, but just the

0:05:50.320 --> 0:05:53.440
<v Speaker 2>knowledge sharing and also bringing up the next generation as well.

0:05:53.480 --> 0:05:55.120
<v Speaker 2>Like all of that stuff really appealed to me. And

0:05:55.160 --> 0:05:58.279
<v Speaker 2>also at the center of it the fact that we

0:05:58.320 --> 0:06:01.680
<v Speaker 2>could democratize it by following this open source process and

0:06:01.800 --> 0:06:03.760
<v Speaker 2>feel like we have some control. We're not at the

0:06:03.760 --> 0:06:06.839
<v Speaker 2>mercy of some faceless corporation making changes and we have

0:06:06.920 --> 0:06:08.840
<v Speaker 2>no impact. Like that really appealed to me too.

0:06:09.640 --> 0:06:14.719
<v Speaker 1>For those of us who are not software phisionados, take

0:06:14.720 --> 0:06:19.120
<v Speaker 1>a step backwards and give me a kind of description

0:06:19.240 --> 0:06:22.120
<v Speaker 1>of terms. What's the opposite of open to proprietary?

0:06:22.200 --> 0:06:23.680
<v Speaker 2>Proprietary is what we say, So.

0:06:24.520 --> 0:06:28.320
<v Speaker 1>Specifically and practically, the difference would be what between something

0:06:28.320 --> 0:06:30.160
<v Speaker 1>that was opened us in something that was proprietary.

0:06:30.240 --> 0:06:32.839
<v Speaker 2>Sure, so there's a lot of difference. So with open

0:06:32.880 --> 0:06:36.839
<v Speaker 2>source software you get these rights. When you're given the software,

0:06:36.920 --> 0:06:39.320
<v Speaker 2>you get the right to be able to share it.

0:06:39.400 --> 0:06:41.839
<v Speaker 2>And depending on the lot, different licenses that are considered

0:06:41.880 --> 0:06:44.839
<v Speaker 2>open source have different little things that you have to

0:06:44.839 --> 0:06:50.640
<v Speaker 2>be aware of. With proprietary code, it's one copyright the company.

0:06:50.680 --> 0:06:53.040
<v Speaker 2>Even a lot of times, when you sign your employment

0:06:53.040 --> 0:06:55.360
<v Speaker 2>contract for a software company and you write code for them,

0:06:55.680 --> 0:06:57.320
<v Speaker 2>you don't own it. You sign over your rights to

0:06:57.360 --> 0:06:59.280
<v Speaker 2>the company, So if you leave the company, the code

0:06:59.279 --> 0:07:01.599
<v Speaker 2>doesn't go with you. It stays in the ownership of

0:07:01.600 --> 0:07:04.520
<v Speaker 2>that company. So then one one company buys out another

0:07:04.560 --> 0:07:06.440
<v Speaker 2>and kills a product that code's gone.

0:07:06.560 --> 0:07:10.480
<v Speaker 1>It's gone. For a business, Why would a business want

0:07:10.560 --> 0:07:14.360
<v Speaker 1>to be have open source code as opposed to proprietary.

0:07:13.760 --> 0:07:15.800
<v Speaker 2>Well, for the same reasons. Like say you're a business.

0:07:16.280 --> 0:07:20.160
<v Speaker 2>You've invested all this money into this software platform, right,

0:07:20.640 --> 0:07:23.800
<v Speaker 2>and you've upskilled your employees on it, and it's a

0:07:23.800 --> 0:07:26.640
<v Speaker 2>core part of your business, and then a few years

0:07:26.680 --> 0:07:30.000
<v Speaker 2>later that company goes out of business or something happens,

0:07:30.160 --> 0:07:33.520
<v Speaker 2>or even something less drastic. You really need this feuture.

0:07:33.840 --> 0:07:36.960
<v Speaker 2>But for the company that makes the software, it's not

0:07:37.000 --> 0:07:39.840
<v Speaker 2>in their best interests. It's not worth the investment. They're

0:07:39.840 --> 0:07:41.760
<v Speaker 2>not going to do it. How do you get that feature?

0:07:42.120 --> 0:07:44.680
<v Speaker 2>You either have to completely migrate to another solution, and

0:07:44.680 --> 0:07:46.600
<v Speaker 2>this is something that's core at your business, that's going

0:07:46.640 --> 0:07:48.760
<v Speaker 2>to be a big deal to migrate. But if it's

0:07:48.800 --> 0:07:52.560
<v Speaker 2>open source, you could either hire a team of experts.

0:07:52.760 --> 0:07:55.320
<v Speaker 2>You could hire software engineers who are able to go

0:07:55.400 --> 0:07:58.640
<v Speaker 2>do this for you. Go in the upstream software community,

0:07:59.120 --> 0:08:02.120
<v Speaker 2>implement the future that you want, and it'll be rolled

0:08:02.160 --> 0:08:04.720
<v Speaker 2>into the next version of that company software. So even

0:08:04.760 --> 0:08:07.920
<v Speaker 2>if that company didn't want to implement the feature, if

0:08:07.920 --> 0:08:10.960
<v Speaker 2>they did it open source, they would inherit that feature

0:08:11.000 --> 0:08:13.520
<v Speaker 2>from the upstream community is what we call it, so

0:08:13.560 --> 0:08:16.200
<v Speaker 2>you have some control over the situation. If it's open source,

0:08:16.280 --> 0:08:20.440
<v Speaker 2>you have an opportunity to actually affect change in the product,

0:08:20.720 --> 0:08:22.760
<v Speaker 2>and you could then pick it up or pay somebody

0:08:22.760 --> 0:08:24.840
<v Speaker 2>else to pick it up, or another company could form

0:08:24.880 --> 0:08:27.480
<v Speaker 2>and pick it up and keep it going. So there's

0:08:27.480 --> 0:08:29.960
<v Speaker 2>more possibilities. If it's open source, it's more like it's

0:08:30.000 --> 0:08:31.560
<v Speaker 2>like an insurance policy almost.

0:08:31.600 --> 0:08:35.160
<v Speaker 1>So innovation from the standpoint of the customer, innovation is

0:08:35.200 --> 0:08:38.200
<v Speaker 1>a lot easier when you're working in an open source environment.

0:08:38.360 --> 0:08:39.000
<v Speaker 2>Absolutely.

0:08:39.400 --> 0:08:43.160
<v Speaker 1>Yeah. So now at RedHat you're working with something called

0:08:43.160 --> 0:08:45.920
<v Speaker 1>instruct lab. Tell us a little bit about what that is.

0:08:46.360 --> 0:08:48.600
<v Speaker 2>So the thing that really excites me about getting to

0:08:48.640 --> 0:08:51.000
<v Speaker 2>work on this project is AI is sort of that

0:08:51.080 --> 0:08:54.200
<v Speaker 2>has been this scary thing for me because it's one

0:08:54.240 --> 0:08:57.360
<v Speaker 2>of those things like in order to be able to

0:08:57.440 --> 0:09:02.520
<v Speaker 2>pre train a model, you have to have unobtainium GPUs,

0:09:03.240 --> 0:09:07.560
<v Speaker 2>you have to have rich resources, It takes months, it

0:09:07.559 --> 0:09:12.520
<v Speaker 2>takes expertise. There's a small handful of companies that can

0:09:12.559 --> 0:09:16.559
<v Speaker 2>build a model from pre train to something usable, and

0:09:16.800 --> 0:09:19.160
<v Speaker 2>it kind of feels like those early days when I

0:09:19.240 --> 0:09:22.000
<v Speaker 2>was kind of delving in software in the same way.

0:09:22.000 --> 0:09:25.320
<v Speaker 2>I think if more people could contribute to AI models,

0:09:26.120 --> 0:09:29.720
<v Speaker 2>then it wouldn't be just influenced by whichever company had

0:09:29.760 --> 0:09:33.000
<v Speaker 2>the resources to build it. And there's been a lot

0:09:33.000 --> 0:09:36.400
<v Speaker 2>of emphasis on pre training models, so taking massive terabytes

0:09:36.480 --> 0:09:40.439
<v Speaker 2>data sets, throwing them through masses of GPUs over months

0:09:40.480 --> 0:09:43.760
<v Speaker 2>of time, spending hundreds of millions of dollars to build

0:09:43.800 --> 0:09:46.640
<v Speaker 2>a base model. But when instruct lab does is say okay,

0:09:46.880 --> 0:09:49.760
<v Speaker 2>you have a base model, we're going to fine tune in.

