▶️ Watch on 3Speak --- This AI has grown a ton since part 1. Enjoy. --- ▶️ 3Speak
Transcript
You'll do a quick video. I'd like to get a video out at least every month or two. It's easy just to work and let time fly by. I believe I was, one of the last videos I was talking about getting AI, getting some kind of AI on Hive, decentralized AI, I was really hype about it. I'm still very hype about it. I've actually been so hype about it. I've been working 10, you know, probably 9, 10 hours a day since, if you do the math, right, how do you reach 1 ,000 hours, right? If you're working 10 hours a day, 10 days, that's 100 hours. So, Lord knows how much effort and time I've put into this. People ask, how do I not get burned out? It's sort of like when you just have a mission, you separate emotion. I learned to do this as a poker player. So, like, it's funny because I've been playing poker for so long, and you would think, oh, just a card game, but it has really instilled in me. And even video games at a younger age, because I used to grind those. I was a world champion Madden player. I was top three in the world three years in a row. Not a lot of people know that. Before there was poker, there was professional gaming before me, before it was actually a thing. And what this has taught me is the grind. When you, you know, when you play poker, yeah, it's fun, but when you play poker, you know, 80 hours a week for years, and you only get five-minute breaks every hour, and you're playing 30 games at a time, it's a grind. Like, if your emotions in that, you're going to quit. You're not going to be able to succeed. That's why they say it's a hard way for easy living, right? People think it's easy, but there's the grind to it. There's that leather-ass grind, and that's instilled in me. When I'm coding, it's the same thing when I'm playing poker. I got 12 coding screens up. I got Claude. I got Codex. I got cursor. I got, you know, Hive AI, DEAI, decentralized AI, whatever we're going to end up calling this. Yes, it's able to work on itself. So, you know, I got a flow. Everything's organized. I learn that from poker as well. You can't just play 30 games at a time without a system, without having meticulous screen setups and all that. So I have a flow. And when I go, it's just like, boom, boom, and I look up, and the day, you know, the sun's coming down, you know what I mean? Stomach's growling. You know, still kind of fresh out of surgery. It's been about pushing three months. I've really been able to get my mentality back, my mental back. And I'm just excited about what I've able to get done so far and the future of this because, you know, I use AI models as much as anybody. I know these things like that. I had to learn the hardest shit. Everything I do is ridiculously hard. Decentralized AI is ridiculously hard. This has been the hardest. Colonel, all these things you can take the hardest thing you can do right now in coding. And that's what I've been... banging on the cutting edge, the cutting edge of cutting edge of making new shit and having to argue with the AI saying, no, it's not possible, and showing them it's possible, because I know it's possible, and to push the limits as far as I've done, and the research I've done, the ingestion I've done of the internet, the amount of terabytes I've gone through, and how many hard drives, and how much information, how I've distilled it, and how I've, I've gone a little bit crazy, maybe. But I've gotten pushing, 50, 70 million of the highest quality lower pairs, 88,000 characters, minimum per, covering everything from chain of thought, deep thinking, how to teach, everything you can imagine, everything is, it's as good as I've, because I've seen it walk like a child, like, I've seen this thing grow up. Old Hive AI up here. It's the name right now. Old Hive AI. I've seen this thing. It was a little baby. It was a little toddler. And it didn't know much. You ask it something and, you know, it'd be kind of dumb. And I was fooling around and I was like, there's something here. And every time I'd add a new lever, add a new feature. And the next big thing was the harness. I built this fucking harness around it that could take any local model and make it 20, 30x better. And it makes it where it has all the tools it needs, and it uses tokens much more efficiently so you don't have to, before half the job is done before it even uses a token. The harness I've made, the sauce is so ridiculous. It's so meticulous. It's so crazy how much better and how I've been able to push that to get this thing. This thing learns every time you talk. Full internet access to any website, unrestricted, has a blob-in-gest mode where it can just absorb shit, like GitHubs and all that and crank it down. It is very fast, very token efficient. I've been testing this, so I have five old computers all around me. I got a Macbook, I got this old -ass fucking everything, all the computers I could have. I'm like fucking some crazy, got these