220.md

00:00 - Car: I'm gonna hit record now. [Sound of a boop] Boop. And we're back.

00:00 - Chris: Car, I tweeted at you, I tweeted at you, and now I'm tweeting at you to your face. We need to do... we need this thing. What is this?

00:13 - Car: I mean, the only thing that's stopped us is just... nothing.

00:17 - Chris: Well, I don't have a hundred million dollars to give you. I've got... I can buy you lunch.

00:22 - Car: We could probably do it for more... for a little... for a little more than lunch.

00:24 - Chris: Do you even have a podcast, first of all? Do you even have a podcast for me to buy? Or did you just make up a podcast for me to buy? You told me you had a podcast. What is this podcast you have?

00:33 - Car: It's called Early Days. You can buy that one.

00:35 - Chris: Well, I don't know that that's an appropriate name. And look at this: TPBN. We need a cool acronym first of all. What the fuck does this even stand for? Technology something? No one knows. Okay. So I think ours is going to be called OAPP, okay? OpenAgents Propaganda Podcast. We'll start with that and then we can change it later if we want to. But working title is OAPP.

00:55 - Car: OAPP. OAPP.

00:57 - Chris: TPBN. Let's just look at... let's just see what we're... what is this? What are these people doing? Signull has good tweets, right? "Short bus at the end of the rainbow." This is... this is an anti-OpenAI podcast by the way. This is an anti-Big Lab podcast. We're competing with these people, okay? We've been competing for years. And like... like they got a propaganda arm, we're gonna make a propaganda arm, except we're not gonna bullshit you and be like "editorial independence." I'm paying Car. This is just gonna be fucking propaganda, okay? Is that alright? Let's just call the spade the spade.

Now, let's take a look at this. Signull has talked to probably at least ten people I deeply respect about this OpenAI move and it was almost universally touted as complete WTF. Blah blah blah blah blah. Significant downside. Look at this, look at this. Let's... let's read this. I have no idea who this is, but I like the words, so let's read it. "I've previously wondered if OpenAI might be like Twitter, another text-centric company that fell backwards into a huge market and never developed into a functional business because of it. If Twitter is a clown car that fell into a gold mine, OpenAI might be the short bus at the end of the rainbow." [Laughs]

01:58 - Car: I mean, gosh. I mean, they... I mean, they canceled their what... their image thing, right? And now Google has it.

02:04 - Chris: Well, they're... they're like shutting down all their bets. They're focusing on one thing, which is like playing catch-up to Anthropic. They're like "we need to spend all of our money on catching up to Anthropic, except we also want to scoop up this thing for a hundred million dollars."

02:17 - Car: Did they ever build out those data centers? Or those infrastructure?

02:19 - Chris: No, they abandoned the Stargate plan. They've reframed it as "Oh no, no, no, we're not building our own thing. That... that press conference we did with President Trump and we're like 'we're gonna build all this shit,' yeah, we're not doing any of that. We always meant that it's just gonna refer to our compute aspirations broadly," which is what? Spending six hundred billion dollars to freaking pay big cloud people? Okay.

So, we're gonna meander a little bit here folks, but on the subject of paying hundreds of billions of dollars for compute. Okay. So... so here's OpenAI. Here's what they project to spend. Something like six hundred billion dollars on compute. And our big thing is like okay, what if... just imagine for a second that you've got hundreds of billions of dollars or just billions of dollars of compute. What if that doesn't go to NVIDIA? What if that doesn't go to the big cloud companies? What if we can have a way to have models trained by retail consumer compute, like some of these other projects are already starting to doing, where they're doing these large decentralized training runs? Here's an example.

Okay, so here's our guy Sero. So we threw a little bit of Bitcoin at Sero here. Amazing open-source AI developer. Follow him and support him if you haven't. But he's... he's here's a tweet that he has and I riffed off this, but this is good. So he says, "This didn't receive the attention it deserved. They... they being this BitTensor subnet, more on that skeptically in a second, but this didn't receive... okay. They pre-trained this model completely peer-to-peer, no data centers. Everything was done over a permissionless network. I have tried the model. It's honestly not a good LLM but that's beyond the point. We need this, we need an alternative."

