Emad Mostaque (CEO Stability AI): Democratizing AI | Stable Diffusion | TransformX 2022

foreign founder and CEO of stability Ai and Alexander Wang CEO and founder of scale AI stability AI focuses on building the foundation to activate Humanity's potential through AI previously a mod LED multiple technology initiatives with multilaterals and governments and was a hedge fund manager and autism researcher Ahmad is a passionate video game player and computational theologist please welcome to the stage and mod mustache and Alexander Wang [Music] foreign y excited for the last session of the day with Ahmad so Ahmad is the uh founder and CEO of stability AI who many of you will know from stable diffusion which blew up the internet uh over the past few months uh and uh and is what probably one of the coolest things happening in AI today so uh thanks for thanks for coming by thanks for having me Alexander I remember we met in London a few weeks ago and you're like well it's been six months since the release and I was like no it's been six weeks what are you talking about yeah you had a you had a good line where you've you've created a a time dilation machine in in your uh in your backyard that's the secret yeah yeah oh sorry sorry proprietary secret my bad my bad um so uh I'm excited to we'll you know we'll just kick off by uh talking through that uh scale Ai and stability AI are announcing a partnership today actually um you know uh both scale and and stability I have very resonant missions you know our mission is to accelerate the development of AI applications in order to build foundations uh to enable human potential and you know we're both aiming to democratize as much of this technology as possible make it accessible to as much of the world as possible which I'm I'm deeply passionate about so I'm super excited about our partnership I think you know we're going to be able to combine scales data engine and all the great models and and research that stability AI is doing to build incredible things um what are you what are you excited about Bob I mean it's going to be awesome right I think that you've seen some very interesting things in the area Paradigm it was bigger and bigger and bigger models like 100 billion 500 billion student trillion parameters but I think a lot of what the data showed is isn't about model size like the chinchilla scaling paper from deepmind which was like is it trading more epochs actually it's better data and it's more diverse data so as you have more diverse data you have more and better outcomes and this is what we've seen like stable diffusion itself is only 890 million parameter model versus you know you'd expect it to be billions of billions it's a hundred thousand gigabytes a two gigabyte file the data isn't amazing but it's better than what was out there and that's why we pushed the lion data set and now I think the potential for there to be global data sets one of the things we're doing is country strategies where we have an India model an at Belize model a malawian model and others combined with feedback loops to make them even smaller these models will be the key thing so one of the things that scale and stability collaborated on is the trlx instruct data set because stability isn't just image we actually are probably we are fully multimodal we do image language audio we're doing protein folding that was just announced it started trading now code models Etc and audio audio Yeah forgot about that we stuck that out because we didn't want some people to come after us oh yeah the artist thing um although we've got some great speech to speech coming the trlx thing allows you to instruct the models down because these models are big neural networks and they're kind of very general but figuring out which things get lit up on the neurons allows you to compress that so the work that kind of you've done into your data engines will really allow us to have personalized models at scale and then on the edge and this is the exciting thing stable diffusion is now working on my phone if you think about how crazy that is like just even a year ago imagine what the future is the future is everybody's got their own models their own data because that's what really makes a meaningful difference a hundred thousand gigabytes of information two gigabytes is knowledge but it's about when you bring yourself into it or your company or your culture you need to have the models and the data to do that I think this is why it's going to be a great partnership yeah I'm super excited you know there's so many interesting jumping off points from uh from from our both not only our partnership but also the incredible work that you've done I I'm curious to just Australia with maybe maybe there's some fun stories you know uh stable if you shouldn't again it is like really hard to to understate how much is broken the internet what are some of the coolest things you've seen as a result of of open sourcing and making this technology very free and open there literally been hundreds of things every time I open up my like inbox I was like look what I've done look what I've done look what I've done I think that one of the coolest things is when someone took stable diffusion combined it with dream Booth which is what Google open source to take image in which unfortunately isn't open sourced and customize it so put yourself into the model so September 17th I believe it was 48 gigabytes to run it and put yourself in the model you can put yourself in AC by September 31st it was eight gigabytes they'd optimize it down all the way so that it was accessible on any video card similarly when we were creating a stable division we're like we're going to hit 24 gigabyte vrams so it's accessible for anyone with a 3090. now it's three gigabytes like this optimization process from the open source Community has been insane beyond that it's some of this kind of taking it and applying it to some of the 3D worlds and the potential that's you can literally see the future happening in front of you that Ready Player One future minus the whiny teenagers and microtransactions it's going to be awesome yeah no it's it's pretty incredible