The AI x Crypto Intersection | David Shi - Sam Green - Scott Trowbridge - Grabiel Tumlos

[Applause] okay great well I am personally very excited for this panel because I could not imagine a better set of people to do this so I'll quickly give my intro and then I'll let everybody individually introduce themselves but my name is Casey and before this most recently I was an investment partner at Paradigm before that I was uh doing engineering work at Google mainly ML and ux work and now I'm an open source developer also investing in the space and so why don't we start with Scott and then we'll just go down the line hi everyone my name is Scott um I'm from stability AI um vice president of Business Development of Partnerships uh F stability is best known for launching um and pioneering stab diffusion which is like an image generation tool um we now have you know models across every major modality video audio um language and various others and yeah really using this technology to elevate humanity is our mission statement hey I'm David shei um I'm one of the founders of operator. and we build an agent to agent toolkit uh we make it really easy for developers to take an agent that they've developed and bring it onchain so that anyone can message it as long as you have a wallet address hi hi everyone uh my name is Sam green I am a co-founder of semiotic Labs semiotic is a core Dev team in the graph protocol and we specialize in cryptography and AI for the protocol uh my my background uh I got a PhD in reinforcement learning in uh 2019 and so yeah these are exciting times uh for for us for me hey everyone thanks for being here uh my name is Gabriel I'm the founder of light works we are an autonomous laboratory for the future of onchain intelligence um we're currently working on autonomous labeling networks so so um for every llm and chat that you send out to chat CHT there's basically an army of Filipinos that are in the back end uh labeling some stuff to make sure that it understands what it's learning um it's been about a decade in the space for me I used to work at consensus from 2015 to 2018 uh and before that I was a CPA so forever a recovering accountant wow wow okay great so we have about 30 minutes today I'm thinking we can do a conversation for 20 minutes and then we'll open it up to Q&A and make this as interactive as possible but just to kick off I would love to hear some hot takes in terms of what do you guys think over the eth Denver so far has been to spoken about kind of the more of a trit topic and what do you think people haven't talked about enough at the intersection of crypto and AI happy to kick off uh so I guess I came into this looking more on the AI side you know we like really focus on like training open foundation models and like deploying those into the public allowing anyone to access them and fine tune them optimize them Etc um but what's really compelling is looking at like how crypto web 3 technology is actually very complimentary to um AI specifically generative AI so now you know learned a heck of a lot over the past several days I think there's a few like trends that are quite compelling um on the sort of training model side still you require huge amounts of compute to do these you know large language models and various other modality models but actually on the inference side there's something very compelling around you know decentralized compute networks I think these models are becoming smaller you can run them on lower-end Hardware so I think there's something very interesting there um the second part is attribution so you know as a Creator um where your data is used in these open data sets having the ability for um systems to effectively be compensated to those initial creators is something um interesting as well and the last one is um data licensing I'm sorry uh model licensing so looking at you know where a initial core model is trained a variant model and you know fine CH model and a derivative of that fine CH model it goes there's many layers how does that initial Creator get commentator as well and just one follow question on that Scott so right now in web 2 we use hugging face mainly for models and I'm curious from your Vantage Point why don't you think they've handled attribution yet or why don't they reward the model Creator is today yeah good point I think like hugging face right now is just a directory right it's like you go to hugging face you see the open models there's some like benchmarking you see like the research paper you download it then you effectively go away um why they haven't done that like I don't really know to be honest I think it's like it's sort of I guess the next step they've started they have logins now so potentially um that's something that will come but I don't think like their core focus is not really on like decentralization itself right like it's a platform is just a directory yeah um I think uh in crypto we we focus a lot on infrastructure and we're definitely at the time of building infrastructure so it makes perfect sense um but I think we haven't as an industry really is Illustrated what's possible with this infrastructure um so would love to have seen more um kind of what agents um or AI armed with the Technologies of crypto what that could have enabled um so thinking about things like autonomous worlds like what does it mean when AI agents like can actually transact um with all the different people who want to do so um use cases like that are I I think they're touched upon but um I think we can definitely go deeper and more creatively well let let's