WEBVTT - Smart Talks with IBM - Hugging Face and watsonx: Why Open Source Is the Future of AI in Business 0:00:04.440 --> 0:00:12.600 Welcome to Tech Stuff, a production from iHeartRadio. Today, we 0:00:12.680 --> 0:00:15.640 are witnessed to one of those rare moments in history, 0:00:16.000 --> 0:00:19.239 the rise of an innovative technology with the potential to 0:00:19.360 --> 0:00:24.080 radically transform business and society forever. That technology, of course, 0:00:24.560 --> 0:00:28.120 is artificial intelligence, and it's the central focus for this 0:00:28.280 --> 0:00:32.320 new season of Smart Talks with IBM. Join hosts from 0:00:32.320 --> 0:00:36.040 your favorite Pushkin podcasts as they talk with industry experts 0:00:36.080 --> 0:00:39.640 and leaders to explore how businesses can integrate AI into 0:00:39.720 --> 0:00:43.040 their workflows and help drive real change in this new 0:00:43.120 --> 0:00:46.800 era of AI, and of course, host Malcolm Gladwell will 0:00:46.840 --> 0:00:49.120 be there to guide you through the season and throw 0:00:49.240 --> 0:00:52.120 in his two cents as well. Look out for new 0:00:52.159 --> 0:00:55.040 episodes of Smart Talks with IBM every other week on 0:00:55.080 --> 0:00:59.320 the iHeartRadio app, Apple Podcasts, wherever you get your podcasts, 0:00:59.520 --> 0:01:03.760 and learn more at IBM dot com slash smart Talks. 0:01:04.840 --> 0:01:08.560 Hello, Hello, Welcome to Smart Talks with IBM, a podcast 0:01:08.560 --> 0:01:14.959 from Pushkin Industries, iHeartRadio and IBM. I'm Malcolm Gladwell. This season, 0:01:15.160 --> 0:01:19.760 we're continuing our conversation with new creators visionaries who are 0:01:19.840 --> 0:01:23.880 creatively applying technology in business to drive change, but with 0:01:23.920 --> 0:01:28.760 a focus on the transformative power of artificial intelligence and 0:01:28.840 --> 0:01:31.840 what it means to leverage AI as a game changing 0:01:31.920 --> 0:01:36.399 multiplier for your business. Our guest today is Jeff Boutier, 0:01:36.840 --> 0:01:40.360 head of Product and Growth at hugging Face, the leading 0:01:40.480 --> 0:01:45.640 open source and open science artificial intelligence platform. An engineer 0:01:45.680 --> 0:01:49.560 by background, he has a self professed obsession with the 0:01:49.600 --> 0:01:54.160 business of technology. Recently, IBM and hugging Face announced a 0:01:54.240 --> 0:01:58.880 collaboration bringing together hugging faces repositories of open source AI 0:01:58.960 --> 0:02:03.800 models with IBM's Watson X platform. It's a move that 0:02:03.840 --> 0:02:08.239 gives businesses even more access to AI while staying true 0:02:08.280 --> 0:02:13.920 to IBM's long standing philosophy of supporting open source technology. 0:02:14.840 --> 0:02:18.720 With open source, businesses can build better AI models that 0:02:18.800 --> 0:02:23.320 suit their specific needs using their own proprietary data while 0:02:23.360 --> 0:02:28.440 browsing a ready catalog of pre trained models. In today's episode, 0:02:28.639 --> 0:02:31.760 you'll hear why open source is so crucial to the 0:02:31.800 --> 0:02:36.400 advancement of AI, how IBM's Watson X interacts with open 0:02:36.440 --> 0:02:40.919 source AI, and Jeff's thoughts on why this singular omnipotent 0:02:41.160 --> 0:02:45.240 AI model is a myth. Jeff spoke with Tim Harford, 0:02:45.440 --> 0:02:49.680 host of the Pushkin podcast Cautionary Tales, a longtime columnist 0:02:49.760 --> 0:02:53.239 at the Financial Times, where he writes the Undercover Economist. 0:02:53.520 --> 0:02:57.320 Tim is also a BBC broadcaster with his show More 0:02:57.480 --> 0:03:01.160 or Less. Okay, let's get to the interview. 0:03:08.600 --> 0:03:11.480 I am a Jeff Boudier and I'm a product director 0:03:11.560 --> 0:03:12.600 at hugging. 