Latest / The Index Podcast / 🎙️ Avi Patel (Kled AI) & Andrea Muttoni (The DATA Foundation): Who Actually OWNS Your Data in the AI Age?
Transcript
- 0:11SPEAKER_02: Hey everybody, welcome to the index.
- 0:14SPEAKER_02: I'm your host, Alex Kahaya, and I have two guests today.
- 0:17SPEAKER_02: I'm really excited to talk to them.
- 0:18SPEAKER_02: I've got Avi Patel, who's the founder and CEO of CLED AI, and now chief data officer at the Data Foundation, which formerly used to be Story Protocol.
- 0:28SPEAKER_02: Um, and I've got Andrea, who's the CEO of the Data Foundation.
- 0:34SPEAKER_02: And I'm super excited.
- 0:36SPEAKER_02: Did I get that correct?
- 0:37SPEAKER_02: Andrea.
- 0:38SPEAKER_02: Well, you're Italian.
- 0:39SPEAKER_01: Yes, no, Italian's a great pronunciation.
- 0:44SPEAKER_01: I tried to get it right.
- 0:46SPEAKER_02: Yes.
- 0:46SPEAKER_02: Yeah, no, I I I kind of I kind of you know, I kind of took a swing and I got it right, which is good.
- 0:52SPEAKER_02: Um yeah, I'm excited to have you guys here.
- 0:54SPEAKER_02: Thanks for being here.
- 0:55SPEAKER_02: The question I always start off with, and maybe we'll start with Avi and then go to Andrea.
- 0:59SPEAKER_02: What gets you out of bed in the morning to do this?
- 1:01SPEAKER_02: And Cléd is, you know, you've had some success, and it's it's a uh hard thing to do to build a startup.
- 1:06SPEAKER_02: And then now doing what you're doing with the Data Foundation, like what's your why?
- 1:10SPEAKER_02: What drives you to do this?
- 1:11SPEAKER_00: I think I'm very stubborn.
- 1:13SPEAKER_00: I had an idea for making licensing easier four and a half years ago, and the industry is very, very tough to work with.
- 1:20SPEAKER_00: I was in music prior to this, so I was I was creating the first music licensing marketplace for three and a half years, um, and ended up getting sued pretty hard by record labels, and the business ended up getting shutting down just because of how complicated licensing is.
- 1:34SPEAKER_00: Um and then we got into training and AI training licenses, and we found much more success there.
- 1:40SPEAKER_00: And we're at the point now where I think if I keep going at this for the next 20 years of my life, I have the the right stepping stone to hopefully fix the licensing problems in every single IP sector, whether that's music or image or video or anything else.
- 1:55SPEAKER_00: Um so that that's what picks me up every day, pretty much.
- 1:58SPEAKER_00: Yeah.
- 1:59SPEAKER_02: And and I think also my if my understanding of CLED is correct, like people AI companies can license data and train data, you know, train AI on that data, right?
- 2:07SPEAKER_02: And they pay for that.
- 2:08SPEAKER_02: And that do those payments go down to the individual person who contributed the data?
- 2:13SPEAKER_00: Yeah, yeah, yeah.
- 2:14SPEAKER_00: I I've been building licensing marketplaces for the last four years.
- 2:17SPEAKER_00: So um this is just the the same individuals sovereignly contribute their their data, whatever they want to contribute, depending on whatever tasks we have available.
- 2:25SPEAKER_00: Um, and then that is sold directly to foundation models and labs.
- 2:28SPEAKER_00: Uh, we work with most of the leading labs, um, and then uh yeah, that that that trickles straight back to the user.
- 2:34SPEAKER_02: Yeah, very cool.
- 2:34SPEAKER_02: I have a lot of thoughts on that, but I want to go get to Andrea and hear what your yeah, your story is.
- 2:41SPEAKER_02: Why are you doing this?
- 2:42SPEAKER_02: What's data foundation?
- 2:43SPEAKER_02: You know, how did you get here?
- 2:44SPEAKER_01: For sure.
- 2:45SPEAKER_01: I always like to say that I was in AI before it was cool.
- 2:49SPEAKER_01: So I was in AI since, I don't know, 2016, 2017.
- 2:54SPEAKER_01: Back then it was Amazon.
- 2:55SPEAKER_01: So Amazon Alexa, you know, voice assistance, conversational AI, a little bit of the right thing at the wrong time, right?
- 3:03SPEAKER_01: If it if it had waited a couple years and now they're trying to catch up with all these new LLMs, but I see a lot of patterns that are very similar.
- 3:13SPEAKER_01: And accidentally, my career has also been kind of weaving in and out of IP.
- 3:18SPEAKER_01: So I started in Amazon uh as actually on the Kindle side, so Kindle books, so working with publishers to get their books on Kindle, so working with a lot of IP.
- 3:28SPEAKER_01: Then I went to Dapper Labs, which was the company that probably in the crypto space did the most with IP, right?
- 3:34SPEAKER_01: Disney, NBA, NFL, UFC.
- 3:39SPEAKER_01: And and so Story Protocol initially was all focused on IP, but we we went like super, super broad, right?
- 3:46SPEAKER_01: Um music, art, you know, RWAs.
- 3:49SPEAKER_01: And we quickly realized that data, AI training data, was where things just resonated the most, where there was this everything else, and kind of linking back to what Avi was saying, right?
- 4:01SPEAKER_01: Like the the reason he pivoted his career was also because the IP industry, the traditional IP industry is very, very entrenched and not interested in modernizing because the entrenchment gives it defense.
- 4:15SPEAKER_01: Whereas AI training data is like this blue ocean, right?
- 4:18SPEAKER_01: It's this blue ocean that's grown massively in the last few years, and it's the right time, right place.
- 4:25SPEAKER_01: So the Amazon Echo was the right thing at the wrong time, uh, in my opinion.
- 4:31SPEAKER_01: It was still cool, still novel, but it was the right thing at the wrong time.
- 4:34SPEAKER_01: Focusing on AI training data is the right thing at the right time.
- 4:37SPEAKER_01: And so that's what that's what gets me up in the morning.
- 4:39SPEAKER_01: And um, yeah, there's still so much to be done, and we can get into it in the podcast.
- 4:44SPEAKER_01: But um, yeah, that's the TLDR.
- 4:46SPEAKER_02: I don't know if you know him or not.
- 4:47SPEAKER_02: Do you know Bid Hon Roy from Bagel?
- 4:50SPEAKER_02: No, he's also an Amazon Alexa alum.
- 4:52SPEAKER_02: I mean, it's a huge company, but yeah, he's an A Bagel is one of my favorite companies.
