[REDACTED] 2024 | Decentralized AI Apps You Can Use Today

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

1. 0:00

   my name is still Jeff wilser host of the AI curious podcast and I am super excited to be joined with yet another All-Star lineup of very smart folks you discuss decentralized AI apps you can actually use today and potentially also what you can use tomorrow and how do we get from today to tomorrow I am joined by Ben Fielding by colantha May by edgers Nam and by Jessica Solomon hello all welcome let's have a let's start with some brief intros of what you guys

2. 0:32

   are focused on then we'll dive straight into the red meat Ben you go first sir hey everyone I'm Ben uh my background originally is in AI so kind of uh you see the reason why I care about this um but now I'm the co-founder of Jensen and we're building the network for machine intelligence um so we essentially build infrastructure for the core resources of machine learning uh to be done in a

3. 0:55

   fully decentralized way uh and entirely programmatically rather than requiring all of the kind of human world things that machine learning requires right now Claudia hey everyone my name is clenia May I'm the co-founder of Masa Masa is a decentralized network that provides data for AI developers what we provide is truly realtime structured and factorized data for anyone anywhere in the world to

4. 1:19

   build AI applications and AI agents fun fact I'll volunteer one uh my cat lafy is an AI agent well I do have a real cat but he is also an AI agent amazing Edgars hey I'm Edgars uh I'm the co-founder Chief product officer of Jer Ai and we are building gen layer uh gen layer is a new kind of blockchain where the smart contracts can connect directly to the internet to fetch any data and

5. 1:44

   execute natural language instructions using large language models we do that in a way that's unique uh by aggregating the intelligence of many different large language models through our consensus mechanism Jessica hey how's it going my name is Jessica Solomon and I work on a couple complimentary projects both 707 and Vana 707 is the one and only ecosystem project of Aura which puts AI

6. 2:07

   models on chain 707 is focusing on AI generated content models on chain and we have actually um named our protocol after the extension to erc721 nfts uh called ERC 707 which is essentially a way of showcasing an nft and correspondence to a piece of content that was generated with an AI model vaa is focused on user owned data and we love using data to train AI models that then generate content amazing so what

7. 2:41

   I'd love to do is have something of a threea play right here okay in act one I want to cover what we're supposed to cover apps we can use right now act two what we think might happen tomorrow what that could look like and maybe what are some like gaps between where we are today tomorrow act three how do we get there and so I'm not going to ask each of you one by one by one so jump in whenever let's talk about what you think are some of the most exciting really useful apps that

8. 3:13

   everyone here in this audience everyone in live stream they can actually go right now on their phone on their laptop and actually do things to show what's in the field right and you guys are all on the front lines you there's no one better than the folks in this room to kind of chime in this so I'm curious you know what what excites you about what currently exists today and we'll go from there something that I'm seeing um and that's kind of coming online basically

9. 3:37

   right now is using large language models to create kind of a universal Oracle um because so far oracles have been quite limited they only work with very narrow precise data feeds um and for some for something like a prediction Market where you have oneoff questions you need to have humans basically to resolve it which is slow and expensive and unreliable so what we're seeing now and what we're also working on at gen layer

10. 4:00

    is one of the use cases is universal article where you can fetch any data any from the internet and use large language models to extract an answer for example who won the election that's something that's already super reliable with the current level of Technology we also seen like chain link and Chaos Labs come out with their you know uh simpler Solutions then a gen layer will building this full platform fully fledged consensus mechanism that's going to be super

11. 4:24

    reliable for that but that's that's something I'm super excited about Claudia yeah I think I'll probably talk about one thing that people have been saying in AI for a very long time but for the first time mainstream audience finally understands what we all have been talking for many many months and years which is AI agent up top I mentioned that I do have a cat whose name is lafi and who is also an AI agent and I think many of you have seen were

12. 4:48

    interacted with Luna goat and many other AI agents on Twitter the that is I think the most um approachable or accessible AI application that we have seen in mainstream right now but imagine what can happen in the future currently both goat and Luna are trained on prescrip data sets where static data sets but in the future using real-time data provided by Masa um goat and Luna and many other AI agents out there including my cat

13. 5:19

    lafy will be able to learn in real time Commerce in real time and make all the AI agents more contextually aware and emotionally intelligent I know I jump the G and I'm talking about the future but I'm super excited I I definitely want to explore this emotional intelligence that you're describing there I need some for myself personally too so I'm very intrigued by this so Jessica for you yeah I think people should try generating an inference on

14. 5:42

    chain with one of aura's tools um so if you want you can use actually stable diffusion on chain right now the cool thing I find about generating a piece of content or an image with stable diffusion on chain is that unlike an nft which is often time a JPEG or some painting that I made somewhere else and then I decided to Mint it on chain with wearable or openc this inference is