old clankers around me, these clunker computers around me and shit. Everybody's working. Everyone's always working. All these computers always working. So you might hear them the fans and shit. And what I did was that's my squad. That's what I practiced on. GPU. And some of them are in the other, on the condo, on a different IP, wireless, land, connected, WAND, all that shit, everything, right? So. They're all connected now. So now I got five GPUs, and I've been able to train, decentralized train on each one, and run interference, and tweak that, and tweak this, and say, oh, well, this computer should have this one, and this one should have some speculative decoding, and this one should be, you know, an ELM, and this one don't got a GPU, so it can do CPU jobs and RAM jobs, and oh, it can also, RAM jobs, that sounds kind of dirty. It can also do hard drive, you can store. It's like, if you don't got nothing, but as store, you can still be a part of this system. It is grandma-proof, literally grandma-proof. Anyone can join. Anyone can earn. You earn for you, so you log on, you have this open, you're earning. You're fucking earning. And then you can use your local AI, and you get free AI to the system. Keep in mind, this is attached to a decentralized rag, a brain, if you will, a giant... fucking brain of all the shit it needs to know, and it has little laura adapters everywhere that help where you're weak. So there's all kinds of little things, the speculative decode, so you get things done real fast. It knows your, it'll so, slow down, slow down. So we got on here, and this is going to be a bunch more fancy. But when you first sign in, It's going to be, it's going to check your computer and it's going to be like, look, this is what you got and this is what you should have. And I'm going to have a bunch of adapters and I'm going to have some nice trained ones. But the whole point of hive AI, the AI, decentralized AI, is to bring your own model. I'm not inventing the LLM at all. I am building the harness, the ecosystem, the infrastructure to make any LLM a super badass. So if you have your super badass LLM that you trained on. whatever you train it on and you want to come and you want to earn and you want to have Because all you're doing is you have your LM at home, like me. Like, oh, I have my LM, this is cool. I'm decentralized, it's private. But, and you might even have you a little cute rag, and maybe you did some super advanced shit and got that shit able to scrape and take all the data in and make sure that data's not bullshit that you're getting in. And maybe you even went a little bit farther, and you forked an ID, and made your own custom ID where you can actually scrape GitHub and code and all that shit from your local files. Maybe even. did all that. If you did, that's, congratulations, you did a lot of fucking work. That's just the stair step here. Because even if you did all that, you plug your little computer into this network, and guess what, now you're connected to possibly hundreds of GPUs running interference, and you're connected to a giant rag, not just your rag, because the great thing about information is, the more of it you have, the better it is, especially if it's... organized and trained. So the decentralized rag is going to live on IPFS, actually had to improve IPFS. I've fixed bugs on IPFS. I've fixed bugs on parallel X, pipeline parallelism. That's one of the most cutting edge things going on right now. Guess what? They had a day one bug that I had to fix on my own. I sent it to their PR, and they actually accepted it on their main live thing. That's how deep I am. I'm deeper than any of these people are pushing because I understand where their limitations are. I know where they went and had to... stop. I'm like, oh, this is where they stopped, because they couldn't have gotten any further, because I had to squash these bugs and do all this other shit. Different tiers, yes, there's going to be P -P-P, yes, get your laughs out, P -P, pipeline parallelism. That's basically splitting shards of a larger brain, so everyone can go. That's called Tier 3. We have Tier 2, which is a L -M version of that. It's like a smaller version, basically, with land. So it's like your land buds. So that's the second brain. The third brain is your big brain. That's your pipeline. That's your P.P. Tier 3, big P.P. Level. Pipeline parallelism. That's where you... you can introduce Wayne. And Wayne is different from Lane. Lane's like local shit. Wayne's like worldwide. So, like, Tier 2 brain is like you and your homies, or maybe you got a data center or some shit, and you bring that Tier 2 brain, right? That's cool. Tier 3 is like everyone can kind of participate and run these shards. First, it's going to be coordinators, Webitrust, all that stuff. But then, you know, the coordinators can then say, hey, yeah, we want these people to run shards as well. And then Tier 1, anyone can get involved in. Anyone can get involved in Tier 