Avi Patel: Building KLED, Selling Data, Buybacks and More | TG Podcast
KLED has moved from promise to consumer traction, with 25,000 app downloads in under four days and close to $1 million paid to data contributors. Avi Patel expects one million downloads by year-end. Thread Guy argued that surviving an exceptionally hostile token launch and continuing to ship can ultimately earn a project and its holders credibility: “You survive it, and then you sit down and you still ship… then you’ll get rewarded for it.”
The core business is an enterprise-driven market for legally licensed, highly specific human data—not a flat-price repository of generic camera-roll content. A dancing video might command a premium from Epic Games for AI-generated emotes but only a nominal rate from Microsoft Azor; KLED therefore wants to become “the biggest data collection engine,” then match each dataset to the buyer with the strongest need.
Superpower’s $1 million LOI is KLED’s clearest proof point for this demand model. Avi described the roughly $300 million healthcare company as intending to allocate that amount toward datasets for an in-house healthcare LLM, with KLED proposing anonymized integrations into portals where patients communicate with doctors. Avi’s pitch is that rare consumer-sourced conversations cannot simply be obtained elsewhere.
Avi said a third of relevant enterprise proceeds goes toward token buybacks, but explicitly called current purchases “nominal buy pressure that means nothing” beside actual execution. Roughly 10 million tokens have already been removed from circulation; his hypothetical is that the mechanism becomes consequential at much larger scale—for example, a $1 billion contract would direct $333 million to buybacks.
The durable AI bottleneck, as framed by Thread Guy and endorsed by Avi, is authentic human data because models cannot endlessly train on their own synthetic output without degradation. Thread Guy called this “AI inbreeding”: repeatedly recycling generated golden-retriever images produces progressively less faithful dogs, while sourcing real images may also cost less compute. KLED says its validation stack checks duplicates, internet plagiarism, EXIF metadata and AI generation, reporting about 95% detection effectiveness and aiming for “literal 99% certainty” that delivered data came from a real user’s device.
Avi sees the token less as an enterprise product requirement than as a distribution and fundraising accelerant. Users do not need KLED to use the app, while green candles, referral activity and millions of impressions create measurable FOMO among creators and Silicon Valley investors. His thesis: “Tokenized startups at early stage will always beat untokenized ones,” though he believes crypto needs several genuine success stories before the stigma disappears.
KLED’s next growth loop combines controlled onboarding with gamified data collection rather than opening an economically unsustainable floodgate. Avi is tripling each waitlist batch as contracts grow and previewed KLED Games: timed prompts, potentially backed by a $50,000 prize pool, would ask users to capture specified real-world images while computer vision verifies submissions. Near term, he plans to spend 15 days pursuing larger San Francisco contracts, ship the game, expand PR and pursue centralized-exchange listings.
1. KLED is trying to replace launch-era skepticism with shipped product
Avi introduced KLED as the data-licensing next vertical to Nitrily, his music-licensing marketplace serving hundreds of thousands of artists. KLED lets people license camera-roll and personal data to AI and technology companies, then returns part of the proceeds directly to those contributors.
Thread Guy described Kled as “probably the most flooded token ever in ICM” and said surviving that early period had hardened him. Avi separately said the team kept building through what he called a nightmare: over four months, KLED published roughly 200 updates and delivered an app that would ordinarily represent “a year’s worth of work minimum.”
The launch produced 25,000 downloads in less than four days, against a stated target of one million by year-end. Avi also cited close to $1 million in user payouts, creators backed by around 500 million followers and meetings that week with OpenAI, Anthropic and Google, while saying he had also been signing LOIs with major labs.
The lean operating structure is part of Avi’s explanation for speed: Eli and two other engineers handle engineering, with an AI engineer planned, alongside a design lead, a 3D creative lead, a UGC operation and what Avi called a “fat legal team.”
2. Data becomes valuable when it answers a buyer’s narrow training need
Avi’s framing: there is no universal price for an image or video because value follows the enterprise use case. TikTok-style dancing footage could be premium input for an Epic Games system generating Fortnite emotes, while the same footage might have little special value to Microsoft Azor.
Baseline quality still matters—resolution, useful text, clean images and accurate labels—but scarcity has shifted from having “a lot of data” to possessing specific data unavailable elsewhere. “Everything’s valuable, like literally all of it,” Avi argued, provided KLED can classify it and find the buyer whose model needs it.
The host’s practical question—why not submit his own streams?—produced a concrete rate: Avi said roughly 100GB of high-quality video could earn about $1,000, depending on volume and demand. A well-shot, free-flowing podcast could train conversational or podcasting systems, while location-specific dashcam footage serves a wholly different buyer.
Enterprise acquisition currently comes through investor introductions, online attention and inbound requests from both data owners and data buyers. Avi argued that crypto Twitter’s visibility carries into Silicon Valley Twitter: “The hype online is literally the same thing for Silicon Valley.”
3. Superpower illustrates the case for anonymized, private-domain data
Avi described Superpower as a roughly $300 million healthcare company working toward an AI “super doctor” trained on real exchanges between patients and leading clinicians. The missing ingredient is not public medical prose but data showing how people actually communicate with doctors.
KLED’s proposed role is to integrate with healthcare portals where people already talk to their doctors, anonymize the data and provide it to Superpower. Avi stressed privacy as essential: Superpower would receive useful conversational patterns without knowing the identity behind a record.
The announced $1 million LOI means, in Avi’s explanation, that Superpower would allocate $1 million toward buying datasets; it is an allocation framework rather than a completed purchase. The larger thesis is repeatability: if KLED can prove that a million users will provide requested data, it can build bespoke integrations for many enterprise contracts.
Legality is central to the sales pitch. Avi argued that labs sometimes source data illegally or from wherever they can find it, exposing themselves to enormous future claims; KLED instead aims to offer “the most legally licensed way possible,” making clean provenance valuable both operationally and during fundraising.