you know maybe maybe uh it'd be cool to talk through stability ai's approach to AI which has actually been extremely um you know I almost like an into a breath of fresh air uh in a community that deeply needed sort of more open source more open development uh more openness in general you know maybe maybe walk us through what is stability's philosophy and approach to research and approach to building uh building your company yeah so I think you know stability came out of like a Luther Ai and the open source research collectives and try to put a bit more order and structure and resource behind that because this technology is just too cool and too powerful this is the reality classic AI was big data and then extending that and talking to Alexander with like the latest kicks or whatever you know there's new models they take principle-based analysis do the Transformer based architectures and the attention I mean a two gigabyte file that can recreate almost any stuff that's insane but it's the human brain it's type 1 versus type 2 brains and so I think what happened understandably maybe is that a lot of companies shy back because it's like if you have an AI that can do anything then it might do something bad so it's too dangerous to release whereas my take was like okay that means Indians can't have it you know like when is it ever going to be diffused to the world when will ever be enough in terms of some of these ethical things and who gets to make these decisions I think it should be open and it should be before this goes massive and someone does it with bad intentions that we get this technology out there to open up the dialogue and I create a stability to basically be Switzerland it was a place whereby we take the heat we take some of the positivity but we collaborate and cooperate with everyone we all do technology transfer we will listen so you know like things like building artist attribution programs and other things you know things about on the Deep fakes obviously that's going to be an issue we just announced a 200 000 prize for the best open source deep fake application so that's kind of where we try to put it in the sensitive coordinate because no one was coordinating this and all the decisions were being made behind closed doors which again is the prerogative of these entities but there was no other alternative yeah and I think one part of it you know to to add on to your point is that you know there were these large the large companies that were doing a lot of the AI development they also had very big businesses that they needed to protect which which created this sort of mixed incentive where you know they weren't they weren't building AI for AI sake per se they're building AI for the sake of the you know ultimately the sake of their larger businesses yeah I think there was that business imperative and then there's also this AGI imperative like you know you have Deep Mind opener and others there was a big focus on artificial general intelligence around these models I think it's fine but I'm more interested in augmented intelligence and AI that could push people forward and I saw a lack of anyone who was thinking how do we bring the whole world up here as opposed to we build AGI and then we can extract massive economic rents from it because the idea of any Corporation any individual like having control of the most powerful technology in the world and having Monopoly on that felt a bit wrong you know it was more like this is the first technology that allows us to go beyond it allows us to communicate like one of the things I talk about is we're having a chat here because it's the easiest thing you know we got a nice vibe text is a bit more difficult you know like you go and you write hard and that's why journalists get paid or writers or whatever visual is the most difficult and the most painful be it art or be it PowerPoints PowerPoints suck right and we've all been there you know so in a few years we'll have our open source PowerPoint replacement I'll get a Nobel Prize and then all this a program will kind of be there it's difficult but suddenly this technology makes it easy anyone can communicate visually in seconds how insane is that and so like I said the idea that this would be monopolized by the entity it feels more like a public good right it feels like the internet part two is what I call kind of the intelligent incident everyone's got this AI at the edge and it allows them to communicate with each other in any way and I think that's been the promise and it's kind of here now even though it does sound a bit like Pied Piper that's all right yeah um we uh scale named a product called nucleus so uh well there you go we're all familiar with the Silicon Valley references yeah yeah I wanted to uh maybe one thing we've talked about this privately but uh what are your thoughts on on AGI as a concept so HR is really interesting right um you get some people who are very religious about it shall we say it's a terrifying thing what if you have an AI that can do everything will it be aligned with you or not be aligned with you will it create Utopia or turn us all into paper clips my basic thinking around this was this and you know a Luther split again into stability Ai and then conjunction that does alignment full time if we train on data sets that are very Western based on the internet which we all know is a bit of a tip you know like remember that tay thing that chatbot she turned into a Nazi in a day um we're probably all gonna die and I'd rather not have that happen and I have this vision of every single country having their own data sets and models you know every single race being represented and diverse like stable diffusion is based on line which I Snapchat on the Internet it's proven not diverse but you can actually see what's in there and you can do better like in Japan a whole series of researchers rejoin the text encoder so that salaryman would no longer mean a happy man with lots of salary but very sad man you know these cultural nuances and making technology that isn't designed to Target us but instead augment our potential I believe is a necessary but not sufficient condition