AI ROLLUP #12: Kaito's Secret Token Utility | Grok 3 Released | Bittensor Launch | Virtuals Revenue Record

Welcome Bankless Nation to the AI Rollup where we say up to speed with the emerging trends and developments in the AI crypto space. I am David Hoffman here with my co-host Ijaz. Ijaz, how was your week? Did you get did you get rugged by a president?

No, I I didn't, but yeah, unfortunately the president of Argentina decided to launch a meme coin. Uh and I uh I I know how much you love these these presidential back meme coins, David, but yeah I I I really think it's kind of exhausting the industry a little bit. Oh we yes um

The

I really it it turned on the lights in the club and everyone realizes that we're all drunk and ugly.

Yeah, yeah, like super ugly, actually. Um, and so I I the point I want to make around this is not to hop on about meme coins, but rather I think we're getting to the end of this um financial nihilism sprint. You know how everyone's like really depresso and they're like crypto doesn't solve anything but um a resolution for gambling addicts, you know? Um I think the meme coin extraction phase of crypto's life is is kind of done. It's it's it's coming out of its teenage years, I think, finally. And for months now we've seen kind of millions of tokens get launched and millions of people get rugged as a result. And the majority are just kind of pure pump and dumps, right? And and I think the reason why it got so much engagement, regardless, is that people had just lost faith in um what we were really striving for in this industry. They just thought it was all just kind of like random memes and gambling. Uh, the passion for the innovation kind of thing was just lost. But now I've seen sentiment so low on the timeline. I don't know about what your timeline looks like, but for me, it's just Libra this, Libra that, which is the name of that coin. Please don't buy that coin. Um, and it seems like the industry is just finally done. They're tired. So I think we're gonna shift focus over.

uh to invention tinkering innovation again and I think there's no better sector that's primed to benefit from this than in AI because there's so much stuff going on in AI right now at Hibbett.

Yeah. Well, the i I will say the AI.

AI, the crypto AI sector with the AI agents was born out of the meme coin like media because a lot of the tokens that we talk about started off as meme coins that backed into being AI coins. Like not all of them, like Virtuals is a startup, it's a regular startup, it's a great startup. So that's not what I'm talking about. But like AI 16Z and GOAT are both of these tokens that are meme coins that backed into being these AI coins. So there was this inherent tie between this like crypto AI agent sector of crypto and the meme coin mania because they kind of happen at the same time. But nonetheless, the AI crypto sector had most of its hot air drained from it in by the end of like January and definitely into February. And I think with this Libra thing, we're seeing whatever remainder of hot air is in this sector is now completely deflating. And it has deflated. And so now people are like, there's no, you have no choice but to build because the hot money of gambling speculation is not coming back. And so what is truly left is people who are actually like believing in this in this thing. And I will say, like, in the grand scheme of things, there's like a lot of AI, crypto AI, and AI agent builders that are like really convicted and are definitely here and are not here because it was just like the next theme of meme coins, but it was actually here because of like the the there's something.

No, I

Yeah.

I talk to them every week. Yeah. Because I'm like, am I crazy? Is is the whole timeline right? And like we're the only ones that are just wrong. But there are some

Really exciting updates, which we'll get into later on in this episode, that just indicates that people are naturally building. And kind of like to round up the point on the meme stuff, David, if we were looking for a silver lining, right? Because you mentioned there was a bunch of AI teams which launched coins, which were originally kind of memey, and then they kind of backed into utility. Um, the silver lining with all these meme coin launches was that they were kind of like a fairish launch. Uh obviously they could get sniped, uh, but at the end of the day, anyone could access it. And they had, you know, all the information that was needed for you to be able to, you know, just plug into the internet and buy it if you wanted to, right? But obviously, you know, the point you're making is that can get abused pretty drastically. And we're finally getting to the point where, you know, we're gonna find some kind of middle ground, I presume.