0:09:49.800 --> 0:09:53.240
<v Speaker 2>On the other end, it takes less compute resources. The

0:09:53.280 --> 0:09:55.559
<v Speaker 2>way we've built in struck lab, you can play around

0:09:55.559 --> 0:09:58.080
<v Speaker 2>with the technology and learn it on it off the

0:09:58.080 --> 0:10:01.520
<v Speaker 2>shelf laptop that you can actually buy. So in this

0:10:01.640 --> 0:10:05.400
<v Speaker 2>way we're enabling a much broader set of people to

0:10:05.480 --> 0:10:08.200
<v Speaker 2>play with AI, to contribute it, to modify it. And

0:10:08.200 --> 0:10:11.160
<v Speaker 2>I'll tell you one story from red Hat Succi, who

0:10:11.240 --> 0:10:15.680
<v Speaker 2>is our chief diversity officer, very interested in inclusive language

0:10:15.679 --> 0:10:18.600
<v Speaker 2>and open source software, doesn't have any experience with AI.

0:10:19.360 --> 0:10:21.400
<v Speaker 2>We have a community model that we have an upstream

0:10:21.440 --> 0:10:24.000
<v Speaker 2>project around for people to contribute knowledge and skills to

0:10:24.040 --> 0:10:26.160
<v Speaker 2>the model. She's like, I want to teach the model

0:10:26.480 --> 0:10:29.440
<v Speaker 2>how to use inclusive language like replace this word with

0:10:29.480 --> 0:10:31.280
<v Speaker 2>this word, or this word with this word. Oh my, oh,

0:10:31.320 --> 0:10:34.199
<v Speaker 2>that's so cool. So she paired up with Nicholas who

0:10:34.280 --> 0:10:37.080
<v Speaker 2>is a technical guy at red Hat, and they built

0:10:37.200 --> 0:10:40.240
<v Speaker 2>and submitted a skill to the model that you can

0:10:40.320 --> 0:10:42.480
<v Speaker 2>just tell the model, can you please take this document

0:10:42.559 --> 0:10:45.000
<v Speaker 2>and translate this language to more inclusive language and it

0:10:45.040 --> 0:10:47.439
<v Speaker 2>will do it. And they submitted it to the community.

0:10:47.520 --> 0:10:49.240
<v Speaker 2>They were so proud. It was like, that's the kind

0:10:49.280 --> 0:10:51.720
<v Speaker 2>of thing that like, you know, maybe a company would

0:10:51.720 --> 0:10:54.160
<v Speaker 2>be incentivized to do that, but if you have some

0:10:54.280 --> 0:10:57.880
<v Speaker 2>tooling that's open source and something that anybody could access,

0:10:58.040 --> 0:11:00.640
<v Speaker 2>then those communities could actually get together and build that

0:11:00.760 --> 0:11:01.960
<v Speaker 2>knowledge into AI models.

0:11:02.200 --> 0:11:06.720
<v Speaker 1>Just so understand, what you guys have is the structure

0:11:06.800 --> 0:11:11.080
<v Speaker 1>for an AI system, and in other cases, individual companies

0:11:11.240 --> 0:11:15.199
<v Speaker 1>own and train their own AI systems. It takes enormous

0:11:15.200 --> 0:11:18.240
<v Speaker 1>amount of resources. They hoover up all kinds of information,

0:11:18.800 --> 0:11:22.000
<v Speaker 1>train it according to their own hidden set of rules,

0:11:22.040 --> 0:11:26.480
<v Speaker 1>and then a customer might use that for some price.

0:11:26.840 --> 0:11:28.760
<v Speaker 1>What you're saying is, in the same way that we

0:11:28.840 --> 0:11:32.680
<v Speaker 1>democratize the writing of software before, let's democratize the training

0:11:32.720 --> 0:11:36.400
<v Speaker 1>of an AI system. So anyone can contribute here and

0:11:36.800 --> 0:11:40.480
<v Speaker 1>teach the model the things that they're interested in teaching

0:11:40.480 --> 0:11:43.400
<v Speaker 1>the model. I'm guessing correct me. On the one hand,

0:11:43.920 --> 0:11:46.000
<v Speaker 1>this model, at least in the beginning, is going to

0:11:46.040 --> 0:11:49.160
<v Speaker 1>have a lot fewer resources available to it. But on

0:11:49.200 --> 0:11:50.959
<v Speaker 1>the other hand, it's going to have a much more

0:11:51.320 --> 0:11:53.280
<v Speaker 1>diverse set of inputs.

0:11:53.760 --> 0:11:56.840
<v Speaker 2>That's right. And the other thing is that IBM, basically

0:11:56.840 --> 0:11:59.840
<v Speaker 2>as part of this project, has something called the granite model,

0:12:00.720 --> 0:12:03.040
<v Speaker 2>and they've donated some granite models. So these are the

0:12:03.040 --> 0:12:06.400
<v Speaker 2>ones that take the months and terabytes of data and

0:12:06.480 --> 0:12:09.360
<v Speaker 2>all the GPUs to train. So IBM has created one

0:12:09.400 --> 0:12:12.520
<v Speaker 2>of those, and they have listed out and linked to

0:12:12.559 --> 0:12:14.760
<v Speaker 2>the data sets that they used, and they talk about

0:12:14.760 --> 0:12:17.960
<v Speaker 2>the relative proportions they used when pre training, so it's

0:12:17.960 --> 0:12:20.079
<v Speaker 2>not just the black box. You know where the data

0:12:20.120 --> 0:12:22.959
<v Speaker 2>came from, which is a pretty open position to take.

0:12:23.360 --> 0:12:25.400
<v Speaker 2>That is what we recommend as the base. So you

0:12:25.480 --> 0:12:28.559
<v Speaker 2>use the instruct lab tuning. You take this base granite

0:12:28.600 --> 0:12:31.280
<v Speaker 2>model that IBM has provided, and you use the instruct

0:12:31.320 --> 0:12:33.640
<v Speaker 2>lab tooling that red Hat works on, and you use

0:12:33.679 --> 0:12:36.760
<v Speaker 2>that to fine tune the model to make it whatever

0:12:36.800 --> 0:12:37.280
<v Speaker 2>you want.

0:12:37.840 --> 0:12:40.760
<v Speaker 1>I want to go back to the partnership between IBM

0:12:40.800 --> 0:12:44.800
<v Speaker 1>and red Hat here with them providing the granite model

0:12:45.400 --> 0:12:48.680
<v Speaker 1>to your instruct lab. Is this the first time red

0:12:48.679 --> 0:12:50.680
<v Speaker 1>Hat and IBM have collaborated like this.

0:12:51.480 --> 0:12:54.040
<v Speaker 2>I think it's something that's been going on. Like another

0:12:54.320 --> 0:12:57.560
<v Speaker 2>a product within the red Hat family would be OpenShift AI,

0:12:57.679 --> 0:13:00.679
<v Speaker 2>where they collaborate a lot with IBM Research team, Like

0:13:00.920 --> 0:13:03.280
<v Speaker 2>bolm is one of the components of that product that

0:13:03.320 --> 0:13:07.280
<v Speaker 2>there's a nice kind of exchange and collaboration between the

0:13:07.320 --> 0:13:08.480
<v Speaker 2>two companies.

0:13:09.240 --> 0:13:11.920
<v Speaker 1>How large is the potential community of people who might

0:13:11.960 --> 0:13:13.560
<v Speaker 1>contribute to instruct lab.

0:13:14.480 --> 0:13:17.040
<v Speaker 2>It could be thousands of people. I mean, we'll see.

0:13:17.080 --> 0:13:20.840
<v Speaker 2>It's early days. This is early technology that was invented

0:13:20.840 --> 0:13:23.320
<v Speaker 2>at IBM Research that they partnered with us at red

0:13:23.320 --> 0:13:26.200
<v Speaker 2>Hat to kind of build the software around it. There's

0:13:26.200 --> 0:13:28.480
<v Speaker 2>still more to go, Like right now, we have a

0:13:28.559 --> 0:13:30.559
<v Speaker 2>team in the community that's actually trying to build a

0:13:30.600 --> 0:13:34.120
<v Speaker 2>web interface to make it easier for anybody to contribute.

0:13:34.280 --> 0:13:35.800
<v Speaker 2>So we have a lot of those sort of user

0:13:35.880 --> 0:13:39.320
<v Speaker 2>experience for the contributor to the model stuff to work

0:13:39.320 --> 0:13:42.040
<v Speaker 2>out that we're still actively building on. But like my

0:13:42.200 --> 0:13:44.840
<v Speaker 2>vision for it even is I like going back to

0:13:44.840 --> 0:13:47.960
<v Speaker 2>that academic model of learning from what others and building

0:13:48.000 --> 0:13:50.880
<v Speaker 2>upon it over time. It would be very good for

0:13:51.040 --> 0:13:53.880
<v Speaker 2>us to sort of go out and try to collaborate

0:13:54.040 --> 0:13:56.520
<v Speaker 2>with academics of all fields, like, hey, you know, the

0:13:56.559 --> 0:14:00.079
<v Speaker 2>model doesn't know about your field. Would you like to

0:14:00.120 --> 0:14:02.480
<v Speaker 2>put something into the model about your field so it

0:14:02.520 --> 0:14:05.880
<v Speaker 2>knows about it? Or even you know, talk to the model.

