NEAR AI Ecosystem - What Did You Ship This Week? #18

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

1. 0:01

   I near AI. What did you get this week? It seems as if Cameron's microphone froze. So, James, welcome to space. Why don't you just uh hit it off and then um we'll bring Cameron right back when he's there. It's all yours. And if you want to present your screen, um, please go ahead and do so.

2. 0:59

   Oh, I'm not going to be showing any slides. I'm just going to be giving a quick overview of Unity Wallet and really what our mission is. Sounds good. Is all yours? Yeah, great. Thank you. So, I'm James Toddano. Uh, I'm COO of Unity Wallet.

3. 1:14

   We're a self-custodial crypto and web 3 wallet. Um, and really our mission is to try to make uh, you know, DeFi accessible to as many people as as possible. Um, the user interface is designed to help make everything, you know, pretty easy and intuitive. Um, and I would also say that in terms of swaps and currencies, we we kind of have, you know, hundreds um, hundreds of tokens, but what we do a little differently within the swaps is is people can select

4. 1:49

   from a whole bunch of partners, including including Near Intense. So users, depending on what they're doing, find, you know, a pretty a pretty good deal in terms of uh, in terms of swaps. And we also have uh, USDT0 bridge in there. Now, um, but in terms of features, we feel that our wallet has, you know, way more features than any of the other wallets out there. You can you can run KYT checks, which allow you to

5. 2:15

   determine if someone you're about to trade with has, you know, has crypto that's been tainted by, you know, deep web activity or whatever. So, we have, uh, blockchain intelligence to help people make better decisions. It's kind of a security layer in a way. We have a whole security center. People can also shop um buy a whole bunch of stuff with their with their Bitcoin. Um so yeah,

6. 2:38

   it's it's really but Dex Dex trading is probably the biggest the biggest part of of what we do. Um and yeah, our mission really is to make navigating and and you know, transacting in crypto as easy as possible through through our user interface. Um, and also to really get users to think about self-custodial and and what it means because a lot of people they don't know that their bank account is custodial. They don't know what that is. So part of what we have to do is educate people um because we don't

7. 3:11

   want them you know we want them to understand what ownership is and the responsibility of ownership but also uh enabling them to be the bank and the customer at the same time. So that's that's really it. Yeah. Shy think it seems as if connection is out again. Yeah, James.

8. 3:52

   Um, okay. I can just go on and jam a bit. So, um maybe can you go in a bit of details like how Unity wallet is um like a key differentiators and how you're using how how you're working with NIA. Yeah, absolutely. So um so really in the last release version 8.7 um so we have we have near near staking in but in the last release we added um near intense in in its beta. So that's kind of one of

9. 4:24

   the that's one of the options um in terms of exchange partners when you're when you're doing swaps. Um so that's there and then also we integrated um wallet selector or you know into uh into wallet connect. So our ecosystem now has access to near near daps as well um which is great. Um and then obviously there's other stuff we're working on on too but I'm I'm you know uh that's coming down the down the pike. Yeah.

10. 5:16

    All right. We're just having some connectivity issues. Uh James, I'm just going to bring you on again. Okay. All right. Super excited. Do you hear me? Yeah, I can hear you. Yeah. How's it been um how's it been integrating near intents? Have you seen any volume been picking up? Um yeah, that's a good question. So, I actually do not know. I'm just going to be completely honest. I haven't uh I haven't looked into it. So, um it's really being handled by the development team, but I should have some insight on

11. 5:43

    that and it's something I'm going to look into. Yeah. Interesting. And any like um call to actions for the audience? Any Sorry, what? Any call to actions for the audience? Anything you want us to do? Should we go test out Unity wallet? Anything specific in specific you want people to play around?

12. 6:01

    Sorry. So I think um yes so right now if you download Unity wallet um and you want to test all the near integrations then that would that would be great and we have you know we are in the process of integrating something else now that I don't think I can I'm not sure I can talk about but that's really exciting too. Um so that's that's it. But yeah really go and download the app. Um, I

13. 6:27

    think one thing that's important is when you're a wallet app and you're decentralized as we are and you're having to use, you know, third party services, it's really um it's really important that we can integrate them into our into our UI. Um, and so that's that's kind of a challenge with a with a self-custodial a self-custodial wallet. But um yeah, it's slowly it's slowly getting uh things are getting smoother in the space which is which is great.

14. 6:58

    Sounds good. Well um keep up the good work, James, and keep building. Um excited to keep following your progress and um thanks for coming on. Thank you. All right, Cameron, let's see. Does it work? It you tell me, man. Does it work?

15. 7:14

    Yeah, it does. It works. I'm going to move to the background again. Awesome. Thank you so much for for bearing with me everybody. Uh this is the 18th episode of what did you ship this week. Shout out to Near Week for pulling through when the internet was being shoddy. Um my name is Cameron Dennis. Uh we do this every week. It's essentially a decentralized engineering standup where people in the ecosystem,

16. 7:33

    different bunch of founders uh show off what they've been building. And so shout out to Unity Wallet. Been working with them for the last couple of months to um integrate near get added to the wallet selector and uh now have near antenna supported inside the app. And so really looking forward to building more agents that can trade on behalf of users and making them available through UNV wallet. Um next up we have uh Open Gradient. Uh let's see. Matt, are you

17. 7:58

    here? Matthew. All right, great. Coming to stage. How you doing, Matthew? Okay. Can you guys see me? Hopefully it's uh it's working well. But uh yeah, I'm doing great. How are you? Thanks for having me on. Living the dream uh specifically with shotty shoddy internet. So um you know you've been building open gradient for for quite a while. We've been you know working together for about a year or so.

