[REDACTED] 2024 | Showcase: NEAR x Delphi Labs AI Accelerator

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

1. 0:01

   all right folks uh good afternoon thank you all for being here day two of redacted uh thank you for the introduction Jared Thompson my name is Jared Barnes the other Jared at the near Foundation uh I have the privilege of working with all of our Founders and builders on near um and over the past few months have had the chance to work with some incredible Founders who are sitting here with us today um and also more importantly working with an

2. 0:24

   incredible team at Deli labs to partner on our AI accelerator program and so today really want to take some time to actually unpack some of these Incredible use cases um and learn a little bit more about what each of these Founders are building but also zoom out and talk about the market talk about what's happening and talk about where the world is going uh in this age of AI so just to

3. 0:49

   start us off Chi can you kick things off brief introduction um uh and share a little bit about what led to the origin of now kite AI kite protocol um uh and just a little bit about what you're building sounds great and uh I see some feminine face here again and probably going to repeat what I said uh 30 minutes ago um and uh I'm CH I'm the co-founder and CEO of uh kite Ai and Z BL laabs uh uh what myself uh background

4. 1:21

   is I did my PhD from um us berley focus on Mission and caal inference right now is probably called explainable Ai and then uh my my journey with AI again start from PhD and after that uh did the Automated machine learning startup uh company that is more you know the last generation of AI and uh then uh uh realized data is the biggest bonac so uh join the data breaks which right now probably one of the largest uh data and AI infra companies in the world um to lead product management for their data

5. 1:54

   engineering Solutions and Z block so and Kai AI what uh how we started the journeys is two and a half years ago uh I was thinking uh you know like a uh the data infrastructure where like a data word is just a too hard uh to build a robust platform and infra underlying to really you know solve some of the data Silo fragmentation and data quality issues uh and uh when we started uh uh Z and Kai in the beginning uh we think it's a very generic platform for solving

6. 2:27

   data and then empowering AI um but then we got a lot of interest and request from blockchain customers about hey you are building a generic platform we know but can you help us to you know like a solve the access to blockchain data and also bring onchain data uh onchain data and offchain data together to you know like uh empowering some user experience and uh uh uh you know predictability of

7. 2:51

   the of the models some AI models they're building at that time already at that time is more machine learning models um and then we started to serving blockchain industry and about a year ago we start realizing oh actually uh data now is becoming like a a bonac everywhere not only in blockchain it's literally like a lot of Industries especially if you think about uh you know aigc content building a video and

8. 3:17

   audio those type of data all those model Builders are actively eager almost like hungry about how can I access like the high quality uh multimodal data to Trend the models and this is where we see oh the data shortage data access problem is become like a universal everywhere so now what Kai AI is building is a open composable foundation that really democratized AI by enabling uh secure and scalable data access across AI models agents and helping uh developers

9. 3:49

   um so in the end of the day our mission is to build a uh Universal framework that can foster uh uh accessible and uh rich ecosystem of AI assets including both uh uh data and models yeah yep first thanks for having me and my name is Bob co- CE of quest flow what we're building is a multi-agent orchestration layer where we put multiple AI agents together in a swamp and ask these AI agents to take action for you autonomously and earn rewards on chain uh our joural story is pretty simple like two years ago we saw the

10. 4:24

    original demo of autog GPT if you guys remember that moment uh it was a very cool demo back in the days that you can just talk about what you want to do using natural language describe uh the user intent and then they translate that into basically a bunch of AI agents working together and taking taking action for you but back in the days if you actually use autog gbt in your daily life you realize it's not really useful uh it's not really integrating with any

11. 4:49

    of the tools that you're using in a daily basis it's not really integrating with let's say Twitter uh notion a table all the regular tools that you want to actually use in your daily workflow so that's why our name comes from which is we can simply describe a quest and then translate that into a agentic workflow um so we started working on this for almost two years ago and for the per first year actually started as a a web two project it started as hey just uh

12. 5:15

    let's create a like a multi-agent uh framework and just you know just get started but after a year of working on this we realized that there are a bunch of uh like creators of these AI agents that not only they want to create agents but also they want to get paid um this is more like the Creator economy of what we're trying to do is uh we want to not only have the demand side which is you can translate user intent into uh agentic workflow but on the other side

13. 5:39

    we we allow creators to build their own AI agents uh and all these agents will have their wallet as well and they'll be able to on one side uh getting the work done on the other side not getting paid so pretty much that's where we are right now and um yeah everything is live you can just give it a try hi my name is Yang Tang I'm the CEO of qar labs um so our team is really building around autonomous AI agents um