- 4:57SPEAKER_02: There's been some really cool stuff, um, AI related.
- 4:59SPEAKER_02: Open source enable people to like monetize open source models without giving array.
- 5:04SPEAKER_02: The the they they invented like a ZKP-based tech for knowing who contributed what parameters to a like training of a model.
- 5:12SPEAKER_02: Yeah, and then like I can train a model, then you can add your little like flavor on top, and then we get compensated based on the amount we've improved the model.
- 5:21SPEAKER_02: So, like if people use our model and it's hosted somewhere, then we both get compensated.
- 5:25SPEAKER_02: Very cool business.
- 5:27SPEAKER_02: Very much needed too.
- 5:28SPEAKER_02: What's the connection between CLED and data now?
- 5:31SPEAKER_01: Yeah, I I can start, and Avi, feel free to chime in as well.
- 5:34SPEAKER_01: Um yeah, so actually it's kind of like a match made in heaven in the sense that uh CLED, a byproduct of what CLED does, right, is just having a massive amount of data, right?
- 5:50SPEAKER_01: And part of the the flywheel of that business model is providing that data to labs to improve models, to improve state-of-the-art models.
- 5:59SPEAKER_01: And one thing that labs look for when they source data from different companies is essentially peace of mind, right?
- 6:06SPEAKER_01: They want to make sure that if they're sourcing data, that data is accounted for, there's there's receipts associated with it, right?
- 6:13SPEAKER_01: Because any data that they ingest becomes their responsibility in some way or another.
- 6:19SPEAKER_01: And so we kind of teamed up because CLED has a lot of AI labs.
- 6:25SPEAKER_01: Um, they work with a lot of AI labs, and they kept getting this feedback of like, hey, like, you know, how how do we get these types of receipts?
- 6:33SPEAKER_01: Like, how do we feel comfortable licensing massively crowdsourced data?
- 6:38SPEAKER_01: And and so we designed our flagship product called Trace, and we can get into that later as well, which is the front end through which all of these data receipts are accessed and can be verified.
- 6:51SPEAKER_01: So any data that gets uploaded to CLED from anyone in the world and and and other apps as well gets registered on the data network as a verifiable receipt, fully auditable, verifiable receipt that contains all of the metadata of that file, the metadata of the user that uploaded that file and contributed that file, terms of service that were signed, privacy policy that were signed, all the lifecycle events in an immutable way.
- 7:17SPEAKER_01: So like there's no tampering with it, right?
- 7:20SPEAKER_01: And so when an AI lab wants to verify a whole data set of data, they literally can have the receipt for every single data item, right?
- 7:30SPEAKER_01: And that's useful for the labs, but why do they want that?
- 7:34SPEAKER_01: Well, because also from a regulatory perspective, there's increasing pressure from various governments around the world to have more transparency on training data.
- 7:44SPEAKER_01: So whether it's the EU, whether it's California, you know, this is just going to keep getting more and more pressure around, hey, AI models can't just be black boxes.
- 7:53SPEAKER_01: You need to kind of be aware of what data has gone in, and you need to be able to show and prove what data goes into it.
- 8:00SPEAKER_01: And and so this is again going back to the point of right place, right time.
- 8:04SPEAKER_01: What we're doing with the data foundation is is really just to provide that transparency layer.
- 8:09SPEAKER_01: But Avi, feel feel free to jump in if you have anything.
- 8:12SPEAKER_00: I I think you tapped on everything pretty nicely.
- 8:14SPEAKER_00: Um yeah, I there's just this like increasing look, open AI, anthropic, all these companies are IPOing like this year, if not next, and uh all of them uh across the board when we were trying to sell to them prior to the this like public audit trail.
- 8:27SPEAKER_00: We're all asking for proof of consent from all of our users that the data that we're sending them is actively cleared for uh rights and licensing to actually use that data for training their models, right?
- 8:38SPEAKER_00: And that is just not a public thing that like has ever been done.
- 8:42SPEAKER_00: These companies like Mercore and Scale, it's just like, hey, trust me, bro, and then I'll sell you my data.
- 8:47SPEAKER_00: And like if my investor list looks cool enough, then you just trust me, right?
- 8:52SPEAKER_00: Like that's basically it, right?
- 8:53SPEAKER_00: Um, but but that barrier for entry is becoming more and more strict over time, and people need like active receipts of every single piece of data that is being sent to a lab with proper contracting, TOS, whatever it is in place, depending on the collection method.
- 9:08SPEAKER_00: So this is exactly what that's built for.
- 9:11SPEAKER_00: Anytime someone uploads to CLED, instant receipt is sent and put on chain so it's immutable.
- 9:15SPEAKER_00: You can never change that at any point in time with a full uh audit layer, basically, showing the TOS was signed, the privacy policy was signed, the KYC status, the tax status of the user, the metadata of the actual image without ever revealing the image or video publicly.
- 9:31SPEAKER_00: So you can you'll never get your data leaked, but a lab can then put these unique IDs for any image or video that we send to them, and they'll get the full information breakdown report of how the user is assigned all the proper stuff, the consent receipts, everything.
- 9:45SPEAKER_00: Um, and they can then buy data from us comfortably.
- 9:48SPEAKER_00: Uh so proper use case stuff.
- 9:50SPEAKER_00: Yeah.
- 9:51SPEAKER_02: I can't help it, Brandon.
- 9:52SPEAKER_02: We talked about this before the show, but I think it's a point in time to bring up like what happened in May with you, Avi.
- 9:56SPEAKER_02: You know, there was like this big viral moment of some company that like got a big big round of funding.
- 10:02SPEAKER_02: I can't remember who the VC was that funded the catalyst.
- 10:05SPEAKER_00: We can put a long blast, I fucking love to tell them.
- 10:07SPEAKER_00: Yeah.
- 10:08SPEAKER_02: That that was uh that was a a kind of a copy of of your business.
- 10:12SPEAKER_02: And I don't know all the details of everything, like I just saw it like I was kind of a spectator like eating pop like popcorn emojis.
- 10:18SPEAKER_02: So watching this happen.
- 10:20SPEAKER_02: Um, but the reason I bring it up, because I remember one of the things you were talking about about that company was data and like where they source it from and the integrity of that data.
- 10:27SPEAKER_02: I think I think you were talking about that.
- 10:29SPEAKER_02: Yeah, and you were like basic DD was like this was bad, you know.
- 10:33SPEAKER_02: And and you know, can you talk about that a little bit?
- 10:36SPEAKER_02: Because I think that's to me, that's the counter case, and that's why that your customers have such a big problem, is because of this exact thing.