18. 8:21

    Um but you've been taking a pretty you know specific focus into portable memory which I think is sort of a prerequisite for AI mass adoption. Uh, this is for those who might not know. Actually, Matthew, I kind of want you maybe to explain it. But the reason why I'm so interested is because AI doesn't always have the large enough context window for you to actually teach it what you know your preferences and everything in order for it to to act better on your behalf.

19. 8:47

    So, um, if you don't mind just giving a quick lowdown about what Open Gradient uh has been building, what you've been focused on regarding portable memory, and specifically what you shipped last week, that would be great. Yeah, exciting. So I guess to give everyone a little bit of context on what open gradient is. So we are an AI research lab focusing on applied AI ML as well as blockchain computing. Today we've built infrastructure with it's an L1 network that allows developers to you

20. 9:15

    know run AI workflows you know from things like inference data prep-processing in a secure and verifiable manner. We have things such as TEES. We leverage uh confidential computing ZKML for you know verifiable ML inference as well. And after we built this network, we started building our own killer use case so to speak which is now that we have this network for verifiable confidential computing, what are the key unlocks we can make in the

21. 9:40

    AI space that um is uniquely enabled by our infra. And one of the things we landed on was um as you had alluded to portable memory and the idea is actually very straightforward today. every time you use an AI app, you know, whether it's chatbt, whether it's, you know, meta using their llama models or, you know, whether you talk to, you know, uh, AI companions on character AI or AI therapists, your memory is kind of siloed on these individual platforms. Chat GPT has no

22. 10:10

    idea about your conversations with your AI therapist and, you know, meta AI has no convers, you know, context on your conversations with chat. So we want to build this kind of uh this memory base of data and context by collecting all of your digital footprint in one place in a way that is secure and fully confidential and allow you to port this memory into any app or into any AI

23. 10:35

    agent. So this way in the future you can one click sign in into some AI app and it will immediately know who you are your preferences you know your history uh as well as your kind of like relationships with people around you and it can more directly service your needs with a personalized response and like you had said uh you know absolutely limited context windows mean that we have to build a lot of novel infrastructure around you know correctly retrieving the best memories for context

24. 11:05

    to generate to generate those personalized responses. That's super cool. Um curious on what you use for like access control management like how are you actually going to you know help the users decide what information to provide certain models or certain platforms uh versus others and sort of like the key management architecture around that.

25. 11:29

    Yeah. So there are basically two levels of access control today. Uh firstly is um you know you can manually kind of actually go and delete you can view your memories and manage them. So this is completely manual and in you know the user's control they can go sort through them they can view them by tags they can delete them they can edit them. Um they can also add new memories manually as well. So that's the first level. Um and

26. 11:54

    if you don't want some memory stored you could very easily remove that from our system. You have full control of all of your data. And the second one is you know beyond that the infrastructure that we have uh allows you to basically generate keys that need to be used to sign any kind of sessions in order for the the session to you know leverage your your memories or leverage your your context in this case. So if you want to

27. 12:20

    grant access to a certain app or an AI for you know it to get access to your memories you need to be you need to be there to manually approve it and then for that session it has access to to your memories and context. Got it. And how do you verify privacy? I am absolutely horrified of giving open AI or any single entity uh everything that I have online. uh I think it makes zero sense to do so and will probably be uh one of the biggest risks to society

28. 12:50

    if this is scaled. Um how do you ensure that information stays secure and private? Yeah, so firstly it's end to end encrypted. Uh and secondly uh the you know all the you know the vector embedding generation and uh the inference happens in trust execution environments and we submit remote addestations to the open gradient network where they can be validated. So you can actually see the code that's running. You can see that hey it's not actually secretly porting this data

29. 13:19

    elsewhere. Um and uh you know that way we get both verifiable and confidential computing kind of guarantees. And I agree that it is very much uh something that will increasingly become important as data for AI becomes uh you know more and more valuable as AI is able to do more and more now. So I think yeah we absolutely want to ensure that users have their data you know completely

30. 13:43

    private and secure unless they explicitly grant access uh to people um or apps. So um I don't think users are very sophisticated most of the time especially with data permissioning when they just like say yes I allow you to have access to pretty much everything.

31. 14:00

    Um, I think Apple's done probably the best job at trying to give people like, you know, or apps access to like very select images or, you know, sort of very detailed information. How is the UX intended to look in this scenario?

32. 14:13

    Because when, you know, an app prompts a user saying, "Hey, can I get access to your photos?" A lot of people just say yes without any idea of exactly like what they're giving up or the amount of information that can be harvested inside of your entire photo reel on your phone.

33. 14:29

    like how are you managing that part because uh Apple seems to be the on the cutting edge here and it's also some of like a unsolved problem. Yeah. No, that's a great question. So firstly uh I think it's a great point that you bring up. You can't really expect users to have the kind of um you know uh time to go in and actually manually view their memories. So something that we actually do is we tag the memories systematically um when we

34. 14:56

    generate the the vector embeddings and we actually censor uh memories that we consider to be you know more private or you know memories that would be considered confidential in many cases. So as an example uh you know things like passwords or private keys or you know very very sensitive personal data you know including maybe like your ID or sometimes like AI might get access to this information and we like censor all that and uh you know don't don't store it because we don't see it being

35. 15:22

    particularly helpful for uh usage of different AI as well. How do you see it if it's encrypted like how can you tell what is sensitive? Yeah, so it's being like decrypted within the TE. um that's how like we you know process that over that data. Um but of course like it's within the you know confines of the trust exeu ex execution environment. So you know it comes with uh you know like segmented uh processor and memory to ensure that you know this can't be accessed by anyone else. Um and

36. 15:54

    yeah, I think this is like a great systematic way for us to ensure that people get the best trade-off between like management of their memories uh done systematically plus of course confidential you know uh data privacy. Yep. That checks man. Um I guess we we got to move on to the next guest soon.