So, download OpenCode, download Pi, pay for open-source, share your AI sessions, learn to do RL. We can't be at the mercy of any lab. Agree. Now, we want to do all that, yes, but we don't want you to have to learn to do RL. Let's like kind of crowdsource that. What we want is for you to be able to like install our app and then you click the button and you're using your compute. So, we're gonna do our own version of this with a hundred times the participants. So... so the Templar BitTensor run, they had seventy people. You contribute your compute, you turn on your thing, you send some compute here and there.

04:26 - Car: I mean, I have it right there on that... that Mac Air right there is going...

04:29 - Chris: Yo, that'll be one of our devices. Like, why stop at seventy people? Well, maybe because you only had seventy people who give a shit about your shitcoin TAO. Well, how about you use Bitcoin because everyone wants Bitcoin?

04:40 - Car: It makes sense.

04:41 - Chris: So let's get seven hundred people. Let's get seven thousand people. Let's get seventy thousand people. Let's pay every single fucking person who's got spare compute sitting on their desk that they're not doing anything with to contribute to a massive fucking training run. Why not?

04:52 - Car: Let those sats flow, I mean.

04:54 - Chris: So we're gonna have to do a little bit of like research about like no one's... no one's done a training run at that scale. No one's... no one's. But so just we'll get more into the details of that. And look, the point is not to build a model that competes with Qwen, but there's things that these models could be specialized to do. Models that focus on local, edge, agentic tool use. And then we've talked about this Perceptapost. Like, there's some pretty advanced shit that we can do that the other labs wouldn't be able to do.

05:20 - Car: This is so cool. Kind of break down like how you're seeing like the Autopilot stuff. I know you just dropped the... the new one... last week. Like how do you see that kind of like working together in conjunction with kind of like this OpenAgents kind of distribution distribution network you're kind of building? Yeah, how do you see it all playing out?

05:37 - Chris: Yeah, so we want people to be able to install a single piece of software that does a bunch of cool shit for you. So, right now, obviously OpenClaw has proven that people want personal... is that the camera right there?

05:48 - Car: That's this one right here. Yeah.

05:49 - Chris: OpenClaw has proven that people want a personal agent. Like fastest open-source project. It's obviously like not very good. And you can see all of the people at least on my timeline like happily switching from OpenClaw to Hermes agent from Nous. Nous, big fans of Nous. Hermes is awesome. Someone today was like, "Why don't you build on Nous? Build on Hermes?" because I was like "Well, we are allergic to Python. But yeah, like we'll have Hermes people be able to contribute to this." But the idea is that we want one app that you install. You could call it a desktop-based super app, by the way, which is exactly what OpenAI's pivoting to. So we're happy to contribute to that.

But it should be as simple as you click a button and you're like going online for compute and... and contribute to that. But we want it to do all the kinds of things that an OpenClaw does. And the beautiful part about everything being open-source and ours being open-source too is that like there's no reason not to just pull in the best code from everywhere else and fold it into one piece of software. We have... alluded to there's some maybe some new open-source code that dropped from last week from some certain big lab coding agent code leaking. You know, we might have looked at it. You know, we may have looked at it. Anyway.

So OpenAgents Autopilot is the core product. The thing that we're launching today and in a... well by the time you're seeing this, it'll be up online. So you'll go to our little website here and you'll install this. You'll copy this.

07:23 - Car: Oh, nice.

07:25 - Chris: You'll copy this and you'll give this to your agent. And this is going to install for you. So we've got Autopilot, that'll be like the overall umbrella product with the agent built in. We're launching right now Pylon, which is the part of it that just sells your compute. So we want you to like copy this, give it to your Claude Code, give it to your CodeX, give it to your Hermes, your whatever, your OpenClaw. And what is this going to do? It's going to install Pylon for you. It's going to look and see what your system specs are. It's going to download the appropriate open model. It's going to put your open model in a distributed inference network. It's going to do two things: your compute's going to be available for running inference for local models and be available for training runs.