for aligned artificial intelligence and again my focus is not building an AGI it's about seeing how can we lift up all the humans and let them everyone activate their potential but I think if people take these additional data sets and this diversity and we widen the discussion train hundreds of thousands of new engineers and get them on the right curriculum to understand the importance of these we're more likely to have a positive outcome than a negative one yeah and and this is something that I mean I agree with you full-heartedly which is you need to unlock the models at an abstraction layer that's not just the sort of like API info model layer which is obviously the the perspective that has existed in the industry to date you need to unlock it at a you know how are the models being built you know the open sourcing of the models what is the data used to to build these models and and unlocking these lower layers of attraction in the infrastructure hierarchy then allows for the adaptation of these models to actually get around all these social issues that we've been talking about for years and years and years yeah and there's this interrogation of how it's done like to be honest you talk to any of the big model people they don't actually understand how the models really work it's like how has it got so smart we kind of know we don't know that's terrifying right and now that we've got this kind of instruct basis so most people are adapting instruct ones like I said we release trlx you can take a 20 billion parameter model and Strike It Down by everything there's actually neurons we're not sure what that looks like in terms of like you just removed a bunch of little brain cells like what does that actually do and people are kind of barreling ahead without giving any transparency on that let's make it so it isn't terrible you know and then maybe we can curb some of these bad outcomes well at the same time promoting some safe things because everyone understands how this is and more and more people are bought into it so one of the things we're doing is sponsoring the faststar AI courses which are one of the best course for converting programmers into ml developers and so we've got stable diffusion there but we'll have that across the entire ml stack to bring in more diverse voices I mean in fact from our first hundred employees at stability we had one PhD which is as differentiators you can get in AI yeah it's pretty incredible um and and I I do think that like the I think you've used the word Collective to describe your team at uh at stability AI it truly is a it's an incredible collection of people very diverse backgrounds and and um that are sort of unified by this Mission yeah I think this is the thing it's a core team but then it's a couple of hundred collaborators and a hundred thousand people in our community and we want these to continue growing because we believe this community should be everywhere from you know Uzbekistan to Indonesia to others because everyone needs to build this technology for themselves again IO for the People by the people isn't just something that we say it's something that we want to enable to happen because nobody knows what's best for people in Vietnam except for Vietnamese right and others and those Technologies should be available to the world yeah yes AI for everyone yeah for everyone yeah for everyone yeah you know I I uh I I kind of wanted to ask you this question because I think uh I think I might know the answer but I think it's a fun question to ask you know Peter Thiel always had this uh you know he begins zero to one with maybe this uh this prompt for startups which is what is uh what is basically your contrarian belief that that other people uh disagree with you on and I you know uh stability is a very in many ways a very contrarian company I'm curious what's your answer to the classic Peter teal question of what's uh what's a truth few others agree with you on people are good and so again like you look at the entire AI ethics thing again everyone's trying hard but there's this inherent belief that if you give people access to powerful technology the outcome will be bad and make the world worse my contrarian belief is like again look at social networks we're all aware of some of the benefits but in general people like these are manipulators and things like that I don't believe that centralized control over information super highways is good I believe instead that we can come together with diversity and build better systems together and that's the way we can scale and that's the opposite of what a lot of other entities believe right yeah I mean it's it's a it's a powerful idea but I think that you know what we've seen from the community the stable diffusion community and uh and all the incredible people have built on top of your platform I mean it's it's incredibly creative it's and it's brilliant yeah this is activation energy that caused that time it's like the early days of web 3 but not crap you know like it's actually useful right now and I mean this is one of the things like we all have dreams of the future like you know just self-driving cars right now 100 billion dollars has gone into that you know and now we're just seeing something here the web 300s of billions of dollars this area is just starting now it's going to be the biggest investment theme in the world because all static content will tell intelligent we have compression which is intelligent so that's crazy and people are being attracted to that like I've had people literally fly over please don't do this and Camp outside our offices and say I want to be part of this you know that's amazing um and again it's something that should be enhanced by building strong communities hopefully where people can collaborate because people want to believe in something bigger and especially changing technology around that yeah it's a it's a very compelling uh very very compelling Vision you know one of the things that we talk about is is this concept of the intelligent internet right and this idea that you know all of us will have a very personalized AI That's attuned to our preferences that will enable