37. 16:14

    Uh it's actually very relevant to knowledge management and you know sub privacy. Before we go is there anything else that you want to flag to the community? Um, yeah, go check out memsync.ai. It's currently kind of in a private alpha beta, but uh memscync is memnc.ai.

38. 16:31

    Uh, if you want to go try it out and try out some of the memory kind of features that we have built, uh, yeah, feel free to do so and we'd love to hear back from you. Thank you so much, Cameron. And, uh, yeah, of course, we'll stay in touch. I want to double click on uh access key management and like how do you actually kind of programmatically give access to certain apps but uh we'll talk about in DM probably. Thank you for joining Matthew.

39. 16:55

    Uh next up we have Matt Hamilton from Asimov Protocol. Uh Matt, you are not a stranger to the show. Um but for those who may not know who you are, please give everyone just a quick overview of what Asimov is, who you are, and what you guys shipped this week.

40. 17:10

    Sure. We we also have Artto on as well if you want to bring Artto up as well. So I'm going to do the uh the usual thing if I'll do the overview and and that and then I'll I'll pass over to Artto to talk about kind of explicitly what we shipped this week. So um let me just uh share my screen here and bring up and get it to find the right window.

41. 17:37

    Okay, here we go. There we go. Hopefully uh you can uh see my screen here. Uh so um Azimov, what is uh Azimoth? So we are a project that is specifically working around what is known as neuros symbolic AI and uh structured um reasoning. So we've just launched uh a blog site.org and one of the first articles on here gives you a good background on what we are actually working on the concept of trustworthy neurosyolic AI. But the

42. 18:19

    whole idea here is to produce something that uh allows AI access to information that does have some provenence uh attached to it and can be verified in uh various ways. You can have attestations on it that it is actually uh valid or has some prominent prominence information to it. And all of that is then anchored to uh a blockchain. So anchored to to near in this case um for uh reasoning right and then as part of that we have the Azimoff platform if you go to.sh

43. 18:58

    and so that gives you a whole sort of rundown of the ecosystem and um again what we're sort of building from a technical point of view including the Azimoth modules. So there's a whole directory as well. directory that has a whole bunch of modules for pulling in information from your bookmarks, from Amazon, Facebook, etc., etc., Instagram.

44. 19:19

    Um, and so you can use all these modules, this entire modular system to bring this information in and convert it into a knowledge graph. The whole idea of knowledge graphs is you can bring disperate data from different places and you can query over it in a coherent fashion knowledge graph. So, I'm going to pass over to Artto now uh to talk about specifically what we've launched because we've got a few modules that we've launched uh this week.

45. 19:47

    Thanks, man. Uh yes, so since the last time we were on the uh pod uh two weeks ago, we've uh published a number of new modules. Um we are now uh moving into the uh PIM territory as in personal information management. And as um you may or may not know the cornerstones of that would be um your your contacts, your calendar, your email, uh your bookmarks, your notes. Uh so those are all of the things that we are prioritizing uh unlocking from your um

46. 20:21

    private confidential data um whatever AI systems you deem trustworthy u which of course typically would be something like a local or llama because you wouldn't necessarily want to sacrifice uh this to the cloud guns right so uh let me speak a little bit about that then um let's see here lots of uh tabs um So we we have made available uh in um early early release form um now a total of 15 modules. We have almost 100 uh in the books. A screen share coming up. Yeah. Uh we have almost 100

47. 21:02

    in the books um that we are um you know finishing up as resources allowance and and and then flipping to public. Um within the existing uh modules we have um an interesting mix meaning uh on the one hand for example we have the near module here for fetching uh blockchain data. Uh on the other hand we have uh kind of aggregator aggregator and break tools um such as API bright data and sub API through which we uh are able to pull in data from hundreds of uh websites.

48. 21:36

    And of course the the point of the platform is that we we pull in this data and uh we uh turn it into a knowledge graph. Uh which means that all of this data is possible to trivially merge. Uh once it's in knowledge graph form um you can just um a graph plus a graph is a graph. Uh so that's that's sort of the magic of knowledge graphs. Um so we have these uh aggregator modules um which uh help for pulling in um for example your Twitter followers, your LinkedIn um

49. 22:07

    connections and and the like things that you can't really get through even any API uh that need uh you know this kind of uh scraper APIs. Uh and then we have a growing list of um modules that um um unlock your your uh personal confidential data such as the IMAP module uh which I'll demonstrate shortly. Um that is then uh a good way to access your personal email. Um but not only the IMAP module, we also have

50. 22:35

    as you can see the inbox and mailer modules. These are for more advanced users. Uh if you know what those things mean, uh you know that you want those modules. If you don't know what what they mean, you really don't need those modules. Um and um we we are sort of um um unlocking these personal information management uh aspects of your life uh module by module aspect by aspect. If