So we're about to kick off our version of this thing I said with the BitTensor thing because all their code's open-source. All of them and PrimeIntellect are doing are implementing DiLoCo originally from DeepMind. All that code's open-source. Cool. Port it all to Rust. Now we have it in ours. We have one single piece of software. You click the button and you're contributing to compute. Let us pay you Bitcoin, okay?

08:23 - Car: It's a no-brainer. I mean, when you think about it, I mean, we used to do this in the 90s, remember, with SETI@home? That used to be a whole thing. Give people their compute... give people that compute back then and now you actually get Bitcoin but you're not searching for aliens, you're searching for Bitcoin.

08:38 - Chris: Yeah, and we're going to call this compute mining because we need a... we need a good phrase. And you know, speaking of phrases, I just got... we just gotta... we gotta tie this back to the main theme of kind of punching OpenAI. Give me... give me a second. We gotta tie it back to... [Humming] Sam Altman will be remembered as an amazingly ambitious narrator of words for AI. Robot taxes, efficiency dividend, right to AI.

So this comes from OpenAI's... so we got... like they go from hilarious propaganda acquisition to a report out this morning from another outlet that they freaking pay and hire and they have disclaimers about how they're like influenced by OpenAI. "Oh, Sama tells me that he feels such urgency about the power of coming AI models that OpenAI is unveiling a new deal for superintelligence - ideas to wake up DC. He says AI will soon be so mindbending that we need a new social contract." Be careful here people, new social contract, "hey government, do our regulatory capture bullshit," and then are these people massive shareholders of that fucking Worldcoin thing? Just scan your eyeballs right here to prove you're human. Sign up for your fucking government commie UBI. Fuck this. What is it? It's commie bullshit. No to all of this. If you want people paid, pay them dividends yourself.

You have all this money. Why did we start OpenAgents? Episode one of our series. It was the day after Sam Altman said at DevDay 2023, "we're gonna do a GPT marketplace because we think agents are the future." Here's our first version of agents.

10:11 - Car: What year was that?

10:12 - Chris: November of 2023. Episode one of our 220 episode series. This is episode 220 by the way. But you click here, there's our repo. And what did they say? They said "we're gonna do revenue sharing." And I said in episode one, like they're gonna half-ass that. There's no freaking way they're gonna do what they should do, which is contributors to agents, contributors to anything relevant here, any workflow that someone's paying for, the contributor should get a share of revenue.

So, fast forward, how many people has OpenAI paid revenue share to? Zero. They paid zero. How many have we paid to? 30. But look, infinity percent more than OpenAI. Error divided by zero more than OpenAI. So you have the money, you could pay dividends. And so a big theme of... of OpenAgents, and I'll touch on one topic and then we'll start wrapping this up. Go to episode 200... this is our sort of like master plan, kind of like what all we're doing here and we're building this big agent network.

11:08 - Car: You tell 'em in the open too, I mean, you gotta respect that.

11:10 - Chris: And there's the problem is that there's been just too much shit. So we're gonna make a podcast, we're gonna like start touching on these topics a little bit more... more... more bluntly. But like, everyone's freaking out about labor displacement blah blah. Okay. So what's going on here? You've got deflation plus dividends. So deflation is like technology makes things better. Life gets cheaper at the same time more people can earn from the network's ongoing activity. But deflation, the benefits of deflation should be broadly distributed to the people, right?

But if you have a company that's passing the benefits only to their own shareholders and they're making their like... like fucking like overseas sheikhs super wealthy, but you're not seeing any of that productivity. Maybe your... your agent is makes you a little bit more productive but like... and then the benefits of that deflation are not trickled down through society because what we have? Inflation and your... your money's being devalued. So we're saying that the missing piece here is dividends. You have to take AI workloads and you pay people Bitcoin. Like, this is the only solution to all these labor problems and like job loss is fucking pay people Bitcoin.

So this is the big theme of... of our... our podcast. And if we need a more serious name for the podcast, I was thinking call it like "Pay me Bitcoin." Pay me Bitcoin. Pay me Bitcoin. Can we pay you Bitcoin? Okay. Does this make sense?