0:14:06.080 --> 0:14:08.440
<v Speaker 2>It got it wrong, let's correct it. Can we lean

0:14:08.520 --> 0:14:10.640
<v Speaker 2>on your expertise to correct it and make sure it

0:14:10.640 --> 0:14:13.760
<v Speaker 2>gets it right and sort of use that community model

0:14:13.800 --> 0:14:17.839
<v Speaker 2>as a way for everybody to collaborate because before instruct lab,

0:14:18.600 --> 0:14:22.040
<v Speaker 2>my understanding is if you wanted to take a model

0:14:22.040 --> 0:14:24.120
<v Speaker 2>that's open source license and play with it, you could

0:14:24.160 --> 0:14:25.840
<v Speaker 2>do that. You could take a model kind of off

0:14:25.840 --> 0:14:28.880
<v Speaker 2>the shelf from Hugging Face and fine tune it yourself.

0:14:29.280 --> 0:14:30.840
<v Speaker 2>But it's a bit of a dead end because you

0:14:30.920 --> 0:14:33.440
<v Speaker 2>made your contributions, but there's no way for other people

0:14:33.920 --> 0:14:36.840
<v Speaker 2>to collaborate with you. So the way that we've built

0:14:36.880 --> 0:14:41.000
<v Speaker 2>this is based on how the technology works. Everybody can

0:14:41.000 --> 0:14:43.360
<v Speaker 2>contribute to it. This is something that it can keep

0:14:43.400 --> 0:14:44.840
<v Speaker 2>growing and growing and growing over time.

0:14:45.160 --> 0:14:48.720
<v Speaker 1>Yeah. Yeah, what's the level of expertise necessary to be

0:14:48.760 --> 0:14:49.480
<v Speaker 1>a contributor?

0:14:50.080 --> 0:14:51.960
<v Speaker 2>You don't need to be a data scientist, and you

0:14:52.000 --> 0:14:54.960
<v Speaker 2>don't need to have exotic hardware. Honestly, if you don't

0:14:54.960 --> 0:14:57.400
<v Speaker 2>even have laptop hardware that meets the spec for doing

0:14:57.400 --> 0:15:01.080
<v Speaker 2>instruct labs laptop version, you can minutes to the community

0:15:01.160 --> 0:15:03.640
<v Speaker 2>and then we'll actually build it for you. We have

0:15:03.760 --> 0:15:06.120
<v Speaker 2>bots and stuff that do that, and we're hoping over

0:15:06.160 --> 0:15:08.680
<v Speaker 2>time to make that more accessible, first by having a

0:15:08.760 --> 0:15:11.360
<v Speaker 2>user interface and then maybe later on having a web service.

0:15:11.680 --> 0:15:14.880
<v Speaker 1>Yeah, so give me an example of how a business

0:15:15.000 --> 0:15:17.040
<v Speaker 1>might make use of instruct lab.

0:15:17.600 --> 0:15:20.160
<v Speaker 2>One of the things that businesses are doing with AI

0:15:20.240 --> 0:15:23.920
<v Speaker 2>right now is using hosted API services. You're quite expensive,

0:15:24.200 --> 0:15:27.040
<v Speaker 2>but they're finding value. But it's hard given the amount

0:15:27.040 --> 0:15:29.320
<v Speaker 2>of money they're spending. And one of the things that's

0:15:29.320 --> 0:15:31.200
<v Speaker 2>a little scary about it too, is like you have

0:15:31.520 --> 0:15:35.960
<v Speaker 2>very sensitive internal documents and you have employees maybe not

0:15:36.080 --> 0:15:38.960
<v Speaker 2>understanding what they're actually doing because you know, how would

0:15:39.000 --> 0:15:42.400
<v Speaker 2>you if you're not technical enough when you're asking said

0:15:43.080 --> 0:15:48.760
<v Speaker 2>public web service AI model information about your copy pasting

0:15:48.880 --> 0:15:52.920
<v Speaker 2>internal company documents. It's going across the Internet into another

0:15:52.960 --> 0:15:56.160
<v Speaker 2>company's hands, and that company probably shouldn't have access to that.

0:15:56.600 --> 0:15:59.560
<v Speaker 2>So what both RedHat and IBM and the space are

0:15:59.600 --> 0:16:02.960
<v Speaker 2>looking at, like, the instruct lab model is very modest.

0:16:03.000 --> 0:16:06.480
<v Speaker 2>It's seven billion parameter model, very small. It's very cheap

0:16:06.520 --> 0:16:09.920
<v Speaker 2>to serve inference on a seven billion parameter model. It's

0:16:09.920 --> 0:16:13.480
<v Speaker 2>competing with trillion parameter models that are hosted. You take

0:16:13.480 --> 0:16:16.320
<v Speaker 2>this small model that is cheap to run inference on,

0:16:16.960 --> 0:16:20.880
<v Speaker 2>you train it with your own company's proprietary data inside

0:16:20.920 --> 0:16:23.480
<v Speaker 2>the walls of your company, on your own hardware. You

0:16:23.520 --> 0:16:26.800
<v Speaker 2>can do all sorts of actual data analysis on your

0:16:26.840 --> 0:16:29.440
<v Speaker 2>most sensitive data and have the confidence that has not

0:16:29.520 --> 0:16:30.400
<v Speaker 2>left the premises.

0:16:31.240 --> 0:16:34.280
<v Speaker 1>In that use case, you're not actually training the model

0:16:34.320 --> 0:16:37.560
<v Speaker 1>for everyone. You're just taking it and doing some private

0:16:37.600 --> 0:16:40.080
<v Speaker 1>stuff on it exactly, which doesn't leave the building. But

0:16:40.120 --> 0:16:44.640
<v Speaker 1>that's separate from an interaction where you're doing something that

0:16:45.360 --> 0:16:46.880
<v Speaker 1>contributes overall.

0:16:46.960 --> 0:16:49.520
<v Speaker 2>Right, And that's something maybe that I should be more

0:16:49.560 --> 0:16:51.800
<v Speaker 2>clear about. Is there's sort of two tracks here, and

0:16:51.920 --> 0:16:55.160
<v Speaker 2>this is very red hat classic. You have your upstream

0:16:55.280 --> 0:16:58.320
<v Speaker 2>community track and you have your business product track. So

0:16:58.360 --> 0:17:02.360
<v Speaker 2>the upstream community track is just enabling anybody to contribute

0:17:02.400 --> 0:17:04.320
<v Speaker 2>to a model in a collaborative way and play with it.

0:17:04.760 --> 0:17:08.600
<v Speaker 2>The downstream product business oriented track is now take that

0:17:08.720 --> 0:17:13.359
<v Speaker 2>tech that we've honed and developed in the open community,

0:17:14.000 --> 0:17:16.480
<v Speaker 2>and apply it to your business knowledge and skills.

0:17:17.520 --> 0:17:21.360
<v Speaker 1>This community driven approach marks a pivotal shift towards more

0:17:21.400 --> 0:17:27.159
<v Speaker 1>accessible AI solutions. The contrast between externally hosted AI services

0:17:27.440 --> 0:17:30.639
<v Speaker 1>and the open model enhanced by instruct lab underscores the

0:17:30.680 --> 0:17:35.760
<v Speaker 1>potential for broader adoption of AI in diverse business contexts.

0:17:36.280 --> 0:17:39.680
<v Speaker 1>She envisions a future in which technological innovation is more

0:17:39.680 --> 0:17:43.960
<v Speaker 1>tailored to individual business needs, guided by principles of openness

0:17:44.119 --> 0:17:49.000
<v Speaker 1>and security. To an imaginary case study, Sure, I'm a

0:17:49.080 --> 0:17:53.879
<v Speaker 1>law firm. I'm an entertainment law I have one hundred

0:17:53.880 --> 0:17:58.360
<v Speaker 1>clients who are big stars. They all have incredibly complicated contracts.

