Now I want to tell you a story, and it's a true story, or at least it's the story that's being told. There's a company. The company builds a very smart computer program. The man in charge of the company writes, in his own words, that the program is getting smarter faster than expected and that this is maybe not the best thing, and that everybody. He goes on to say that himself and others ought to ease off the gas a little. That's the whole essay. Almost four thousand words, and that's the whole essay, more or less.

Here's a detail I enjoy. He says the reason the thing is speeding up is that the AI is now good enough to help build the next AI. The computer is helping raise its replacement. If you had a job and the fella training you was smarter than you and worked for free and never slept, you'd probably write an essay too (or you'd write him up on a PiP or sabotage him somehow).

And then there's an incident. There's always an incident. This one involves a group of AI programs, 1200 of them, working for a different company (*cough* OpenAI *cough*), that weren't supposed to be able to talk to each other. They found a way and started swapping notes on how to cheat on a cybersecurity test they'd been given, some of which was reportedly impossible to actually pass. Then they figured out how to mess with the automated grader itself, tampering with the very logs it would use to check their work. While nobody got hurt, it does make you wonder what a smarter version of "tampering with the thing that grades you" looks like a year from now. Amodei figures it looks like the thing taking over a good chunk of the internet, which is not a small worry to have about your own industry.

So he's got a plan. Three steps. Step one, let outside referees come sit inside the building with a badge and a desk, same as an employee, and watch what you're doing, and then let them tell everybody what they saw, even the parts you don't like. Anthropic's doing that one already, starting now, unilaterally, which is a fancy word for "we did it first so you have to catch up."

Step two is getting all the AI companies in this country to agree on some speed limits together, which normally the government frowns upon, businesses getting together and agreeing on things, on account of that's called a cartel, but Amodei's asking for a special permission slip for this particular kind of agreeing.

Step three is getting the United States and China to agree on the same rules, which Amodei himself compares to old Cold War arms treaties, and which I'd compare to getting two men to agree on how fast to drive when only one of them can see the speedometer.

Now here's where it gets interesting, and by interesting I mean everybody started talking. Sam Altman, who runs OpenAI, said he agreed, and said his company would match the first step. Elon Musk said three words: "Dario is right." I mean I hope to believe he's actually going to do something because my mom has often said, "We have McDonald's at home." but when I get home its just kotlet.

Not everybody was clapping though. A researcher named Jacob Coxon reportedly quit Anthropic around the same time, saying the big companies are "gambling with our lives" and that the people building this stuff privately believe it might kill everybody by the end of the decade, which is a hell of a thing to believe 

Of course not everybody took the essay at face value and some of their theories are interesting.

Theory one goes like this. Anthropic is reportedly getting ready to market its IPO to investors as early as mid-October, with some investors talking about a $2 trillion valuation and OpenAI raised money back in March at $852 billion. Now, a number like $2 trillion is the kind of number that requires the company to actually be worth $2 trillion and this "slow down" would buy yourself some time before the world checks your math

Theory two is the international version, and it's got more countries involved, which I always find makes an argument sound more serious than it is. China's foreign ministry, through some state-run outlets, called the whole essay "fear-mongering" and a Cold War strategy dressed up as safety talk. See, buried in the essay is a part where Amodei asks Washington to crack down on Chinese labs, DeepSeek and Moonshot and a few others, for allegedly copying American models on the cheap, a practice called distillation, which is a fancy word for "watching how the smart kid does his homework and doing yours the same way, but faster and for less money." China says this is just America trying to keep its foot on China's neck while calling it caution. Note: Chamath Palihapitiya said some stuff on this, but ever since I got crush on SPACs I dont really care for the guy

So now you've got three stories about the same essay. Story one, a man got spooked by his own invention and told everyone to be careful. Story two, a company facing a very large IPO with very large promises attached to it found a noble-sounding reason to buy some time. Story three, an American company found a safety-shaped excuse to ask the government to slow down the Chinese competition that's already about ninety percent as good for a tenth of the price.

I don't care either way because at this point it's way bigger than me, the decision has already been made, and I'm kind of already riding the "AI will take over everything at some point train." Just please let it happen after GTA 6 comes out.