0:03:12.280 --> 0:03:16.800 Face, So I'm immediately intrigue. Hugging Face. Is this a 0:03:16.880 --> 0:03:18.840 reference to the Alien movie or something else? 0:03:20.200 --> 0:03:24.040 It is not, and it may be not obvious to 0:03:24.160 --> 0:03:27.800 a listener, but hugging Face is the name of that 0:03:27.960 --> 0:03:30.799 cute emoji, you know, the one that's smiling with his 0:03:30.960 --> 0:03:34.120 two hands extended like that to give you a big hug. 0:03:34.360 --> 0:03:37.640 That's hugging Face. So basically we name the company after 0:03:37.760 --> 0:03:39.120 an emoji. 0:03:40.000 --> 0:03:42.400 And it is I saw your website and it is 0:03:42.440 --> 0:03:45.480 a very friendly emoji. So that's that's nice. So tell 0:03:45.560 --> 0:03:47.880 us a little bit about hugging Face and about what 0:03:47.920 --> 0:03:48.320 you do that. 0:03:48.960 --> 0:03:53.440 Of course, hugging Face is the leading open platform for 0:03:53.800 --> 0:03:59.000 AI builders, and it's the place that's all the AI 0:03:59.080 --> 0:04:04.600 researchers use to share their work, their new AI models 0:04:04.880 --> 0:04:09.600 and collaborate around them. It's the place where the data 0:04:09.840 --> 0:04:15.080 scientists go and find those pre train models and access 0:04:15.160 --> 0:04:19.000 them and use them and work with them, and increasingly 0:04:19.040 --> 0:04:23.360 it's the place where developers are coming to turn all 0:04:23.400 --> 0:04:28.480 of these AI models and data sets into their own applications, 0:04:28.520 --> 0:04:29.680 their own features. 0:04:30.360 --> 0:04:33.039 So it's like the I don't know, the Facebook group 0:04:33.120 --> 0:04:36.000 or the Reddit or the Twitter for people who are 0:04:36.040 --> 0:04:40.280 interested in particularly generative language AI, or all kinds of 0:04:40.360 --> 0:04:41.920 artificial intelligence. 0:04:42.160 --> 0:04:46.400 All kinds of AI really, and of course generative AIS 0:04:46.600 --> 0:04:51.320 this new wave that has caught the world by storm. 0:04:51.680 --> 0:04:55.360 But on Hiking Face you can find any kind of model, 0:04:55.680 --> 0:04:59.560 the new sort of transformers models to do anything from 0:05:00.000 --> 0:05:04.080 translation or if you wanted to transcribe what I'm saying 0:05:04.120 --> 0:05:07.560 into text, then you would use a transformer model. If 0:05:07.560 --> 0:05:10.880 you wanted to then take that text and make a summary, 0:05:11.320 --> 0:05:15.000 that would be another transformer model. If you wanted to 0:05:15.360 --> 0:05:19.400 create a nice little thumbnail for this podcast by typeing 0:05:19.440 --> 0:05:23.280 a sentence, that would be another type of model. So 0:05:23.360 --> 0:05:26.480 all these models you can find. There's actually three hundred 0:05:26.640 --> 0:05:31.119 thousands that are free and publicly accessible. You can find 0:05:31.160 --> 0:05:34.840 them on our website at Hikingphase dot co and use 0:05:34.920 --> 0:05:37.520 them using our open source libraries. 0:05:38.360 --> 0:05:41.160 And so this is this is fascinating. So there are 0:05:41.160 --> 0:05:44.520 three hundred thousand models. Now when you say model, I'm 0:05:44.560 --> 0:05:46.400 thinking in my head, oh, it's kind of like a 0:05:47.360 --> 0:05:50.080 computer program. There were three hundred thousand computer programs. Is 0:05:50.680 --> 0:05:52.359 that roughly right or it not? 0:05:52.440 --> 0:05:57.839 Really, it's a general idea. A model is a giant 0:05:59.680 --> 0:06:05.120 set of numbers that are working together to sift through 0:06:05.760 --> 0:06:08.960 some inputs that you're going to give it. So think 0:06:09.000 --> 0:06:13.480 of it of a big black box filled with numbers, 0:06:14.440 --> 0:06:19.240 and you give it as an input, maybe some text, 0:06:19.960 --> 0:06:23.880 maybe a prompt, so you're asking, you're giving an instruction 0:06:24.120 --> 0:06:26.719 to the model, or maybe you give it an image 0:06:26.800 --> 0:06:31.240 as an input, and then it will sift through that 0:06:31.400 --> 0:06:35.400 information thanks to all of these numbers, which we call 0:06:35.440 --> 0:06:39.880 in the field parameters, and it will produce an output. 