51. 22:59

    you look at we have also the vcard module. This is um the current way to bring in your contacts as in you can export vcard from anywhere including your Apple contacts or including your Google directory. But uh soon enough we'll also be rolling out the uh direct Google integration using the oorthth authentication. Uh so it'll get even easier. But uh um this these are in any case um a a sufficient list that you can

52. 23:24

    begin to do many many interesting things which we're going to demonstrate over the coming weeks. Um and one more mention here the uh chromium module. So the chromium module marks out from uh Brave uh Chrome and Chromium and Microsoft Edge as well. Um and and of course there's a long list of chromium based browsers. Um so uh this this means that um uh if you wanted to build for example a um alternative to the for yourself for the pocket service

53. 23:56

    that Mozilla is sunsetting right now to uh much dismay. Um well, this would probably be a good way to to get those bookmarks and and do something with them. Um so so that's uh that's the kind of stuff we're doing and um you can follow us on GitHub at Asimov modules as modules. Please give us some some love over there. Uh we also have the Twitter asimov_modules where we are automatically uh tweeting

54. 24:25

    out the module releases. As you see it's quite busy. um and uh of course the individual um individual modules live in in uh particular um repositories over there. So let me let me do a very brief um live demo of how you can utilize these tools.

55. 24:45

    Um I think um good thing to mention here is that because we are building um we are building in Rust we have a unique approach where um all of these modules are actually um static binaries built for every platform and architecture. And that means by the way that they will work still 20 years from now. No matter what happens to the platform stack, no matter what happens to language runtimes, those kernel APIs, APIs are stable and these binaries will continue tracking. So this is a bit of

56. 25:16

    um um alternate approach to uh the kind of dependency hell that characterizes a lot of AI frameworks today uh and and language ecosystems. Um so all of this um asimov tools that you install as mod. All right. So we we don't see your screen. And if you're sharing your terminal, we don't see your terminal at the moment.

57. 25:36

    All right. It's just a Yeah, it's now. Okay, cool. Yeah. So, um once you have the asim of tooling installed, what you do is you you'd say uh module install imap. And of course, I could, you know, give it a little coposity, of course. Um good like a good Unix tool. We don't uh output anything when everything goes right unless you want vaposity. Um and then once you once you install that you get uh um you know um programs that implement particular patterns such as

58. 26:05

    fetcher cataloger. Um and these as I said are static binaries and they can be used standalone. Uh so I'll just show if I if I write here IMAP cataloger give it an IMAP URL. The platform is fully based on URLs. Everything is a URL. Uh so this just goes to my local IMAP server and pulls in uh you know the the emails uh that I copied off a mailing list dump for airflow. Um and then of course to turn that into a knowledge graph uh or any other format

59. 26:36

    can specify what you what you need and and you get the same thing uh like that. So so that's super simple. uh you want a particular message. Well, you got the I'm a fetcher which takes a longer URL and you can get the uh original or you can get again the knowledgraph version and there's currently a bug in this pre-release that doesn't have the body included there but will be will be later today. So um so this these tools um are absolutely standalone. They they modules

60. 27:06

    they they do not require the rest of the platform. They do not require the command line tooling. uh they can be utilized on any host anywhere standalone to unlock your data. Um so this is why we think that our tools u our modules are going to be integrated uh across the industry into lots of frameworks because they will be the by far the easiest reliable way to unlock a lot of these data sources.

61. 27:32

    So that's so Arto just just to be clear your target market here are the the existing Asian frameworks. Uh well look I I don't personally um you know necessarily care about the agentic uh uh framing here. Um we we are really just um uh enabling anybody who needs uh data u to to get it from various sources and to very importantly do a way that it can all be unified and merged locally.

62. 27:58

    You can pull in public data uh and you can pull in your confidential data. You can unify them and you can feed them to agendic systems if you like or just your local LLM. Um and uh we we'll be demonstrating a lot of interesting uh applications of that over the coming weeks.

63. 28:14

    Super cool stuff. Um shipping love to have you on the show so consistently. We do have to move on to the next guest, but is there anything else besides following modules on GitHub that you want to flag? Uh no, just uh please give us some love.

64. 28:30

    We have uh uh not too many followers as yet. So, as modules, as platform on Twitter. Great. Awesome, guys. Well, thank you so much for joining. Uh, we are going to be moving on to our next guest. We have Naveen from IQAI. Naveen, how are you doing? Let's see. Where did you go?

65. 28:53

    Hey, Naven, can you hear me? Hey, how's it going? Living the dream, hanging out. Got myself a room with consistent internet, which is great. Um, are you going to be demoing anything today? Hey, can you hear me? Hey Naveen, can you hear me? Test test.

66. 29:31

    Now I can hear you. Thank you. Yeah. Can you hear me? Okay. Yeah. Loud and clear. Um can you just quick beautiful There's your beautiful face. There you are. Um so real quick, um for everyone who doesn't know what IQI is, can you just give a quick lowdown of what you're building, who you are, and what you shipped this week?

67. 29:51

    Absolutely. Uh let me present Yeah. So I run IQAI. So we're the agent tokenization platform. So basically we've built a platform where you can launch tokenized agents and we've also built a framework as well to build those tokenized agents.

68. 30:19

    So you can see the platform here. So there's a lot of different agents that have been launched. So one of them for example is Sophia which is the AI editor of IQ wiki the largest blockchain encyclopedia. Um so you can see that here. So you can see all the edits it's making. So if you build an agent the cool thing you can do as a developer is you can make sure to put your developer logs so your AI's activity so everyone could see what your AI agent is doing. And you can see the

69. 30:47

    tokenized agent. You can see that there's trading available. Um you can see of course the price chart data all of that. Um but what's really interesting is that we recently moved so this is kind of what we shipped recently is we moved our framework onto the ADK.