sort of a a new a new form of the internet what does that look like to you you know one click deeper how do you think that how do you think changes the interaction models with our with technology so I think this has been coming already so one of the ways that I describe stable diffusion is a generative search engine like you don't need to use Google image search anymore because you can just make it an image will be solved in a year not by one model that's zero shot but by putting this in a pipeline at the right place and having human in the loop interactions you know like the work that you do at scale AI understanding how humans do that at a scale and having these engines will allow us to have even better experiences that understand what we want right so there's a combination of these things well that looks like practically is like if you look at the MacBook M1 16.8 of the chipset is a neural engine that's not really used why because once they get to 80 penetration obviously they'll go straight to Transformer type engines and serial go from Siri one which is a bit crap I think it took me somewhere else to Siri five no offense to anyone from Apple but you know this is happening at scale everywhere intelligence at the edge and the models again are getting smaller and smaller so you can do that so openly I did a wonderful job with gpt3 it's 175 billion parameters instruct gpt's 1.3 billion parameters you know we're building to show scaling models all the way up to 67 billion parameters you'll be able to chop and change and kind of adjust those language models down to just a few hundred million like we just released one of the best Korean language models with our polyglot software this will enable us to have models on the edge that translate and remove the boundaries between humans and computers and I think that's insane because then you can put yourself into that and you can have your own knowledge and you can create your own worlds like I don't believe in this metaverse concept where you all go and visit somewhere that you pay 40 seven and a half percent to create anything you know instead it's about everyone having their own worlds their own Communications and the AI helping you bridge that between each other because communication is difficult now and like I said we've added that extra bit of text and now visual so you can really express yourself and communicate better and find the things that you want yeah I mean it's it's uh and it's already really happening I mean I think that everything you're describing is it doesn't even you know it's it's inspirational but it doesn't feel that part you know it's not far away it's uh it's like a stone throw away yeah I mean like you can download stable diffusion right now it's two gigabytes and it runs on your MacBook M1 and you can customize it with all your own images like look it's Alexander on the moon or kind of this or that and that's amazing because it can create anything like again anything is just really hard to wrap your head around and you're like I've just downloaded this game from EA it's like 400 gigabytes or something like that yeah this file can create anything what does the future look like with this new alien technology we're not sure but it should mean that we can enhance our ability to communicate to interact on others and again it's no longer a web 2 or classical internet where all the AI and value was in the middle and it directed you and served you ads and fed on your attention because the objective function is basically manipulation instead it pushes it out whereby you have your own AIS and that's beyond a Siri type thing because it can literally be on the edge like the big model still needs to be trained in the center and we still need to have National models and cultural models every company will have their IP turned into models but that's a very different Paradigm it's kind of what web3 wanted to be but didn't quite get there I think right yeah I mean I think it's it's an interesting point that you're bringing up which is sort of you know again referencing Peter Thiel he had this quote that uh you know if the if the if AI if if crypto is libertarian AI is communist something along those lines um and and it's probably actually flipped flipped where crypto seems extremely centralizing but uh with stability uh stable diffusion and Sibley I've demonstrated is that artificial intelligence that has this ability to be incredibly decentralized incredibly liberating yeah I think and this is the beauty of this technology whereby like you're doing all the computation up front so I mean to be honest we optimize stable diffusion so it's only 150 000 a 100 hours whereas kind of our open clip model was 1.2 million Etc it's done in advance so then it means you have to do very little computation at the edge to actually create anything and it's kind of the flip of what happened before with big search engines and data sets we're looking up the data was very cheap but then kind of doing the influence was hard it's the other way around so you can actually go the other way in stable diffusion as well you can put in a prompt uh say an image and you can get the prompt but that's highly computationally intensive and that flipping of the Paradigm is what allows it to you said flip from crypto everything one way in AI to the other way around um which is very ironic I think yeah for everyone and I think this is like I I wanted to stress this again because it's so everything we've been talking about is so um in many ways Counter Culture obviously to right now the the the big theme in AI is hey you know scale's all you need and just keep scaling up the models and you know magic stuff will happen and and I think a lot of what you're talking about was like let's build smaller models let's build models that are more democratized let's build models that have the ability to impact a broader set of people and you get same levels of performance if you sweat the details in the data up front yeah and I think this is a thing like you know scale AI right it was scaling to Big models now now it's scaling to the world is scaling to a billion people this is the one to a billion