0:06:40.480 --> 0:06:43.039 So when I told you, hey, we can transcribe this 0:06:43.279 --> 0:06:47.200 conversation into text, the input would have been the conversation 0:06:47.800 --> 0:06:50.440 in an audio file, and then the output would have 0:06:50.480 --> 0:06:53.479 been the text of the transcription. If you want to 0:06:53.560 --> 0:06:57.599 create a thumbnail for this podcast episode, then the input 0:06:57.640 --> 0:07:00.520 would be what we call the prompt, which is really 0:07:00.520 --> 0:07:05.040 a text description like a Frenchman in San Francisco talking 0:07:05.080 --> 0:07:10.720 about machine learning, and the output would be completely original image. 0:07:11.280 --> 0:07:14.600 So that's how I think about what an AI model is, 0:07:15.080 --> 0:07:19.200 and I think what we're starting to realize is that 0:07:20.080 --> 0:07:24.280 this is becoming the new way of building technology in 0:07:24.320 --> 0:07:28.320 the world. It has been for the field of dealing 0:07:28.440 --> 0:07:32.400 understanding generating text for quite some time, but now it's 0:07:32.520 --> 0:07:36.440 sort of moving across every field of technology. We have 0:07:36.840 --> 0:07:40.960 models to create images, as I say, but also to 0:07:41.120 --> 0:07:46.600 generate new proteins to make predictions on numerical data. So 0:07:46.640 --> 0:07:51.160 every kind of field of machine learning is now using 0:07:52.600 --> 0:07:56.480 this new type of models. But what's interesting is that 0:07:57.080 --> 0:08:00.720 if you're, say a product manager at a company, and 0:08:00.760 --> 0:08:03.840 you say, hey, I want to build a feature that 0:08:03.960 --> 0:08:07.320 does this. A few years ago, the approach would have 0:08:07.360 --> 0:08:11.360 been to ask a software developer to write a thousand 0:08:11.400 --> 0:08:14.840 lines of code in order to build a prototype. And 0:08:14.920 --> 0:08:18.360 the new way of doing things today is to go 0:08:18.520 --> 0:08:23.040 look for an off the shelf pre train model that 0:08:23.080 --> 0:08:27.200 does a pretty good job at solving exactly that problem, 0:08:27.320 --> 0:08:30.400 so you can create a prototype of that feature fast. 0:08:30.440 --> 0:08:33.000 So it's a new approach of building tech. 0:08:33.200 --> 0:08:36.320 I'm not a programmer, but I'm aware that there was 0:08:36.520 --> 0:08:39.080 this idea of open source code, and now we have 0:08:39.160 --> 0:08:42.120 open source models. So what does it mean for something 0:08:42.120 --> 0:08:43.040 to be open source. 0:08:43.640 --> 0:08:49.400 Open source AI actually means a lot of different specific things. 0:08:50.080 --> 0:08:54.280 It's the open source implementation of the model. So if 0:08:54.320 --> 0:08:58.600 you use the Hugging Phase transformers library to use a model, 0:08:58.640 --> 0:09:03.000 you're using an open source code library to use that model. 0:09:03.080 --> 0:09:06.320 Just to end up on the transformers. These are these 0:09:06.400 --> 0:09:09.640 kind of ways of turning a picture of a dog 0:09:09.760 --> 0:09:12.440 into a text output that says, hey, this is a 0:09:12.440 --> 0:09:15.079 picture of a dog, or this is a French text 0:09:15.080 --> 0:09:17.920 and with the transformers helping you turn it into English text, 0:09:18.000 --> 0:09:19.880 or it's doing all of these things that you've been describing. 0:09:19.960 --> 0:09:23.920 That's the transformer is the kind of the engine at 0:09:23.920 --> 0:09:24.760 the heart of that. 0:09:25.559 --> 0:09:29.960 Yes, exactly. And we call them transformers because they correspond 0:09:30.000 --> 0:09:33.920 to this new way of building machine learning models that 0:09:34.080 --> 0:09:38.800 was introduced by Google actually with a very important paper 0:09:39.120 --> 0:09:41.920 called Attention is All You Need and that was published 0:09:41.920 --> 0:09:46.440 in twenty seventeen by researchers out of Google Deep Mind. 0:09:47.400 --> 0:09:50.680 Well that's just six years so new. 0:09:51.960 --> 0:09:55.240 It is very new, and ever since the piece of 0:09:55.480 --> 0:10:00.920 innovation of like new model architectures has real really accelerated. 