4. Contract scale drives both user economics and the token flywheel
Avi said a third of relevant contract proceeds goes into token buybacks. He said approximately 10 million tokens had already left circulation, with execution including Jupiter and Phantom Swap, and emphasized that purchases could occur at random times rather than on a predictable schedule.
His concession was unusually direct: present-day buybacks are “literally meaningless” compared with founders shipping useful products. The mechanism matters at scale, not because a small market buy creates a temporary candle; his hypothetical endpoint was a $1 billion data contract producing $333 million of buybacks.
In a later exchange, Thread Guy referred to Avi putting $33,000 into the chart. Avi said there might eventually be a more effective process, but confirmed that KLED had used Jupiter and Phantom directly.
Asked whether eight- or nine-figure contracts were realistic, Avi said they occur routinely and claimed Whimo had bought “I think $4 billion” of dashcam footage. The host pressed him on the source of that data, and Avi admitted, “I actually don’t know the buyer for this one,” while maintaining that autonomous-driving systems require enormous, area-specific datasets.
His stronger concrete logic was geographic: expanding into Los Angeles requires footage from Los Angeles roads, and KLED says it already has roughly 4,000 Uber drivers there uploading. Avi described KLED’s own contract progression as approximately $50,000, then $350,000, then a $1 million LOI.
5. Authenticity becomes more valuable as synthetic media pollutes training sets
Discussing Sora 2, Avi called it “really, really good” but argued that better compute applied to broad public datasets is only one stage. KLED’s enterprise portal could instead surface extremely narrow material—his example was thousands of gymnastics videos uploaded by a coach—for the next model’s specific weaknesses.
Avi first offered the thesis that labs “slopifying” social media and the internet could increase the value of provably human material. Thread Guy then supplied the “AI inbreeding” explanation: train a model on generated golden-retriever images, feed those outputs back into training, and each generation may drift further from a real dog. He argued that real data may also be more sustainable than repeatedly generating synthetic examples.
KLED says its ingestion system detects duplicates across accounts, searches the internet for plagiarism, inspects EXIF metadata for origin and alteration, and uses Mistral to flag AI-generated files. Avi put current effectiveness at about 95% and the delivered dataset’s target assurance near 99%.
For KLED Games, validation would also be capture-first, “like BeReal,” rather than letting players upload arbitrary saved files. Someone asked to photograph nearby cars would use the live camera, closing the obvious loophole of downloading ten car images from Google.
6. The token is a consumer-acquisition and investor-FOMO mechanism
Users do not need the token to access KLED, although Avi expects future token-locking features. The bigger near-term obstacle is crypto usability: he claimed to have personally walked perhaps 10,000 newcomers through SOL and Phantom, while KLED works toward embedded wallets for non-crypto-native users.
When the host asked whether creators were being paid to promote the app, Avi said the vast majority were friends in his network rather than paid placements; he cited one Instagram Story receiving about 100,000 impressions from an account with roughly 30,000 followers. Future UGC opportunities remain part of those relationships.
Avi said the token’s chart acts as a FOMO benchmark for venture investors. He described investors watching candles, sending messages after price increases and seeking allocation or offering introductions—even though company investors do not receive tokens. He also said the token had become worth slightly more than the company’s current roughly $40 million valuation and referenced a planned $100 million raise.
His broader claim is that KLED accumulated traffic in the millions, 30 million Twitter impressions and rapid app adoption partly because the company and token community grew together from inception. An already-established company launching a token later might receive less incremental benefit: “Tokenized startups at early stage will always beat untokenized ones.”
7. KLED Games and relentless delivery are the next attempt to widen the moat
Avi is throttling waitlist admission because admitting more suppliers than contract revenue can support would underpay contributors. Each batch is being tripled, while referral codes both advance users in line and keep KLED circulating daily across social platforms.
KLED Games would open a parallel path: a hypothetical $50,000 contest sends prompts every 30 minutes—such as capturing 20 local car images—with only seconds to respond. Everyone supplies data, the last surviving player takes the main prize, perhaps 100 others receive smaller rewards, and the frantic real-world participation creates viral clips.
The host’s AI-talent thesis led Avi to say there are “literally no good AI projects on chain” beyond a small handful, call most products vaporware and argue that 99% of ICM projects have become “coming soon farms.” Only repeated delivery, he said, will attract serious traditional AI and Silicon Valley builders back into crypto.
On competition with Launchcoin, Avi denied that its valuation creates KLED’s ceiling and said he no longer wants to farm drama: “Putting out an article with whatever big lab for a billion dollars—that’s engagement farming.” His next 15 days are focused on San Francisco contracts, followed by KLED Games, broader press and outreach for centralized-exchange listings.
Full transcript
How are you, twin?
Been good. How are you?
I've been good, man. I feel like it's been a long time coming, dude.
Yeah, we've known each other for, what, like three months now? It's been a long time coming, for sure. I feel like you could have had me on Crash Out number 2, like weeks and weeks ago.
I was a little afraid of you, dude, honestly, to bring you on for mid-Crash Out. I didn't know what to expect, you know?
Yeah. Well, dude, I'm here now. I'm happy to talk about anything, whatever you want to. I feel like you've calmed down a little bit too.
Maybe. Dude, you know what it is? I think I've just been hardened. I think what happened, which I think is actually really important for all the ICM founders, is that if you just sit down and launch, and you go through—I think Kled was probably the most flooded token ever in ICM. I think I can comfortably make that statement.
But if you go through that and then you survive it, and you sit down and you still ship and actually prove on your promises, then you'll get rewarded for it. So I've just been hardened through that early stage, which was just a nightmare, dude. Honestly.
Kled holders have been through it, bro.