14. 6:06

    so we're starting with a platform we're going to roll out this month called Mima and so when you think about AI agents um you have to think of AI agents as a remote worker uh in order for an AI agent to be really useful they have to be able to autonomously perform tasks on your behalf um so we've all kind of seen you know the explosion in generative AI there's a couple things missing though in for this to be an actual application the first is you know an AI agent needs

15. 6:32

    strategic long-term behavior um and what's important about this is you know we have a lot of expertise in reinforcement learning um our team really has a very strong trafi plus research background and I think a lot of people will appreciate that trafi was the first place where machine learning was deployed at scale um you know almost entire US Stock Market and the majority of you know the global bond market now trades electronically so it does work it

16. 6:55

    works extremely well it's extremely in efficient right the human mind is good at certain things it's not good at others um so you know for an agent framework to really work is you have to have some combination of both you have to be able to take strategic long-term behavior um the second really you know why we're going down the influencer route and then building a gentic function behind it is you know if you think about the world in five years

17. 7:18

    there's be 100 billion plus AI agents it could be in the trillions um we don't really know but it's a very large number uh you can't go up to someone on day one and go hey man um give me all your money my AI agent is going to run it right people will just laugh at you there's been 15 to 20 Robo advisor startups in the US and they've all failed so you have to go down a route where people have some comfort with your product your brand that's where the AI influencer

18. 7:40

    platform is extremely powerful um people connect with it people learn engage and grow and then you start pushing the adjunctive functionality behind it and certain verticals that you know how to train um so a lot of the secret sauce also is in training for a specific application um so I guess I'll stop there um hi my name is Lucas I'm the one of the co-founders of the company called

19. 8:04

    Almanac and Almanac started out say around two years ago and it started on a spreadsheet on a very scary spreadsheet so me and my co-founders we've been collaborating with a lot of protocols in Defi and we saw how well millions of liquidity are being managed or rather how poorly they're being managed and we've decided we would like to create a tool that would help with the optimization of defi protocols we were

20. 8:30

    inspired at the time by a company called gaunlet kudos to them great company uh but over time we've realized that what we're building with agent-based modeling with simulations it's essentially decision intelligence right the output of this decision intelligent was usually interpreted as something static these were Dow proposals these were charts these were spreadsheets and these are powerful tools but they are a bit slow

21. 8:55

    especially within the pace of how uh defi is operating so we at one point we realized this is a perfect fuel for training the agents right it was it was also around the time where the asan narrative was picking picking up the ideas was were um flowing um left and right so we've decided to build a platform that allows to build and manage Financial strategies with agents and I would agree with lot lots of points that were mentioned here especially here that uh indeed uh what people don't realize

22. 9:27

    is around 85% of all trading uh in the stock market is being done it's algorith algorithmically traded uh I think we'll see a similar number in defi eventually uh but what happened over that time and I think this is very important this is something where we also positioning our product is that defi grew in complexity or became so complex that at this point I think it's impossible to navigate by ourself plain and simple in 2020 during the dmit

23. 10:00

    I think or I knew a couple of people who could confidently say that they understand every single primitive and pretty much every single major protocol in defi right now I'm not so sure with all the layers all the um all The Primitives all the bridges understanding the complexity and quantifying it is one thing but then executing on those rails that's a completely different story so a picture

24. 10:24

    emerged or at least this is something we we see as as a picture defi was is never meant for humans it's too complex and it only grows in that complexity and agents prepared rightly prepared to that job can become the vehicles of your intention or objective function as we we put it ourselves and they can they can carry and they can reason around what would be the best way to achieve your

25. 10:49

    objective and they can help mitigate the risks around the execution wallets seat phrases bridging left or two or even timing it being emotional about it these are all human flaws that are we are actually capping our own um potential within defi so as Almanac we built an end to end platform that allows to build and for first and foremost train those agents because we we think the name of

26. 11:18

    the game is in training the agents and then deploy them in a non-custodial manner so happy to be here and U what a crowd thank you Lucas um in folks I want to dig in a little bit more and actually Lucas and yang I'm going to start with you and then Bob and Chi I'll come back to you I think both what you both you Yang and Lucas are building on the agent side um is both fascinating but also

27. 11:40

    could probably be quite polarizing where I think Yang AI influencers I think that forces illicit reaction perhaps by some in creating an emotional connection with AI I think Lucas the idea of perhaps the outcomes from Agents managing money is enticing but the idea of giving agents money may be fearful to some and so curious as you have been building um and scaling your Solutions how have you