- 10:43SPEAKER_00: Yeah, look, these data marketplaces that are coming up are all fake.
- 10:47SPEAKER_00: And I'll say this in the best way possible.
- 10:50SPEAKER_00: Like, they're not making money by selling the data they're collecting on their apps.
- 10:54SPEAKER_00: They're literally going out and manually doing transactions with these labs by like being this broker.
- 11:01SPEAKER_00: Like they'll hire a sales team, they'll go to their local mom and pop shop or enterprise or startup that they know and say, hey, give me all your data, I'll sell it for you, and I'll claim this as revenue as my business.
- 11:11SPEAKER_00: And all these VCs think it's coming from the app, right?
- 11:14SPEAKER_00: If you look at any of these apps, they have none of the infrastructure in place to possibly sell the data.
- 11:22SPEAKER_00: It requires a ton of things.
- 11:24SPEAKER_00: One, you need to actually know your shit when it comes to the legality of selling data, right?
- 11:27SPEAKER_00: Like Latham and Watkins is our is our law firm.
- 11:30SPEAKER_00: They have audited every form and every single jurisdiction and law in every region across the world thoroughly.
- 11:38SPEAKER_00: I've spent hundreds of thousands of dollars with them in order to make sure that our TOS privacy policy and every act and thing we do is clean, right?
- 11:45SPEAKER_02: Like otherwise, because you have like PTSD from where you started, which was in music, getting mass sued, and you're like, that's not happening again.
- 11:52SPEAKER_00: That's never happening again, right?
- 11:53SPEAKER_00: So like you, I can't see.
- 11:54SPEAKER_02: Litigation's not fun.
- 11:55SPEAKER_00: Yeah, like like look, opening eye, for example, approached us pretty early, right?
- 11:59SPEAKER_00: Like, and I didn't want to sell data to them that early because if I did, you ruin the relationship with the lab really fast if you give them bad things, right?
- 12:07SPEAKER_00: So I've spent the last six months of my time getting all of these ducks in a row, which includes one, obviously the TOS privacy policy, everything else like that.
- 12:16SPEAKER_00: Two, it includes fraud detection.
- 12:18SPEAKER_00: Every single piece of data that is sent through CLED is run through an extensive fraud protocol, which by the way, isn't 100% there yet.
- 12:25SPEAKER_00: It's what we're working on right now.
- 12:26SPEAKER_00: We're gonna get it to like 90%, 80% accuracy, and then it's there.
- 12:30SPEAKER_00: Then you can comfortably sell data to large labs.
- 12:32SPEAKER_00: But until you do that, you can't trust that the data someone is giving you is gonna comply with all of the requests the lab is asking for.
- 12:39SPEAKER_00: Like we had to ban the entire country of Nigeria, for example, which was a super viral thing that happened on Twitter from using the CLED app because of a 95% mod rate.
- 12:48SPEAKER_00: Like we were a task request for uploading like videos of you getting your haircut, and they would literally upload a two-second clip and then copy the clip and change the color over overlay on the video, and then do that 30 times over so that they could like scam our system, basically.
- 13:05SPEAKER_00: And let's say a lab is asking for that data, and I just send it to them without thinking, right?
- 13:10SPEAKER_00: And I which is what you have to do.
- 13:11SPEAKER_00: You have to send it very fast.
- 13:12SPEAKER_00: So it's not like I can check every video one by one manually.
- 13:15SPEAKER_00: I'm the one person.
- 13:15SPEAKER_00: I can't do that.
- 13:16SPEAKER_00: There has to be an automated fraud system that is cleaning all this data.
- 13:20SPEAKER_00: So we've built all of that in-house.
- 13:22SPEAKER_00: Um, and then lastly, it's this just proof that you know your shit.
- 13:25SPEAKER_00: You have to look buttoned up, right?
- 13:27SPEAKER_00: Like if you're not buttoned up, you will not be selling to these labs.
- 13:30SPEAKER_00: And a part of that means that your audit trails are really clean, which just doesn't, it's not possible right now.
- 13:36SPEAKER_00: Trace is the only tool on the planet because we specifically designed it for this that actually audits the data that is sent through CLED or any data marketplace.
- 13:43SPEAKER_00: Anyone's on the protocol, right?
- 13:44SPEAKER_00: Um, and then two, it's like having like your software.
- 13:47SPEAKER_02: Trace is on is traces on data on the data network.
- 13:50SPEAKER_02: Is that what you guys call it?
- 13:51SPEAKER_02: Because it used to be Story Foundation, but now it's the data network.
- 13:53SPEAKER_01: So so like the company is called the Story, the Data Foundation.
- 13:57SPEAKER_01: The data foundation, right?
- 13:58SPEAKER_01: Yeah, the network that is running and that is operated, right, is the data network.
- 14:04SPEAKER_01: And the product that Avi's talking about is called Trace.
- 14:08SPEAKER_01: So the rebrand is great.
- 14:10SPEAKER_02: The rebrand is great, but the thing you're on now, the data network, obviously, right time, right place.
- 14:15SPEAKER_00: I wouldn't know what's so interesting is exactly what we designed for.
- 14:18SPEAKER_00: Yeah, like we we like I remember when we were I was having the conversation with like SY and like Andrea like five months ago when we were talking about like we've been building this for a while before we even launched it.
- 14:27SPEAKER_02: Yeah.
- 14:28SPEAKER_00: And the specific thing I said off rip is I've been approached by hundreds of these fucking dog shit web3 protocols, and I refuse to associate my company with something that is not an actual good use case for blockchain.
- 14:43SPEAKER_00: So we literally sat down for like two months straight and brainstormed the best possible use case for on-chain activity, and that is simply just receipts.
- 14:52SPEAKER_00: Receipts are the cleanest way across the board to provide an actual use case layer to anything on-chain.
- 14:58SPEAKER_00: You it's actually genuinely a use case for this for this technology.
- 15:02SPEAKER_00: And what is better than providing receipts for training data, which is just the most influential thing when it comes to actually selling data across the simple, it's not easy.
- 15:12SPEAKER_02: I really liked Ocean Protocol.
- 15:14SPEAKER_02: Like, I don't know if you guys remember them, way back when, like 2017, and they were data marketplaces.
- 15:18SPEAKER_02: Like data markets were very interesting ideas, they were overly complex.
- 15:23SPEAKER_02: And yeah, all you want is provenance and traceability, right?
- 15:26SPEAKER_02: And so a receipt gives you that.
- 15:28SPEAKER_02: It says, this is what it's from, it was legit, and it's good to go, and you can use this, and it's licensed properly, so you can use it in your your training.