people moment right because these Transformer models and others they've only been used by a few million people we've already got 1.5 or maybe two million users of dream Studio but we think there's about 20 million users of stable diffusion already within a few kind of weeks you know like revenue is always nice got revenue all these kind of things but as you see Integrations into canva and Adobe and we've got the video games company and other things a billion people uses technology the next couple of years and then it's about how people use the technology because again Big Data age was about that inference to Target you and who you were kind of mattered now with the instructor models it's about how you use the models to make them more efficient to understand it just like we specialize so it's like maybe the model's almost like our training takes it to a kindergarten or a high school level soon and how you're going to specialize after that it's not one model to rule them all although you can Auto regress them and things like that it's multiple models for different things in the right place at the right time to make your life better yeah and I think this is the criticality of you know a coordinate sort of our partnership obviously is is this instruction component or or reinforcement learning with human feedback um and uh and I think it's it's exactly this point which is how do we take the technology and make it useful to people and and have people sort of guide the algorithm to do things that are helpful and useful to them rather than you know you've painted perfectly like rather than the other model or some Central Authority that's deciding how the model should be fit to you and yeah and I uh I and and the fact that and and I want to stress this again the fact that we're talking about it being on the edge it being fully democratized you know you can have your instructed model on your phone takes away the sort of component of a central Authority that's controlling you know your experience with AI yeah and you might still have big Central models that cooperate with the edge models but again this is why we created stability so that we could standardize a lot of this this is basically 5G but for information as it were so that's why we're taking this to countries around the world we're helping stat standards and we're Cooperative in that I know you know we put out the ethical use license for this you know we are listening to people around attribution standards we have the content Authority with Adobe and others kind of coming to have verifiable output of this generative technology that needs to be coordinated because otherwise this isn't coming anywhere it'll be a mess but if we do it right then we can really uplift the entire Humanity which I think is pretty crazy and pretty awesome it's the first time we've ever had the tools to do that and it's possible just with it like I mean obviously you all will do incredible new research and there'll be better things in the future but just with the technology today I think there's just an incredible opportunity to scale it you scale it to the billions of people in the world and it's just going to you know it has this immense opportunity and potential for good it's that leapfrogging thing like mobile phones like right now one of our projects that we don't talk about much because we've got big announcements coming is that we've been employing tablets to refugee camps around the world and we're teaching children literacy and numeracy in 13 months on one hour a day we're going to scale that incredible to tens of millions of kids but what does the future of Education look like it's AI in everything and it's reinforcement learning with human feedback every AI I want to call it one AI per child but I'm going to pushed back against that you know um let's go to the Future and take Ai and bring it back with us and that AI is not controlled by any one company it is a common good how do we build that for the best that's what we want to collaborate on yeah um you know I wanted to one of the things that I I'm uh we've talked about before that I I actually I find very compelling is sort of your thoughts on the future of media and you all announce that your launch event earlier this uh this week uh a partnership with you know media companies partnership with media companies in building um you know AI generated content so so what do you think broadly speaking all stable diffusion as well as generative AI in general will do for for media so yeah I think what you have right now is it's a bit like the internet right where you had the static internet and then asynchronous internet and now it's expanding out into all types of content because you can suddenly compress it down to this degree all content will be interactive and dynamic in the future and that has massive implications right it's like if you want Mario drinking a cup of coffee with I don't know like Master Chief you can't do that right now but with these Technologies Microsoft will have their own model Nintendo will have their own model and Starbucks will have their own model and those will be input in something like Photoshop and you can pay them a small license for each of those and they've been verified as something that's proper that's kind of insane you create whole worlds again it's ready player one experience it's not a metaverse it's a Multiverse of all these different things although yeah I'll probably have to have a chat with Jensen about Multiverse versus Omniverse Avengers versus Ben 10. so yeah I think that's what it's going to be there won't be a single company in media in the next five years that will not have a foundation model compressing their media because you know text to video is coming you know that text to audio is here text to images here and as you combine them all that's super powerful for new experiences that interactive and can tell stories yeah and I do think that this is a a great thing to to hone in on because I think and and again we've talked about this before but but stories are at their core a sort of like primitive for Humanity you know like we communicate in stories stories are