They've been through it. Yeah, which is great because they have the diamond hands. These guys are wizards. Seriously, they just don't know what the sell button is.
Yeah. You've mellowed out. I like it. I like this new arc that you're on.
Look, I figured it was a good time to bring you on now. I'm just going to be honest: the chart—I was like, "Yeah, let's do it right now." You have some announcements, things are going well, the market's kind of turning, and people are looking for more risk-on stuff to bid. They're looking at alts and Kled.
I don't know. You've been here grinding and locked in. Do you want to start with a quick introduction into who you are and where we are with Kled? I think most people here are familiar with you.
Yeah, I'm happy to. My name is Avi. I am the founder of Kled. I've owned businesses for the last three years, but I primarily work in the intellectual property world.
I own Nitrily, which is a music licensing marketplace that works with hundreds of thousands of artists. It basically lets you license any song out there.
Kled is our next vertical, which is data licensing. It basically lets you sell your camera roll and your personal data to the largest AI companies and tech companies that need it, and then we pay the users directly.
What started this whole run is that, for months and months and months, we've just been proving that the model works. We've paid out close to $1 million to users so far.
More importantly, we launched the mobile app less than 4 days ago, and it just took off. We're at 25,000 downloads now, which I haven't announced. We should hit 1 million by the end of the year.
We're going viral on most platforms. We're backed by creators with around 500 million followers, and we're backed by some of the biggest venture capitalists in Silicon Valley, like K5 Global, who's invested in X, Uber, and you name it.
Now I'm in this run in Silicon Valley. I'm in San Francisco right now meeting with all the big labs. I have OpenAI this week, Anthropic this week, and Google this week, and I've been signing LOIs with a lot of big labs.
I announced one of them today, which was Superpower. They're building a $300 million healthcare company that's going to be using some of our training data to build their own in-house healthcare LLM.
I should just be sticking it out. I'm excited to see what the rest of the week looks like.
Dude, you know, every time you name-drop some big lab or someone backing you, you get in trouble, right? It never goes well. Every time you—
I think at the start it was problematic because literally none of this was announced, right? But we went for three months where we just kept saying all these things, and now finally I get to show it all to everyone. Now it's real, which is the good part.
Yeah. Advisor list, all that. When did you launch the app, and why did you launch it? 25,000 downloads in 4 days feels like a lot.
Yeah, it's a lot, especially considering the fact that it's a pretty crazy concept, right? Selling your personal data is not something everyone's open to, but we just made it very friendly for everyone.
We launched the app literally less than 4 days ago. It was just a really hard app to build because the UI is probably one of the cleanest UIs possible. We've made it really user-friendly.
We've sped up pretty much everything. Data storage and labeling data are just not the easiest things to do, especially at scale. We host petabytes of data from users, and that's what took the longest to crack down on.
My team's also just a bunch of perfectionists. They literally refuse to launch anything that's not perfect and up to my standards, definitely.
So we got it out. It took some time, but the market clearly reacted positively toward it.
You said you've paid out $1 million to people providing the data. Are people actually using this thing who aren't Kled holders? Are there random consumers who are actually interacting with it?
I've been going hard on marketing on Instagram and everything. We have a ton of big creators posting videos about Kled, like Noah Boat with 10 million followers and Kios with [__] 20 million followers.
Are they paid videos?
No, these are just friends of mine. I text them and tell them, "Hey, help us out here, and in the future we can help with UGC opportunities."
The vast majority of it is just friends in a network. They've made a ton of videos, and then it's gone viral on my personal Instagram as well. One of my stories had 100,000 impressions about how we're paying out for camera rolls, and I have around 30,000 followers. That's just an Instagram story, so we've had a ton of inbound from there.
Now we're basically trying to make it frictionless so that users have wallets in-house on the app. You don't have to deal with anything complicated, so if you're new and not crypto-native, you can still use the app very easily.
So far, I've personally probably onboarded around 10,000 people to SOL or something crazy. I've sat down and walked normal people, big investors, and big creators through how to make a Phantom wallet and things like that.
That's what we're doing right now. We're going to make it really, really friendly for most people.
Do you need the token to use the app or the product?
No, but we're going to have token-lock features in the future, which will be nice.
How do you value different data and content? How do you decide what's worth more than others? What do you put the most emphasis on?
It depends on the enterprise contracts we get. Let's say, for example, Epic Games comes to me and says, "I need videos of TikToks of people dancing so I can train an AI that will make emotes for Fortnite game characters," something like that. They'll pay a lot more for that because there's a premium and a need for it.
Versus if you sell that same dataset to Microsoft Azor, they have literally no need for that, so they'll just buy it for a nominal rate.
It depends on the enterprise contracts, which is why our goal is just to become the biggest data-collection engine. Depending on what these opportunities look like in the future, we can position this data and sell it to whoever accordingly.
Overall, higher-quality videos are always good. Higher-quality text documents and higher-quality images matter, and then it's all on us to label everything properly.
In reality, no matter what a picture or video is, they all have very specific use cases for training. These big labs are at the point where it's no longer about positioning yourself to have a lot of data; it's about positioning yourself to have specific data.
You can plug your data into any of these plays and print a significant amount of money, then pay that out to users accordingly. Everything's valuable—literally all of it, no matter what it is.
How do you get these enterprise contracts?
It's all networks. Our investors have thankfully made a lot of introductions. I'm in San Francisco right now, and these labs introduce themselves to you.
We have a lot of hype on Twitter, which people don't understand also drives crypto Twitter and Silicon Valley Twitter.
I was hit up by Nick Shvchenko this morning saying, “Look, I wanted to do something with OMI and get data to you guys.” I was speaking with three a16z partners yesterday. The hype online is literally the same thing for Silicon Valley.