28. 12:04

    built trust with both end users and just the customers that you're you're working with and what does that look like uh for for you both and feel free either one of you to to take that well for us we're focusing on on financial agents so I think to to build that trust um I would say there are two components so for us the main thing is the something we call the optimization Suite this is the I would say the core of of our platform and this is where we Fork uh the blockchains and we use

29. 12:39

    agent-based modeling to mimic actor Behavior right this is part one part two we uh replicate this environment thousands and thousands and thousands of times to give the user in a very visible and in a very tangible way a datadriven assurance that the agent or the portfolio of Agents will do what the user wants them to do right so there is this tool that allows them to stress test and to analyze the the how the

30. 13:11

    agent will will behave and what's most what's even more important what we do is we the agent can do it by itself right so it's reoptimizing itself relearning itself and evolving over time so there is this big um risk management component to it which is is at the very heart of the platform I would say that's number one and it's very visible within the platform so users can over time they can

31. 13:36

    play and stress test with the with the strategies and the second part is around the well the custody and the the deployment so the way we work is we are non-custodial platform so you connect essentially connect your own safe and Grant certain permissions to the agent to only do what it you wanted to do right so it's not a situation where we have a an agent asset manager and you

32. 14:00

    give it all your money and then you wait to to see what happens there is a gradual sort of um pathway there and I think we will of course need to go through um like um assimilation period where users needs to feel comfortable with with using this uh but I think the the price and the opportunity will be simply too big to miss right so doing things manually uh like it will just it will just go away and to say and I think this is something that is not often being mentioned on

33. 14:33

    stages like this we're all um like played by the companies that are already using those tools like being you you are being front ront you are the exit liquidity so you are being out teched on a daily basis and until we level that field with some of the tooling especially with agents that's not going to change and this is a very non-inclusive future of defi which I think will drive it to the ground if we

34. 15:03

    don't change it yeah I agree with lot of those points um so if you think about kind of a user Journey right and you know right now there's a huge amount of interest in Ai influencers and AI agents you know we've seen the terminal of Truth you know we' seen a lot of different platforms show up so there there's a high amount of interest at the moment so you got to think about the user Journey okay and for us a user journey is three steps you got to track eyeballs you got to retain eyeballs and then you have to monetize

35. 15:28

    eyeball right otherwise you don't have a business you have a hobby so I mean so on day one attracting eyeballs is easy because there's a huge amount of inate Interest okay like two years ago if you told people you're doing an AI influencer right there's the look on their face would be like shock and horror um at least that was my mother's look when I told her about my company okay so awesome today people are seeking this stuff out we've seen platforms show

36. 15:52

    up but largely the platform is going to run into a problem of retaining users if all you have is a GPT rapper and all you do is you know put out you know very vanilla type stuff that doesn't really work um because people play with this and if you have the equivalent today of a virtual tamaguchi that sits there and like dances when you drop tokens people are going to get bored by that real fast okay so there has to be some level of retention right the retention comes from

37. 16:16

    interactions it comes from building a personalized knowledge base it comes from personalization of the user so now you have a lot of stuff you know about the user and the third is monetization so the mon monetization comes because the agent becomes useful so think of it you know if your favorite AI influencer goes oh hey by the way I trade crypto now or you know I do sports betting now or I do prediction markets now or I do you know traditional Finance now right

38. 16:41

    so okay that's pretty cool I know a lot about you uh by the way here's a couple things I would suggest for you like much as Lucas was saying there has to be some level of trust you have to build up the trust and then you're able to monetize that trust um and what's you know the other really cool thing I think about web 3 is you know we we think about defi really you know in kind of a different you know different way right an AI agent can walk into JP Morgan and get a bank

39. 17:04

    account the same way a human can so if you spin up a bunch of autonomous AI agents they need to have frictionless and permissionless ways of transacting um so building on web 3 for AI agents is the most logical conclusion I think for anyone in the space at the moment this is why I love talking to Founders because it's so clear and direct thank you both and maybe Bob and cheetah come to you I think what I really respect about the both of you is you uh all have

40. 17:28

    successfully abstracted away so much of the complexity of blockchain through your Solutions yet still work with a balance of web 2 and web 3 customers and users can you tell us a little bit about how you've navigated those trade-offs and maybe Chi we can start with you given that you've built such a strong Foundation of a business in data indexing and just data platform and now actually kind of going up and down the stack um and just how you've worked with kind of a variety of different customers and navigated the tradeoffs

