Ten31 Timestamp: In It For The Tech

**Marty Bent** (0:00)
John, it's over, Bitcoin's dead.

**John Arnold** (0:02)
For the 285th time, we have to, unfortunately, attend Bitcoin's funeral. It's as sad as it always is.

**Marty Bent** (0:09)
The eulogies are getting better, though. They're getting better, it was fun. It was a nice college try. We kid, we kid.
Obviously, a lot of people out there wondering what's going on with Bitcoin, what is happening? We'll get to it, but I think we want to start on another side of the market, AI. I think a continuation of our conversation from last week, but first, before we get into any AI dividend funds for localities, we'll talk about the progression that Anthropic is really pointing out here in terms of what the models can do. It was actually funny, I was at a lunch on Friday, where we were talking about this and naturally got on the conversation of humans thinking linearly and not being able to think in exponentials. And if what Anthropic said last week is true, which was in this When AI Builds Itself blog that they published, we may be getting closer to some weird moment where maybe AGI, who knows, but basically Anthropic saying that what are the 80% of the code for new models is being written by Claude itself.

**John Arnold** (1:12)
Yeah. Well, we say, you said, we're starting on the other side of the market and we'll get to Bitcoin. But I think it's worth remembering as we talk about all these that increasingly, I think, not to go to Pepe Silvia on it, but I think it is all increasingly connected given how big AI is becoming on various dimensions. But yeah, I thought it was good to start here just because I think this maybe didn't get enough airplay in mainstream press, at least as far as I saw. This is the kind of thing that recursive self-improvement, this has been like the goal that a lot of us probably only really started paying meaningful attention to AI and progress in AI over the last few years, as it's become mainstream and more involved in our lives. But this is the kind of thing that if you go back and look at what people in five square miles in San Francisco were talking about in the late 2010s, early 2020s, this is the hurdle that you want to get to.
If you're extremely, call it AGI-pilled, and whether you're positive or negative on that future, if that's what you're looking for, you're looking for the moment where AI can be the best AI researcher, right? So you know why you take humans out of the loop or minimize their involvement, and AI analyzes itself to build like the next, one generation of a model analyzes itself to build the next generation of itself, right? And obviously, that can lead to some potentially frightening conclusions and some paperclip maximizer type futures that we could think about. But despite all that, I didn't really see this picked up on a lot of mainstream press headlines. And to some extent, they're talking their own book here, right? This is a marketing document from Anthropic. They're incentivized to tell you that they're just on the cusp of getting to this big milestone. But there's a good amount of data in there that points to the fact that this probably is happening. And it seems important.
So to the extent that this is not just marketing BS, it validates a lot of what we said last week about the pace of progress here, if you want to call it progress, potentially leading to some meaningful, discontinuous changes in the economy in the next few years. And that will have a lot of downstream implications that we have talked about and that we will talk about.

**Marty Bent** (3:24)
Yeah. And it's funny too, not only did this blog post come out, and I agree, I don't think it was picked up by enough people. I was happy that it was recognized at lunch on Friday when I brought it up. But another, you mentioned Human in the Loop, taking them out for the model building themselves. But I'm not sure if you caught this, but you had Boris from Anthropic and Peter Steinberger from the OpenClaw project basically coming out and saying the same thing, but on the prompting side, like they don't actually prompt agents anymore. They simply set up these loops of agents prompting each other, and they basically set it off once. And the way they're looping architecture works, the AI just builds everything for them. They don't even prompt anymore.

**John Arnold** (4:06)
Yeah, that's wild.
It's only going to get weirder. And I think to be fair and not get too sensationalist, there's always a diffusion curve, and the vast majority of people have largely never even heard of this stuff. There are enterprises everywhere that will be slow to adopt for reasons that we've talked about. They may get disrupted, and they may ultimately be dinosaurs that go extinct. But the call here is not that in 12 months, every white-collar worker will be without a job. But on the margin, you could start to really see if this continues and accelerates, even with the diffusion delays that you would expect to see, some meaningful impacts on at least segments of the economy, which maybe flows into the next slide nicely with a long-time political allies, Bernie Sanders and Donald Trump, both arguing for the government or the public, as it were, to take a stake in leading AI labs so that the public at large can benefit from all of this crazy growth that we're seeing here.