0:17:59.160 --> 0:18:03.600
<v Speaker 1>I feed a thousand of my company's contracts from the

0:18:03.680 --> 0:18:07.520
<v Speaker 1>last ten years into the model, and then every time

0:18:07.560 --> 0:18:10.160
<v Speaker 1>I have a new contract, I ask the model, am

0:18:10.160 --> 0:18:12.840
<v Speaker 1>I missing something? Can you go back and look through

0:18:12.880 --> 0:18:15.240
<v Speaker 1>all our own contracts and show me a contract that

0:18:15.520 --> 0:18:18.800
<v Speaker 1>is missing key components or exposes us to some liability.

0:18:19.680 --> 0:18:23.280
<v Speaker 1>In that case, the model would know my law firm

0:18:23.520 --> 0:18:26.880
<v Speaker 1>contracts really really well. It's as if they've been working

0:18:27.400 --> 0:18:30.800
<v Speaker 1>out my law firm. They're not distracted by other people's

0:18:30.840 --> 0:18:36.600
<v Speaker 1>particular styles, or a bunch of contracts from the utility industry,

0:18:36.720 --> 0:18:41.360
<v Speaker 1>or they know entertainment law contracts exactly.

0:18:41.480 --> 0:18:43.520
<v Speaker 2>Yeah, you can train it in your own image, your

0:18:43.600 --> 0:18:47.400
<v Speaker 2>style of doing things. It's something that your company can

0:18:47.440 --> 0:18:50.879
<v Speaker 2>produce that is uniquely helpful to you. No third party

0:18:50.880 --> 0:18:53.159
<v Speaker 2>could do that because no third party understands how you

0:18:53.200 --> 0:18:56.919
<v Speaker 2>do business and understands your history and your documents. So

0:18:56.960 --> 0:18:59.560
<v Speaker 2>it's sort of a way of getting value out of

0:18:59.600 --> 0:19:01.320
<v Speaker 2>the stuff if you already have sitting in a file

0:19:01.359 --> 0:19:03.520
<v Speaker 2>cabinet somewhere, it's very cool.

0:19:03.800 --> 0:19:06.800
<v Speaker 1>Yeah, give me a sort of a real world case

0:19:06.800 --> 0:19:10.000
<v Speaker 1>study where you think the business use case would be

0:19:10.080 --> 0:19:14.680
<v Speaker 1>really powerful. What's a business that really could see an

0:19:14.680 --> 0:19:18.680
<v Speaker 1>advantage to using instruct lab in its way.

0:19:19.160 --> 0:19:21.399
<v Speaker 2>The demo that I've given a couple of times at

0:19:21.440 --> 0:19:25.040
<v Speaker 2>different events used an imaginary insurance company. So you say,

0:19:25.040 --> 0:19:28.960
<v Speaker 2>you have this company, you have to recommend repairs for

0:19:29.080 --> 0:19:32.480
<v Speaker 2>various types of claims. You've been doing this for years,

0:19:32.560 --> 0:19:35.400
<v Speaker 2>you know. If you know the windshield's broken and you've

0:19:35.400 --> 0:19:38.240
<v Speaker 2>gotten this type of accident and it's this model car,

0:19:38.480 --> 0:19:40.200
<v Speaker 2>these are the kinds of things you want to look at.

0:19:40.920 --> 0:19:43.679
<v Speaker 2>So you could talk to any insurance agent in the

0:19:43.720 --> 0:19:46.400
<v Speaker 2>field and be like, oh, you know, it's a Tesla.

0:19:46.520 --> 0:19:48.639
<v Speaker 2>You might want to look at the battery or something like.

0:19:48.720 --> 0:19:52.120
<v Speaker 2>They'll have some latent knowledge just so you can take

0:19:52.160 --> 0:19:54.880
<v Speaker 2>that and train it into a model. Honestly, I think

0:19:54.920 --> 0:19:58.160
<v Speaker 2>these kind of new technologies are better when they're less visible.

0:19:58.760 --> 0:20:01.199
<v Speaker 2>So say you have the as agents in the field

0:20:01.280 --> 0:20:03.280
<v Speaker 2>and they have this tool and they're kind of entering

0:20:03.280 --> 0:20:06.240
<v Speaker 2>the claim data. They're on the scene at the car,

0:20:06.840 --> 0:20:09.480
<v Speaker 2>and it might say, oh, look, I see this is

0:20:09.520 --> 0:20:12.080
<v Speaker 2>a Ford Fiesta. These are things you want to look

0:20:12.080 --> 0:20:15.280
<v Speaker 2>at for this type of accident. As you're entering the data,

0:20:15.720 --> 0:20:17.600
<v Speaker 2>it could be going through the knowledge you had loaded

0:20:17.600 --> 0:20:20.080
<v Speaker 2>into the model and be making these suggestions based on

0:20:20.119 --> 0:20:23.080
<v Speaker 2>your company's background, and hey, you know, let's not make

0:20:23.080 --> 0:20:25.640
<v Speaker 2>the same mistake twice. Let's make new mistakes and let's

0:20:25.680 --> 0:20:28.560
<v Speaker 2>learn from the stuff we already did. So that's one example,

0:20:28.600 --> 0:20:31.159
<v Speaker 2>but there's so many different industries in ways that this

0:20:31.240 --> 0:20:34.000
<v Speaker 2>could help, and it could make those agents in the

0:20:34.040 --> 0:20:35.520
<v Speaker 2>field more efficient.

0:20:36.240 --> 0:20:38.680
<v Speaker 1>Have you had anyone talk to you about using instruct

0:20:38.720 --> 0:20:40.240
<v Speaker 1>lab in a way that surprised you.

0:20:42.280 --> 0:20:46.960
<v Speaker 2>I mean, some people have done funky things, but sort

0:20:46.960 --> 0:20:49.360
<v Speaker 2>of playing with the skills stuff. That's where I see

0:20:49.359 --> 0:20:52.360
<v Speaker 2>a lot of creativity. The difference between knowledge and skills

0:20:52.400 --> 0:20:55.720
<v Speaker 2>is that knowledge is pretty pretty understandable, right, like oh,

0:20:55.800 --> 0:20:59.679
<v Speaker 2>historical insurance claims or you know, legal contracts. Skills are

0:20:59.680 --> 0:21:02.600
<v Speaker 2>a little different. So whenever somebody submits a skill, sometimes

0:21:02.720 --> 0:21:04.680
<v Speaker 2>it tends to be really creative because it's not something

0:21:04.720 --> 0:21:08.040
<v Speaker 2>that's super intuitive. Somebody submitted a skill. I don't know

0:21:08.040 --> 0:21:11.040
<v Speaker 2>how well it worked, but it was like making ASKI art,

0:21:11.240 --> 0:21:13.640
<v Speaker 2>like draw me a I don't know, draw me a dog,

0:21:13.680 --> 0:21:15.359
<v Speaker 2>and would do like an ASKI art dog. I mean,

0:21:15.400 --> 0:21:17.960
<v Speaker 2>it's stuff that you can do programmatically. One that was

0:21:17.960 --> 0:21:21.879
<v Speaker 2>actually very very helpful was you know, take this table

0:21:21.920 --> 0:21:25.120
<v Speaker 2>of data and convert it to this format like, ooh,

0:21:25.160 --> 0:21:26.800
<v Speaker 2>that's nice. That actually saves me time.

0:21:27.359 --> 0:21:29.880
<v Speaker 1>How far away are we from the day when I,

0:21:29.960 --> 0:21:34.679
<v Speaker 1>Malcolm Globwell technology ignore Amus can go home and easily

0:21:34.720 --> 0:21:36.480
<v Speaker 1>interact with instruct lab.

0:21:37.840 --> 0:21:39.600
<v Speaker 2>Maybe a few months, a few.

0:21:39.400 --> 0:21:42.080
<v Speaker 1>Months, you're gonna say a few years.

0:21:42.720 --> 0:21:45.879
<v Speaker 2>No, I think it'd be a few months. Wow, I

0:21:45.880 --> 0:21:48.600
<v Speaker 2>hope it's power open source innovation.

0:21:49.040 --> 0:21:53.000
<v Speaker 1>Yeah, oh that's really interesting. Yeah, I'm always taken by surprise.

0:21:53.320 --> 0:21:55.960
<v Speaker 1>I'm still thinking in twentieth century terms about how long

0:21:56.040 --> 0:21:59.200
<v Speaker 1>things take, and you're in the twenty second century as

0:21:59.200 --> 0:21:59.880
<v Speaker 1>well as I could tell.

0:22:00.000 --> 0:22:04.560
<v Speaker 2>The instruct lab core invention was invented in a hotel

0:22:04.680 --> 0:22:07.720
<v Speaker 2>room at an AI conference in December with an amazing

0:22:07.760 --> 0:22:10.880
<v Speaker 2>group of IBM research guys December of twenty twenty three.

0:22:11.200 --> 0:22:13.879
<v Speaker 1>Wait back up, you have to tell the story.

