Layer 1 Stage | AI and Crypto: The What, Why, and How

Captions as YouTube publishes them, usually machine-generated, so product and people names can be misheard.

1. 0:04

   hello everyone can everyone hear me great so today I'm going to be talking about AI crypto as as was said and um I'm going to do a little bit of a you know 30,000 feet out view of why is this space interesting and um kind of take it from first principles you know what are the things to be looking at in this in this um category so what's the motivation what we have today is is this

2. 0:35

   you know massive usership in AI trap GPT has hundreds of millions of users using it every week uh products are coming out left and right and tons of money is entering the space people can't seem to get their hands on hardware and there we have a great backdrop for a very very exciting new technology and new product products but what are the problems there's a lot of problems evident today

3. 1:06

   with AI the first is that the infrastructure is highly centralized model storage model ownership and governance is all stored on Central servers and uh places that can be you know taken down computation on models is also centralized they're behind big public clouds and and big GPU services access to training data and most models are also censored you if you try any of the even the open source models they're highly sensored and it's become this cat and mouse game to try to get you know deep interesting

4. 1:43

   information out of these these models this infrastructure is also highly permissioned there's high cost to using and improving AI it takes advantage or it's advantaged by centralized players it's uh these things are are very expensive to train and operate they also have centralized permission apis to access anything so again it's another Gateway for you accessing these AI models then finally these systems don't

5. 2:14

   provide you any privacy any computational integrity and any censorship resistance so this creates a good backdrop for where I think crypto can have an interesting place for AI then finally governments incumbents large corporations are attempting to now regulate the space to try to cut out any new players to from entering the space and um I I think it kind of goes from a philosophical point to show you know they're kind of trying to attempt to say you know we know what's in your best interest but I think people know what's

6. 2:49

   in their best interest the future impact here is really large with enough data you can replicate any anyone virtually or physically you can have a politician say what they want to say have your favorite actor or movie star write what they want to write and it's believable by a lot of people so this will leads to a lot of scams increasingly as we become more digital and online it's another vector and entry point for these models to

7. 3:19

   become highly you know susceptible in the um in the world we have there's a really great potential uh to create highly engaged in new forms of media and propaganda and chat the 2024 us selection ads are already using AI so we're already moving in this direction of AI starting to control these these forms of media which is really really big and so I think an important thing here is who's controlling the models who's controlling what goes into them who's controlling the governance aspect

8. 3:53

   and who controls who builds the products on top um these are a couple slides or a couple images from a talk I really like called the AI dilemma they basically talk about social media was this great you know web 2 Innovation allows anyone to publish create all these great things you can connect to your friends you can talk to people you can get jobs but it also created all these bad things you know everything from information overload to you know these these highly addictive online

9. 4:25

   platforms today AI has a lot of great promises but it also has a lot of potential issues everything from creating highly engaging fake robotic uh characters and chats that that engage people to building new types of scams and so with all of this it's kind of a scary picture I think crypto is the thing that will help solve a lot of these issues and and really create the AI future that

10. 4:56

    we want AI is this highly centralizing technology it takes advantage of large swats of data collected into these big models and they're they're improved from that standpoint crypto is this highly decentralizing technology it brings power back to the people and it allows people to have sovereignty over their information their data and how they transact and so the obvious thing is to combine these two things combine this power of

11. 5:26

    intelligence with this trustless you get from crypto crypto AI at least the aim is to go from won't be evil to can't be evil and not allowing big corporations or governments or really anyone tell you what to do and and how to transact if we've already decentralized money and music and property records and compute why not AI so this is how I think about the AI stack and this is something that um we've kind of stolen from Ilia and credit to him in this this mental framing at the bottom right we have

12. 6:08

    compute so everything from verifiable inference to training privacy of user data and censorship resistance all through cryptography and Game Theory on top of that is the models the things that are actually being used and interacted with by the end applications on top of this is the data what are the things that are actually feeding the models and and trying to make these things open and then finally are the applications on top what are the

13. 6:40

    things people are building on top of these models alongside the left here is is governance which is all of these pieces of the stack have an element of governance and crypto governance is a really great way to manage you know how should a a certain AI model you know be steered how should interact with people and and things of that nature so a big question is uh what are the use

14. 7:10

    cases and I think a really good interesting mental framing is to say AI is really good at finding information or creating information creating properties of programs that were not explicitly programmed if you take a swath of data and it gathers you some insights on it how are ways we can apply that in the crypto world that would be interesting one is very simple so it's validity safety privacy being able to

15. 7:41

    interact with an llm or diffusion model in a way that's private or is um you know keeping your data Sovereign uh is already a great start and as we go down there are examples in every subcategory within defi you can build automated LP strategies that are run by you know by an AI model you can build new ways to parameterize risk and to adjust say the interest rate parameters in a in a lending protocol in nfts uh it's things like appraisal

16. 8:13

    improving and kind of enhancing the interactivity of the nft in gaming it's building autonomous worlds it's building personalized experiences allowing games to take your information and improve how the game is actually done within social building trustless recommendation systems ways to recontextualize information imagine if your favorite author or your favorite writer you know you don't like the writing style but you like their content you can take their content and have it

17. 8:46

    Rec contextualize to you in a way that might be better for you maybe it's simpler language maybe it's in the form of Heming way or or really anything like that you know the world is your oyster and finally governance being able to have Dows build automated governance proposals which today is a highly human you know human Focus human intensive task and very laborious and as the Dow

18. 9:11

    space grows it's going to be difficult to keep up with the number of humans that are needed to staff Dows we should let AI help what we're doing with governance and this is just the beginning there are many many many more applications I think my slide was cut off but um one one other example it's kind of an interesting sci-fi example is imagine you can take a human language intent you driven by an llm and and then turn it into you know self assembling smart contracts so an example would be I

19. 9:45

    could type into a chat bot and say you know I want you to build me synthetic 2x levered exposure to Apple with a with near tokens as collateral normally that would take five or 10 steps the synthetic exposure might be in a different protocol the Apple might be in a different protocol near is on a on a different you know different protocol and so what if you could have an llm that generates these inputs feeds them through one of these new intent protocols and then self assembles the

20. 10:15

    smart contracts and generates you that payoff all autonomously the future of Finance you know really can be driven by by AI so for those in the audience that are interested in building in this space outside of the use cases that I mentioned there's also a lot of really interesting research directions to to think about one is how do we do proofs of validity how do we prove that the

21. 10:40

    compute done for a model was done correctly there are there's a a sub sector of the space called zkl that's quite popular and there's also a lot of interesting teams building in the optimistic proof side of this space decentralized training how can we build these these trustless training mechanisms and create you know the foundation models of tomorrow and build transparent Foundation models AI model governance how do we govern how an AI interacts with a human and how it should

22. 11:12

    behave you know for a specific human and their preferences and then finally privacy in AI ZK proofs only go so far you know it can really only prove information you know prove uh things about local information um so we'd have to look to the the next generation of Technologies fhe trust's execution environments and maybe MPC so I didn't talk much about what what we're doing uh ritual is the

23. 11:43

    company that I'm working on and the first product we came out with yesterday is called infernet uh it sort of comes from from the term inference so infernet is an SDK that allows any smart contract developer to use an AI model in their smart smart contract there's a bit of an outline here showing how that actually works but the end goal here is to make it extremely simple for people to use or build any of those applications I talked about just a few slides

24. 12:17

    ago and that's it for more information it's my email uh my Twitter and then the Twitter handle of our company thank you all