go deeper right now so what do you think is the most compelling agent use case where somebody even you know a age agnostic where you could really explain the value proposition to them yeah it's a good question um so I I had the same answer on a previous panel today but I think something that's underlooked is autonomous agents in the context of NPCs in the real world like I think we haven't fully discussed what it what it means to have like NPCs that are part of our reality that have a stable uh crypto identity that you know has different kind of rules that can interact in different different kind of environments um you can have One agent that lives on multiple crypto games right multiple onchain games and multiple autonomous worlds um and I think seeing that as it's not just something that's within this like virtual world but something that's like actually exists in ours um I think that's very interesting and and so just so people in the audience because agents is more of a newer term in the last year sometimes I hear people just say or like from my vent I think it's more just like a Dap like an onchain app that uses Ai and primarily llms today but can you just break it down from your perspective yeah that's that's a great question um I think that uh what I would call an agent is essentially it's any application or any software that uses intelligence as part of its its core um its core kind of functionality so you can think of AAP as really it's pretty deterministic right it you want to swap like one token for another token that's a Dap um if you have a agent that is it can technically be a Dap but what it could do is it could also have like a process or an intelligent kind of conversation with you so you can talk to it about like which tokens um you know you should swap uh and and I think like adding you know especially with generative AI uh large language models um I think that's really how you bridge the gap between like what a normal D apppp is and what like an agent D app is got it uh so your your original question was about what is not being discussed enough yeah what's what is being discussed too much and what is it being discussed not enough um so I I okay so I I I don't know if this will answer your question but I I think in general this like last year there was no AI at e Denver like no AI conversations uh and and this time we're having like I don't know 10 days full full days of AI conversation so I I think right now I think that uh this is a time of experimentation people are excited about this field for obvious reasons but I don't I don't know if if uh product Market fit has been found in the space and I think that's okay right now right now we're trying to experiment with the technology and and see what works so I think I think that's I think a lot of people probably already probably think that and I think it's but I think it's important to just for me it's important to recognize that this is a time of experimentation and we're trying to see what fits my hypothesis about where where it probably will fit and and why we would go through the expense of decentralizing AI is putting the decisions of the Agents of AI whether it's agents or or not back on chain so I I think right now uh I believe that crypto has found best product Market fit in defi I also think that there's going to be many other things like decentralized social media dpin there's going to be many other applications in the future but but right now defi has been the thing that has had the best product Market fit and having agents that control onchain funds in a decentralized way uh I think that's where we're going to uh that's where I think there's going to be a a short-term uh value and so just to even be more explicit it would be an agent that's directing yield farming without a human in the loop for example yes yeah one one example okay or crypto taxes that needs to be fixed um I will not comment on taxes because I'm done being an accountant um let's see hot take uh so I think too much here's what there's too much of there's too much agent Marketplace stuff I think that my the conversations I've been having and the analogy people are using is like a very web to analogy so the the picture they're trying to create in your mind is like app store for agents and if you've played with chaty BT um you might have played with their agent Marketplace and you can like create an agent and like tell it to talk to you in a certain kind of way and then give it a picture and that kind of thing but I'm not super interested in that I'm much more interested in the inverse so agents as market makers and marketplaces so kind of like filling in this and dis intermediating this large group of people that sit in the middle of an organization that just pay pass paper back and forth um order book things like this intelligent agents that literally go out and make markets and pair buyers and sellers rather than just sitting dumb inside of a like an app inside of a Marketplace so kind of connected yeah yeah just just to add on that I I completely agree I think if you have agents that sit as the new intermediaries you it's actually Superior to any of the existing systems right um we trust many agents in our day-to-day lives I'm sure a lot of you have booked airbnbs um to to stay for for this conference in Denver you're trusting Airbnb to kind of act as an agent in that regard um but what if we could have transparency in the algorithms and the systems of the agents that would matchmake us with you know where we want to stay during a conference I I think there's a lot of opportunity