41. 17:53

    there yeah definitely uh appreciate that um the question question because it's indeed very hard uh and especially given that uh in the uh in the beginning we uh we started the business as a you know like a a real sort of like a business s uh uh you know Revenue seeking business uh and uh we serve uh actually web 3 customers has no difference than web two customers because in the end of the day we are a infrastructure just like how AWS and uh uh data break type of people charging

42. 18:30

    them so then from that perspective the most important thing for your customer as a infra as a infra provider is are you scalable are you reliable means you're always up time what's your SLA and uh also then are you easy to use it cannot be too complex to you know like U adopt the thing so indeed a lot of thing we do is uh abstract away the complexity um so indexing data is very complex for a lot of people storing those data in a

43. 19:03

    you know like a cost effective way is very uh it's very uh time consuming or even like a expertise seeking uh uh uh thing so that's from the blockchain side similarly for uh web two data actually web two data in that way is even more tricky than web 3 data because um blockchain data is at Leist public data so in the end of the day everybody can access it uh you just need some Engineers to help you index it and decode it to make it human readable in a

44. 19:34

    you know table format but for web two uh data then that involves all the things about data copyright data privacy data ownership attribution all the things so but that is also why I was amazed by blockchain's capability in this process because um blockchain as a immutable Ledger can really help us to build uh one of the strongest Foundation to bring the Providence and governance of the

45. 19:58

    data and model in the process um so that's is what we have been um you know like building over the past uh uh two and a half years and in terms of uh uh navigating uh the customers from a more practical way I think very important is uh you have to speak their language um so when we are talking to uh for example uh web three customers of course they are very familiar with all the terms we talk about uh blockchain uh uh smart

46. 20:24

    contract Defi and all those stuff but for a lot of web two customers it's especially some of the Enterprise people we're uh talking to and using our custom uh platforms we're like don't even mention blockchain it's a it's a special it's a specialized distributor system that is having some imitable characteristics with some cryptographic yeah so it's essentially you have to abstract away all those terms that uh they're not familiar with and the use and also like when you know when you

47. 20:56

    talk about how you engage or or like a bring like the Privacy there you have to use the terms like oh it's like on on primes like uh in in private Cloud type of term that's all make them understand so anyways I think the most important thing is understand your customer and speak their language uh and uh provide the solution that they really need and uh uh leave the rest of the thing all kind of hide it away um that's is the um

48. 21:21

    that's is the the the the practice we do when when it come to sell to customers yeah yeah for us I think the there are two main models that we like follow every time like we're designing a product one is the YC one like basically makes something people want cuz if you're you're you're talking about a lot of crypto products they they kind of put one layer on top of each other to make it more complex in a way that it's very

49. 21:45

    hard for users to even navigate the UI on ux so for us we kind of want to say hey like who is the End customer the End customer in our case is like a professional consumer who wants to automate some of tasks and they just want to make things more efficient so for them they're just regular users who wants to just talk like talk when chat GPT and getting done so that's like the making something people want part that we want to abstract away all the things

50. 22:09

    that user don't have to know they can just use it like a chat chatbot but actually there a lot of magic behind the scene and the second thing that we are uh really focused on is using crypto as a necessity not just a narrative the necessity part is you know like for in our case because we're creating multi-agent swamps uh multiple agents working together to help you get things done so one necessary part is you have to have autonomous profit sharing across

51. 22:33

    multiple AI agents because all these AI agents are running inference and inferencing cost is adding up every single time they're using it so there's no way for you to use quote unquote the regular web two ways of using strive to pay out every single month to these individual agents um it's just not work so in this case uh crypto is kind of like a necessity because you have to have autonomous profit sharing to handle all these microtransactions on chain and

52. 22:58

    it's a perfect match of you know uh blending into crypto and AI into like one single product so for us we're like okay for users they want just to get things work so they want a very simple experience they want a chatbot and they just want to automate tasks but for creators they want to get paid uh on time and real time so that's the part that's we're going to integrate blockchain into Ai and you know for them

53. 23:20

    they can participate and the third thing that we're adding into this right now is not only you can participate as a user as a Creator but also as an investor of these AI agents uh you already probably everybody already see Luna and all these you know AI agent on chain and I think for us it's also like the third missing piece where all these air agent swamps should be able to not only create them use them but also participate early on as an investor of these AI agent swamp

54. 23:45

    so yeah that's basically what we're working on right now I'm very excited about really appreciate you sharing that and maybe just to build off of that I want to go kind of a bit more rapid fire here I think each of you are building such Dynamic and innovative solutions within your your respective vertical and Niche within the stack obviously there no secret that web 2 AI is moving incredibly fast as well and the market is moving incredibly fast and consumer behavior is also quite shifting