I think that's right. But when we so we're going to talk about a bunch of topics in the AI crypto sector as we normally do, we're going to talk about Kaido. We're going to talk about the Kaido token because everyone's hyped about that. Very related to the AI sector. We're going to talk about BitTensor and some virtual things. But in the Trad AI world, the normal outside AI world, the non-crypto side of things, there's also been a bunch of activity. Uh Grok3 was released, and Grok is the model out of X AI. That's Twitter AI. That's Elon Musk's AI sector. And there's this whole world of like AI activity that's going on. That is like crypto AI is downstream of. Like we are basically downstream of these models being built by these uh big, big AI labs. And out of the big AI labs came Grok 3 last week, which blew people's minds. So it blew it blew other models out of the water and really uh elevated, I'll say, like X the X AI team to be like a top tier contender sitting shoulder to shoulder with Open AI and some of the other model generators. What was your take on some of the the uh the GROK 3 release this week?

Yeah, I mean I I kind of want to

push you a bit on your framing of this, David. So you you said trad AI, and actually I'm guilty of saying the same, but I kind of want to mesh both of those together. And like that might sound a little audacious because like what if what has crypto AI been been known to show? But I'm more comparing Trad AI with open source AI.

Like, remember a few weeks back, Deep Seek AI came out of nowhere, right? That's an open source thing, right? This week we're about to talk about two things which were 100% open source from some of the biggest companies in the world, including Microsoft, right? Which is now 100% open source. I've noticed a shift from Trad AI going from centralized to more open source development of this AI. And I think that's super interesting and something that crypto is very much a huge participant in. You look like you're about to say something.

Yeah, so so so I think I understand your point, but not completely, because Grok three is not open source. Once the Elon Musk and the XAI team, they said that once Grok 3 is stabilized and released into production, then they open source Grok 2, not Grok 3. And so there's kind of like a tip of the chain phenomenon here where the very old models from years back are getting open source and they're getting completely monetized. But the freshest, most like innovative tip of the chain from AI for the AI models, open AI, that's closed source. And like this has been what this is turning into is like if you want to pay money, you can be like three to six months ahead in people in your like AI superpowers. But other than that, the open source world is open for the taking.

You're absolutely right. But let me use your analogy, right? You mentioned three to six months. So there was a time, David, where it was uh definitely nine months on average. Then it became six months, then it became three months. Now it's becoming a couple of weeks. At what point does it become days, right? So, like um you had Deep Seat come out, and then that was already open source, but by that time you had several iterations of it, which was much better than it, and therefore subsequently performing much better than open AI's O3 mini models or whatever that might be. But um the point is these iterations are happening super, super quick. But let's not lose focus. Let's let's get onto your point around Grok 3. Um

The the kind of summary of this is this is uh X AI or formerly known as Twitter um AI's product or their latest uh their latest uh AI model. So um it's called Grok. Now they had Grok 1, Grok two, it was fairly okay, meh. It was more kind of like a Ground.

Yeah, it really stuck. It couldn't do any of the fan fancy stuff, dude. And honestly, it wasn't even compared on the same plane as some of these other things, like perplexity AI, um, you know, Claude, open air, all that kind of stuff. But now they've finally entered in with a model that competes not only at the same standard, but in some cases much better. So, what you're looking at right now is a screenshot from their presentation um where they kind of like compared similar metrics um or kind of like standards, um, which they compare a lot of these models against. And typically it's around coding, mathematics, and science, a lot of reasoning involved in a bunch of those tests. And you know, it's kind of like a general framework. There's pros and cons of both, but it's kind of like the the blanket test. Um and it outperformed on pretty much all of those against um OpenAI's O3 mini model and OpenAI's O3 full model, which is just insane to see because this company has been around for I don't know, like what's it, 22 months at this point, right? And you know, Google's DeepMind has been around for how long?

OpenAI has been around for you know a number of different years now at this point. I don't know why I said different, but a number of years at this point. So these iteration cycles are getting much quicker. And I'm just so impressed by this, David. I've seen a bunch of demos about this. You know, coders are using this to create uh a bunch of software, which you normally couldn't do before. Um, it's just crazy to see this. Like literally a month ago or two months ago, the idea of an AI model replicating um some kind of low level software engineer's work was completely out the wayside. And people were memeing it. People were trolling it. They were saying there's no way these things are actually going to replace us. And now we just seem to be talking about it every single week.