1, anyone get involved in Tier 2 or 3 as well. There's no restrictions. Tier 1 is dense models. Everyone's running their dense model based on the computer that they have. and what they can bring. And we're like Indians, you know, we don't even waste the bones. You know what I mean? I'm talking about Native American Indians. We don't waste the bones, you know what I mean? Like, we make the bones into weapons and jewelry and shit. We don't waste nothing on here. So if you don't got nothing but hard drive or CPU, you're still useful. Okay, let me take a trick. I have bronchitis, by the way. I've had fucking bronchitis. So if I start sniffling, cough in a whole bunch, don't mind me. Oh, my robots stopped. Got to get the robots back to work. Hold on. There you go. And, um, all right. So, there's so much to talk about. I don't even know where to begin. Let's just thinking of the people like, what's this hive mine? I don't know if, um, Hold on, I'm getting messages and shit. Trying to give a talk and getting these... So people have probably heard of TIG, the innovation game, right? It was a cool idea for algorithms and all that. Basically, I took that same principle, but I applied it to mining and improving the rag and the code itself. So if you see these challenges, like 27 of them right now, there's going to be some more. And these revolve around compression, density, expanding answers, getting higher scores, improving core code, so it took the idea of improving the algorithm but to improve the actual ecosystem. So instead of even needing a core dev, you have what you call champions. And these are going to be higher bounties, like the higher you get, because the new champions always get a bigger and bigger and bigger and bigger and bigger bounty, right? That's the whole idea. Because if you're a champion, your code, and one of the cool things is your actual code that you made, And, you know, people are going to use their clods and their codexes and all that. That's fine. I'm burning their tokens to improve the system. And it's not going to get easy, right? Because everyone's going to be doing that. Well, you're going to stand out. You're going to have to do more than that. And that's when those real champions are going to come. You're going to be immortalized in the living code. So. When you win, you're also a champion, goes on, and your code is what is used inside the AI system. So that's really cool, because it basically took that whole idea, and it was like, well, let's just improve the blockchain with it. And now you can have champions, so now I can come here, and I can just plug in play and mine crypto while improving the protocol, Bada Bing, Bada boom. And this is super robust. Like, it's not, like, people are like, oh, how? How do you know if the code is like, trust me, you're going to see. Like, this thing is overkill. This thing is so smart, it knows every little thing about how to self-improve from A to Z. So there's already a brilliant system in place that can take the code and make it live itself. One of the greatest joys I have with this is improving Hive AI, essentialize AI, with... But with itself, I'm sitting, it's like, I imagine, like, you're operating on yourself, and you're, like, you're adding, like, a, like, a fucking metal rib or some shit, like, your jacks off of Mortal Kombat, like, adding some fucking laser eyes to yourself and, like, improving yourself, if you think about that, that's what it is, because this AI, I use it to improve itself, sure, there was a bug that it solved that Codex and Claude could not solve, and it was one of the happiest days of, like... of my coding career. It's like, a little tear almost came out of my eyes. Like, wow, that's amazing. Of course, Codex and Claude could have, when I said it, they didn't one-shot it. They did not one-shot it. But, of course, they would have got it on a third, force try, or whatever, if you would have kept digging. So it's already on that par, and it's... It's interesting because it has superpowers that the other AIs don't quite have access to because it has so much information on very niche things. Like, you know, you go ask Grock and knows a lot about celebrities and all that. Like, it's hard to know everything. You know, it's hard to know about Justin Beaver's weekend and, like, hardcore cutting edge new shit when it comes to coding. Like, you're going to miss some shit. And I noticed that because when I go and ask, because I'm not just a researcher, I've, I've, I've, Flip that topic upside down. This has made me so powerful that I've known, I can always test, always new information, always new features, always new things, and the thing that I've made, the blob, I call it, it ingests everything and spits out the skeleton that makes the bones into a fucking necklace. Like, I find every little kernel of ideas out there, and I mission match them, I synergy them, I picture book them, yeah, these are the names of my shit, but this is what gets me to put