We get a lot of cold inbound from people who want to monetize data—people who not only own data but also need it. It’s really not easy to get data. The internet has basically been scraped; everyone has taken data off the internet at this point, and these consumer data sets that are sourced naturally are very rare because they’re not available anywhere.
We’re giving you something competitive compared to anyone else. If Google has our data set and Anthropic doesn’t, Google is winning because they can tell their investors, “Hey, we have rarer data than Anthropic.” That means we can train our models better than Anthropic, as long as our compute is the same. That’s how it goes.
Why am I not submitting my streams and my videos? Is that—
That’s a great question.
What’s the Rasper [?] thing?
Yeah, Rasper [?] does it. We pay him tomorrow, I think.
How much are you paying him?
I don’t know the number off the top of my head. Eli has it all in-house. It’s literally a system where they check every part of the video, and I’ll get it straight into enterprise contracts.
Depending on how much he uploads, if you upload 100 gigabytes, you get paid $1,000 for high-quality content. Think about that for people in third-world countries. A lot of our users are based in India and other places that aren’t necessarily the best.
These guys have solid iPhones and will spam pictures all day long, record dashcam footage, and get paid rates that American people don’t even get paid. It’s pretty lucrative if you’re recording all day long and just uploading.
Hey, Rasma [?], you’re in the chat. Tell us how much you’re about to get paid.
What do you classify as high quality? Would this count as high quality?
This would definitely count as high quality. You’re also doing a lot of conversational things, so we’d literally plug this into someone looking to train podcasting models or conversational models. That’s what this is for.
Someone building humanoid AI agents could use it. There are a hundred different use cases. Yours is definitely high quality. It’s really nice camera quality, and the conversation is always free-flowing. You can plug that into pretty much anything.
Damn. I’ll let you know when Rasper [?] says how much. That’s awesome, though.
Tell me about Superpower, the company you just announced a partnership with. What is that?
Superpower is a $300 million healthcare company. They’re very big. Their whole gist is basically to give consumers the best healthcare experience possible, and a large part of that is letting consumers have access to the best doctors in the world at any point in time.
Their whole intention is that, at some point in the future, they’re going to have to train their own LLM. You’ll open your phone, go on the Superpower app, and type to any doctor you want. Not any doctor, but an AI agent trained on conversational models of people talking to their doctors.
You’re basically trying to create this super doctor that you can talk to and get advice from at any point in time. What that requires is data on how people communicate with top professionals and top doctors, and there’s no way to source that other than from consumers because no one has access to any of that.
What we’ll do in-house is set up integrations with these healthcare portals where people already talk to their doctors. Then we’ll make all the data anonymous so that you won’t know who it is. Privacy is obviously very important.
We anonymize the whole thing and provide it straight to Superpower to train their LLM. That’s what their data needs are for.
There are a ton of these use cases, but Silicon Valley is seeing that Kled is really good at sourcing data very quickly for very specific things because we have this big community of people who are willing to do it.
As long as we keep proving that there’s a big community of people—a million users, for example—who are willing to give whatever data they want, we can get contracts for anything. We can provide that integration to our users, they can supply the data, and these big companies can pay us for it.
The announcement said you received a $1 million LOI. What does that practically mean?
Basically, they’ll allocate $1 million toward buying data sets. We have a pretty interesting flywheel where a third of that goes into buybacks for the token.
Even though these nominal buybacks don’t matter at this stage, they do matter.
Nice little bag.
Yeah, it’s a nice little bag.
Yeah, that’s not nothing. That’s definitely not nothing.
Think about when we get a billion-dollar contract. Let’s say this keeps going, which it will. If we get a billion-dollar contract, a third of that going toward buybacks is pretty big.
It doesn’t matter at this stage, but with any buybacks, I hope everyone understands this: the flywheel at the stage that all ICM companies are at is literally meaningless. It’s nominal buy pressure that means nothing.
It’s way more important that the founders are shipping really good updates and getting all these people, but we have the flywheel in place. It’s all already set up. We’ve taken 10 million tokens out of circulation with it.
This matters a lot more when we get to that billion-dollar contract and we’re buying back $333 million of tokens.
How realistic is it to scale to get these? How realistic is an 8-figure contract or a 9-figure contract? Do these already happen with other companies, or is this uncharted territory?
All the time. Whimo just bought, I think, $4 billion of dashcam footage.
Whymo bought $4 billion of dashcam footage?
Yeah. These contracts are massive, and they happen every single day. AI has grown. I have it in our white paper: AI is the most explosive industry in the world right now.
The only things that improve AI are compute and data. Basically, all your spending from VC and revenue is going toward compute and data. Data is simply our game. That’s what we’re best at.
We’ll hopefully be signing some of the biggest dashcam providers and getting all this in-house from enterprises and consumers. These contracts are pretty realistic. We started with a $50,000 contract, and we’re already at $350,000. It’s been pretty parabolic up until now.
Now we’re at $1 million, so these contracts just get bigger and bigger over time. I’m going to keep stacking as many of them as possible, and that goes straight to users.
Is that where Waymo spends most of its money? They spent $4 million. Do they just need unlimited dashcam footage?
That’s how they have to do it. It’s area-specific, too. Let’s say they want to expand to Los Angeles; they need a bunch of Uber drivers in Los Angeles, which I just so happen to have. I have 4,000 of them uploading to Kled.
That data can be bought. The same thing applies if they want to expand to some rural town. They need people recording themselves driving on the streets all day long so they can train their cars not to crash. That would be bad.
Data is really important for them.
Who did they buy from? Who did they give $4 million to? Who sourced all that data?
I actually don’t know the buyer for this one, but traditionally, over time, they’ll partner with Tesla. They’ll partner with dashcam companies. They’ll partner with people who have uploaded to public drives. Whatever it is, it’s dog-eat-dog. They’ll get it from anything. They don’t care where it’s from; they just need to get it somehow.