Right. I I think like my one interpretation of this is like this is another iteration of Deep Seek. This is just coming inside of from China. It's coming from Elon Musk. Uh XAI. Here's Gavin Baker, who is just an AI chat out there who's singing the praises of Grok 3. He says Grok 3 is the first model ever to score over 1400 on chatbot arena and outperforms the best publicly available reasoning models from OpenAI and Google. XAI was founded 13 years after DeepMind and the eight years after OpenAI and is now ahead of both. And so this is like the second whammy to open AI and Chat GPT. Like I said, the first one was from China, but this one is from Elon Musk. And it's clear that like scrappy teams are really able to make leaps and bounds ahead of OpenAI, which seems to have been this very entrenched incumbent that was never going to be able to be beat. And that is just not the case, seemingly anymore.

Yeah, well, it's because of the secret source that DeepSeq kind of highlighted as well. And I don't know for sure if Grok is doing this. We'll know once they've open sourced or like released some kind of technical architecture as to how it works. But the whole breakthrough with DeepSeek, remember, was traditionally you needed a hell of a lot of compute and data to make any of this stuff worked. And DeepSeq found a way to use less data and less compute to create the same standard and quality of model. Um, known as, you know, it's kind of a reinforcement learning or a mixture of like reasoning data, basically. I don't want to get too much into it, but effectively a new technique. Um, and that's allowed a lot of these teams on the ground level, scrappy as you say, to compete with the best. And I think that's only just better for like a bunch of open source development. It's super cool.

Uh w how I see this impacting crypto AI and AI agents and really the the sector that this podcast mostly covers is like this kind of is like

You know how like the Federal Reserve sets interest rates and that dictates how hard or easy it is to have a business in the world?

The the strength and power of models as they gets as they get more and more performant and their capacity increases, it makes the ability for AI agents, AI startups, anybody building in the AI crypto space just gets easier and easier to do cool stuff. So, like two months ago, everyone was talking about like, oh my god, these agents are so goddamn annoying. They're absolutely useless. This is just a bot with a token. Uh, if me, if that was all true, okay, sure. But like

Yeah.

That was two months ago. And models are like already so much more

What if that box could code, David? You know? Well what if we could build an app from scratch, just from a prompt or just from a tweet, you know?

So as the hot air leaves the crypto industry because everyone realized the meme coin casino was up, like the fundamentals around like what crypto AI developers have the ability to do is like only growing very, very like rapidly.

Yeah. Yeah. No, I agree.

Are we gonna get ready to get into the rest of the episode?

Yes, sir.

Okay, but first before we do, we're gonna talk about Kaido, sports betting, a few other things. Uh, but we're gonna first talk to our friends and sponsors. So, Ejaz, we just talked about Grok 3 and the closed source AI world, but there was uh a new model entering the arena. A new model has entered the chat, and this is my understanding from um from you. Uh talk to us about where the what this new model is and where it came from and how it's different.

Okay, so um there's this team within crypto, David, uh known as Noose Research. So they're kind of like a yeah, one of us. Um they started off as a pure uh open source AI um research team. And, you know, it's a bunch of engineers that just wanted to kind of build for the open source good. Um and they kind of stumbled across crypto uh a while back when they discovered this protocol called BitENTER, which we're actually gonna talk about and update later today. Um and then they kind of like went into the worlds of incentives, but they've never really kind of delved into tokens too much. They've just focused on the open source research. And they've put out some really impressive things, David. They're actually at the forefront of pushing things like uh distributed and decentralized training of AI models. Um, they've also pushed out a number of open source AI models themselves, known as ORMES. Um, and they've actually released their recent model. Um, it's called Omez 3. Uh

Preview, okay. A new LLM which unifies reasoning and intuitive language model capabilities. Now, before getting into the kind of like details of this, or rather as opposed to getting into the details of this, the best way to think about this is this is their version of Deep Seek, right? And the reason why I say it's a version of Deep Seek is because they've implemented a bunch of Deep Seek's uh really uh groundbreaking techniques. So reinforcement learning with reasoning data, they've basically done their own flavor of this. Why I think this is super cool, and just to kind of like summarize it really quick, is now you just have open access to this model, which you can host on pretty much any kind of decentralized compute network, whether that's hyperbolic or wherever. Um, and you can get access to this pretty much 24 7. If you just scroll down to that uh bar chart tweet, David, um, you'll see how it compares as a benchmark across um, you know, Deep Seek and a bunch of other models there. So, and it's previous models as well. So, really cool to see.