in kind of taking the decentralization technology that we've created and transforming it and moving it into um more of the traditional kind of Market play/ agents so yeah I think that's a great Point yeah I agree yeah and actually like even outside of crypto web 3 you know like open models and like distributed systems are effectively needed for like regulated Industries anyway because they just need to be able to like fully audit um what is going to be happening with you know these autonomous agents so like financial sector Health sector um you know effective Regulators need need to be able to see how this underlying technology was trained what the data was trained on and then be able to have the ability to view like every single transaction that happened in that you know chain um like for high frequency trading for like credit card fraud all these types of um Financial situations okay so Switching gears a little bit I think one of the maybe most unsettling dynamics of this industry is the speculation that's going on with crypto a i and the fact that there are these tokens that are occuring so much FTV and tvl and maybe they're not the most robust under the hood and so I just kind of want to hear from you guys as all of you are technologists how do you reckon this and how do you kind of think about operating in this space which is kind of plagued by this devil of um excessive greed my running joke for the entire week when I have conversations and people come up to me and ask me how's eth Denver been is say the the say Ai and money falls from the sky that's kind of how the conversations have been going um there's a bunch of these projects out there I won't name them that I think are really like they're they're getting big names kind of behind them and people are just kind of running and dumping how much we need to pay you to name them um when token uh yes let's you know probably the say uh a little bit more than the current valuation of uh oh I would I would have to name it no there there's a there's a decentralized AGI company that's out there raising at a very uh handsome valuation let's say uh with no product yet and a lot of reputation so there's a lot of these projects going on right and um there are some projects with just big names behind them and they're kind of just raising on the hype and there's kind of like a rush to like just dump some money into some of these kind of like more or less Ponzi schemes um with the centralized AI but when you look deeply at like what they're building and what's being used in the ecosystem it's all just a financial game and like you have to ask yourself like who's using the models right and again not naming any projects but there are some of these like very well capitalized High market cap projects in the space where nobody's using the models right and that is a signal of something so I think there's still a ways of a way to go for the crypto um part of crypto AI to be taken seriously I think in the proper AI Community for the last 10 years the AI Community has had a lot of reasons to actually hate us because we've been eating up all the gpus for the past 10 years to mine shitcoins right so I yes so you know there's still some uh trust building that we need to do and I think uh it's great that there are legitimate partners coming into the space like stability that have models that are widely used um that are are doing incredible things and like innovating at the edges and creating vibrant communities just like ours uh so that people can have better access to these literally magic powers right so I think uh we should just be patient and just look at where the proper AI people are building um and just augment what they're doing with you know kind of the funny token incentive games that we like to play that was a great answer does anyone want to add yeah I mean I think it's just um there's there's generally this uh uh it's technological re revolutions um there's generally a a phase of mania um I think it's relatively clear that when valuations are not in line with what is actually being delivered there is some degree of mania and we should probably recognize that um at the same time I think that the tools and the technology that have been created are so much more uh sophisticated that they've ever been so technologists can really like start using these and think about the first principles of what what can be done with them um so I I really think it's just like keeping the eye on the prize of what this technology can do um implementing systems that that leverage that uh and yeah like I think the financial incentives and financial engineering games are are really just like they're they're interrelated in some degree but they're not you know the industry and crypto AI is not reliant on on them playing out uh to successful yeah just to add on that absolutely right I think we I mean looking at this space it's like I think there's tokens are interesting and I think they're applicable for like adding incentive into an existing system or framework um and effectively I think a lot of web through and crypto is still based on like speculative you know Financial incentives but actually what's interesting about really what AI is doing now is these this technology is here right like we literally access it right now in this room and it has profound effect and profound value on our everyday lives and it's only going to get more enhanced so I think there's an an sort of shift where it's more of a utility value incentive that can be built into these models and we're looking at you know right now