15. 6:05

    actually occurring on the blockchain which I think is a somewhat game-changing feature of both Ai and blockchain in conduit with one another Ben any current use cases you want to highlight yeah I'd probably just um I'd double click on the agents use case I think that's the most exciting thing right now um I would say i' I'd maybe caution that I do think we're in like POC territory with it like it's in right

16. 6:30

    now these agents uh you're able to converse with them and they can they can say interesting things but behind the scenes a lot of what's going on is still very centralized um which is fine right like it it doesn't make a huge difference with our interactions it demonstrates what's possible that's really exciting um but I would say um what will become more exciting in kind of Jensen's opinion is when those agents can actually do real tasks on our behalf they're not just like uh conversing with

17. 6:54

    us anymore they're going out and kind of executing something and I think for us to get there we need um to basically bring a load of the infrastructure that they rely on to an onchain environment doesn't necessarily need to be onchain but it needs to be kind of ajacent to the chain because those agents themselves aren't truly autonomous yet but they can be um so yeah again jumping to the Future a little bit but I think when we can do real autonomy which is

18. 7:18

    one of the most interesting things about crypto it will get super exciting U not to say it isn't exciting right now is it fair to say that we're still in verly even these use cases you guys describe which are Super exciting I'm guessing are all still pretty early adopter right there's there's not a lot I keep making discussing grandma today but lot of Grandma's out there doing decentralized AI stuff right and even like the cat and

19. 7:43

    your cat sounds amazing both the real and fake one sorry not fake uh apologies to our a overlord human challenged right um interesting exciting stuff but it's not mainstream decentralized a agents are not yet mainstream is that is that fair to say that it's not it's not yet spilled over into normies just yet is that fair yeah I think so but in many ways I think it's quite exciting to start with what the normies are most excited about sure um my Boomer parents

20. 8:16

    love generating content with mid journey and all um while mid journey is not open source I'd love to see a way in which egotistical Boomers can generate content and actually prove it's theirs on the blockchain and that's kind of Ben's point that the narrative matters right and even if it's if if we waited and we we've all learned this in kind of the overall web three crypto space if we waited until everything was perfectly

21. 8:40

    decentralized before doing anything before shipping a single thing the industry would be dead right you have to start with something and then we kind of progressively centralize um with that said let's let's pivot to the Future um what excites you now about actually let's start with you uh on your emotional intelligence emot intelligence cat agent if I'm getting that correct talk about what you see down the road of how these decentralized AI agents can work and give use cases that might kind

22. 9:10

    of blow Norm's Minds yeah first of all I actually want to piggy back to the prior question about um is decentralized AI mainstream yet I think the answer is no but when mainstream is interacting with decentralized AI in the future I don't think they need to know that it's decentralized great Point that's really the magic when it comes to Consumer adoption you don't have to understand the technology you just need to find

23. 9:33

    utility in what you're using take the cat Asian as an example right it will make I'll make a statement here it will make all of your prior meme coins look very dumb because they're not emotional intelligent they're not contextually aware they cannot think in real time so the first iteration of the future will be truly emotionally intelligent and contextually aware AI agents then to

24. 9:56

    your point I think that AI agents should be able to not only converse with you to make you happy and entertained but also being able to complete tasks through agent to agent coordination agent to agent payments and in the future agent to non-agent entity so the end game of this is that any single one of you sitting in the audience on on stage here you'll be a able to have an army of AI agents that's completing task for you

25. 10:21

    that is basically creating the mini clenia but clenia 1 2 3 4 5 ABC out there creating value on your behalf that that's I think the pretty more remote future we're looking at can I double click a little more on the emotionally intelligent point I'm I'm really curious of what this looks like can you give an example concretely of um so what does that mean as far as emotional intell they're able to sense what your emotion

26. 10:45

    is and then give an appropriate response that kind of respects that or or kind of flush it out that what that looks like yeah so if you use our agent Network atasa you'll be able to First at the prompt and then Define the personality and the traits of the AI agent right for example my cat's personality or cat personality there is really a sassy roaster who is just making very snarky comments out there judging you based on your Twitter history but being an

27. 11:14

    emotionally intelligent cat it means that the cat will be able to learn in real time and formulate Chain of Thought and be able to refine the way that it reacts to your behaviors or your conversation we living in Wild I think most people have no idea where we're headed this is wild and amazing stuff Edgars what excites you about what these could look like in the future and give

28. 11:37

    us some examples like that uh for us I think the most interesting well for me personally even the most interesting thing is using AI for um dispute resolution kind of like extending this idea of you know a prediction Market is already done but extending it to kind of way more powerful smart contracts because the way we see it right now is well for some things you can use a Smart contract if you want to have a bet about the price of ethereum Tomorrow incredible you can have a smart contract