0:10:01.240 --> 0:10:06.000 But it really started from this inflection point that came 0:10:06.120 --> 0:10:10.400 from this paper and its implementation in what is now 0:10:10.440 --> 0:10:16.240 called Transformer models, the transformer that has conquered every area 0:10:16.360 --> 0:10:18.080 of machine learning since. 0:10:18.280 --> 0:10:21.840 Okay, so say turned up. So you've got this library 0:10:21.840 --> 0:10:26.120 of Transformer models and that open source, and that means 0:10:26.280 --> 0:10:28.480 that means what anyone can use them for free, or 0:10:29.240 --> 0:10:31.320 that anybody can implement them for free. What does it mean? 0:10:32.840 --> 0:10:35.800 So again, there's lots that go into it, but the 0:10:35.840 --> 0:10:40.240 most important thing is for the model itself to be 0:10:40.480 --> 0:10:44.600 available so that a data scientists or an engineer can 0:10:45.000 --> 0:10:49.400 download them and use them. And also there are a 0:10:49.400 --> 0:10:54.240 lot of considerations about how you make them accessible, and 0:10:54.280 --> 0:10:58.240 a very important one is whether or not you give 0:10:58.480 --> 0:11:03.520 access to the training data, all the information that went 0:11:03.679 --> 0:11:07.920 into training that model and teaching it to do what 0:11:08.720 --> 0:11:09.640 it's trained to do. 0:11:09.800 --> 0:11:12.800 So I might have fed millions of words into a 0:11:12.920 --> 0:11:16.040 into a language transformer, or I might have fed millions 0:11:16.040 --> 0:11:18.640 of photographs into a into a picture transformer. 0:11:18.720 --> 0:11:22.160 Yeah, yes, and now it's trillions and that and the 0:11:22.520 --> 0:11:26.160 accessibility of that training data is very very important. 0:11:27.160 --> 0:11:32.960 What's the relationship between the hugging face libraries and GitHub, which, 0:11:34.080 --> 0:11:38.360 if I understand GitHub correctly, it's this the repository of 0:11:38.400 --> 0:11:42.360 open source code lots and lots of lines of code 0:11:42.360 --> 0:11:47.280 and routines and programs that are shared and updated and tracked, 0:11:47.320 --> 0:11:50.480 and they're all available on GitHub, which sounds similar to 0:11:50.520 --> 0:11:52.959 what you're doing with hugging face for AI. So what 0:11:52.960 --> 0:11:55.600 what what is the interaction or the relationship there? 0:11:56.200 --> 0:11:58.640 Yeah, I think you nailed it on the head there. 0:11:58.679 --> 0:12:02.839 So hugging phase is to AI what GitHub is to code, right, 0:12:02.840 --> 0:12:08.959 It's this central platform where AI builders can go find 0:12:09.440 --> 0:12:14.720 and collaborate around AI artifacts, which are models and data sets. 0:12:14.760 --> 0:12:18.719 So it's quite different than software, but we play this 0:12:18.840 --> 0:12:23.079 central role in the community to share and collaborate and 0:12:24.080 --> 0:12:28.880 access all of those artifacts for AI, like GitHub offers 0:12:28.880 --> 0:12:29.839 for code. 0:12:30.679 --> 0:12:33.600 And that community must be incredibly important. I mean, the 0:12:33.640 --> 0:12:36.240 open source is nothing if you don't have a community 0:12:36.280 --> 0:12:38.640 of people working on it. So how have you been 0:12:38.679 --> 0:12:41.800 able to foster and nurture that community. 0:12:42.400 --> 0:12:45.760 Well, I think it goes to the origins of the 0:12:45.840 --> 0:12:49.960 transformer model and hugging and face role into that. So 0:12:50.600 --> 0:12:55.160 when the first sort of open model came out, it 0:12:55.280 --> 0:12:58.440 was called Bird and it came out of Google. The 0:12:58.480 --> 0:13:02.720 only way you could would access it was to use 0:13:02.920 --> 0:13:07.360 a tool called TensorFlow. But it happened that most of 0:13:07.400 --> 0:13:12.840 the AI community was using a different tool called PyTorch, 0:13:13.960 --> 0:13:18.920 and something that Hugging Face did is to make that 0:13:19.000 --> 0:13:25.480 new model Bert accessible to all PyTorch user and they 0:13:25.480 --> 0:13:28.680 did it in open source. It was a project called 0:13:29.200 --> 0:13:32.720 Bert's pre Trained PyTorch or bird pitworch pre trained. 