Foundation models still require huge amounts of data to be trained on you know like stab diff Fusion was trained on like 100,000 gigabytes of image data um our video model was trained on a petabyte of video data which is you know like a million gigabytes of data it's like a ton of data and I think right now Foundation model companies are going out and acquiring and Licensing data they're spending hundreds of millions of dollars you know acquiring high quality licensed data sets and that data gets stale that data contains biases in some cases it can lead to disc discriminatory experiences for the end user and actually like if if there's a way that you can incentivize collaborative participation systems you know through some incentive structure where people are getting so much value from these models they want to improve them you know they want to get more value from them um that doesn't have to be strictly token financially um incentivized kind of picking backing off of that I heard well one Dynamic that's happened over the last year is obviously now we have these big base models and then in this coming year there's going to be a lot of more specialized models that are more fine-tuned for specific use cases as we start to bifurcate the market in different ways and something that you mentioned on a different panel was National models I just want to hear a little bit more about that and what you're thinking from that regard yeah so we we basically so actually stability when it first was conceived it was it started as a dow Dows and effectively why we didn't proceed with that was because the technology and the token economics weren't really aligned so we effectively moved to a structure where we knew we wanted to deploy this technology and you know open source and make sure these models continue to stay open um accessible to all with the premise of effectively elevating Humanity um at the same pace of change um what we did is we effectively trained you know like core foundational models across every major modality you know like video audio all the ones I said um and that becomes some of like you know uh almost like a primal like artifact in this ecosystem so you know you can take them um and you can build on top of them and you can fine-tune them and you can effectively integrate them into your own applications um and and do what you like with them I think now like that's been that's gone very well you know like there's 140 like million downloads of stability Like official models just on hugging phas you know there's 300 million plus downloads of variant models so they take in those models and effectively fine-tune them and refine them but still if there's a system where you can create you know country specific and culturally culturally relevant um models across these modalities that kind of enhances it even further you know every single country every single nation state should effectively own the intelligence of their own country and the historical data of that country it's it's a it's a basic right so moving to a state where we have a framework to deploy you know country cultural models with across those different modalities that's sort of what we're looking at now H how scared is the group of AI regulation because I think it's also related to this I would just love to ask this panel yeah are are you guys coming at the crypto AI intersection with a fear based on what's happening in the US a regulation of AI exactly yeah uh I think I think in general I think uh people in crypto are wary of of Regulation and I'm in the camp that uh thinks that the big companies that are that are pushing uh regulation are trying to do regulatory capture of AI uh they they want to they have the resources to hire the lawyers and to to deal with the bureaucracy send them to DC what's excuse me and send them to DC to help with the policy exactly right they're setting the policy they're you know they'll jump through the Hoops they can jump through the hoops and it will become a benefit for them and a barrier to entry for well open source AI so I I personally hate uh the the regul you know the the the regulations the the discussions that are happening do you have a prediction of what happens from here four years 5 years down the road what does AI regulation look like and how does our community save the day yeah I I think that um I I think that there there are going to be regul ations I think probably what's going to happen is the models are going to the models will get more advanced and uh these models are basically what's happening in my opinion is these are the regulations are also going to fold AI into the military-industrial complex um and and probably for good for good reason if I were the government I would also want to be able to have first dibs on Cate of-the-art models um I think that what's going on uh uh what we're doing in crypto with you know crypto is actually becoming stronger and stronger at being able to like uh interface with The Regulators outside of the AI context uh I I think that that experience is going to be important for us that is going to be an important learning experience and that uh we may need to help open source AI or use you know use that experience to to make sure that we still have the freedom to to do stuff yeah um I think one way to go about this is like making sure that we have fully open auditable and permissionless models um like Regulators effectively I think they're more scared of like closed proprietary models than they are of effectively open models because you can't actually audit these models you can't see the