- 15:35SPEAKER_02: And I I wonder one question I have for you guys, because like obviously the big labs are the big buyers right now, that's probably where all your revenue and usage comes from.
- 15:44SPEAKER_02: But what's the rest of the market gonna look like?
- 15:46SPEAKER_02: Because we have like I see what I see happening is like individuals using local models on their computers and then like their own at-home racks and then and training them.
- 15:56SPEAKER_02: And I don't know if they're gonna be data buyers.
- 15:58SPEAKER_02: And then there's enterprises that have their own data that they'll likely do it through their own agents.
- 16:02SPEAKER_00: I think I think like all the individuals when they're making their models will have their own agents running, and they'll these agents will autonomously start transacting uh and and purchasing data off of these marketplaces um on their own, individually, if they're training their own open source or individual models.
- 16:16SPEAKER_00: So a ton of use cases, but even past AI, our market includes like hedge funds, for example, right?
- 16:22SPEAKER_00: Like we have a feature in our app that that's coming out in our next update where you can link like your Netflix and Spotify and like all these other accounts to your CLED app.
- 16:31SPEAKER_00: And then every single month, we'll just track like what songs you listen to, like what movies you're watching, everything, um, and we'll pay you, we'll basically pay for your subscription costs for these companies for free.
- 16:42SPEAKER_00: And then in turn, we can sell this data to somewhat like a hedge fund.
- 16:45SPEAKER_00: And the reason why hedge fund will buy is let's say a hedge fund notices that across 40,000 users, the app usage of Netflix is going down for two months straight.
- 16:54SPEAKER_00: They'll short the stock.
- 16:55SPEAKER_00: Like it's quite literally an alpha that they couldn't have before.
- 16:58SPEAKER_00: That's just one example.
- 16:59SPEAKER_00: Then you can sell to like government organizations.
- 17:01SPEAKER_00: Like, for example, we have a task in the app that says take pictures of potholes all around New York City.
- 17:06SPEAKER_00: Um, and that that is going to go straight to the New York government, and they can then clean up every single pothole in New York City with an exact geotag location, right?
- 17:16SPEAKER_00: Or you can sell to enterprises like uh fashion brands, for example.
- 17:19SPEAKER_00: And like let's say someone ate uh ate food at this place, like what is their most common association to like what clothes they are they're likely to have worn or want to wear, right?
- 17:27SPEAKER_00: Like there are a ton of random use cases, but the the market of data was a trillion dollar market prior to AI even being a thing.
- 17:35SPEAKER_00: Um Facebook was doing this for years, so tons of tons of like buyer side.
- 17:40SPEAKER_00: What's also interesting is the sell side.
- 17:41SPEAKER_00: It's not just CLED that is going to be selling data.
- 17:44SPEAKER_00: Like you see this with Pokemon Go, for example.
- 17:46SPEAKER_00: They started, they started getting data sales.
- 17:47SPEAKER_00: They're selling all of their user data, for example, probably for billions, to several of these not only foundation models and labs, but everyone else.
- 17:54SPEAKER_00: And I get hit up by all of my friends and random mom and pop stores.
- 18:00SPEAKER_00: All of them, they're like, hey, can I get into like data selling?
- 18:02SPEAKER_00: Like I would love to sell data and provide data to these foundation models and labs because it's just a multi, multi, multi-billion dollar market that very little people are capturing on, right?
- 18:12SPEAKER_00: So all these people that want to get into this need this audit layer, need this like tool to basically clean up their data and then provide receipts for where the data is coming from.
- 18:22SPEAKER_00: Otherwise, you cannot sell that data.
- 18:24SPEAKER_00: So the protocol in its ultimate vision is going to be taking in data from not only data marketplaces, like explicit data marketplaces, say CLED, but enterprises across the world that just want to start getting into data selling and need this like almost like Vanta-like tool to actually audit the data that they're they're giving out to labs.
- 18:44SPEAKER_02: So that's the Andrea, what what's what's been your perspective with like enterprise, for example?
- 18:48SPEAKER_02: Have you have you talked to like other companies?
- 18:50SPEAKER_02: Do they want to get their data on this marketplace?
- 18:53SPEAKER_01: Yeah, so on trace, CLED is obviously a massive user of trace, the largest user by far for now, but we all we already have other companies using trace, right?
- 19:02SPEAKER_01: And going back to the go going back to the question, which ties into this one as well, Avi covered the supply side really well and the the non-AI demand side.
- 19:14SPEAKER_01: I think on the AI-based demand side, right?
- 19:17SPEAKER_01: Right now demand is concentrated a lot on the big AI companies, right?
- 19:22SPEAKER_01: The big foundation models, the big world models.
- 19:25SPEAKER_01: But I do think that we're in this mainframe era of AI.
- 19:29SPEAKER_02: And what do you mean by that?
- 19:31SPEAKER_01: Well, I mean that right now everyone's using, even if they're using the if they're building their own app, they're just like wrapping around Opus or apping or wrapping around GPT-5, right?
- 19:41SPEAKER_01: And now we're starting to see like more and more, as you were saying, Alex, right?
- 19:44SPEAKER_01: Like more and more uh, especially with China competing uh with with open weights, right?
- 19:51SPEAKER_01: It's becoming easier and easier to fine-tune your own models.
- 19:54SPEAKER_01: I feel like companies will start to become mini AI companies as well, right?
- 20:00SPEAKER_01: So they're gonna start to train models specifically focused and based on their data, but that data won't be enough.
- 20:08SPEAKER_01: So they they'll have good context because, for example, like a manufacturer can train a customer's service agent on all of the history of their specific customer support, but that doesn't mean that that is enough data to make the model exceptional, right?
- 20:24SPEAKER_01: And a foundation model might just be too too broad, right?
- 20:29SPEAKER_01: So you're gonna have a fragmentation of models, but that fragmentation will also lead to a much more varied demand for specific types of data.
- 20:38SPEAKER_01: And that type of data will also be very esoteric, right?
- 20:41SPEAKER_01: Will be very like edge cases that you can't find easily on the internet.
- 20:45SPEAKER_01: The internet has been scraped.
- 20:47SPEAKER_02: But you have to get those companies to contribute that data, right?
- 20:49SPEAKER_02: Like I think to the network to your network, right?
- 20:51SPEAKER_02: So like telecoms, like I could think of niches, right?
- 20:54SPEAKER_02: Like telecoms like who sell like have sell care, like sell iPhones or whatever, like Verizon.
- 20:59SPEAKER_02: Yeah, you know, uh, they probably have big enough data sets, so maybe it's not a good enough example.