powerful on their own and and anything to empower storytellers I think is is always great for Humanity and making people believe they're storytellers I'm like forget about consumption let's go for creation creation beats consumption any day and again we're moving from an instant of consumption to one of creation where you can tell your stories people don't believe they can be visual sure if you're an artist or something you do or if you want to master prowl Point teller presentation guy of which there are probably some in the audience here the vast majority don't believe that and suddenly they become capable of doing that like I genuinely believe this is one of the biggest moments since the Gutenberg Press allow people to write and spread media that way because now we can create visuals and communicate visually without any barriers like we're one part of that but it's going to be the community taking it forward and thinking about what's new ways to communicate visually textually audio to create narragut narrations and narratives that can go around the world as opposed to going through centralized entities which then optimize for engagement so one of the issues with YouTube and why kind of some of the extremism went big on that is because YouTube optimized for engagement which was extreme and so extreme stuff was engaging it wasn't anyone's fault no one did that deliberately but it was just a byproduct now we can think about new ways to exchange stories and information and things like that and find the things we need and I think this is the last time that will happen it's an inflection point in the way we are is almost a race which is kind of crazy yeah and and uh you know I think it's sort of like almost uh to your point on the intelligent Internet it's like now we have computer reading computer right uh and and human augmented computer reading computer right and you know all these new capabilities that exist to empower just a much more vibrant internet yeah and you know it can be used to empower or it can be used to enslave right almost like these are really persuasive Technologies and so again this is why we need to it's almost like again we rewind to the start of social network and we have the opportunity to also build the countermeasures and the counter balances and make them open and available to everyone because no technology that's powerful is only good or bad because it's powerful it's about how the person using it does it so as a society we have to come together to ensure that it's overwhelmingly good as opposed to bad but you only do that by making it a more diverse discussion and bringing in more people to opine on that and then build just build build build totally you know there's there's tens of thousands of members of the AI Community sort of listening right now and and um and and listening to this conversation um I think one of the one of the great calls to action should be to all the members of the AI Community how can we ensure as an open community that this technology is used for good and that that there's um we don't face a lot of the negative externalities that you know people have been afraid of what would what was your recommendations to the audience be or what would your what would your sort of advice be to to the builders in the audience it's literally just join the communities and make your voice heard you know it's self-organizing so if there's something that worries you organize it speak up and people will listen otherwise there's no way to have that communication unless we do what we're doing right now everyone is a part of the community because this is technology for Humanity we do not have all the right answers and we're trying to figure out so it's been eight weeks and two days the amount of feedback we've got we're revising a lot of different things but again it's not we it's kind of us we're just one point that helps coordinate a lot of the resources around this it's the community of this building as a community that will feedback and then they will decide for every single country race culture et cetera they'll make their own decisions on how these things are but as the AI experts who've been privileged enough to have the education to have thought this through and other things should speak up and guide it you know if there's something you're not happy about or something you think is worrying speak up formalize it form groups let's go yeah and and uh last question for me uh this has been this has been awesome Ahmad by the way um uh what are what are some of the things in the future that you're you're deeply excited about and try to be as specific as possible about uh about what you're looking forward to seriously I hate PowerPoint again no knock on Microsoft they're doing the best they can and they'll have intelligence there but like I am excited about the ability to again tell the stories like I started my career in film organized I've Been instrumental film award and I always thought what if I could make that and have this entire narrative to tell the story what if I could recreate the last season of Game of Thrones and make it not crap you know these kind of things I think that's nearly within our power and in the next five years I think we'll get there and again that's insane I don't think anyone's expecting it but you should because you know with the power of community it's going to come to be yeah it's going to be incredible I think the script too I'm excited for uh for stability or the community on top of stable diffusion and your other products to build the uh script to movie pipeline that'll uh make it so that all of us amateur playwrights can uh can become movie Producers yeah tell the insane life story of Alexander yeah not not too interesting yet but uh but uh we're working on it you got some time we got some time yeah Ahmad uh thank you so much for everything you've done for the AG Community it's genuinely uh it's been absolutely incredible and uh and thank you so much for being with us today my pleasure thank you yeah you're ready

Channel: Scale AI

Views: 46,117

Keywords:

Id: k124oUlY_6g

Length: 31min 46sec (1906 seconds)

Published: Sat Oct 22 2022