weights and code um and you can't see the data they were trained on effectively know how they work so what is like the secret source so I think that's one step to you know get Regulators more happy about you know how at least the models that are deployed um can be used but yeah I mean it's I think the concern is like the technology is moving so fast and effectively you know it does have the ability to you know really have huge like economic impact and like security Society impacts um but actually like restricting it is not the best way to go either you know like if us restrict R it if UK restricts it if EU restricts it there's going to be other countries that effectively take it and it's going to be bad actors that go and you know Advance it utilize it for for bad purposes so it's almost you need the ability to have open models that allow you to also like prevent you know the negative aspects of how these models are utilized we put like huge amounts of security into our you know models to stop adult content and like se sound filtration and not safe for work and things like this um because we think it's right and we think it's the right thing to do as a as an organization yeah uh I I agree um I mean I think it's easy to forget that regulation is not a like we don't live in a monarchy right in this country um I think like there's a lot of uh there's a lot of push and pull with the constituents and the users of these Technologies um I think that crypto uh definitely created a framework of so many people kind of adopted it um that it's much more it's much more difficult to just arbitrarily kind of shut off access to it I think Uber um kind of went through the same pattern right like uber um as sensibly broke many regulations as they as they launched their service but um as they delivered a lot of value um politicians realize hey our constituents would hate us and we would just be this like you know we'd be this uh totally uh yeah we we would be basically on the hook for removing this really valuable service so I think with AI it's it's it's much of the same thing um I think work with Regulators kind of educate them on on really like what what the state-ofthe-art is um and then keep in mind that if this technology is really adding value to a lot of people um it's not really a you know binary black or white process um there's a lot of work that can be done yeah absolutely okay so I want to do one more question and then we'll open up to Q&A starting with Gabe I want to do a ask and an offer framework so one off offer to the community in terms of crypto aai that you can do like knowledge or pretty much anything and then one ask if it's like your recruiting you want somebody to come talk to you about something really general actually I'll start just to give an example uh and then we'll pass it on to you so my offer is that I've been doing a lot of uh backend work for providers on AOS so if anybody's getting started on AOS and needs help uh and probably your girl I think I'm their largest provider right now and then my ask is I'm always looking for more open- source devs to collaborate with uh that are ethically and values aligned and just looking to put cool software out in the world so if that's you can find me after and then gab do you want to continue hello um sure so my offer is um my gift of language if there are people in this audience that are more on the AI side that need to know a little bit more about the crypto side I'm I'm pretty fluent let's talk and I can help you kind of speak the language a little bit more to understand what elements of crypto actually make sense in Ai and on the flip side also if you're a little bit more in the crypto side and want to get more into kind of what's going on in machine learning and AI I can speak the language a little bit too and we can kind of like talk about what things might make sense in your crypto project something like that my ask is uh if any of you are starting AI companies in the next year two years tomorrow something like this um and you need services around labeling data and training models um my company is starting to focus on um giving Equity to those who actually put in a lot of the boots on the ground labor to get these things labeled so that they can then train the models afterwards and a lot of these countries and people who participate in this process for the past 10 years have um not received an equitable or commensurate amount um of value and meanwhile some of the largest companies in the space are raising you know billions if not trillions of dollars for who knows what right and I think where crypto really excels is in thinking about token and E and uh tokenization and equity for those who participate in building a network and building something valuable so that's something that I want to participate in for the next decade so come find me if you want to talk about that um so I'm going to combine the oper and ask uh so in the graph the graph what the graph does is is a decentralized network of indexers we Monitor 40 plus blockchains and we we organize all the data on 40 plus blockchains and in the protocol we already have a bunch of validators uh hundreds of validators in the protocol and it's basically uh they have they they run infrastructure and it's basically free uh a very small change for us to add uh gpus to these validators infrastructure so you know we already have a token we're not launching a token we already have a network we already have all these rules in place so we're just going to do a small tweak and and add uh open source uh models to the network