0:13:33.240 --> 0:13:35.360 So this is like being able to play my Zelda 0:13:35.400 --> 0:13:39.440 game on an Xbox or a PlayStation, right or am 0:13:39.480 --> 0:13:41.120 I not really understanding what's going on? 0:13:41.559 --> 0:13:43.920 No, That's exactly what it is. And the thing is 0:13:44.120 --> 0:13:48.080 everybody was using the game Boy and so it became 0:13:48.440 --> 0:13:53.200 a very popular and from there the community sort of 0:13:53.280 --> 0:13:56.839 gathered to make all the other models that were then 0:13:56.960 --> 0:14:01.360 published by AI researchers available with that library, which was 0:14:01.440 --> 0:14:07.000 quickly renamed from bird bretrain Bytorch into Transformers to welcome 0:14:07.120 --> 0:14:12.280 like all of these different new models, and today that's 0:14:12.440 --> 0:14:17.440 open source library. Transformers is what all AI builders are 0:14:17.559 --> 0:14:20.880 using when they want to access those models, see how 0:14:20.920 --> 0:14:22.400 they work, and build upon them. 0:14:23.720 --> 0:14:26.880 What's striking about this field is that it's changing so fast, 0:14:26.920 --> 0:14:30.720 it's improving so quickly. So how do open source models 0:14:31.440 --> 0:14:35.320 keep up with that? How do they get iterated and improved? 0:14:35.440 --> 0:14:38.400 Actually? It's not so much that open source is keeping 0:14:38.480 --> 0:14:41.440 up with it. It's actually open source that is driving 0:14:42.160 --> 0:14:45.600 that is driving this piece of change. And that's because 0:14:46.320 --> 0:14:51.680 with open source and open research data, scientists researchers can 0:14:51.800 --> 0:14:55.480 build upon each other's work, they can reproduce each other's work, 0:14:55.760 --> 0:14:59.760 they can access each other's work using our open source library, 0:15:00.000 --> 0:15:02.320 et cetera. So in a sense, it's not really that 0:15:02.720 --> 0:15:07.320 open source AI is a new idea. It's rather the opposite. 0:15:07.480 --> 0:15:11.600 There's been a blip of time in which closed source 0:15:11.840 --> 0:15:15.560 AI seemed to be the dominant way, but it's really 0:15:16.120 --> 0:15:19.840 a blip. In fact, you know, none of the incredible 0:15:19.880 --> 0:15:24.480 advances that we're marvel about today would be possible without 0:15:24.680 --> 0:15:27.680 open source. We're standing upon the shoulders of fifty years 0:15:27.680 --> 0:15:32.120 of research and open source software. So I think that 0:15:32.120 --> 0:15:35.000 that's really important. If it wasn't for that, we'll probably 0:15:35.000 --> 0:15:39.880 be fifty years away from having these amazing experiences like 0:15:40.040 --> 0:15:45.840 JGBT or stable diffusion, et cetera. So it's really open 0:15:45.880 --> 0:15:50.240 source that is fueling this pace of change, all these 0:15:50.280 --> 0:15:53.800 new models, all these new capabilities. To give you an example, 0:15:54.120 --> 0:15:58.640 so Meta released the Lama large language model just a 0:15:58.680 --> 0:16:02.960 few months ago, and ever since, there's been this Cambrian 0:16:03.120 --> 0:16:07.520 explosion of variations and improvements upon the original models, and 0:16:07.560 --> 0:16:10.600 today there are over a thousands of them that we 0:16:11.160 --> 0:16:16.560 host and track and evaluate. So yeah, open source is 0:16:16.600 --> 0:16:20.280 really the gas and the engine for that. 0:16:21.560 --> 0:16:24.400 Jeff just made it clear that it is open source, 0:16:24.640 --> 0:16:28.640 not closed that sets the pace for AI innovation. If 0:16:28.680 --> 0:16:33.240 that's true, then forward thinking businesses shouldn't shy from leveraging 0:16:33.320 --> 0:16:37.680 open source AI to solve their own proprietary challenges. But 0:16:37.880 --> 0:16:42.800 how businesses can face serious obstacles when trying to adopt 0:16:43.040 --> 0:16:47.600 open source technologies, like complying with government regulation or making 0:16:47.640 --> 0:16:51.880 sure their customers data stays protected. In the next part 0:16:51.920 --> 0:16:56.200 of their conversation, Jeff and Tim discuss how IBM's collaboration 0:16:56.360 --> 0:17:00.520 with hugging Face empowers businesses to tap into the open 0:17:00.560 --> 0:17:04.879 source AI community and how the watsonex platform can enable 0:17:04.920 --> 0:17:08.720 them to customize those AI models to their needs. 