Sometimes they’ll even do it illegally. That’s the problem, and it’s what’s causing all these companies to face big lawsuits.
The big differentiator here is that Kled is providing legal data. You’ll get it in the most legally licensed way possible, and you won’t get sued for $10 billion later, which is really important to these companies because it’s a value proposition when they’re raising money.
It’s also a value proposition because they’re not getting sued for $20 billion by some random company, like Disney suing OpenAI.
Billions and millions of dollars of losses are flying left and right every single day, basically.
Okay, so you're putting $33K into the chart. What's your process? You're just on-ramping it into crypto and swapping the coin, basically?
I'm sure there's a more effective way of doing that at some point in time.
Just throw it in a Phantom wallet and raw-dog it.
That's basically it. Yeah, that is how it goes.
Whenever you feel like it, too.
Yeah, whenever. You just never know when everything goes. You'll never know, and that's what I want, right? I want our holders to be in here because they have conviction, not because they're timing certain events.
I'll do crazy random shit that's really going to impact the company's chart because it's impactful to the company, and I'll do it at random times. We have a Kled holders group chat that everyone's pretty aware of. It's not on Telegram, and it's not private. Anyone can join it, basically.
In there, I literally say, "Be on your toes. I'm literally just going to drop shit at any point in time, any time of the day, and you'll just never know." It might just—you get my point. But yeah.
Yeah. Are you on Jupiter, or are you just raw-dogging Phantom Swap?
We've done Jupiter, and I've raw-dogged it as well.
That's awesome.
Yeah.
That is awesome. This is off topic, but what do you think of Sora 2? Another reason I wanted to bring you on right now is that Sora 2 just launched, and it's pretty crazy. What do you think of that?
Sora 2. Yeah.
Yeah, Sora 2. Yeah, yeah.
Yeah, it's really good. It's really, really good, right? But again, right now they're competing with Google pretty heavily, and Google's really kosher about how they license data. OpenAI currently is not, which is why their data needs are very, very high right now.
They're either fully complicit in crime, which kind of seems like they are, but even if they're not, they still need more data. Sora was trained off some of the biggest public datasets in the world, and all they've really done is improve compute.
I can only imagine how good it's going to be once we start introducing very specific, trainable datasets to Sora. If you go on our enterprise portal right now and search, "I need 5,000 videos of a girl on a high beam doing gymnastics," we literally have that there.
There's a user who uploaded thousands of videos of his—he's a gymnastics coach. I'm sure he's going to reply to this—and he's uploaded all of his stuff. We have thousands of videos of gymnastics, for example.
If Sora is going to try to improve its next model, it can come to us with very, very specific things to train on and help itself out, right? Instead of training on these general datasets, you're now really specifying things.
If someone types into Sora, "I need a gymnast doing the typical thing," and they need to build an LLM that's training on all of that at the same time, it gets very complex. Data is fundamental to why Sora is getting better, and data and compute are fundamental to why it's getting better.
Also, maybe this is cope, but I have this thesis that as all these labs spend trillions of dollars slopifying Twitter, social media platforms, and the internet more broadly, there's an increased value on—
Human, exactly. Provably human stuff like this. Anyone not watching, that's a scientific concept. It's called AI inbreeding.
If you have a picture of a dog and ask ChatGPT to generate an image of a golden retriever, it generates 20 images, and then you plug those 20 images back into ChatGPT and say, "Train off these images that you just generated," the next time you generate an image of a dog, it's going to look less like a dog. It's just going to keep happening over and over again.
The need for natural images that are very real is always going to be there. You can't solve inbreeding. It's not a solvable concept, right?
Data is always going to be needed. Even if you solve inbreeding to some extent, generating those images is so compute-intensive and difficult that it's more sustainable, cost-wise, to get real user data instead.
That's at the heart of all of this. You can't generate it with AI. You need it from normal human people, and the slop platforms are not going to plug their own slop back in. They need real stuff to plug back in.
So when you go talk to these serious AI people in San Francisco about what you're doing, and then you say the word crypto, do they get soft? Is the narrative changing around that? What's the vibe?
You don't have to say crypto, right? My holders don't care if the labs are interested in my token. The holders care that the labs are giving me a billion dollars to buy data. That's all they care about, right? You don't even tell people you're working in crypto.
Yeah, I don't have to. You know what? The best part is that sometimes they find out, right? I had a founder a couple of days ago say, "Holy shit, you just put out this announcement, and now your company's valuation is—basically, your crypto coin is now worth more than your company's valuation, right?"
We'll be raising $100 million, but at our current $40 million valuation, the coin's worth a little bit more, which makes sense, obviously, for the big raise that's coming up. That's really interesting to them.
You'll have people asking for allocation so they can then incentivize me by saying they'll make introductions, et cetera. But I think that process has to happen naturally.
I think crypto will be fully destigmatized if there are Kled and 5 other success stories. Right now, it seems like there's just Kled and maybe a couple more, but it needs to happen naturally. You can't force it.
I'm not going to mention it to these guys. I'd much rather just get these big market buys, you know.
Okay. Well, you definitely mention it to investors who are investing in the company. Do they feel weird about the fact that there's this liquid token that's worth more than the company's current valuation? Are they getting tokens? Is that taken into account? I was always curious how that works.
They don't get tokens. They just get FOMOed like crazy. It's a massive FOMO mechanism.
Right before the stream, I was with Keenan from K5 Global, and the way he literally greeted me was, "Patel mode." These investors are looking at the charts.
That is insane.
These investors are looking at the charts for sure. The amount of Silicon Valley investors who follow me from Twitter, for example, is surprising because you'd think my entire audience is crypto-based, at least on Twitter. My Instagram's a little different, but it's very high.