70. 31:02

    So we moved it to uh to work with agent development kit which is Google's uh Google ADK. So their platform for building multi- aent solutions and kind of building exciting agents. And then what we shipped this week specifically is a MCP for connecting agents to an MCP for connecting agents to near intents. With near intents growing so fast, we figured this is a must have for any tokenized agent builder. Uh

71. 31:33

    basically it allows the agent to you can build something like where you just do prompt to transact. So you just prompt you put in the prompt and then the agent operates for you. I can actually show you guys the video. So, start that.

72. 31:48

    So, yeah, you can see we're starting out with the near wallet, but you'll see the next phases. Yeah. So, you can see they're swapping one year to USDT using near intense. Right now, it's processing and then boom. Yeah, there's the quote and then you can put in the address. So right now it wants you to proceed with the swap, but for future you can just do it automatically, but that's kind of how we built it so far.

73. 32:26

    Hey, um just real quick for everyone in the audience um and myself uh where are these price feeds coming from? Is this all from Near Intense directly or Yeah, like how do you know the price of year? I believe Yeah, I believe it's all pulling from Near Intense directly. Got it. Cool. Through the oneclick API.

74. 32:45

    Exactly. Exactly. Nice. Yeah, this is super cool. Uh what next? Like what do you where does this go from here? There's a lot of really cool stuff we're working on. So another thing, this was just executed. So we built also a uh we also built a MCP specifically for connecting the agents to the near blockchain in general. So that way so the first step is near intense is the swapping between chains, swapping between tokens, but also now you know this new MCP is going to allow you to

75. 33:18

    use the full near suite. So the full near protocol, all blockchain operations, account management, smart contracts, everything. Um, so that's going to be exciting. We'll probably have a demo for that probably next month. So that's going to be really cool. Um, some other really cool stuff that we're working on is that uh we also just recently we moved all of our backend agent software. So for our

76. 33:42

    internal agents like you saw Sophia, we moved that all to Gemini and so we're working with Google Cloud on some cool stuff. So that's why also the TypeScript. So the ADK platform, the framework is specifically designed to work very easily with all of the Gemini suite and all of that. So that's pretty exciting. Some other big stuff is you're going to see some MCPS combining robotics and AI. So, I don't want to give too much away because we'll

77. 34:07

    probably demo it actually uh next month on the town hall and probably also as well on near week as well on the on this stream as well. Uh we're going to have something very cool connecting robotics near and you can see how this can kind of f form the foundation for that, right? because you first you're going to need the swapping those elements, the near blockchain, and then some really cool stuff coming as well.

78. 34:31

    That's super cool stuff, man. Um, we do have to move on to the next guest. Is there anything else you want to add before we do so? Uh, yeah, definitely just check us out. So, we're at iqai.com. Uh, check out adk.iqai.com to see the the framework and everything.

79. 34:48

    And, yeah, happy to connect and thanks for having us on. Thanks for joining. See you next time. All right, next we have another familiar face, but this time with some pretty cool announcement. We have Jordan from Public AI. Jordan, how's it going?

80. 35:06

    Great to be here. I'll share my screen so we can show off some of the cool stuff we've been shipping. Beautiful. But real quick, for those who don't know you, who are you? What is public AI and what do you ship? Excellent. Yeah, thanks for the space.

81. 35:21

    Um, real brief, we'll do a a context on what public AI is and then we'll go into the features that shipped this week. Uh, at at the core, what a what we're building at public AI is the human layer of AI. And what that means is as this AI revolution is sweeping across industries and being integrated um across the board, we're identifying where there's human value in these systems and using web 3 primitives,

82. 35:51

    capturing that value and enabling everybody on earth um around the world to participate. And that means that it's not only developers who can uh be part of this AI revolution but also everyday contributors can bring their expertise and experiences uh in a valuable way.

83. 36:08

    And where we're starting is with data. Um data is one of the components of AI that um we see as being in high demand and will continue to be in high demand. The other bottlenecks in the AI industry include things like compute and memory resources which um are becoming more and more commoditized. And as the uh breakthroughs happen with different research, the the value of those um commodities is becoming less and less as

84. 36:41

    it becomes more efficient to be able to train and run inference on models. But no matter what you're doing, whether it is training or inference, data is always valuable um as far as like fine-tuning models for applied ML purposes. Um that last mile data to get your performance up to snuff as far as like reliability and cost effectiveness of running models in production. And then um even for

85. 37:06

    inference like some of the previous presenters mentioned, it's super valuable to have relevant data when you're um using these systems. So this is an overview of how we're building the human layer of AI. At the very top we have our clients that are coming to us for this niche on demand data. And then below that what we have is the the public AI platform. And what that comprises of is the data hub which is a

86. 37:31

    DAP where users can go and upload data, vote on each other's data and then also um provide feedback once the data has been applied to models. So we've got human in the loop across the entire pipeline uh for AI models training and and then testing them. And then we also have data hunter which is a chrome extension which can be used for further data fetching and also offloading the mechanical component of data verification. So um our custom AI models

87. 38:02

    that are used to verify the authenticity of uploads u making sure that it's not AI generated etc um can be part of the network that's making all of that happen. So that's how people can get involved. That's where the community of 1.5 million plus data builders is hanging out uh doing real work every day. And then below that we have the blockchain component. So those web3 primitives I mentioned underly all of

88. 38:28

    this technology where we have a reputation system in place uh to identify high quality contributors over time, consensus voting which is a lot like Dow voting for that data verification uh state collateral to ensure that participants are making meaningful contributions and skill tree KYC uh to identify expertise in the community through quizzes and things like that. So that's a a quick background on like how we're bringing this to life. uh how things are going. Uh it's great.