0:17:09.400 --> 0:17:11.920 Just want to ask about the partnership between hugging Face 0:17:11.960 --> 0:17:14.720 and an IBM. How did that come about? 0:17:16.680 --> 0:17:23.280 Well, it came through a conversation, a conversation between our CEO, 0:17:24.080 --> 0:17:29.320 Clement de Lange and Bill Higgins IBM, who's really really 0:17:29.400 --> 0:17:34.280 close to all the amazing research work and open source 0:17:34.400 --> 0:17:39.399 work that's happening at IBM, and that conversation sort of 0:17:39.680 --> 0:17:44.240 sparked the evidence that we needed to do something together. 0:17:44.840 --> 0:17:48.840 We share a lot of values in terms of the 0:17:48.880 --> 0:17:53.600 importance of open source, which is fundamental to us, with 0:17:54.000 --> 0:17:58.800 the importance of doing things in an ethics first way 0:17:58.920 --> 0:18:04.040 to enable the commune to incorporate ethical considerations in how 0:18:04.520 --> 0:18:09.760 they're building AI. And we sort of have a different 0:18:10.040 --> 0:18:14.119 audience to start with, which is all the AI builders 0:18:14.240 --> 0:18:18.840 use hiking phase today to access all the models we 0:18:18.960 --> 0:18:22.879 talked about, to use them using our open source and 0:18:22.920 --> 0:18:27.320 build with them. And IBM has this incredible history of 0:18:27.440 --> 0:18:32.920 working with enterprise companies and enabling them to make use 0:18:32.960 --> 0:18:37.000 of that technology in a way that's compliant with everything 0:18:37.040 --> 0:18:40.800 that an enterprise requires, and so being able to marry 0:18:40.840 --> 0:18:45.000 these two things together is an amazing opportunity. And now 0:18:45.040 --> 0:18:49.280 we can enable the largest corporations that have sort of 0:18:49.520 --> 0:18:54.920 complex requirements in order to deploy machine learning systems and 0:18:55.720 --> 0:18:59.080 give them an easy experience to take advantage of all 0:18:59.119 --> 0:19:01.600 the latest and great is that AA has to offer 0:19:02.119 --> 0:19:02.920 through our platform. 0:19:04.480 --> 0:19:08.040 Let's talk about this idea of a single model or 0:19:08.080 --> 0:19:11.600 a variety of models, because what I've been hearing you say. 0:19:12.160 --> 0:19:14.000 You've been saying, oh, there are lots of models, there 0:19:14.040 --> 0:19:18.119 are hundreds of thousands of models available on hugging Face. 0:19:18.280 --> 0:19:21.640 But you've also said there's this single thing, the transformer, 0:19:22.280 --> 0:19:26.720 and they're all transformers. So if they're all basically the 0:19:26.760 --> 0:19:31.480 same thing, why can't you just build one super clever 0:19:31.560 --> 0:19:32.640 model that can do everything. 0:19:34.760 --> 0:19:39.679 That's a really interesting idea and very much a new idea. 0:19:40.520 --> 0:19:44.400 The reason we have over a million repositories three hundred 0:19:44.480 --> 0:19:48.119 thousand free and accessible models on a hiking Face platform 0:19:48.560 --> 0:19:52.320 is that models are typically trained to do one thing, 0:19:52.680 --> 0:19:55.920 and they're typically trained to do one thing with specific 0:19:55.960 --> 0:20:02.439 types of data. And what became new and evidence in 0:20:02.480 --> 0:20:04.920 the research that came out over the last couple of 0:20:05.000 --> 0:20:09.120 years is that if you train a big enough model 0:20:09.600 --> 0:20:14.680 with enough data, then those models start to have sort 0:20:14.680 --> 0:20:18.720 of general capabilities. You can ask them to do different things. 