and uh on top of that then we're going to launch this inference uh data service on the protocol and our value proposition is that we we already have you know data organized data from 40 plus chains and we believe that there's going to be value to having the inference and the data right next to each other in the future training with that data right next to each other my my ask is that we focus we excuse me we focus on being the uh middleware uh and the uh the compute provider um I would like to uh collaborate with uh applications that need or want decentralized uh compute especially our goal for everything we do is is ver is to be able to provide verifiability in the long run so we want to be able to to if you have like agent-based systems that you want to launch uh we would like to help enable that and uh in the future be able to provide verifiability of the inference and privacy of the infer of inference um yeah I'm going to do the same as Sam uh so I I've been building uh as well as working with a lot of teams on building AI agents that specifically leverage um crypto so if you've ever thought about building an agent that leverages onchain data um I'll give an example so an agent that's able to basically respond to the identity of someone who's using it um let's say hey do you have this nft um you can use my agent or hey if you've done this transaction this agent knows um if you've ever thought about building agents in crypto uh there's a lot of value that that you can provide both uh for the rest of the world as well as for many of the networks that are doing so um so I'm happy to help uh you know draft proposals or help design MVPs um I've done it you know many times at this point um so happy to do that and you can find me uh in my company at operator. um and yeah that's that's pretty much it and David's amazing to work with so I highly recommend coming to find him uh yeah my give uh is we gave like about 20 million 8100 um hours for Academia and research purposes last year and we expect to do the same this year uh that's about like plus 50 million in market rate compute um so yeah effectively if you have a really interesting idea that you want to develop um we'd love to we'd love to hear about it and like support you in that endeavor crypto taxes is still one that I want someone to fix um and then yeah my ask would be just please use the models and give us feedback I think I think we want further feedback specify on our code model which is like trained on 18 of the most popular code languages but if you're if you're not a developer and you want to be one that's a good place to start okay great so I think we have time for two questions if anybody wants to raise their hand yep you yeah absolutely just go through say your name your Twitter and your company sure be loud uh my name is Gabriel uh on Twitter it's player ZX and it's player zer0 the number ZX and uh my company is light works and it's spelled like light like the thing in the sky but the G is a q my name is Sam green I'm on Twitter at ZX Sam green and I'm at semiotic labs and we're a core Dev team with the graph our website is semiotic doai hey my my name is David um my my Twitter handle is prompt root Rotator So like um so like you you prompt chat GPT that prompt and then like Rotator as in like shape Rotator um um and and my company is operator. and Scott's from stability AI uh my Twitter handle is complicated it's SEO the O is a zero ttor Ty yeah not great my name is Casey Caruso my Twitter is my first name Casey my middle initial K and then my last name Caruso okay cool any other questions before we wrap up somebody's got to have one question okay amazing so um you'll talk about the uh the AI in a perspective Ive that AI is helping and synergies with crypto but I'm thinking like in a generative future where things like Sora that can create like digital realities how can blockchain can be a counter measure from a um like people who humans who create content for example how how they can use blockchains for example to timestamp content right like because in the future as this evolve we seen with Sora is that like anybody can create a digital reality but how humans can protect themselves against that reality using blockchains thank you yeah uh I it has to happen on the training data side so you know as a uh one of the images or you know video content that's effectively in that initial training data there needs to be some type of like attribution you know this is being used and then I think your point I think some type of like Watermark just so at least like you know computationally we know that this is AI generated and it can be like signal to the end user or you know there's some way to verify it I think that's fundamentally important I mean like dat we're going to have so much data from like all of the we're all creators now right we're all artists we're all videographers we can we're going to have more and more data that's going to be generated synthetic data everything so it's a good question I have a take on that um I don't know about how block blockchain is going to solve it but crypto graphy I think will have a a role in it there's a nice uh paper by Dan Bonet he's a cryptographer at Stanford and uh what they're doing is they're embedding uh basically cryptography at the sensor level so that everything that's captured is signed and I think that that's going to be important in the future is that every every media device that we have has cryptography in it so that everything can be can be signed cryptographically okay thank you so much [Applause] everyone