They look at the charts and use them as a benchmark to measure FOMO. When they see a massive green candle, I get 50 texts from everyone—people who never supported us or doubted us way before that—saying, "Oh my God, this is the greatest thing ever. You proved it all out," or whatever it is.
That's why the chart does matter, and the investors actually want to see the chart. That's the best part of ICM.
I mean, I hit you up after a couple of green candles.
Yeah, exactly. Eric Zhu, who refused to tweet about us, hit me up. Everyone who said, "No, Avi, I'm not supporting this shit," is now spamming the timeline.
Whether it's because they have a bag, who knows, right? Or it's just, "Avi proved it out, so now it's okay." That's where it's at right now.
Do you have motion on Instagram?
I do have motion on Instagram, yeah.
For what? How?
I've done a lot of social hacking. The reason I have motion on Instagram is that every big influencer on planet Earth follows me there.
Okay, I'd say 30K is decent motion.
If you scroll through the 30K, though, it's just blue checkmarks the whole way down. Not Meta Verified blue checkmarks—blue checkmarks from people with 10 million followers, the whole way down, basically.
Every big creator—three days ago, the entire David Dober team followed me, other than David. He'll do that in 3 days when I sponsor his shit Halloween party or whatever it is.
Yeah, we'll see.
Adam Wed follows me. I had an entire spreadsheet 2 years ago when I was doing artist outreach, with thousands of people I knew. It had all the big influencers organized by follower count, what they do, everything, and I had a VA who ran the whole thing.
I don't do that anymore because it takes way too much time, but Instagram is just packed with celebrities and influencers. That's it.
You're like the L.A. name-drop final boss.
Yeah, probably. Yeah.
You are. It's okay, which you are. You are.
It works.
Well, you asked, right? Like, I told—I asked. I asked: how was the— That's all I have. How was the sperm racing?
Uh, dude, the Speed one was a mess, man. We hosted at UCLA. We went to UCLA.
Okay. What?
We hosted—we set everything up at a frat, and the competing frat heard about it and called the police. UCLA basically shut it down, and within an hour and a half, me and everyone who was there had to find a venue to host the new event. We went to a new venue, moved everything in an hour and a half, and organized security while people were crowding the place because it was leaked somehow that it was happening.
We organized buses for all the USC and UCLA kids. Speed finally showed up. He was on stream for 10 or 15 minutes. K's logo was on all the racers, whatever it was, and then Speed left the stream because he wasn't feeling well or whatever it was. There was a lot of chaos and [__].
I just said hi on the way out, but I know his team now, at the very least, which is nice. And yeah, that was that. David's was cool. It was fun. These are my boys, right? I like supporting them.
David's was really cool. That one went very well, and the UCLA-USC one before that went well. You were at that one. I might have seen you there.
Yeah, I was there. On the—what was it? DYOR Speed. They had the Solana logo in the middle. That [__] was lit.
That was lit. Yeah, it was for Speed. But yeah, Jakey—the streamer Jakey, the crypto streamer Jake—was supposed to get a picture with a Solana phone with Speed in it as well, and that would have really done it. But the problem is Speed left early. Otherwise, that would have all happened. There would have been a lot more that happened if Speed decided to stay.
I met Speed in a camera store in L.A. He was actually hella cool.
Really? Yeah. Cool.
Okay, I had a goal coming into this that I was not going to ask you about Ben Pasternak. I just wasn't going to do it. I'm just like, I don't care.
Yeah, I honestly don't care.
I don't know the extent. I know you guys go back and forth on Twitter sometimes, but I'm curious. Not to do price talk, but Kled's at $52 million and Launchcoin is at $106 million. I think that's the closest our ratio has been. Do you feel like Kled is boxed in and has a ceiling, which is how successful Launchcoin is in this ecosystem?
So again, look, I wish Ben and the entire team infinite success. Seriously, because we all want ICM to win, right? Whether that's them or whoever's launchpad, I wish every founder success.
I just couldn't give enough of a singular [__] about the drama anymore. It's a waste of time—a time suck. I don't give a [__]. There's no ceiling because I'm not part of Believe, and I'm not part of any of this, which is what everyone's realized, which is good.
There can be an ICM runner organically that outcompetes Launchcoin. And look, it's no shade to Ben. Again, no drama at all anymore. I don't care about that anymore, right? But my goal, obviously, is that if Ben's at $100 million, I want to be at [__] a billion, right? That's the whole point.
So it gets competitive, sure, but yeah. No drama. Peaceful. Me just in San Francisco, signing labs. That's the type of vibe right now—just mellowed out all the time. Yeah, that's basically it.
WLED PR team.
No, Wav PR team, bro. The Kled PR team is saying to do much worse than that, but there's no point, guys. I don't want to engagement-farm this [__] anymore. It's just a waste of time.
You know what's engagement-farming? [__] putting out an article with whatever big lab for a billion dollars. That's engagement-farming right there. That's what really instantly sends one candle to $500 million, whatever you guys want, right?
But this engagement-farming stuff for mindshare is just useless. I'm over the drama. There's no point. I wish everyone success. Seriously.
I love how you're making announcements and watching the 1-minute chart. It's kind of a movie.
Yeah. Yeah.
Do you—I was always curious. When the ICM stuff first happened, before I even knew him, I remember asking Bobby from Dupe if there was any meaningful increase in traffic on the Dupe website after there were token holders, because I know people in the Solana trenches would try anything. They'll play any—
I put out a case study. Maya Hedge is here with me in San Francisco, literally. We were just at lunch. Maya Higa, you know.
Yeah, yeah. I do know Maya Hedge, actually. Yeah, 100%.
I was just with Maya and K5. We were meeting with investors, right? Maya is helping us with the tokenized equity, and he actually wrote a case study on this.
Kled's traffic is really good for a company with nothing—just a founder, or just a project that's not really public-facing—and just a founder who is supposed to be a public-facing product to launch a coin, because you will literally build your entire company from the start off the coin.