0:20:19.000 --> 0:20:22.480 You can even train them to respond to instructions. So 0:20:22.600 --> 0:20:26.840 with the same model, you can say, hey, summarize this paragraph, 0:20:27.240 --> 0:20:30.960 translate this into English, start a conversation in French, and 0:20:30.960 --> 0:20:34.560 pivot to German. And so these are general sort of 0:20:34.680 --> 0:20:42.000 language capabilities. And I think when CHGBT came online and 0:20:42.320 --> 0:20:47.000 the world sort of discovered these new capabilities, there was, 0:20:47.560 --> 0:20:50.480 at least for a short period, this sort of idea, 0:20:50.600 --> 0:20:54.480 this sort of myth that the endgame of all this 0:20:55.440 --> 0:20:59.199 is maybe one or a handful of models there are 0:20:59.400 --> 0:21:03.640 so much better than anything else than exists, that they 0:21:03.640 --> 0:21:06.280 can do anything that we can ask them to do, 0:21:07.080 --> 0:21:10.560 and that's the only model that we will need. And I, 0:21:10.800 --> 0:21:15.080 for one, think it is a myth. I don't think 0:21:15.119 --> 0:21:19.200 it is practical for a variety of reasons. Say you're 0:21:19.600 --> 0:21:23.760 writing an email and you have like this great suggestion 0:21:23.920 --> 0:21:28.199 of text to sort of complete your sentence, Well, that's AI. 0:21:28.640 --> 0:21:31.159 That's a large language model, that's a transformer model that 0:21:31.200 --> 0:21:33.840 does that. So there are a ton of existing use 0:21:33.880 --> 0:21:37.520 cases like this, and these use cases are powered by 0:21:38.320 --> 0:21:41.280 specific models that have been trained to do one thing 0:21:41.400 --> 0:21:44.479 well and to do it fast. If you wanted to 0:21:44.600 --> 0:21:51.200 apply these sort of all knowing, powerful oracle type of model, 0:21:51.600 --> 0:21:55.639 you would not be able to serve millions of customers 0:21:55.680 --> 0:21:58.359 through a search engine. You will not be able to 0:22:00.080 --> 0:22:04.119 complete people's sentences because the amount of money that you 0:22:04.160 --> 0:22:07.400 would need, the number of computers that you would need 0:22:07.640 --> 0:22:13.240 to run such of service just exceeds what is available 0:22:13.359 --> 0:22:18.760 on the planet. So one reason for which it's not 0:22:18.880 --> 0:22:24.359 a practical scenario is that it's just very expensive to 0:22:24.600 --> 0:22:27.440 run those very very large models. 0:22:27.760 --> 0:22:29.920 What I'm hearing is it's like, look, if you want 0:22:29.920 --> 0:22:33.679 to screw in a screw you need a screwdriver. You 0:22:33.720 --> 0:22:37.720 don't want an entire tool shed full of tools if 0:22:37.800 --> 0:22:39.960 the task is to screw in a screwdriver, and sure 0:22:40.040 --> 0:22:43.240 you could bring the toolshed that are all the tools. 0:22:43.280 --> 0:22:47.320 There's a screwdriver there, but it's not necessary. It's incredibly expensive, 0:22:47.320 --> 0:22:52.119 it's incredibly cumbersome, and that cost exists even though maybe 0:22:52.200 --> 0:22:54.879 is the user who's just typing in a into a 0:22:54.920 --> 0:22:57.680 prompt box. The user may not see it, but it's 0:22:57.680 --> 0:22:58.720 still very real. 0:23:00.040 --> 0:23:03.480 That's right. And then another one is performance. So taking 0:23:03.520 --> 0:23:06.760 the screwdriver example, so and by the way, like we're 0:23:06.800 --> 0:23:09.560 not quite there at this moment where we have this 0:23:09.720 --> 0:23:13.240 all knowing, powerful oracle that is still sort of a 0:23:13.320 --> 0:23:16.919 sci fi scenario, but we have screw drivers, but we 0:23:17.040 --> 0:23:21.680 also have the leatherman, right, the multitol Swiss army knife. 0:23:21.920 --> 0:23:24.919 And that's sort of the moment that we are in today. 