0:22:14.119 --> 0:22:17.080
<v Speaker 2>This group of guys we've been working with, they were

0:22:17.119 --> 0:22:19.479
<v Speaker 2>at this conference together, and it's a really funny story

0:22:19.560 --> 0:22:22.399
<v Speaker 2>because you know, it's hard to get access to GPUs

0:22:22.720 --> 0:22:24.520
<v Speaker 2>and like even you know, you're at IBM and it's

0:22:24.520 --> 0:22:27.280
<v Speaker 2>hard to get access because everybody wants access. They did

0:22:27.320 --> 0:22:30.240
<v Speaker 2>it over Christmas break because nobody was using the cluster

0:22:30.320 --> 0:22:32.399
<v Speaker 2>at the time, and they ran all of these experiments

0:22:32.480 --> 0:22:34.280
<v Speaker 2>and I'm like, whoa, this is really cool.

0:22:34.680 --> 0:22:38.640
<v Speaker 1>And wait. Their idea was we can do a stripped

0:22:38.640 --> 0:22:44.000
<v Speaker 1>down AI model, and was the idea and even back

0:22:44.040 --> 0:22:45.280
<v Speaker 1>then combine it with grantite.

0:22:45.320 --> 0:22:48.080
<v Speaker 2>What was the original the original idea. It's sort of

0:22:48.200 --> 0:22:51.800
<v Speaker 2>multi there's like multiple aspects to it. So like one

0:22:51.800 --> 0:22:54.280
<v Speaker 2>of the aspects it actually came on later, but it

0:22:54.320 --> 0:22:56.600
<v Speaker 2>starts at the beginning of the workflow. Is you're using

0:22:56.680 --> 0:23:00.639
<v Speaker 2>a taxonomy to organize how you're fine too the model.

0:23:00.680 --> 0:23:03.240
<v Speaker 2>So the old approach they call it the blender approach,

0:23:03.640 --> 0:23:06.000
<v Speaker 2>to just take a bunch of data of roughly the

0:23:06.040 --> 0:23:07.800
<v Speaker 2>type of data that you'd like and you kind of

0:23:07.840 --> 0:23:10.399
<v Speaker 2>throw it in and then see what comes out. Don't

0:23:10.480 --> 0:23:12.879
<v Speaker 2>like it, Okay, throw in more, try again, see what

0:23:12.960 --> 0:23:16.399
<v Speaker 2>comes out. They had used this taxonomy technique, so you

0:23:16.440 --> 0:23:20.560
<v Speaker 2>actually build like a taxonomy of like categories and subfolders

0:23:20.600 --> 0:23:22.959
<v Speaker 2>of like this is the knowledge and skills that we

0:23:23.000 --> 0:23:25.680
<v Speaker 2>want to train into the model. And that way you're

0:23:25.720 --> 0:23:29.000
<v Speaker 2>sort of systematic about what you're adding, and you can

0:23:29.040 --> 0:23:31.560
<v Speaker 2>also identify gaps pretty easily. Oh, I don't have a

0:23:31.600 --> 0:23:33.720
<v Speaker 2>category for that, let me add that. So that's like

0:23:33.920 --> 0:23:36.200
<v Speaker 2>one of the parts of the invention here.

0:23:37.000 --> 0:23:41.919
<v Speaker 1>Point number one is let's be intentional and deliberate in

0:23:42.000 --> 0:23:43.240
<v Speaker 1>how we build and train this thing.

0:23:43.440 --> 0:23:46.719
<v Speaker 2>Yeah, and then the next component would be Okay, So

0:23:47.000 --> 0:23:49.560
<v Speaker 2>it is actually quite expensive. Part of the expense of

0:23:49.600 --> 0:23:53.119
<v Speaker 2>like tuning models and just training models in general is

0:23:53.160 --> 0:23:56.359
<v Speaker 2>coming up with the data. And what they wanted to

0:23:56.400 --> 0:23:58.520
<v Speaker 2>do is have a technique where you could have just

0:23:58.560 --> 0:24:01.679
<v Speaker 2>a little bit of data and expand it with something

0:24:01.680 --> 0:24:05.080
<v Speaker 2>they're calling synthetic data generation. And this is where it's

0:24:05.080 --> 0:24:09.040
<v Speaker 2>sort of like you have this student and teacher workflow,

0:24:09.640 --> 0:24:14.359
<v Speaker 2>so you have your taxonomy. The taxonomy has sort of

0:24:14.359 --> 0:24:17.320
<v Speaker 2>the knowledge like a business's knowledge documents, their insurance claims,

0:24:17.600 --> 0:24:20.840
<v Speaker 2>and it has these quizzes that you write and that's

0:24:20.880 --> 0:24:22.800
<v Speaker 2>to teach the model. So I'm writing a quiz based

0:24:22.920 --> 0:24:24.520
<v Speaker 2>just like you do in school. You read the chapter

0:24:24.720 --> 0:24:26.800
<v Speaker 2>on the American Revolution, and then you answer a ten

0:24:26.880 --> 0:24:30.160
<v Speaker 2>question quiz where you're giving the model quiz. You need

0:24:30.200 --> 0:24:33.360
<v Speaker 2>at least five questions and answers, and the answers need

0:24:33.400 --> 0:24:36.000
<v Speaker 2>to be taken from the context of the document, and

0:24:36.119 --> 0:24:39.600
<v Speaker 2>then you run it through a process called synthetic data generation,

0:24:39.880 --> 0:24:41.919
<v Speaker 2>and it looks at the documents, so we'll look at

0:24:41.920 --> 0:24:44.680
<v Speaker 2>the history chapter, it'll look at the questions and answers,

0:24:45.040 --> 0:24:47.680
<v Speaker 2>and then it'll look to that original document and come

0:24:47.760 --> 0:24:50.280
<v Speaker 2>up with more questions and answers based on the format

0:24:50.320 --> 0:24:52.680
<v Speaker 2>of the questions and answers you made. So you can

0:24:52.760 --> 0:24:56.080
<v Speaker 2>take five questions of answers amplify them into one hundred

0:24:56.359 --> 0:24:59.120
<v Speaker 2>questions and answers two hundred questions and answers. And it's

0:24:59.440 --> 0:25:02.560
<v Speaker 2>a second model that is making the questions and answers,

0:25:02.680 --> 0:25:05.639
<v Speaker 2>so it's synthetic data generation using an AI model to

0:25:05.680 --> 0:25:08.480
<v Speaker 2>make the questions. We use an open source model to

0:25:08.520 --> 0:25:11.480
<v Speaker 2>do that. So that's the second part. And then the

0:25:11.520 --> 0:25:14.520
<v Speaker 2>third part is we have a multi phase tuning technique

0:25:14.520 --> 0:25:18.360
<v Speaker 2>to actually take the synthetic data and then basically bake

0:25:18.440 --> 0:25:20.760
<v Speaker 2>it into the model. So sort of that's the approach.

0:25:21.440 --> 0:25:24.520
<v Speaker 2>A general philosophy of the approach is using granite because

0:25:24.560 --> 0:25:27.280
<v Speaker 2>we know where the data came from. Another approach is

0:25:27.280 --> 0:25:29.640
<v Speaker 2>the fact that we're using small models that are cheap

0:25:29.640 --> 0:25:32.160
<v Speaker 2>to run inference on. They're small enough that you can

0:25:32.160 --> 0:25:34.520
<v Speaker 2>tune them on laptop hardware. You don't need all the

0:25:34.520 --> 0:25:38.679
<v Speaker 2>fancy expensive GPU mania you're good. So sort of like

0:25:38.720 --> 0:25:41.879
<v Speaker 2>a whole system, it's like not any one component. But

0:25:41.960 --> 0:25:44.480
<v Speaker 2>it's sort of the approach they took with somewhat novel,

0:25:44.760 --> 0:25:47.440
<v Speaker 2>and they were very excited when they saw the experimental results.

0:25:47.840 --> 0:25:50.600
<v Speaker 2>There was a meeting between red hat and IBM. It

0:25:50.640 --> 0:25:52.760
<v Speaker 2>was actually an IBM research meeting that red hatters were

0:25:52.800 --> 0:25:56.040
<v Speaker 2>invited to, and I think the red Hatter's involves sort

0:25:56.040 --> 0:25:59.640
<v Speaker 2>of saw the potential, WHOA, we can make models open

0:25:59.680 --> 0:26:03.680
<v Speaker 2>source finally, rather than them just being these endless dead forks,

0:26:04.600 --> 0:26:06.760
<v Speaker 2>we could make it so people could contribute back and

0:26:06.800 --> 0:26:09.560
<v Speaker 2>collaborate around it. So that's when red Hat became interested

0:26:09.560 --> 0:26:12.719
<v Speaker 2>in it and we sort of worked together, and the

0:26:12.800 --> 0:26:15.840
<v Speaker 2>research engineers from IBM Research who came up with the technique,

0:26:15.960 --> 0:26:18.520
<v Speaker 2>and then my team, the software engineers who know how

0:26:18.600 --> 0:26:23.560
<v Speaker 2>to take research code and productize it into actually runnable,

0:26:23.640 --> 0:26:28.520
<v Speaker 2>supportable software, kind of got together. We've been hanging out

0:26:28.520 --> 0:26:31.200
<v Speaker 2>in the Boston office at red Hat and building it out.