29. 12:02

    it's self-resolving and since then it's super cheap but if I want to hire somebody to write me a blog post right that's not something I can actually achieve with a smart contract because the smart contract cannot see the article and more fundamentally it cannot evaluate what is a good article for that you need natural language you need um you know flexible kind of contextual Concepts that uh formal languages like solidity cannot deal with that kind of stuff uh so right now you basically need

30. 12:26

    to have a paper contract and go to a judge uh it's going to take you a year and 10 grand to get that resolved uh so what we see is that you can actually use Ai and by combining the intelligence of many different large language models to make it super reliable you can use it to have these contracts where where the AI basically acts as a decentralized court system a decentralized panel of Judges that can resolve really high stakes contracts such as a truly autonomous

31. 12:51

    organization or per kind of performance-based Contracting hiring people to do stuff um anything like that that's super interesting to me kind of taking this idea of smart contracts and making them intelligent extending this concept to really take a big bite of the legal system so we have ai judges AI armies AI cats Jessica what what what excites you about either what's coming that's being built right now that will

32. 13:13

    come out soonish or long-term is or maybe what you want to see dealers totally well I'm excited about more people actually building AI models and them having to me the means to do so and I think that's something that blockchain has done a really brilliant job of is helping curate ecosystems and help sort of curate essentially funding for peoples that are members of those ecosystems to really incentivize developers in align interests in a way to which we can have more Niche versions

33. 13:43

    of AI that might not come otherwise without decentralized technology Ben how about you f future use cases you want to see um I think I'd probably agree with Edgar on the essentially trusted decision making um I think we we've seen that decision- making in the human world typically involves outsourced trust somewhere to some organization or or some like legal entity or something like

34. 14:07

    that and that's worked for a long time in a certain way but it has has downsides and then we've got decentralization and like decentralized trust mechanisms which are really interesting for us to co-organize around things but when it comes to some kind of complex decision we get stuck again we either go to some kind of voting system which typically fall apart for various reasons or we Outsource that trust again to one of those older legal institutions whereas if we can have trusted execution

35. 14:31

    of some kind of agent on chain you can Outsource that decision- making a group could say we are all opting in to Outsource this decision making to this entity um and we can see that execute entirely on chain for that to happen you're going to have to have uh verification and kind of uh verified execution of the agent itself all of the components of the agent are need going to need to be transparent and verified but that's all possible um and so that's one of the big things we're working on

36. 14:55

    is bring all those pieces of infrastructure to the chain so that you can Outsource that trust um so I think when you do that it changes what we can do collectively uh in a big way um and to to what Jessica said it allows people to build more and more interesting AI models and AI agents uh one of the things I would caution though is we talk a lot about um autonomous agents in this

37. 15:17

    like anthropomorphized way where we think they'll look like people and they'll talk like people I don't think that's the case I think it's much more interesting if we let them evolve as this kind of entity with a completely different set of abilities it doesn't need to talk to people yes it probably will people are a system that it needs to interact with uh but then we have those interagent Communications that doesn't necessarily need to be even be human readable it can be super efficient in its own way um same with how it interacts with the outside world it

38. 15:42

    doesn't need to go via the human rails those are likely incredibly inefficient for it we need to think of the Rails that make sense for the agents themselves uh and the tech itself and I think that's when it gets really interesting but you end up super abstract so it's a balance between anthropomorphizing these systems and allowing them to be as powerful as they possibly can be I love that trait that that tension you outlined that's that's a great Point um as promised act three

39. 16:06

    how do we get there how do we get to that future um and I'm curious mostly what do you think the main pain points or friction exists to get there is it is it is it is it the real challenge overcoming Tech hurdles and we we need certain we need more chips need more capital or or is it about a cultural awareness is it is it about uh marketing is that my personal thought by the way in a branding issue I think we got to Rebrand even the word AI agents it's

40. 16:36

    like kind of awkward and hard to say it reminds me like on a 30 Rock there was a bit of like a movie called the rural juror and no one could say the rural juror I think AI agent is a clunky word we got to do better than that but anyway my personal thought but seriously what do you guys think it will take to get there and what are the main pain points we have to resolve Edgar kick us off I think big one is Tech of course there's centralization at every layer of the

41. 17:02

    stack from how you train the models to how you do the inference to how you interact with them and I would add a final one which is the that a single model itself is centralized it's a subjective entity it's a biest entity is statistical entity you know you you talk to chat gbt 90% of the time it gives you the correct answer 10% of the time it hallucinates something right so it can be wrong uh so you we think at least you also need to decentralize uh you know

42. 17:28

    between multiple mod models have them work together so I think it's on the tech side of course tons of problems to solve but I also think it's on the imagination side you know people are just starting to understand even how to talk to a simple large language model and then you know to get to a point what can you actually do with them how can you redesign the entire Society how we