0:23:24.960 --> 0:23:28.600 But now if I'm trying to open up my computer, 0:23:29.200 --> 0:23:32.439 turns out that it requires a specific kind of screw 0:23:32.600 --> 0:23:36.760 like these tiny little tork screws, and having a torqu 0:23:36.800 --> 0:23:40.520 screwdriver will get me much further than trying to use 0:23:40.760 --> 0:23:43.399 my leather man, where maybe I'll get the knife blade 0:23:43.440 --> 0:23:46.520 and it will mess up the screw and maybe eventually 0:23:46.520 --> 0:23:49.160 I'll get to what I need. But my point is 0:23:49.280 --> 0:23:54.399 that if you take a very specifically trained model for 0:23:54.480 --> 0:23:57.960 a particular problem, it will work much better. It will 0:23:57.960 --> 0:24:02.760 give you better results than a very very generalistic, big 0:24:02.840 --> 0:24:05.800 model that can do a lot of things. And so 0:24:05.880 --> 0:24:10.119 for things like search engines or things like translation, for 0:24:10.280 --> 0:24:15.000 things that are very specific, companies are much better off 0:24:15.119 --> 0:24:19.680 using smaller, more efficient models that produce better results. 0:24:19.480 --> 0:24:24.000 That's really interesting. And presumably then being able to know 0:24:24.040 --> 0:24:26.800 which model to use, or being able to know who 0:24:26.840 --> 0:24:30.640 to ask which model to use, becomes a very important capability. 0:24:31.480 --> 0:24:35.000 Yes, and that's what we're trying to make easy through 0:24:35.040 --> 0:24:35.800 our platform. 0:24:37.160 --> 0:24:41.160 So tell me about how this works with IBM's what's 0:24:41.160 --> 0:24:44.760 an X platform? How do you see hugging faces customers 0:24:44.800 --> 0:24:45.640 benefiting from that? 0:24:47.560 --> 0:24:51.640 The end goal is to make it really easy for 0:24:51.760 --> 0:24:56.240 what's an X customers to make use of all the 0:24:56.320 --> 0:25:00.600 great models and libraries that we talked about, all the 0:25:00.600 --> 0:25:03.320 the three hundred thousand models are today on hugging face 0:25:04.160 --> 0:25:08.440 and to do this we need to really collaborate deeply 0:25:08.520 --> 0:25:12.080 with the IBM teams that build the What's and X 0:25:12.160 --> 0:25:17.360 platform so that our libraries, our open source our models 0:25:17.760 --> 0:25:21.480 are well integrated into the platform. If you are a 0:25:21.640 --> 0:25:24.560 single user, if you are a data science student and 0:25:24.600 --> 0:25:26.680 you want to use a model, is we make it 0:25:26.720 --> 0:25:29.399 super easy, right. We have our open source library. You 0:25:29.440 --> 0:25:32.159 can download the model on your computer and run with 0:25:32.240 --> 0:25:37.320 it then. But in enterprises there is a vast complexity 0:25:37.560 --> 0:25:42.800 of infrastructure and rules around what people can do and 0:25:43.400 --> 0:25:47.600 how the data can be accessed, and all this complexity 0:25:48.280 --> 0:25:52.879 is sort of solved by the Watson X platform. 0:25:53.560 --> 0:25:57.520 This season of the Smart Talks podcast features what we're 0:25:57.520 --> 0:26:00.399 calling new creators. Do you see yourself as being a 0:26:00.440 --> 0:26:01.440 creative person? 0:26:02.359 --> 0:26:05.960 Ah, I think it's a requirement for the job. I mean, 0:26:05.960 --> 0:26:10.720 we're in such a new and rapidly evolving industry that 0:26:11.000 --> 0:26:15.000 we have to be creative in order to invent the 0:26:15.080 --> 0:26:19.640 business models the use cases of tomorrow. My role within 0:26:19.680 --> 0:26:24.680 the company is really to create the business around all 0:26:24.840 --> 0:26:28.840 the great work of our science and open source and 0:26:28.960 --> 0:26:33.080 product team, and by and large, the business model of 0:26:33.240 --> 0:26:38.200 AI within the whole ecosystem is still something that companies 0:26:38.240 --> 0:26:43.200 are trying to figure out. So creativity is really important 0:26:43.320 --> 0:26:47.120 to really have the conversation with companies, understand what they're 0:26:47.160 --> 0:26:49.240 trying to do, and then build the right kind of solution. 0:26:49.840 --> 0:26:54.080 So that's like where creativity comes into play. 0:26:54.800 --> 0:26:59.000 And one of the things that you've you've been talking 0:26:59.040 --> 0:27:02.520 about is just this growing number of models, this growing 0:27:02.600 --> 0:27:09.040 number of capabilities, this growing number of use cases enormously 0:27:09.080 --> 0:27:15.000 exciting but also I think completely bewildering for most people 0:27:16.000 --> 0:27:20.640