to the point where I understand all of this shit and how all these things work together. So, getting back to this little feature that's actually quite big, these, when it comes to actually coding, it's beaten other systems or is on par with them. And that is just with my setup. Once you get hundreds of GPUs, because keep in mind, Exing out of the TIG game, you understand that this is just a really cool way to mine crypto from the thing. Anytime you use this, it's going into the rag, and once it learns something, it gets better and smarter. So, and there's a way, you have the second brain. I know people probably heard of the second brain from Obsidian that everyone's using on Clod. I use it, too. It's very cool. If you're not using Obsidian for your Cloud or Codex, you absolutely should, so your memory file doesn't get all. clunked up. Learned all that on Hive. Hive has a memory file like that, but it automatically keeps it dense. You don't have to fucking say, hey, why is my memory file bloated? It's fucking high the guy knows. It's like my memory file can't be bloated, bro. I have a second brain built into me. That's the whole point. Like, the IDE is the home. Everything's built into it. It has to act that way. And it has the second brain right here. And the great thing about the second brain is you can mint golden books to... The rag. So let's say you got a bunch of fucking cool shit in your second brain, and then it triggers. It's like, oh, you're eligible to try to mint a golden book, which will go into the laura, or which will go into the rag as a laura pair. And I might not be saying these things, right? You know what? I focus on how to get shit done. I don't care about, I'm not like Googling what the word is, so I can hear it. And, but, oh, okay, I don't give a fuck what people think. Let my products speak for themselves, because obviously I know what the fuck I'm talking about. I can call a giraffe a graph. I don't go to fuck, because if I know 10,000 damn things about the giraffe, you better shut your ass up. So you mint, and it goes into the rag. And if it's approved, you earn crypto. So it's like, you have this little obsidian on your little codex, and it's like, oh, that goes cool shit. But you're not earning crypto with it, and you're not making a system smarter. You're making the centralized AI smarter. And I don't even think they're a threat, because they're such a loss leader. They lose so much money, it's going to be a, like, they're going to milk it very soon. Like, they can't just keep losing money forever. It's an attention grab, and eventually it's going to become surveillance state, and they're going to be floated up by secret programs. That's how the fucking... The show is going to keep going on. And nobody's going to like that, and it's going to lose quality, and like they always do. Facebook started out, cool, and then it became a fucking surveillance to everyone does it. This skips all of that. So your second brain, actually, you can mint to the things. So it's like you're constantly absorbing information, and then it has hardcore duplicate check. So it's like you're not putting new shit or old shit on the rag. That's already there. It already knows. It's smart enough. But it's also smart enough to, like, say, you have 60% of a pair. 40% duplicate. Like, you can then improve your pair over time to get it to go back in. So there's all kinds of ways to earn crypto. Hive makes it perfect with incentivized Stacore Distribution. You can upvote. You can earn for participating. So, yeah, I'll just... I just do a little GitHub, this one, this one, show a little bit of the capabilities, summarize this, and then you hit send, and it says 282 plus that's way off. It's well over 10 million right now. So now look, it's searching, you see it, it's using its tools. And now it is spinning out. And look how fast that was. Dude, this is a, this is a Gwen 2.514B LLM on a 16 gig hard thing. This is one PC right now. I got my other PCs doing other shit right now. Let's read it. I ask them to summarize it. I have messengers. It's essentialized end-to-end encrypted messaging. PWA built, I mean, y' guys can read it, but look at that. Zero back in, end-to-end, group chats, text, boom. It uses tools built in, and went and got it in what, like five seconds on the internet. I can edit my own files, I can do anything with this. And the cool thing is you're just going to log in and you're going to have like a little thing here and you're just going to be earning credits when you're not using and when you use it, your credits are going to go down. But you're going to have free credits based on what you provide, what you're coming into the system. So it's like you get to use your LLM, but it's better because you're attached to the rag and the harness and everything else and all the other GPUs to help run interference and all that shit. And all the other people, like, as they joined, they're bringing more information into the reg, more…