I think the thesis here is that tokenized startups at an early stage will always beat untokenized ones. I think Kled is the perfect example of that, because traffic is in the millions, with 25,000 users in 4 days on just the mobile app, let alone the waitlist, which has grown significantly.
Payouts to users. We did 30 million impressions on Twitter alone, right? If you search Kled on Twitter, it's flooded with positive [__] by the second. I had an investor search Kled on Twitter, and it's just flooded with positive [__]. Every 20 seconds, there's a post, and they're like, “Holy [__], I've never seen this before.”
Silicon Valley runs on FOMO. It's the biggest FOMO driver of all time, right? So that's why it's important. I think Bobby's probably seen an impact, but because his was established before he launched the coin, it might have been different. But yeah, I'm sure you've seen the impact.
That's—I like that take. The communities will just use anything. They'll pump any product. Also, I see this picture of you and Speed. Eric tweeted, “Patel mode is hard.”
Yeah, it's funny. Yeah, I mean, dude, I think the token's really, really useful for a lot of people. I think everyone in the startup scene should be doing something similar. I think it's really dumb not to, but it's not obvious yet.
Kled will be the success story and make it really obvious. I think that's the best way to look at it.
Someone keeps asking in the chat, “Please ask when waitlist people can get in.”
Guys, look, I'm literally tripling the amount of people I let into this waitlist every single batch. Once these big contracts ramp up in size, then it's much easier, right? Because you have to throttle this. If I don't have enough money to pay people, then some people are going to get paid a lot less than what they deserve because there are simply too many people inside the app, right?
It's not because the app isn't ready. The app is clearly ready. We have thousands of users in it, right? But it's really important. It also drives a ton of FOMO.
If you go to the timeline and look, we have this social-sharing thing for getting referrals. There's a ton of people spamming their referral codes for Kled, and it helps them get earlier on the waitlist. It also makes sure that we're viral every single day, and I think that's important.
So yeah, we'll triple down on the waitlist every time a new batch opens up, and more people will be let in. More importantly, I'm introducing Kled Games, which we talked about, but we haven't talked about publicly, so this is actually the first time.
Imagine we had a $50,000 prize pool set aside. Every 30 minutes on your phone, you get a notification from Kled saying, “Upload 20 pictures of cars in your area,” and you have 10 seconds to do it. If you upload them in time and we recognize that it's true, then you're in the next round. Basically, you survive, and the last person standing gets all $50,000.
We collect data from everyone. We clip-farm the [__] out of it, because imagine a high school kid, for example, is in the final round. He's in math class taking a test, gets a notification, jumps out of his [__] classroom window to sprint to the parking lot to take pictures of cars because that's what it takes, and starts spam-celebrating. That's the TikTok clip that goes viral everywhere.
We can ramp up these prize pools over and over and over again. We can let every user into that because we collect data from everyone. One user wins, and we can probably have 100 people who get marginal prizes as well. It's going to be a really cool game that'll probably go very, very viral, and the winners also get access to the waitlist.
So we can do these mechanisms, and that's actually not hard to do. I just have to do image-recognition and video-recognition software, and that takes—it's very fast as well. Once that's done, we can release that, and everyone's technically off the waitlist. Then, once the contracts come in, we can let everyone into the main app as well.
That's kind of a movie. When does that launch?
Eli is finishing a Wallers protocol integration right now for decentralized storage. After that, he has a big update he's pushing tomorrow, which is very exciting. He's pushing a massive update for Kled tomorrow. The day after he's done with Walrus, he can get started on the image-recognition software.
Once the image-recognition software is done, our front-end guy is a freak. He'll finish up the front end, and then we'll deploy it.
Whoa, that's kind of a movie. How do you determine if any of this data—these images—is AI-generated?
We have—
Do you miss any?
We're literally at 95%. Eli has set up a system that will detect duplicates across every single account and check the entire internet's history for plagiarism across the board when we're first searching images and—
Exactly.
We'll also double-check the EXIF metadata. If there are any alterations in the image, you won't get paid for it. We'll check the metadata of the image itself. If it's not off your device, you won't get paid for it.
If it's AI-generated, we use Mistral to check whether it's AI-generated, and you won't get paid for it. We have 10 layers of checks for plagiarism, so that anytime an AI lab gets our data, it's with literal 99% certainty pretty much the safest data, straight from the user and off the real device.
We let anyone upload anything, sure, but every single time we'll filter out the bad stuff.
So even if you went on Google and searched for cars and just saved 10 images, you would reverse-search them and be like, “This is not original. This is already on the internet”?
Well, think of it like BeReal. It's not going to be like that. It's not going to be “upload.” It's going to be you—
Like, take a picture. Yeah, yeah, yeah. And same with the videos.
Yeah.
Damn. BeReal was a movie.
Yeah, it was. It was.
That was peak social media.
It definitely was, dude. That—
Hey, do you subscribe to or believe in the great AI talent migration into crypto and more financial products?
What is the great AI talent migration? Do you just think more AI engineers are coming here? There's the thought that AI is a talent vacuum, and definitely from a venture perspective, all these labs are raising at ridiculous valuations. A lot of that money is going to trickle back into crypto products because—
Think about it: the upside to building a successful financial product in crypto is—
It's like pumping liquidity—$2 million a day in perpetuity and a deca-billion-dollar valuation in year 1—versus AI slop company number 75 that makes no money and has zero clear plans for monetization at any scale.
Got it.
I think if AI is doing very well, which it is, then obviously things trickle back into the system. But there are literally no good AI projects on-chain. They just don't exist, other than a very small handful.
Until you see good developers taking crypto seriously, this is just not going to happen. Hopefully Kled spearheads it, right? Most of these products are vaporware. They're literally just useless vaporware, and it's sad to see that people are bidding them because they're simply bored of the other conditions in the market, which are nonexistent. So, yeah, that's the current state of it, I think.