- 21:03SPEAKER_02: But generally speaking, what I'm thinking is like in your like the I need to augment my training data with more data, and that's why I'm gonna go buy data.
- 21:10SPEAKER_02: And I mean my competitors potentially would have had to put that data or their users.
- 21:15SPEAKER_02: I guess that's how you get around it.
- 21:16SPEAKER_02: Back to the Netflix thing, right?
- 21:17SPEAKER_02: It's the individual users that can contribute the data.
- 21:19SPEAKER_02: So I don't want to answer my own question.
- 21:21SPEAKER_02: Another thought that's been going through my head while you guys have been talking, and it's connected to my work.
- 21:25SPEAKER_02: We've made this platform called Corbett's, and it's a it's an gentic operating system.
- 21:28SPEAKER_02: Okay, it gives you like all the controls you need to deploy agents at scale in an enterprise, like identity, uh runtime level permission, like permissioning system, and audits.
- 21:38SPEAKER_02: And the audits are get backed.
- 21:39SPEAKER_02: We store everything as cryptographically signed.
- 21:41SPEAKER_02: So like any action the agent takes, it's tied to an identity, which is tied to a human.
- 21:46SPEAKER_02: And and we re record that data.
- 21:48SPEAKER_02: And the reason is obviously like you want to know what it did.
- 21:50SPEAKER_02: You don't want to just like get the inference back of what it did.
- 21:52SPEAKER_02: You want like an actual log of what it did so you can make it better.
- 21:56SPEAKER_02: I have actually thought that that data is gonna be super valuable to these companies.
- 22:00SPEAKER_02: And it's occurring now, like could be valuable in this way.
- 22:03SPEAKER_00: That's like I could I could put you in touch with my or the companies could.
- 22:06SPEAKER_02: You know, we're open source, so they can self-host it too, and then they can monetize it.
- 22:09SPEAKER_02: But I think what I want to ask you guys is how do you think about agentic data?
- 22:15SPEAKER_02: Because it's different than what that's coming, like the that's ramping, right?
- 22:19SPEAKER_02: Is like actions of agents and them creating data because they're using inference to create that data, and it kind of is like that was already trained on other data, and like is it right?
- 22:31SPEAKER_02: Is it wrong?
- 22:31SPEAKER_02: You know what I mean?
- 22:32SPEAKER_02: Like you can get the actual audit logs if you're using our tool of what it did, you know.
- 22:36SPEAKER_02: But is it a is it a weird complication?
- 22:38SPEAKER_02: How do you handle that?
- 22:40SPEAKER_00: It's that it's definitely coming.
- 22:41SPEAKER_00: Right now, Trace obviously is supporting like way easier modalities, right?
- 22:46SPEAKER_00: Like it's almost trace supports the what right now, right?
- 22:49SPEAKER_00: So, like when I when I say what, I mean it's supporting image, video, audio, um, text, all of these like very simple modalities to understand.
- 22:57SPEAKER_00: It doesn't currently support the why.
- 22:59SPEAKER_00: The why is like how something gets to an end result, which by the way is not actually what a lot of labs are buying.
- 23:07SPEAKER_00: Like labs are actually kind of stupid, like they're like a little lower stage than that.
- 23:11SPEAKER_00: They're still on the what.
- 23:13SPEAKER_00: Um so like Google and nano banana is just buying so many images in bulk, right?
- 23:18SPEAKER_00: They're not buying the why an image um if put into this model improves it, kind of shaped that way thing.
- 23:26SPEAKER_00: Yeah, yeah, yeah.
- 23:26SPEAKER_00: It's not the but the why is gonna come.
- 23:29SPEAKER_00: The why is something that like I know Anthropics looking for the why right now.
- 23:33SPEAKER_00: Like they love the why.
- 23:34SPEAKER_00: Um OpenAI loves the why.
- 23:36SPEAKER_00: Like this, some of these like bigger players, they love the why, and some of the teams at these bigger companies also love the why.
- 23:41SPEAKER_00: Um, but it it is a slightly smaller market, and I think as Trace evolves over time, we'll get into how we shape the why and put it into this uh into this tool.
- 23:51SPEAKER_00: I assume it'd be some form of like text documentation line by line.
- 23:55SPEAKER_00: Yeah, but we also have historic receipts.
- 23:58SPEAKER_00: So like theoretically, you could put like a historic receipt of like what an agent does end to end and just constantly update a a single block with all of those like receipts.
- 24:08SPEAKER_02: Yeah, I mean it is just text.
- 24:09SPEAKER_02: I mean, it's just JSON, right?
- 24:10SPEAKER_02: Like text file, like so it's you know, you can get via API.
- 24:13SPEAKER_02: I just was more wondering if like the shape of that data, because it came from inference, like it's a layer on top, right?
- 24:20SPEAKER_02: It's already and so it is the actual like tool calls.
- 24:24SPEAKER_02: Like every every tool call, every inference call, every turn, every prompt, everything that it got back is like all recorded.
- 24:30SPEAKER_02: And then the reason why this matters because like what we're seeing is people don't most companies want workflows that are agentic.
- 24:38SPEAKER_02: They don't want just straight up like ChatGPT, like broad thing, going back to your point, Andrea, of like point-specific models.
- 24:44SPEAKER_02: They want a workflow that is agentified and it has a bunch of agents like running at different steps to move it along.
- 24:51SPEAKER_02: And I think there's just interesting data getting created from that, right?
- 24:55SPEAKER_02: Like, because then it's also interacting with customers as well, and or or you know, in inside their organization or even externally outside their organization with customers or other businesses.
- 25:04SPEAKER_02: Yeah.
- 25:04SPEAKER_02: And I just think that the world, like that just I again, I'm new to like what you guys do specifically in this stack, but it just feels like the shape of that gets really interesting and the velocity of data is like gonna explode.
- 25:16SPEAKER_02: Yeah, right.
- 25:17SPEAKER_02: And and I think how do you even deal with that?
- 25:19SPEAKER_02: It's like huge scale.
- 25:21SPEAKER_01: Actually, it's not only is it a like that level of magnitude is going to explode, but also the data that is produced, going back to your question, will have several uh orders of use, right?
- 25:36SPEAKER_01: Like the direct logs, right, will be useful for reinforcement learning or pattern checking or cluster analysis of, hey, like how are people behaving and you know what, how should the model um improve in certain certain directions?
- 25:52SPEAKER_01: Not only will it become also a signal for what data is missing, right?
- 25:56SPEAKER_01: Let's say a workflow is completed but it doesn't perform well.
- 26:00SPEAKER_01: That is a signal that that type of workflow needs more data to to improve.