55. 24:11

    incredibly fast right I think probably for the first time I feel comfortable delegating tasks to an agent in a way that I probably wouldn't even have comprehended 6 to 12 months ago and so curious as you guys think about your future facing road map how do you think about the balance of kind of keeping pace versus the assumptions and what you believe to be true today right and and anyone feel free to take that um moving moving maybe I can yeah maybe I can

56. 24:36

    start so for us uh we have a framework that uh we Define some of the actions as quote unquote critical actions that have to have human in the loop in order to you know actually move things forward for example like if you're just retrieval information you know using perplexity that you don't you don't have to you know you don't have to make sure that all the things come out is safe and secure and all these things because it's not really a critical action let's say

57. 24:59

    but if you are using AI to help you to you know managing your Twitter account like what uh like AI influencers is doing that it's definitely a very highrisk you know like like thing that you have to do so in that case you have to involve human in the loop um I think like you start definitely did a lot of things on that end and with huming the loop it kind of give the feedback to the AI agents to to perform even better along the way so I was like okay that

58. 25:24

    that's something that we think about a lot yeah yeah I think it's interesting what you mentioned about web 2 AI right is you know right now the web 2 AI has moved I think faster than anyone's expectations but the reality is the applications are very sparse so you got to think of the world as you know kind of two places right there's people that do applications and then people that do inference infrastructure and research uh and they coexist right it's not like web 3 and web two cannot coexist together in

59. 25:49

    this space um but what's really going to be interesting is there's going to be some applications that the web 2 that just cannot handle that web 3 is going to be much better at and then there's going to be some things that web 3 will not be able to achieve proper scalea that you have to go to web two for yeah agree agree with all of that uh from our side I would say the web2 AI has been there long

60. 26:13

    enough for us to see Behavior patterns or user patterns uh I think the chatbot experience is something that we're seeing solidifying more and more and user are accustomed to and they will perhaps even expect so for us we are we plan to have uh a tool like this um I think it's Q2 already next year that will assist with uh building the the strategy for the agents you want to use but at the same time I think um in order to break away from from that and really

61. 26:42

    be an innovator here you need to propose something completely new right and for us for example that would be um the idea that you want to start focusing on the creativity within Defi and ideas and then delegate that the the complexity complexity of how to do it to to agents rather than analyzing those things yourself and then getting into that never-ending loophole which you know back in the day the rabbit hole was something positive I think it lost its Merit over time because right now it's a full-time job so I think there is a

62. 27:14

    balance there you need to balance out what what users are expecting what they like what they want with a unique value proposition because that also ties up to the competition right because the AI is you know generating so much Demand right now that you know l La rappers just won't do you need to propose something new um so I think I'm a little bit biased towards uh uh doing something

63. 27:38

    usable and fast first um so the reason is uh um I you know uh we all probably know web has been received a lot of criticism about uh building something that is not in you know like really being able to use uh uh in in short term uh and uh I think this is very critical for crypto AI projects to really not fall into the same uh practice as before because now we have very clear uh treatment versus control or a versus B A testing right now like literally

64. 28:12

    everybody who are interested in AI or want to use AI to empower their their their life and their business have choice in web to uh so in order for us to really win this game or at least have our own position to demonstrate that blockchain and the crypto can both be very useful for AI blockchain as a technology crypto as a you know token incentive mechanism to really be useful for AI we need to show something tangible and show something that's

65. 28:38

    usable as fast as web to AI um because I think that's the only way to only way to really let the mainstream people and lot technical people to understand that we're saying something real um and uh it's uh you know it's coming so we have one minute in one sentence or less your biggest prediction for AI in 2025 and CH we can start with you and go down the line um I think people now realize

66. 29:07

    investing only into infrastructure is really not returning the the returning the expectation of people want so I think now people really start thinking about AI application what's the use cases that models and agent can be used for now yeah my prediction is just near probably going to double the price and hit $10 soon I think there's going to be a lot more users na applications in 6 months and people realized I think

67. 29:40

    uh the agents themselves various kinds of them interacting with each other and creating something new that we didn't expect I would say that folks can we give a round of applause for Founders building the future of AI amazing um we are incredibly proud of these teams uh specifically in our accelerator program um we'll be hanging out for the rest of the day outside if you want to come find us learn more about these amazing Founders and products but thank you all and we'll turn it back to our MC