0:26:31.560 --> 0:26:34.800
<v Speaker 2>April eighteenth was when we went open source and we

0:26:34.840 --> 0:26:37.320
<v Speaker 2>made all of our repositories with all of the code public,

0:26:37.359 --> 0:26:39.600
<v Speaker 2>and right now we're working towards a product release, so

0:26:39.640 --> 0:26:40.600
<v Speaker 2>a supported product.

0:26:40.720 --> 0:26:42.800
<v Speaker 1>How long did it take you to be convinced of

0:26:43.760 --> 0:26:47.040
<v Speaker 1>the value of this idea? I mean, so people get

0:26:47.040 --> 0:26:51.240
<v Speaker 1>together in this hotel room. They're running these experiments over Christmas.

0:26:51.480 --> 0:26:53.439
<v Speaker 1>Are you aware of the experiments as they're running them?

0:26:54.800 --> 0:26:57.240
<v Speaker 2>I didn't find out till February.

0:26:57.280 --> 0:26:59.399
<v Speaker 1>So they come to you in February and they say, MO,

0:27:00.720 --> 0:27:02.800
<v Speaker 1>can you recreate that conversation?

0:27:03.840 --> 0:27:08.280
<v Speaker 2>Well, our CEO, Matt Hicks, and then Jeremy Eater, who's

0:27:08.320 --> 0:27:10.960
<v Speaker 2>one of our distinguished engineers, and Steve Watt, who's a VP,

0:27:11.160 --> 0:27:13.720
<v Speaker 2>were present, I think at that meeting. So they kind

0:27:13.720 --> 0:27:16.000
<v Speaker 2>of brought it back to us and said, listen, we've

0:27:16.000 --> 0:27:20.440
<v Speaker 2>invited these IBM research folks to come visit in Boston,

0:27:21.160 --> 0:27:23.600
<v Speaker 2>you know, work with them, like, see, does this have

0:27:23.640 --> 0:27:25.880
<v Speaker 2>any merit? Could we build something from it? And so

0:27:25.920 --> 0:27:29.040
<v Speaker 2>they gave us some presentations. We were very excited when

0:27:29.040 --> 0:27:32.520
<v Speaker 2>they came to us. It only had support for Mac laptops.

0:27:33.119 --> 0:27:35.240
<v Speaker 2>Of course, you know, Red Hat were Linux people, so

0:27:35.320 --> 0:27:37.120
<v Speaker 2>we're like, all right, we've got to fix that. So

0:27:37.280 --> 0:27:40.000
<v Speaker 2>a bunch of the junior engineers around the office kind

0:27:40.000 --> 0:27:41.600
<v Speaker 2>of came and they're like, okay, we're going to build

0:27:41.640 --> 0:27:43.720
<v Speaker 2>Linux support. And they had it within like a couple

0:27:43.760 --> 0:27:46.600
<v Speaker 2>of days. It was crazy because this was just meant

0:27:46.640 --> 0:27:49.200
<v Speaker 2>to be, Hey, guys, you know what, these are invited

0:27:49.240 --> 0:27:52.760
<v Speaker 2>guests visiting our office. See what happens. And we ended

0:27:52.840 --> 0:27:56.240
<v Speaker 2>up doing like weeks of hack fests and late night

0:27:56.280 --> 0:27:58.919
<v Speaker 2>pizzas in the conference room and like playing around with

0:27:58.960 --> 0:28:01.880
<v Speaker 2>it and learning, and it was it was very fun.

0:28:01.960 --> 0:28:02.680
<v Speaker 2>It's very cool.

0:28:02.800 --> 0:28:05.800
<v Speaker 1>Anyone else do anything like this is.

0:28:05.800 --> 0:28:08.359
<v Speaker 2>Not my understanding that anybody else is doing it. Yet

0:28:08.960 --> 0:28:11.920
<v Speaker 2>maybe others will try. A lot of the focus has

0:28:11.960 --> 0:28:15.600
<v Speaker 2>been on that pre training phase. But for us, again

0:28:15.680 --> 0:28:19.280
<v Speaker 2>that fine tuning, it's more accessible because you don't need

0:28:19.320 --> 0:28:21.920
<v Speaker 2>all the exactic hardware. It doesn't take months. You can

0:28:21.960 --> 0:28:23.840
<v Speaker 2>do it on a laptop. You can do a smoke

0:28:23.920 --> 0:28:26.240
<v Speaker 2>test version of it in less than an hour.

0:28:26.760 --> 0:28:27.920
<v Speaker 1>What is the word smoke test.

0:28:28.080 --> 0:28:30.480
<v Speaker 2>Smoke test means you're not doing a full fine tuning

0:28:30.520 --> 0:28:33.600
<v Speaker 2>on the model. It's a different tuning process. It's like

0:28:33.640 --> 0:28:35.680
<v Speaker 2>kind of lower quality so to run on lower grade

0:28:35.680 --> 0:28:37.880
<v Speaker 2>hardware so you can kind of see them didn't move

0:28:37.920 --> 0:28:39.440
<v Speaker 2>the model or not, but it's not going to give

0:28:39.440 --> 0:28:42.040
<v Speaker 2>you like the full picture. You need higher end hardware

0:28:42.080 --> 0:28:43.880
<v Speaker 2>to actually do the full thing. So that's what the

0:28:43.880 --> 0:28:46.160
<v Speaker 2>product will enable you to do once it's launched. Is

0:28:46.880 --> 0:28:48.840
<v Speaker 2>you're going to need the GPUs, but when you have them,

0:28:48.880 --> 0:28:50.480
<v Speaker 2>will help you make the best usage of them.

0:28:50.760 --> 0:28:53.560
<v Speaker 1>Yeah, yeah, and no, there's all the detail. I want

0:28:53.600 --> 0:28:55.960
<v Speaker 1>to go back to sure in order to run the

0:28:56.040 --> 0:29:01.920
<v Speaker 1>tests on this idea way back when they needed time

0:29:02.000 --> 0:29:05.240
<v Speaker 1>on the GPUs, So this will be the in house

0:29:05.640 --> 0:29:09.800
<v Speaker 1>IBM and they were quiet at Christmas, So how much

0:29:09.920 --> 0:29:12.920
<v Speaker 1>time would you need on the GPUs to kind of

0:29:13.040 --> 0:29:14.040
<v Speaker 1>get proof of concept.

0:29:14.440 --> 0:29:16.760
<v Speaker 2>Well, what happens is and it's sort of like a

0:29:16.800 --> 0:29:19.120
<v Speaker 2>lot of trial and error, right, And there's a lot

0:29:19.120 --> 0:29:22.720
<v Speaker 2>about this stuff that like you come up with the hypothesis,

0:29:22.800 --> 0:29:24.760
<v Speaker 2>you test it out, did it work or not? Okay,

0:29:24.920 --> 0:29:26.920
<v Speaker 2>it's just like you know in the lab, but you know,

0:29:27.120 --> 0:29:30.920
<v Speaker 2>buns and burners and beakers and whatever. So it really depends.

0:29:31.000 --> 0:29:34.360
<v Speaker 2>But it can be hours, it can be days. It

0:29:34.400 --> 0:29:36.520
<v Speaker 2>really depends on what they're trying to do. And then

0:29:36.520 --> 0:29:38.880
<v Speaker 2>sometimes they can cut the time down, you know, with

0:29:38.920 --> 0:29:40.560
<v Speaker 2>the number of GPUs you have. So like I have

0:29:40.600 --> 0:29:43.400
<v Speaker 2>a cluster of agpus, Okay, it might take a day,

0:29:43.480 --> 0:29:45.440
<v Speaker 2>but then if I can get thirty two, I can

0:29:45.480 --> 0:29:47.240
<v Speaker 2>pipeline it and make it go faster and get it

0:29:47.280 --> 0:29:49.280
<v Speaker 2>down to a few hours. So it really depends, you know.