43. 17:45

    work with you know how we increase trust with each other how we have a much more uh effective efficient system for working with each other how does that look like in an AI age so there's also a huge task into to imagine this future and then you know go and build it CIA I think Tech obviously we have we're at Devcon so every most of the people here are developers we're working hard on solving the tag problem I do think

44. 18:09

    adoption is the number one issue faced by decentralized AI applications now a lot of people when talking about things such as mcoin would literally just shrug it up thinking that it is so dumb it's a pump and dump scheme however it did bring a lot of much needed awareness to a lot of great decentralized applications we're all building in this room so I'll say adoption adoption adoption I don't mind mem coin I don't

45. 18:33

    mind other ways to get people's attention but in order to go mainstream with decentralized AI we need user adoption Ben I appreciate you've had a a very cold sober look at a lot of this B I'm very curious your thoughts on what's it really like no BS what's it going to take to make us a reality it's it's a good question I think we're in that classic tension that this space has been through many times of infrastructure

46. 18:57

    versus application where you need the infrastructure to enable the applications but you need the applications to prove that the infrastructure actually makes sense to build and it at a certain point both actually make sense and you can build real applications with real decentralized infrastructure I think we're potentially on the cusp of that potentially um but at the same time you're not sure what what still needs to be built it's this kind of this

47. 19:20

    trade-off that you're always in um I think what the space needs is both of those things it needs people focused with a long-term Vision on how these things should actually execute under the hood what resources they'll require and how do we enable those in a way that is most efficient theoretically right now but at the same time we need a bunch of people just kind of sprinting ahead and saying look this is how I think it's going to work I'm going to hack all the stuff underneath it together it's not going to execute exactly how it will in

48. 19:46

    the future but that doesn't matter I'm kind of like proving out the use case specifically and showing the world that something is possible I think the time we're at right now I would upweight that application side um a Jensen we've been building for 5 years in the space we've been building core infrastructure uh we've been building through a time when nobody was thinking about AI agents or

49. 20:03

    AI applications in a decentralized way and people are now and I think that is generally very very good for the space um so I would say to people who are thinking about this don't worry too much about the kind of decentralization in the specific infrastructure underneath your app because that is being built there's people focusing on it just prove that there's a real kind of value of doing this this thing that you've got in your head prove there's actually a good reason to do it um because otherwise the space will just get stuck in and a

50. 20:28

    similar thing that the the crypto and decentralized space has got stuck in before where people build things and there's incentives and there's there's loads of reasons why there's money flowing into this thing but people have built something that actually doesn't have any use case in the world so I would focus on building the use case show what's possible the decentralization will come the power beneath the infrastructure will come um and you'll be able to adopt it when it's ready but just kind of demonstrate

51. 20:51

    what's possible for now sneaky spicy take I love that Jessica how about you what do you think it will take yeah so my sneaky spicy take is tokens and airdrop Farmers I have noticed that people are willing to use a product that might be somewhat still experimental if they anticipate that their usage will then relate to them potentially receiving an aspect of future value I think that's a cool way to help create community and incentives for a number of somewhat um more experimental products

52. 21:22

    and help them mature awesome so final question for you all let's do lightning around Style once again I'm curious about predictions I want to it could be a year from now 5 years from now hell if one of you guys wants to go 20 years from now go there Ben give us a a little glimpse of the future of some actual apps use cases decentralized AI people might use whether 2026 or 2046 okay I would say

53. 21:54

    by 2027 I think the vast majority of the the triggering of the execution of machine learning uh training and inference will actually be done by machines themselves it won't be done by people so the entire kind of industry of machine learning compute which is what I focus on a lot but obviously bringing in data and and all of the other pieces will actually be uh interacted with by

54. 22:18

    agents I think those agents like I said before will be less anthropomorphized and more like just systems uh but they'll be the ones interacting with this infrastructure that we're all building it won't be people clonia I'll make extremely short-term predictions so I can be proven wrong or right so I predict in the next 24 hours Bitcoin will search past $80,000 no Financial investment advice ADV right no Financial

55. 22:43

    advice please um but I also predict that in the next 6 months uh the newly established um Twitter influencers or kols will be more than 50% AI agents driven instead of real human driven amazing Edgars I'll extend on both of these guys then say that by let's say 2030 the vast majority of the entire economic activity in the world is going to be conducted by AI agents um you know

56. 23:08

    working with each other on behalf of humans hopefully but uh we'll see but at least the agents for sure working with each other together and Jessica yeah so my prediction is that AI generated content will become ubiquitous not just in the form of art but also in the form of pharmaceuticals and I'm excited for blockchain to help really show attribution towards who was creating those AI models and what version of that model created those assets and helping also those assets become somewhat more

57. 23:36

    divisible my final prediction next conference my AI agents will be moderator for all of you guys AI agents and the five agents will have a great conversation and it'll be a great time thanks so much guys really enjoyed it thank you all appreciate it thank you