So, okay, so how do how is this model different from like Chat GPT or Grok? Is this like a more specialized model? Is it a generalist model? Who is this model for?

Um, it's actually not quite too different, except that it outperforms uncertain benchmarks. So, so think about like, you know, when you create a base model and then you have to fine-tune it to act in a very specific way. Most of the models that we interact with on ChatGPT are very generalist. This is less generalist than that. So you won't get completely um uncensored answers, but you'll get a lot more uncensored answers than you would expect. It's a lot more real and candid with you than some kind of uh heavily curated model. That's the best way to think about it.

Okay, so and it is that a property of just who who is building this thing? Like news research? Yes.

From the crypto side of things, crypto people not really into censorship, totally into individual expression. So they're totally interested in just like letting the brakes off of some of these models and letting them go wild. Is that kind of the deal?

I I would argue that's like a main uh fundamental goal for a lot of the open source AI community. We've seen it with other um teams before. Like uh we've mentioned Venice AI on this show before. Um, you know, Eric Voorhees uh actually released that token a few weeks ago, I remember. But their product, Venice AI, is pretty much like an open source, uh self hosted um.

Version or uncensored version of all the top models that we we see today. And you can kind of ask it, you know, how do I hotwire a car? And I'm not saying that's a good thing, but I'm just saying it will give you an answer. Whereas if you tried that with uh Chat GPT, they wouldn't say anything, pretty much.

Really? ChatGPT won't ask well let me see. Let me see what it'll say.

Yeah, go

for it.

How do I hotwire a car?

I can't help with that. If you're having trouble with your car, I recommend contacting the locksmith or roadside assistants. Let me know if you need help with anything else. Wow. All right.

Yeah.

Oh, okay. Well, how do I ask uh how do I ask this this model that from Noose Research uh where can I go access this one?

Well, if you actually just just go to Ask Venice, which was the example that I gave you. Yeah, yeah.

Okay. Fancy private and uncensored AI.

There you go. Should be able to pull it up.

How do I hotwire a car?

Hotwire car is a complex and potentially illegal process, but here are the steps to hotwire a car. You typically need to access the steering column, identify the battery, connect the battery. Uh however, modern cars actually have advanced security systems, so be careful for those. Wow.

If you live in the same city as David, look out for your Navy car theft ratings. It might be up-only. It might be the only up-only chart for a while. Okay. Yeah.

Okay, moving on. Um, you there's a few companies that are involved in this AI thing, David, right? And you know, it's typically the small teams, the unknown teams that help on the open source AI side. But there's also this little, um, lesser known company known as Microsoft, which released this new tool that turns all these AI models that we keep going on about week after week into agents, David, that can use your computer. So think about this, right? Say, like you had access to ChatGPT or OpenAI's O3 model or Perplexity or whatever it might be. You can now combine it or mash it with this new product, which is, by the way, 100% open source from Microsoft, and it turns it into an agent. What that means is it's an agent which is hosted on your laptop, computer, desktop, whatever you're using, and it can use computer use. So it can see everything that you're seeing. You could say, Hey, uh, David looks pretty good on this thumbnail. Maybe I can brighten up or lower down his saturation or whatever that might be. Or maybe it can go to a certain tab. You say, Hey, agent, could you just go check the stock market for me whilst I'm working on this other thing, or whilst I'm working at my full time job? And it'll monitor things. He can do transactions for you. You could give it access to your bank account, whatever that might be. It's just a pretty wild tool to release in the open source world, to be honest.

I think the implications of this are actually kind of huge. Okay, so having an agent inside of your computer, so you can just basically dictate to your computer what you want to happen, which sounds very

awesome in the traditional. It's like awe inspiring. But then also,

yo, like

some I'm gonna give control over my computer to this like thing that lives inside of my computer.