0:29:49.360 --> 0:29:52.440
<v Speaker 2>But it's like everybody's home for the holidays. It's a

0:29:52.480 --> 0:29:55.040
<v Speaker 2>lovely playground to kind of get that stuff going fast.

0:29:55.800 --> 0:29:59.360
<v Speaker 1>Let's jump forward one year. Tell me the status of

0:29:59.400 --> 0:30:02.880
<v Speaker 1>this project, tell me who's using it, tell me how

0:30:02.920 --> 0:30:08.960
<v Speaker 1>big is it. Give me your optimistic but plausible prediction

0:30:09.240 --> 0:30:12.960
<v Speaker 1>about what instruct lab looks like a year from now.

0:30:13.880 --> 0:30:17.280
<v Speaker 2>A year from now, I would like to see kind

0:30:17.320 --> 0:30:23.720
<v Speaker 2>of a vibrant community around not just building knowledge and

0:30:23.760 --> 0:30:27.440
<v Speaker 2>skills into a model, but coming up with better techniques

0:30:27.480 --> 0:30:30.040
<v Speaker 2>and innovation around how we do it. So I'd like

0:30:30.080 --> 0:30:33.200
<v Speaker 2>to see the contributor experience as we grow more and

0:30:33.240 --> 0:30:35.960
<v Speaker 2>more contributors to be refined. So like a year from now,

0:30:36.160 --> 0:30:39.280
<v Speaker 2>Malcolm Gladwell could come impart some of his wisdom into

0:30:39.280 --> 0:30:41.640
<v Speaker 2>the model and it wouldn't be difficult, it wouldn't be

0:30:41.640 --> 0:30:44.560
<v Speaker 2>a big lift. I would love to see the user

0:30:44.600 --> 0:30:48.680
<v Speaker 2>interface tooling for doing that to be more sophisticated. I

0:30:48.720 --> 0:30:52.240
<v Speaker 2>would love to see more people taking this and even

0:30:52.360 --> 0:30:54.600
<v Speaker 2>using it. Maybe you're not sharing it with the community,

0:30:54.640 --> 0:30:57.560
<v Speaker 2>but you're using it for some private usage. Like I'll

0:30:57.560 --> 0:31:01.080
<v Speaker 2>give you an example. I'm in contact with a fellow

0:31:01.160 --> 0:31:03.880
<v Speaker 2>who is doing AI research and he's working with doctors.

0:31:03.920 --> 0:31:06.880
<v Speaker 2>They're GPS in an area of Canada where there's not

0:31:07.000 --> 0:31:09.719
<v Speaker 2>enough GPS for the number of patients. So you know,

0:31:09.800 --> 0:31:13.640
<v Speaker 2>anything you can do to save doctors time to get

0:31:13.680 --> 0:31:15.960
<v Speaker 2>to the next patient. It's like one of the things

0:31:15.960 --> 0:31:18.840
<v Speaker 2>that he has been doing experiments with is can we

0:31:18.960 --> 0:31:22.719
<v Speaker 2>use an open source, licensed model that the doctor can

0:31:22.800 --> 0:31:24.720
<v Speaker 2>run on their laptop so they don't have to worry

0:31:24.760 --> 0:31:27.280
<v Speaker 2>about all of the different privacy rules, Like it's privates

0:31:27.280 --> 0:31:31.360
<v Speaker 2>on the laptop right there, take his live transcription of

0:31:31.360 --> 0:31:35.040
<v Speaker 2>his conversation with the patient and then convert it automatically

0:31:35.080 --> 0:31:37.440
<v Speaker 2>to a soap format that can be entered in the database.

0:31:37.680 --> 0:31:40.240
<v Speaker 2>Typically this will take a doctor fifteen to twenty minutes

0:31:40.320 --> 0:31:44.080
<v Speaker 2>of paperwork. Why not save them the paperwork at least

0:31:44.080 --> 0:31:45.320
<v Speaker 2>have the model take a stab.

0:31:45.520 --> 0:31:48.120
<v Speaker 1>Does the model then hold on to that information and

0:31:47.520 --> 0:31:50.880
<v Speaker 1>he interacts with the model again when well, that's.

0:31:50.760 --> 0:31:53.040
<v Speaker 2>The thing not within struct lab. Maybe that could be

0:31:53.040 --> 0:31:56.640
<v Speaker 2>a future development. It doesn't once you're doing inference, it's

0:31:56.680 --> 0:31:59.160
<v Speaker 2>not ingesting that what you're saying to it back in.

0:31:59.480 --> 0:32:01.760
<v Speaker 2>It's only the fine tuning phase. So the idea would

0:32:01.760 --> 0:32:05.360
<v Speaker 2>be the doctor could maybe load in past patient data

0:32:05.640 --> 0:32:08.320
<v Speaker 2>as knowledge and then when he's trying to diagnose maybe

0:32:08.480 --> 0:32:10.960
<v Speaker 2>you know what I'm saying. Like, But the main idea

0:32:11.040 --> 0:32:13.440
<v Speaker 2>is somebody might have some private users. I would love

0:32:13.520 --> 0:32:17.720
<v Speaker 2>to see more usage of this tool to enable people

0:32:17.720 --> 0:32:20.080
<v Speaker 2>who otherwise never would have had access to this type

0:32:20.080 --> 0:32:22.840
<v Speaker 2>of technology who never like you know, a small country

0:32:22.960 --> 0:32:27.080
<v Speaker 2>GP doctors, it doesn't have GPUs. They're not going to

0:32:27.160 --> 0:32:29.320
<v Speaker 2>hire some company to custom build them a model. But

0:32:29.400 --> 0:32:31.160
<v Speaker 2>maybe on the weekend, if he's a techie guy, he

0:32:31.200 --> 0:32:32.200
<v Speaker 2>could say with that.

0:32:32.280 --> 0:32:34.560
<v Speaker 1>Sim well, I mean the more you talk, the more

0:32:34.600 --> 0:32:38.480
<v Speaker 1>I'm realizing that the simplicity of this model is the

0:32:38.640 --> 0:32:41.400
<v Speaker 1>killer app here. Once you know you can run it

0:32:41.400 --> 0:32:45.160
<v Speaker 1>on a laptop, you have democratized use in a way

0:32:45.200 --> 0:32:48.680
<v Speaker 1>that's inconceivable with some of these other much more complex.

0:32:49.560 --> 0:32:52.720
<v Speaker 1>But that's interesting because one would have thought intuitively that

0:32:53.200 --> 0:32:55.720
<v Speaker 1>at the beginning that the winner is going to be

0:32:55.760 --> 0:33:01.400
<v Speaker 1>the one with the biggest, most complex version, saying actually, no,

0:33:01.600 --> 0:33:07.000
<v Speaker 1>there's a whole series of uses where being lean and focused,

0:33:07.280 --> 0:33:11.120
<v Speaker 1>focused is actually you know, it enables a whole class

0:33:11.120 --> 0:33:14.480
<v Speaker 1>of uses. Maybe another way of saying this is who

0:33:14.520 --> 0:33:16.960
<v Speaker 1>wouldn't be a potential instruct lab customer.

0:33:17.360 --> 0:33:20.160
<v Speaker 2>We don't know yet. It's it's so new, like we

0:33:20.200 --> 0:33:22.680
<v Speaker 2>haven't really had enough people experimenting and playing with it

0:33:22.720 --> 0:33:25.400
<v Speaker 2>and finding out all the things yet. But that's that's

0:33:25.400 --> 0:33:27.120
<v Speaker 2>the thing that's so exciting about it. It's like, I

0:33:27.160 --> 0:33:28.640
<v Speaker 2>can't wait to see what people do.

0:33:29.080 --> 0:33:30.840
<v Speaker 1>Is this the most exciting thing you've worked on in

0:33:30.880 --> 0:33:31.320
<v Speaker 1>your career?

0:33:31.640 --> 0:33:32.080
<v Speaker 2>I think so?

0:33:33.320 --> 0:33:37.160
<v Speaker 1>I think so, Yeah, Well, we are reaching the end

0:33:37.160 --> 0:33:39.960
<v Speaker 1>of our time. But before we finished, we can do

0:33:40.000 --> 0:33:44.280
<v Speaker 1>a little speed round. Sure, all right, complete the following sentence.

0:33:44.920 --> 0:33:47.560
<v Speaker 1>In five years, AI will.

0:33:47.640 --> 0:33:52.120
<v Speaker 2>Be boring, it will be integrated, It'll just work, and

0:33:52.160 --> 0:33:54.400
<v Speaker 2>there will be no now with AI thing. It'll just

0:33:54.440 --> 0:33:55.040
<v Speaker 2>be normal.

0:33:56.680 --> 0:33:59.840
<v Speaker 1>What's the number one thing that people misunderstand about AI?