89. 38:57

    We're seeing a lot of demand from clients and have already delivered data to enterprises and uh also small and medium businesses, startups, etc. And then the community continues to grow at an astronomical pace. We've got contributors from every country uh every continent and uh a large majority of those are web 3. But it's great to see that there's interest in this from the web 2 community as well on boarding even more people um to this AI3 revolution.

90. 39:30

    And a little bit about the team uh couldn't be joined by everybody today, but uh wanted to give shoutouts to Dr. Steven Wong and Professor Suie who really are the origin story of what public AI is and where it came from. they were running an AI research lab. So they had firsthand experience um with the pain of getting data for these purposes and decided to try a decentralized approach after having some success moved to the Bay Area and Kenji

91. 39:55

    and I joined to help drive the business and grow side of things. So uh myself uh as as Cameron said familiar face uh been around the near ecosystem for quite a while having built early DAO and NFT infrastructure and then Kenji joined as the operational co-founder. He is um well suited to that domain. He was the Cso at Caro which is the largest used car marketplace in Southeast Asia um and has been helping out a ton on the

92. 40:26

    operational side of things tokconomics etc. So, that's all been kind of abstract, but I want to underline the fact that this is all very real. We have a platform that's running today and um you can go participate at beta.publica.io.

93. 40:43

    One of the biggest campaigns we're running is the native language audio collection. So, you can go to the site, upload audio in your native language, and get paid about a dollar a day for that if you do 140 uploads, which takes less than 30 minutes. Um, we're in between seasons right now, but you can still get on there and uh set your mother tongue and do the quiz to qualify for this when it goes live next week.

94. 41:06

    And we had over 5,000 participants in the first season that all got paid out for their real work. And because this is a live system, we're always iterating on things to make it easier for our contributors. So, one of the things that we noticed with the last run, the last season of this uh mother tongue campaign was that people would come into a few but then weren't committed to getting their full daily

95. 41:31

    uploads done. So, we came up with a gamified interface that shows you some targets to hit for your daily recording. And along the way, you can see how much you're going to earn when you hit the next target. In this case, we're about to hit 20 uploads where we earned 25 cents. And if you complete the whole day's worth of uploads, then you get your dollar.

96. 41:53

    We've also got a campaign live right now to go and check out where you can flex your status on X. After you do a quiz to learn about public AI and all the mechanics, then you can uh tweet how excited you are about this and uh earn up to 100,000 points a day for um yapping about public AI. So this is our kind of version of Kaido or something like that. And um you only get one submission a day. So don't upload junk. Make sure that you make it count.

97. 42:26

    Another big thing is this voting system that we have. So whenever there's votes, there will always be disputes. And we just got a dispute system online so that as uh disputes occur, they can be bubbled up through our credentialed contributors. So folks that have contributed for a long time across multiple campaigns get guard status. And then um a level above that is our judges. So when you hit dispute over any

98. 42:50

    of the um claims you've participated in voting for, say you're very confident in your um judgment that like uh a hero upload campaign was just algorithmically generated content or off-topic, not about public eye, something like that.

99. 43:06

    But it turns out that the community vote was in the other direction, then you can open a dispute and have that bubbled up. And after a judge looks at it, all of the stake points and reputation involved in that vote will be flipped and um rewards will be distributed as as uh as determined by the judge.

100. 43:30

     Another thing to call out is we did have some participants in the last season of audio that were uploading TTS uh text to speech and AI generated audio. Uh that's not why clients come to us. Uh we've got very clear guidelines on what's acceptable and what's not as far as the uploads go. So we do ban bad actors in the system. Don't come and think this is an easy farm. Um you're going to get banned.

101. 43:57

     And on the other side of that, we also have lots of ways to level up uh for the people who are committed to doing real work and building the future of AI with the humans in the loop. Um if you have contributed to multiple campaigns and you're uh committed to the mission of public AI, then you can apply for guard status on Discord. And that comes with elevated privileges not only in the community but also on the data hub platform where you'll get paid more for your votes and be able to uh address

102. 44:28

     disputes and things like that. Another earning opportunity is to use the referral system. So we've got this idea of like seven degrees to any expert. So uh we encourage everybody to rope in all their friends uh let them know about this opportunity and that they too can be part of the AI revolution.

103. 44:47

     very commonly requested feature across the board. Um, in crypto, it's a it's a reality that sometimes people lose access to their wallets. So, we shipped the ability to disconnect wallets. You do have to bind your email first so that we have a tag on you um and and your contributions. Uh, but after you do that, you're able to disconnect wallets, reconnect wallets um so that if you need to shuffle things around, that's no

104. 45:11

     problem. This one was a commonly requested feature from the community, so a lot of people are happy about it. And then last but not least, this week we went live with the checker for datab eligibility. And this is an NFT collection that we're launching in partnership with hot protocol. So as part of their hot craft NFT marketplace and launchpad, which is chain abstracted, powered by near intense, were the first collection to launch on there. And the really exciting thing

105. 45:42

     about that is that when you show up to Hotcraft, you don't need to um necessarily use a near wallet. You can log in with EVM or Salana, grab your databy, trade it, buy for soul, buy for ETH, sell for soul, sell for ETH. Um and and it all works seamlessly. So um really fun use case and really excited for the mint of this on July 15th. And that's all I got. call to action.