12:23 - Car: No, this makes a ton of sense. I think... I think you're exactly right on the pay the dividends part because no one really sees where the... you know, just the the work... the work that has to be done in order for things to move forward. And I think people are underestimating that... the workforce part about this. And I feel like with what OpenAgents is doing, it's a little bit more focused on paying the actual end user for what they're contributing to the network. Which it's how it should be.

12:52 - Chris: In conclusion. So yeah, just imagine like the billions and billions and billions of dollars of spend on compute for inference and training. Just imagine like a bunch of that, most of it goes out to actual plebs. Plebs, you.

13:09 - Car: This is really early too. I mean, that's a thing I think that most people don't realize is that you're literally getting in on the ground running. Like, you know what I mean? Like imagine if you were there at the inception of something like the Bitcoin network starting out. This... this could potentially be that all over again.

13:24 - Chris: Yeah, it's the kind of like appeal that like shitcoiners will make to pump their token. Except we're Bitcoiners, there's no token. The token is called Bitcoin. If you're an accredited investor, you may reach out to invest in Open... Okay, so let's close on this note. You know, a lot of people on Twitter talk about like "what's your moat as a company and who's gonna be investable blah blah." Let's tell you about the moat of OpenAgents, which is like interesting because we're building on open protocols and open source. So it's like it... it takes a little bit of thinking. But we've been thinking about this for three years.

So here's what we've come up with. "What's your moat? Our moat is dominant network effect where contributors of compute but then also data, labor, liquidity, risk verification participate primarily in our open ecosystem because they get paid the most Bitcoin. We strongly doubt anyone is going to get humans and agents paid more Bitcoin than us, and even if we do bait other players into paying you even more Bitcoin, hey, good for you, good for us anyways. Which means we have no reason not to call our shots here sharing alpha in all of our 220 videos, telling you precisely all the epic shit we're building. You are incentivized strongly to help us because you'll get paid the most! Not worried about anybody copy pasting our software or copycatting the ideas, which is inevitable anyways, but because the value is in the network, the networks can't be copied. The biggest and best network will form, can only form, on anchored or adjacent to the most neutral open permissionless network, which is called Bitcoin."

We're just building the agentic infrastructure around that, interoperable with everyone else who's also building the Bitcoin stack. So, we are launching our GPUtopia-style Bitcoin for compute network next week. Aka you'll be able to go to OpenAgents.com, you'll see this little landing page, you'll click this button, you'll feed this to an agent. What should happen, and if there's a bug please tell us about it, but like what should happen is your agent is like [Sound effect] you're now online, here's your Bitcoin wallet, and then you're just counting your sats. And you're gonna see in these little dashboards that we put up here how your compute is used in training runs. And then we'll just figure out how to make that better and better and better until we're running circles and beating the shit out of these companies that have raised 200,000 times more money than us. But not for long.

15:24 - Car: Bullish. Share the... the YouTube.

15:26 - Chris: The YouTube. Did we fix the typo here? Refresh. OpenAgents, there we go, one word. Alright, so we're gonna ramp up our social... we got two subscribers. Two. Who wants to be the third?

15:36 - Car: Hey, you gotta grow... you gotta start somewhere.

15:36 - Chris: Car, thanks so much. Final words to you. What do we think?

15:40 - Car: I think what you're doing is beyond next level. I mean, we've been watching you doing this from the very beginning in '21, '22. Everything you've said that you were going to do has come to fruition. And if I'm somebody who's looking at this right now, it's like who's... who's the guy that's actually calling his shots and then executing on those shots? And you're like one of the few founders that I know doing this in AI that's actually doing that.

16:03 - Chris: I paid him way less than a hundred million dollars to say that. Very good. We're building...

16:08 - Car: It's true though. It's true. And all the proof of work's out there. It's not like you can't... you can't... there's no one that can take that away from you, all the work's out there.

16:11 - Chris: Yeah, we do... we do have a... we do have a lot of a lot of videos on our Wikipedia here. Okay. It's happening. OpenAgents layers on Bitcoin based here in Austin. We're hiring. See our website and stay tuned. We'll see you soon.

16:25 - Car: Stay tuned for the... the propaganda podcast. [Laughs]