- 26:04SPEAKER_01: So it's gonna like self-signal, okay, we need this type of data more, more.
- 26:08SPEAKER_01: For example, anthropic, you know, open and I are always looking for experts to uh comment and review and correct answers, right?
- 26:17SPEAKER_01: Uh whether it's financial models, whether it's legal opinions, whether it's like that that is very, very precious edge case, high expertise data.
- 26:26SPEAKER_01: And then there's also like the second-order effects of what that data represents, right?
- 26:32SPEAKER_01: So it's not only going to be useful to tell the model what data to source to get better at doing those types of tasks, but it could also be analyzed orthogonally, right?
- 26:44SPEAKER_01: A bit like how Avi was saying with Netflix, you sure, like it's interesting to see what people watch, but a hedge fund doesn't like may not care what people are watching.
- 26:53SPEAKER_01: They're just caring about, okay, are people watching more or less?
- 26:56SPEAKER_01: Or what's the what's the rate of growth?
- 26:58SPEAKER_01: Or like even something as tangential as, oh, like where is the majority of content that is watched being being produced, and should we like buy infrastructure in that country, right?
- 27:09SPEAKER_01: For example.
- 27:10SPEAKER_01: So like there's so many layers to data that um again, going back to your question, like agentic data, is that useful or not?
- 27:17SPEAKER_01: Absolutely, because any agentic created data, at the end of the day, is just a proxy for what the human wanted to do, anyways, right?
- 27:25SPEAKER_01: So it really does encapsulate human signal.
- 27:28SPEAKER_01: So yeah, absolutely.
- 27:29SPEAKER_02: I think that's a really interesting point.
- 27:31SPEAKER_02: The human signal that that that's like the boils it down really nicely is a data point you could extract out of it that's still kind of the what and not the why, Avi, to steal your your phrase.
- 27:43SPEAKER_02: So I I'm curious, what do you guys think the next six months in AI looks like?
- 27:47SPEAKER_02: What's what is what are we what's gonna happen that no one else is seeing that you're seeing?
- 27:52SPEAKER_02: Give me the alpha.
- 27:53SPEAKER_00: I I actually think what everyone thinks they're seeing is uh an increase of synthetic data usage and an increase of these world models and a lot of narratives that are pushed by the CEOs of some of these companies.
- 28:08SPEAKER_00: Um and the alpha here is that none of that is true, and all of it is still going to constantly rely on human data.
- 28:17SPEAKER_00: Um synthetic data is years, like multiple decades out from ever replacing human data to a consistent degree.
- 28:27SPEAKER_00: And you don't even have to look at this in terms of just taking my word for it.
- 28:32SPEAKER_00: You can look at the size of the teams at these companies.
- 28:35SPEAKER_00: OpenAI, for example, synthetic data team is much smaller than their human data team.
- 28:39SPEAKER_00: And their focus and spend towards human data is much, much higher, magnitudes higher than their synthetic data team, right?
- 28:46SPEAKER_00: And uh, we're like on a recursive loop of like three, like where like if you put like a dog into a like mid-journey or like one of these one of these models, and then you tell it to Joe generate me an image of a dog, and then you take that AI generated image and then you plug that back into the model, and you do that like a couple times over, this model just breaks and lose all concept of what a dog is, right?
- 29:07SPEAKER_00: That's still happening.
- 29:07SPEAKER_02: It's like an alien out the other side.
- 29:09SPEAKER_00: It's like something delivered in them.
- 29:11SPEAKER_00: It turns into like fluff, like uh you it has no idea what it's even looking at at some point.
- 29:15SPEAKER_02: What was that meme of like it was uh you know, it's like the prompt for the for the video one was like Will Smith eating pasta or something?
- 29:22SPEAKER_00: Yeah, yeah.
- 29:22SPEAKER_00: Um, so yeah, like you need to get that to the point where it's doing it hundreds of thousands of times over before it breaks, before synthetic data is like the thing that everyone relies on, and we're just not anywhere near there.
- 29:36SPEAKER_00: And even when we get there, even when we hit it, which is by the way, when we hit AGI, because then it becomes self uh a self-loop, then you still need human data because you need survey data on how humans are interacting with the world.
- 29:49SPEAKER_00: Um and and all this human data is what all the companies in the world are collecting and are selling, and that is the the layer we are building for.
- 29:57SPEAKER_02: I actually think synthetic is just not a layer that unless the unless the AGI decides to eliminate us from the equation.
- 30:05SPEAKER_00: Yeah, which by the way, yeah, like well it's either world ends or it's data.
- 30:09SPEAKER_00: It's like other here.
- 30:11SPEAKER_02: Yeah, I like I guess I like those odds.
- 30:13SPEAKER_02: Uh what about you, Andrea?
- 30:14SPEAKER_02: Like, what are you seeing in the future?
- 30:16SPEAKER_01: Yeah, well, it's hard because time dilation is a thing in AI, right?
- 30:21SPEAKER_01: Like, what could happen next week could be what could happen in six months, or it could be what could happen in five years.
- 30:26SPEAKER_01: Like, it really it really depends.
- 30:28SPEAKER_01: Like, you do remember in January or in February, it felt like the world was accelerating by 10x.
- 30:34SPEAKER_01: Every other day, there was like a new announcement.
- 30:36SPEAKER_01: It was just impossible to like do anything.
- 30:38SPEAKER_02: And then Open Claw went viral, and we were like, what happened?
- 30:40SPEAKER_02: The world changed.
- 30:41SPEAKER_01: Yeah, it was the craziest time.
- 30:43SPEAKER_01: Yeah.
- 30:44SPEAKER_01: But but I I do think there's there's several things happening in parallel.
- 30:49SPEAKER_01: So I'll start with the more exciting things, and I'll I'll go to the more boring things, but that are just as important.
- 30:56SPEAKER_01: So on the exciting things, I do feel like robotics, right, is going to slow, it's like it's reached, it's it's gonna have its inflection point.
- 31:07SPEAKER_01: Uh, it still hasn't had the Chat GPT moment yet, right?
- 31:10SPEAKER_01: And may that may not be in six months, it could be in one year, it could be in two years, but uh that that is the next race that everyone is you know trying to position themselves as good as possible for.
- 31:22SPEAKER_01: Because you could envision so many use cases for robots, right?
- 31:26SPEAKER_01: Uh right now, all of these workflows are somehow trapped behind a screen and very limited access to autonomy in the real world, unless it's through a human.
- 31:36SPEAKER_01: But in the future, it's really can be an energetic workflow, right?
- 31:42SPEAKER_01: Like you give a task and the robot can literally go to the DMV and sort things out, or it could literally go food shopping, right?