0:34:00.440 --> 0:34:03.960
<v Speaker 2>It's just matrix algebra. It's just numbers. It's not sentient.

0:34:04.240 --> 0:34:07.560
<v Speaker 2>It's not coming to take us over. It's just numbers.

0:34:07.760 --> 0:34:10.799
<v Speaker 1>You're on this side of you're on the team humanity. Yeah,

0:34:10.880 --> 0:34:15.760
<v Speaker 1>you're good. What advice would you give yourself ten years

0:34:15.760 --> 0:34:17.720
<v Speaker 1>ago to better prepare for today?

0:34:18.280 --> 0:34:22.120
<v Speaker 2>Learn Python for real. It's a programming language that's extensively

0:34:22.239 --> 0:34:25.000
<v Speaker 2>used in the community. I've always dabbled in it, but

0:34:25.160 --> 0:34:26.760
<v Speaker 2>I wish I had taken it more seriously.

0:34:27.040 --> 0:34:28.960
<v Speaker 1>Yeah, did you say, who had a daughter?

0:34:29.560 --> 0:34:30.520
<v Speaker 2>I have three daughters?

0:34:30.600 --> 0:34:33.320
<v Speaker 1>You have three daughters. I have two. You're if you

0:34:33.440 --> 0:34:37.319
<v Speaker 1>got three year you're you're on your own. Are you

0:34:37.360 --> 0:34:38.600
<v Speaker 1>making them study Python?

0:34:39.719 --> 0:34:42.759
<v Speaker 2>I am actually trying to do that. We're using a

0:34:42.800 --> 0:34:45.919
<v Speaker 2>microbit micro controller tool to do like a custom video

0:34:45.960 --> 0:34:49.280
<v Speaker 2>game controller. They prefer Scratch because it's a visual programming language,

0:34:49.280 --> 0:34:51.080
<v Speaker 2>but it has a Python interface too, and I'm like

0:34:51.200 --> 0:34:52.360
<v Speaker 2>pushing them towards Python.

0:34:52.719 --> 0:34:57.040
<v Speaker 1>Good chat box and image generators are the biggest things

0:34:57.040 --> 0:34:59.520
<v Speaker 1>in consumer AI right now. What do you think is

0:34:59.560 --> 0:35:01.520
<v Speaker 1>the next business application?

0:35:03.000 --> 0:35:08.360
<v Speaker 2>Private models, small models, fine tuned on your company's data

0:35:08.960 --> 0:35:10.640
<v Speaker 2>for you to use exclusively.

0:35:11.360 --> 0:35:14.720
<v Speaker 1>Are you using AI in your own personal life these days?

0:35:14.960 --> 0:35:16.759
<v Speaker 2>Honestly, I think a lot of us are using it

0:35:16.800 --> 0:35:18.000
<v Speaker 2>and we don't even realize it.

0:35:18.440 --> 0:35:18.680
<v Speaker 1>Yeah.

0:35:18.800 --> 0:35:22.360
<v Speaker 2>I mean, I'm a ficiano of foreign languages. There's translation

0:35:22.480 --> 0:35:25.680
<v Speaker 2>programs that are built using machine learning underneath. One of

0:35:25.680 --> 0:35:28.040
<v Speaker 2>the things I've been dabbling with lately is using tech

0:35:28.080 --> 0:35:31.120
<v Speaker 2>summarizations because I tend to be very loquacious in my

0:35:31.200 --> 0:35:33.759
<v Speaker 2>note taking and that is not so useful for other

0:35:33.800 --> 0:35:36.399
<v Speaker 2>people who would just like a paragraph. So that's something

0:35:36.440 --> 0:35:39.239
<v Speaker 2>I've been experimenting with myself just to help my everyday work.

0:35:39.400 --> 0:35:43.640
<v Speaker 1>Yeah. We hear many definitions of open related to technology.

0:35:44.239 --> 0:35:47.480
<v Speaker 1>What's your definition of open and how does it help

0:35:47.520 --> 0:35:48.000
<v Speaker 1>you innovate?

0:35:48.400 --> 0:35:54.239
<v Speaker 2>My definition of open is basically sharing and being vulnerable,

0:35:54.360 --> 0:35:57.239
<v Speaker 2>like not just sharing in a have a cookie way,

0:35:57.320 --> 0:35:59.560
<v Speaker 2>but in a you know what, I don't actually know

0:35:59.600 --> 0:36:02.360
<v Speaker 2>how this works? Could you help me? And being open

0:36:02.600 --> 0:36:06.040
<v Speaker 2>to being wrong, being open to somebody helping you and

0:36:06.120 --> 0:36:08.400
<v Speaker 2>making that collaboration work. So it's not just about like

0:36:08.440 --> 0:36:12.040
<v Speaker 2>the artifactor opening, it's your approach, like how you do

0:36:12.080 --> 0:36:12.880
<v Speaker 2>things being open.

0:36:13.120 --> 0:36:16.680
<v Speaker 1>Yeah yeah, well I think that wraps us up. How

0:36:16.680 --> 0:36:20.000
<v Speaker 1>can listeners follow your work and learn more about granted

0:36:20.120 --> 0:36:21.040
<v Speaker 1>and instruct lab.

0:36:21.320 --> 0:36:24.200
<v Speaker 2>Sure, you can visit our project webpage at instruct lab

0:36:24.320 --> 0:36:27.440
<v Speaker 2>dot ai, or you can visit our GitHub at GitHub

0:36:27.520 --> 0:36:30.839
<v Speaker 2>dot com slash instruct lab. We have lots of instructions

0:36:30.920 --> 0:36:33.600
<v Speaker 2>on how to get involved in an instruct lab wonderful.

0:36:33.960 --> 0:36:40.000
<v Speaker 1>Thank you so much, Thank you, Malcolm. A big thank

0:36:40.040 --> 0:36:43.840
<v Speaker 1>you to MO for the engaging discussion on the groundbreaking

0:36:44.160 --> 0:36:49.040
<v Speaker 1>possibilities of instruct lab. We've explored how this platform has

0:36:49.080 --> 0:36:53.439
<v Speaker 1>the potential to revolutionize industries from insurance to entertainment law

0:36:53.760 --> 0:36:56.640
<v Speaker 1>by using an open source community approach that makes it

0:36:56.680 --> 0:36:59.759
<v Speaker 1>easier for more people from all backgrounds to fine tune

0:36:59.800 --> 0:37:06.360
<v Speaker 1>my for specific purposes, ultimately making AI more accessible and

0:37:06.480 --> 0:37:11.640
<v Speaker 1>impactful than ever. Looking ahead, the future of AI isn't

0:37:11.680 --> 0:37:16.760
<v Speaker 1>just about technological efficiency. It's about enhancing our everyday experiences

0:37:16.800 --> 0:37:20.799
<v Speaker 1>in ways that were never possible before, like streamlining work

0:37:20.880 --> 0:37:25.040
<v Speaker 1>for doctors to improve the patient experience or assisting insurance

0:37:25.080 --> 0:37:30.480
<v Speaker 1>agents to improve the claims experience. Instruct Lab is paving

0:37:30.520 --> 0:37:34.839
<v Speaker 1>the way for more open, accessible AI future, one that's

0:37:34.880 --> 0:37:41.360
<v Speaker 1>built on collaboration and humanity. Smart Talks with IBM is

0:37:41.400 --> 0:37:46.080
<v Speaker 1>produced by Matt Romano, Joey Fishground and Jacob Goldstein. We're

0:37:46.280 --> 0:37:50.040
<v Speaker 1>edited by Lydia jen Kott. Our engineers are Sarah Bruger

0:37:50.360 --> 0:37:54.840
<v Speaker 1>and Ben Tolliday. Theme song by Gramscope Special thanks to

0:37:54.880 --> 0:37:57.400
<v Speaker 1>the eight Bar and IBM teams, as well as the

0:37:57.400 --> 0:38:00.840
<v Speaker 1>Pushkin marketing team. Smart Talks with ib VBM is a

0:38:00.880 --> 0:38:05.640
<v Speaker 1>production of Pushkin Industries and Ruby Studio at iHeartMedia. To

0:38:05.719 --> 0:38:11.360
<v Speaker 1>find more Pushkin podcasts, listen on the iHeartRadio app, Apple Podcasts,

0:38:11.440 --> 0:38:16.200
<v Speaker 1>or wherever you listen to podcasts. I'm Malcolm Gladwell. This

0:38:16.320 --> 0:38:19.960
<v Speaker 1>is a paid advertisement from IBM. The conversations on this

0:38:20.080 --> 0:38:35.960
<v Speaker 1>podcast don't necessarily represent IBM's positions, strategies or opinions.