106. 46:09

     Pop on to beta.publicai.io. Uh get started. Make sure you get through the quizzes and everything else you need to contribute. And yeah, excited to see you in the community. Jordan, as always, you have so much to offer. Um and for everyone around the world that speaks a language, which is everybody who speaks can make money. So, uh real sort of intersection of crypto AI, paying people using crypto to actually incentivize the things that AI

107. 46:35

     needs, i.e. more quality data. So, this is incredible. Uh, one thing I was a little uncertain about, when does season 3 uh, for the language competition start? Like, when does it formally start? Yep. It'll we have it on the books for um, early next week since we have the we're probably starting Monday next week is is um, what we're looking at. Yeah, good to know. Uh, we'll we'll we'll pull the strings to get the, you know, the

108. 47:03

     marketing funnels aware. So, u, thank you so much. This has been so good. We have one last speaker. Jordan, thanks for coming. Thank you. Uh, Ventali, are you on? Yeah. Hello. Hey. Hey. Welcome. First time. Love the love the San Francisco background, which is not a fake background. Uh, which is nice.

109. 47:26

     Um, sorry, I'm trying to open the share screen. Um, okay. Just go to present and then make sure just to share the window. Unless you want to share the whole thing. Oh, see you. Um, while we wait, uh, Ventali and I have been discussing, uh, you know, working more closely together in your ecosystem. Uh, I think she's building a super super cool for come on and and share more about it. Um, but I guess I can kind of talk about what what I've been shipping this week. Uh well I

110. 48:04

     actually took off yesterday one of my first days off in a very very long time which was nice. Um but before but besides that uh been working with all these projects here lining up a lot of the the the projects that I mean really what I do is I manage all the integrations uh sometimes investments sometimes uh just general partnerships uh with all these teams. Now right now I'm specifically focused on uh the different parts of the DCML stack. This

111. 48:28

     is the decentralized confidential machine learning. Uh this includes everything uh from aggregating compute. We're primarily interested in H200s uh because they have uh potential TE support um doing a lot of work with uh model training and fine-tuning uh work with teams focused on eval benchmarks.

112. 48:47

     Uh and then portable memory is one of them and uh lastly like all the TE hosting with follow network and others. Um, but Ventelli is back. So, uh, hey, hey, welcome back to the stage. Sorry about that. Um, oh, all good. I I can stall as long as you need. Um, can you see?

113. 49:08

     Are you ready to rock? Yes. Okay. So, um, we're working on this. It's a very simple product. Um, it's just a monetization platform for AI devs. So, um, I have a lot of friends. I'm based in San Francisco. Uh, they're AI founders. and um it's very hard for them to convert their users to paid users and uh because they're you know so many different apps etc. So we are we're thinking about how to essentially help

114. 49:35

     them monetize and so um right now this is only a beta version. We just launched actually two days ago. Um so we're testing with a bunch of developer friends. Um so right now what this developer platform can do is uh can do LM uh model routing. So similar to open router um that's one thing. The second is we're working on billing. Um so people have probably heard of like cursor that sort of um you know annoying all this develop uh user community

115. 50:06

     because he's shifted to um usage based payments um immediately and uh we're trying to make billing much more flexible. Um so um developers can sort of um say like oh I want to charge for example per GPU usage or number of API calls or a combination of both. It's uh so we're making a very flexible billing um product as well. And the third thing we're doing is actually the core um probably one of the core product is the

116. 50:36

     ads function. So uh if you're building a chatbot, if you're building an AI campaign, uh if you're using LMS like we are um helping you inject ads into your um product. So if you um let's say if you're a chat uh like similar chat bots um we help inject product recommendations into the text responses that uh you give to to your users and um and help to you know make revenue for you from day one. Uh so that's what we're doing right now. Um I can show you guys. So if you log in

117. 51:18

     um we're using preview and preview I think supports near right now. Um so um you can add um let's say like an AI app um and test key and then you can you know create a API key that can access right now we support over 80 models um and then you'll be able to um manage your keys here can create multiple keys um for developer or staging or production and then right now for billing we support stride but we're um connecting with near

118. 52:07

     like stable coin so in future uh the problem is that we're going to use stable coins because strike right now takes right like 2 um 9% plus like.3 per transaction so it doesn't make sense for like small uh transactions uh which I think most developers are doing right now is if you if you write like pay like $10, Stripe would charge you like um 0.5 or something uh dollars which is uh pretty pretty high. Um so we're going to connect with um Stablecoin at Stablecoin

119. 52:40

     uh in a week or two. Um and um we're testing the product. So we'd love to um you know get your eyes on the product and tell us you know if anything is feels weird um we would love to get your feedback and um right now we're giving 10% um discount or we're adding so so you can get 10% more credits uh if you use the routing um so if you pay like $10 you get 11 dollars credit um and uh

120. 53:10

     you know early users will love to get your feedback um so yeah that's that's what Sudo is about super awesome stuff. Um, when you say AI developers, you're talking everyone. People training new models, people creating new agents, literally everyone out there. Yeah. As long as you're using like LM.

121. 53:30

     Um, and if you're using, I guess, different models, then you can use the routing service that we have. And all the features are like independent of each other. So you can use apps without using um billing or um uh authentication or or routing service or you can use routing. So um you know it's it's there's no like locked in to anything.