- 31:49SPEAKER_02: Please God, go to the DMV for me.
- 31:52SPEAKER_01: So I think that is like a macro thing that is going to really not only shape AI, but also shape the world and shape the data demands.
- 32:03SPEAKER_01: Um, there's gonna be more and more demand for you know POV footage, for edge cases, for various activities.
- 32:11SPEAKER_01: Right now, a lot of the data that's being sourced is really focused on like factory workers.
- 32:16SPEAKER_01: But if you can imagine everyone owning a robot for general purpose tasks, there are gonna be so many weird edge cases, right?
- 32:22SPEAKER_01: Because humans are very varied.
- 32:24SPEAKER_01: So that that's on like the okay, like where AI is heading.
- 32:27SPEAKER_01: I feel like that is like the next ChatGPT moment, and and we're still not there yet.
- 32:31SPEAKER_01: But yeah, every new video I see of robots doing something, I'm always more and more impressed.
- 32:35SPEAKER_01: The other big thing I I think is also the legislation.
- 32:39SPEAKER_01: And again, I started with the exciting stuff.
- 32:41SPEAKER_01: Maybe, you know, I'll I'll leave it up to I'll leave it up to the listener to uh to decide whether it's like positive exciting or negative exciting.
- 32:48SPEAKER_01: It's definitely exciting.
- 32:49SPEAKER_01: The boring stuff is more around the legislation, um, that tends to move a lot slower, definitely is not able to keep up with the pace of innovation.
- 32:57SPEAKER_01: And I know that the EU is the favorite continent to pick on uh when it comes to legislation and like you know, the US innovates and the EU regulates.
- 33:06SPEAKER_01: And I say that as someone who lives in Italy, right?
- 33:10SPEAKER_01: Um but the legislation will also, I think, be a tailwind for products like trace.
- 33:18SPEAKER_01: Because, as we mentioned earlier in the podcast, there is more and more demand from governments, from regulations to have order, right?
- 33:27SPEAKER_01: Like it's it's been the wild west.
- 33:29SPEAKER_01: It's been the wild west of data.
- 33:30SPEAKER_01: The internet has been scraped, Meta has been sued for billions of dollars for downloading every book ever written.
- 33:36SPEAKER_01: Anthropic now has a lawsuit for downloading every every book ever written for$1.5 billion.
- 33:41SPEAKER_01: Now the legislation is catching up to what has already been done, to what has already been done because they were like, it's there's not enough time to ask for permission.
- 33:50SPEAKER_01: We're just gonna ask for forgiveness.
- 33:51SPEAKER_01: And the legislation is gonna get more and more complex, more and more sophisticated, right?
- 33:55SPEAKER_01: And that will trickle down to we need to make sure that any data has some receipts, has some control, and and I think trace again, right place, right time, um, potentially even early.
- 34:07SPEAKER_02: Yeah.
- 34:07SPEAKER_02: I think that's that's part of what we're hoping with Corbett's too, because you know, we we handle compliance and governance and policy control in the OS, right?
- 34:14SPEAKER_02: At the at the at that level.
- 34:16SPEAKER_02: And like in Europe, especially, this is the the line is about to get crossed in August, right?
- 34:21SPEAKER_02: You have to have something in place that's been mandated that does governance and has identity and has all that stuff.
- 34:28SPEAKER_02: Yeah.
- 34:28SPEAKER_02: Um, and you can see like big investments going into that space um in terms of like enterprises building out these systems.
- 34:35SPEAKER_02: I think actually Europe is a dark horse for open source.
- 34:38SPEAKER_02: I think that that um, like I was talking to another founder the other day.
- 34:42SPEAKER_02: There is a lack of talent, you know, in AI for like training.
- 34:44SPEAKER_02: I think there's a big opportunity for like a European startup ecosystem to get built just around training models and data centers, also.
- 34:53SPEAKER_02: Like they don't have enough data centers, apparently.
- 34:55SPEAKER_02: But I do think that that that that the need for sovereign AI is big there because of these regulations and because they're all starting to realize that like if they outsource it to the US, they don't they're kind of bent over a barrel, you know, like they don't have control.
- 35:11SPEAKER_02: And I just find that interesting.
- 35:13SPEAKER_02: Like the discourse between like the China models, these open source models, and that's been happening like the last couple weeks and the American ones, and like look, they shouldn't be doing distilling, like that's just classic, you know, theft of IP.
- 35:26SPEAKER_02: But one could also argue that, like, to your point, there was a lot of theft of IP to get the American models where they are too.
- 35:32SPEAKER_02: So, you know, I don't know what you want to do about that.
- 35:36SPEAKER_02: But it's interesting for me as a guy who like believes in open source software to see Europe going down that route and think, like, oh yeah, maybe that's gonna be like a bastion in the West for open source.
- 35:45SPEAKER_02: We're at the top of the show, and I always ask this what have I not asked you that I should have asked or that you guys wanted to talk about?
- 35:50SPEAKER_00: We covered a lot.
- 35:51SPEAKER_00: We we I think we actually hit most most of the things that are good to touch on.
- 35:56SPEAKER_00: I think look, growth is going to be a growth is gonna be like one of these where once we hit this inflection point of like fraud getting good enough, uh, like at least on the CLED side, we can then start just like super hardcore ramping up revenue.
- 36:12SPEAKER_00: Um and like we're already profitable, but I think like these data businesses get really big really fast, and I'm excited for that.
- 36:18SPEAKER_00: And simultaneously, on the trace side, once CLED um and uh any of the other ecosystem apps that are on there have proven that they can become just unicorns, which I promise you I will this year or very early next year on my mother.
- 36:37SPEAKER_00: I will do that.
- 36:38SPEAKER_02: I let's go on fire.
- 36:40SPEAKER_00: I love it.
- 36:41SPEAKER_00: Easier for for other people to to to get into the pace of just like also follow like I I I want to be first at everything I do.
- 36:49SPEAKER_00: Um and uh I think the fact that I have the privilege of working with uh the foundation here um to build this um and being first here as a company to be on the protocol means that when the thousands of other data marketplace start copying us because this will become standard practice, um we will be known as first.
- 37:10SPEAKER_00: Um so yeah, I look, I once we start selling to all the foundation labs, and in it is a part of their daily process to go to trace and verify and audit the data.
- 37:20SPEAKER_00: Um, that is going to spread like wildfire in the valley.
- 37:24SPEAKER_00: And it's gonna become really easy to convince anyone to start going under this protocol and spending the money and making it worth their time simply because CLED has just done it first.
- 37:35SPEAKER_00: So that's my job for the next couple of months.