Do you think during the 2024 AI season in crypto—there was, I don't know how much attention you paid to it, but there was this 3-month period—
Zero. Literally zero. I'm the most unc—
You didn't look at anything?
I didn't look at anything. I don't know anything about any of these, so educate me. Tell me what happened.
There was this 3-month period where crypto was always a little bit late, right? There was a mainstream trend that happened, and then it trickled down in a lighter version on-chain. There were about 3 months where I was extremely excited about crypto AI, myself included.
There was this really small subset of developers who seemingly were a relatively big deal in Web2 AI and were coming and launching crypto projects—ex-OpenAI, all this stuff. It went really high. AI coins on Solana probably topped out at around $15 billion, versus the launchcoin ecosystem—ICM topped out at around $1 billion or $2 billion—so it was substantially bigger.
ICM was a little bit euphoric for a couple of days. That was basically it for 3 months, and then it kind of just disappeared. Most of the coins got crushed, and most of the developers left.
It felt like this little pocket where, for the first time in a long time, there were actual people from Web2 coming into crypto and launching. Even if they were just GPT wrappers, it was at least interesting.
We reminisce on it all the time, I feel like, and we're like, “When is it going to happen again? When are the AI devs going to come back and launch in crypto simply because they can make more money?” But there's basically been no indication that that's going to happen again. So I guess it's TBD.
Yeah, TBD for sure. I think crypto's tired of a lot of shit. I want to preface this before I say it: I think 99% of every project has failed to deliver on anything they said at the very start.
Specifically in ICM, there are very, very few founders who have delivered. Bobby obviously has, and Chris from Polycule obviously has, but 99% of these projects are coming-soon farms. They're actually coming-soon farms. They'll just play narratives when hype is riding high, farm the coming-soon forum, and then it goes up.
Yeah, these are disgusting, disgusting projects.
This isn't directed toward any one person. I think everyone who is building here just needs to deliver more. If that's the case, the market will react accordingly and collectively bring us up. That will then cause more actual, traditional AI Silicon Valley founders to come back to crypto.
Unless you're literally as crazy as me, no one is going to do that. It's not realistic, right? Hopefully Kled is the example that will make that happen. The reality is that there just needs to be more people actually giving a shit, and that is not the case right now.
Honestly, man, I root for you, Avi Patel. In my heart, I root for you as a non-Kled holder. As a former Kled holder, I honestly root for you, dude.
I appreciate that, dude. I think there was a lot of shit-hate at the very start, but all I'm happy to see is that if you search Kled on Twitter and scroll down that list, there is zero FUD. That's one of the most surprising things I've ever seen. I never thought the day would come.
I feel like you finally took a breath, man.
Yeah.
I feel like it wasn't the breath, though, because drama never stops in Patel land, right?
What really happened is that we sat down, shipped, and delivered. We put out 200 updates in the last 4 months. The app is not an app worthy of 4 months of work. It's a year's worth of work minimum for a really good team.
We just delivered on the stuff that we said we were supposed to deliver on, and that's what did this.
Well, yeah. Frank just made this comment too, and I think it's true: people's opinions happen based on how the first couple of weeks go, but on a long enough time frame, it's hard to keep the grifter label when you just keep shipping, keep launching cool stuff, keep coming on, keep doing media, and keep launching. Eventually, things turn around.
Yeah, it's literally not physically possible. There are very few people who would stick it out like we have and fight through it all. I try to shield my team from this as much as possible.
It's a lean team. That's why we're shipping fast. Eli and 2 other engineers are handling all of our engineering, and Eli's doing pretty much all of it for the most part. We're going to have an AI engineer come in-house.
Then we have a design lead, which is really important because it's a consumer app, so it has to be really nice and functional. We also have a 3D lead handling animations, creatives, and editing for content creators, and then our fat legal team. That's basically it.
Yeah. And then we have a UGC team, obviously.
You probably have a monster legal team, bro.
Yeah, it’s Latham & Watkins. It’s literally the best lawyers in the world—arguably the best lawyers in the world. They run everything.
Dude, this was sick. As a wrap, what are you most excited about that you’re working on, launching, or making happen? What’s your focus from here?
I mean, I like to think as short-term as possible. We’ve just had 1 day in San Francisco. I haven’t even announced what happened on day 2 or day 3.
I will not announce what happened.
You made it sound like you were about to do it. I was asking. So, I’ll drop it on the timeline randomly and no one will expect it.
But yeah, look, I have 15 days in San Francisco here. I’ll just do as much damage as physically possible here, and then I’ll go back to New York, do the development rounds, and get games out as fast as possible. Hopefully, start securing these really large contracts, get PR and press, and go more viral.
We’ve established the groundwork. Now, let’s just hit the ground running and use the momentum properly. I’m also focused on CEXs. I think a lot of ICM founders haven’t focused on it, but we’re at the market cap now where I can actually get the CEX listings.
I’ll be hitting up Arjun from Kraken, whoever, and just see what happens. That’s the main thing for Kled right now.
He’s a former guest on the Thread Guy stream. If you need an intro—
I have his number. We’ve texted back and forth.
I’m just saying. I’m just saying. He appeared on the stream.
I might, if it’s a warm intro. Yeah, sure. I mean, if it’s like, “Hey, Avi’s crushing it. Go list this shit.”
Yeah. Maybe $100 million, buddy.
Maybe a couple racks.
Avi, you’re the GOAT, dude. It was honestly a long time coming. Congratulations on the last couple of weeks. You’re a fun interview and stream guest, so I appreciate you coming on.
I appreciate it. Okay, sick. I’ll talk to you soon.
Thanks, bro. Have a good one. Peace.
You too.