122. 53:53

     Super cool. And for those um who might not know you, who are you and you know what kind of led you to uh to building this? Yeah. So, um I've previously built two startups and one is actually a consumer AI um um startup um and where we do uh fiction generation um for um users. So, you can go there and generate your fiction um in future. I wanted to you know generate videos and entertainment content like TV shows uh that's very personalized and customized.

123. 54:26

     So, that's my first um startup. And then um the second one is a ZKVM company. uh it's one of the the the major competitive DKVMs out there and um and um so I've been the space for a couple years and um this AI product is um we we really think that payment is uh the the one of the right uh only product market fit of this space and um right now there is this increased demand of like LM usage and um we just think that this is

124. 55:03

     a very good entry point um and uh there's a lot of pain points of developers um getting a lot of inference costs and um etc. So we just you know run into this problems like friends run into this problems all the time and um we think it's a good opportunity there's a huge space so um that's why we're doing this Awesome.

125. 55:32

     Might be my internet, but sound a little robotic at the end. Probably my fault. Um, but it's super cool. Uh, where can people learn more about pseudo? Uh, I know you're winning more AI developers and users. What is the website that they should go to?

126. 55:46

     Um, so the website is pseudoapp.dev. Um and in future uh actually we're also working on a user consumerf facing app store. Um so if you use if your app uses our dev platform uh connect with our user off then your app will be shown on our app store where users can use uh right their their pseudo account to access and discover your app and so we're also working on that but this developerfacing uh website is called pseudoapp.dev dev um and in future the the consumer one will be pseudoapp.ai

127. 56:24

     AI, but this is the this is the website that beautiful. Well, cool. Um, super stoked to hear more about the progress as you know as the months progress and um, thank you for joining. Thank you so much. Yep. See it. So, that is all we have today. Uh, thank you all for bearing with the Wi-Fi connectivity issues. And so uh yeah, I guess very very briefly I'll finish uh what I was talking about when what when am I shipping this week.

128. 56:54

     Um again there's DCML decentralized confidential machine learning. There's a blog post that highlights pretty much all of this. Um I can drop it uh to nearweek and they can post it later. And then uh and this is all about how do you actually train uh models in trusted execution environments. So you can do things like content authentication uh essentially verifying that the model that you're using is uh or the data that is used to train a model is well

129. 57:20

     accurate and correct then eventually help monetize some of these models uh you know to sort of bridge the gap because right now there's pretty significant issue around uh monetization in AI. people spend tens of millions of dollars training one of these LLM only for the next week for it to be out competed uh where a lot of that work essentially is made irrelevant and so we do need to solve this monetization issue which is why I love what uh pseudo is

130. 57:44

     doing so that's DCML decentralized confidential machine learning and then there is AITP AI transfer protocol you go to AITP.dev dev uh to learn that it's the entire spec of how agents can communicate with other agents, pay other agents, audit other agents uh and so in the world where agents are going to connect with other agents, you can ensure that uh or have a little bit more accurate like faith that uh they're doing a good job and they're not going to like leak your private keys. And so

131. 58:15

     next week I'm very excited for some of the teams that are uh working on very important parts of this to present their progress. Um, and then there's a actually two more buckets that uh I don't really talk about a ton because I haven't well one of the buckets uh I don't really own or I don't really do as much work with and that is uh the shade agents. This is primarily led by the proximity labs team and this is where you can have an agent that gets verified

132. 58:39

     on chain. Uh it's all running a TE to ensure that the code uh you you know you can have some sort of at the station saying hey this agent is uh doing the things it's supposed to do and is running the code it's supposed to run. Uh which is very very valuable and it can also sign transactions across any blockchain. Uh we've had Owen from Proximity Labs come on a few times to talk about the progress and hopefully he'll come on I think next week. Um so that's a another third bucket and then

133. 59:07

     the very last bucket is uh I do a lot of work with the house of stake. Uh the house of stake is uh the near a near ecosystems uh primary dowo that is going to manage a very large token treasury to accelerate the near ecosystem. Um I'm one of the uh delegates or you know endorsed delegates by the near foundation and we are going to govern a uh about $10 million treasury. Now uh

134. 59:32

     the work I'm doing specifically on that front is related to AI governance that is actually using AI to help accelerate human decision-m within the DAO. uh this if for anyone who has participated in any type of board meeting, any type of uh I guess like a political campaign, people don't love participating in governance. So, how can we use AI to help people make better decisions um and eventually giving uh an agent or a network of agents uh a seat on the DAO?

135. 1:00:03

     And so they're actually going to be able to vote uh on behalf of their constituents or their you know people who delegate to them uh to push their their sort of uh interest forward. And so uh this is a part of a much bigger vision of how do we create an AI CEO. Um so we do think that uh the house of stake is a perfect place to experiment with AI governance or agentic governance

136. 1:00:26

     the way I like to say it. uh we are not governing how to use AI or what you know not governing the the model itself we're actually using the model to support with governance and so u hopefully there's going to be a lot more to come on this front that can be public about uh but yeah that's pretty much what I've been focused on there's a lot of projects in every one of those and so uh I'll continue to bring them on to the show as uh as we continue that's about it uh we

137. 1:00:50

     hit the hour mark uh thank you all for joining and see you next week and shout out to your week for coming in and uh ensuring a smooth transition while my internet sucked. So, thank you all for joining and hope to see you soon.

138. 1:01:06

     Thanks for watching. Subscribe to Nearweek.