- 37:37SPEAKER_02: Very cool.
- 37:38SPEAKER_01: Yeah, on my end, I feel like we haven't really touched on numbers, and I it would be good for the listeners to have an idea of what we're talking about.
- 37:46SPEAKER_01: Like, and I and I'll I'll start by you know shilling all of CLED's progress.
- 37:50SPEAKER_01: So uh CLED recently reached 500,000 contributors worldwide.
- 37:54SPEAKER_01: Cool.
- 37:55SPEAKER_01: And has already quite some time ago passed 1 billion data contributions, right?
- 38:01SPEAKER_01: Which is a massive, massive number.
- 38:04SPEAKER_01: On trace, we're ingesting all new data contributions going forward, and we're also processing the backlog, and we're now at 100 million data receipts and counting, uh, which for us is like the top line, you know, progress KPI.
- 38:20SPEAKER_01: And and yeah, as Avi said, I I feel like we're we're still early in this growth curve in general for crowdsourced data.
- 38:28SPEAKER_01: Um the more AI becomes competent, the more AI becomes autonomous, self, you know, the more that you can lean on it, the more it will like we will start to shift between commanding it and just providing it context.
- 38:44SPEAKER_01: And a lot of that context is really just data.
- 38:47SPEAKER_01: So having efficient ways to source data for different tasks, whether it's an agent requesting it, whether it's a human requesting it, is going to be very, very, very important.
- 38:56SPEAKER_01: And and having that, you know, not only marketplace like CLED is, but also that trace, that trail of okay, who contributed what data, right?
- 39:05SPEAKER_01: Going back to what you were saying, Alex, around if different people contribute to uh to a workflow, then they should, in the right proportions, get some contribution back.
- 39:17SPEAKER_01: That that's gonna happen at at a broader scale around like, okay, how much are you contributing to to AI in terms of what data you contribute, in terms of how you use it, what you teach it.
- 39:27SPEAKER_01: Yeah, I I I just feel like we're we're still early in this transformation, right?
- 39:31SPEAKER_01: And and data is gonna be at the core of everything.
- 39:34SPEAKER_02: I wonder if there's a really cool model around open source too, here where like I'm just keeping thinking like OpenClaw and like Hermes agents, you know, and like getting a cle a CLED plugin in there, you know, so like I'm running my my thing and then you guys are getting the data and I'm making some money off of it, it's like covering my inference cost.
- 39:50SPEAKER_02: Or like, yeah, you know, like because there's gotta be just a massive amount.
- 39:53SPEAKER_02: Think about that.
- 39:54SPEAKER_00: There's a play on both the buy and sell side here.
- 39:56SPEAKER_00: We're actually like the end result of all this fraud stuff being all set up and done is that we can build an enterprise like we already actually have it built, um, but an enterprise portal where any lab, for example, or any uh any any agent, for example, um can just go on there, uh request a data set, wire the money up front, and then that gets streamed straight into the app with all the proper instructions and the quality detection in place so that any data that's uploaded is clean, and then you can just train off of that data immediately.
- 40:26SPEAKER_02: Yeah, you would be crazy is to like have a have like an agent train a model for me.
- 40:31SPEAKER_02: Yeah.
- 40:32SPEAKER_02: You know, like actually just on a loop, just start training a model on a specific thing and pay for that autonomously.
- 40:38SPEAKER_02: I mean, you could do that with there's like you know, MPP and X42, you could like try to create like a an experiment for that.
- 40:44SPEAKER_02: Um I don't think like I don't think the market's there yet for that, but it's that that's those are the rails, right?
- 40:48SPEAKER_02: Like long term that I think could get used.
- 40:51SPEAKER_00: It's just like robotics, like robotics is gonna take uh like another year or two to like get to a point where egocentric data, for example, um that is like semi-disorganized.
- 41:01SPEAKER_00: Um right now the expectation for egocentric data is that it has to be perfect to train on, and you just can't do that across thousands and thousands of users with just an iPhone.
- 41:11SPEAKER_00: Yeah, but we'll get to a point where the models are training well enough that like you can just do it on an iPhone.
- 41:16SPEAKER_00: I think it's gonna take like two years.
- 41:18SPEAKER_00: But similar thing here where I think agents are gonna take like two years to get to the point where they can transact properly on things like this and train their own models.
- 41:26SPEAKER_00: But we will get there.
- 41:27SPEAKER_00: There's infinite possibilities.
- 41:28SPEAKER_02: Well, that the technology is there.
- 41:30SPEAKER_02: Like, I mean, literally, my company we spent like the last year building in a framework for agency commerce called Fairmeter, and it's like agnostic in terms of the standards.
- 41:37SPEAKER_02: And like we powered probably almost 60% of the market, most of it on Solana.
- 41:42SPEAKER_02: Um, but the market's tiny, you know, it just wasn't a place where we were able to generate revenue in the near term.
- 41:47SPEAKER_02: And so, you know, thus the agentic operating system that we're doing, because that's what like customers were willing to pay us for right now, and where we could get value, like capture value as a company.
- 41:56SPEAKER_02: But I but this is CLED as a marketplace for data for agents that are like permissionlessly able to just pay for it on these rails, is like the first thing that I've thought that I've seen that like actually makes any sense for specifically for its own inference, right?
- 42:12SPEAKER_02: Like I'm gonna train a model so that my agent can get better actions out of this specific data set.
- 42:19SPEAKER_02: Yep.
- 42:19SPEAKER_02: Um, and I think that makes a ton of sense.
- 42:21SPEAKER_00: Yeah, yeah.
- 42:22SPEAKER_00: I I I very much see a future where like this is why we're building the enterprise portal.
- 42:26SPEAKER_00: I think I think you could just go in and they request something, and then you have this app that's just like collecting it from thousands of people at your command, and it becomes pay-to-play.
- 42:35SPEAKER_00: Like if you're not paying enough money, for example, then that the the task isn't as enticing to users, and then you just figure it out.
- 42:42SPEAKER_00: Um, and then simultaneously you could have on the sell side agents selling their own data, for example.
- 42:47SPEAKER_00: Um, and then you can pay for your own agents.
- 42:50SPEAKER_00: Yeah.
- 42:50SPEAKER_02: Well, thank you so much, guys, for coming on the show.
- 42:52SPEAKER_02: It's been fascinating, and I would love to check in with you guys again in like six months and see if any of these predictions we talked about came out.
- 42:58SPEAKER_00: Yeah.
- 42:58SPEAKER_00: So thanks for having us.
- 43:00SPEAKER_00: Thanks for having us, man.
- 43:01SPEAKER_00: Yeah, it was great.