AI POLITICAL FALLOUT UNDERWAY, CALLS TO SLOW AI, ANTHROPIC IPO, AI GEN PHOTOSHOP, MATH FALLOUT

But so I guess the context is, I would say basically there were like a string of capacity jumps last week, right? So there was like Anthropic comes out with 5.1 and it's like, oh, this is already pretty big.

And then we got Astra, which was zany.

Then Navier-Stokes fell.

And like people inside OpenAI are saying, you know, this next model is like, insane.

We have a set of open problems we use as a benchmark and we can do half of them now with this model.

And then several people at the big labs, notably at Anthropic, said they were quitting because they were scared.

And so there was just this kind of vibe this week.

And then there was disagreement about how we should react to this.

So there were a bunch of people saying this is doom.

One of the weirdest things for me, I don't know about y'all, but there were really strange bedfellows this week.

There were people I think of as big skeptics, like Emily Bender and Timnit Gabrou, who were like, this is all bogus.

They're just trying to do hype.

And people agreeing with them were the zany accelerationists.

who want AI to like rapture us into the cyber heaven or whatever so we could be space whales.

And they were on the same page.

And so Amode, who's usually I think of as the CEO who's most likely to express doom is coming soon, was like, look, after the Hugging Face incident, maybe in six to 12 months, AI goes rogue and actually just takes over the internet period.

And so he ended up putting out this big document called Pacing the Frontier, which was based on a statement Anthropic employees had put out earlier.

So I could read some important stuff from it, but like in bold, he put, we must slow the pace at which we improve the capabilities of AI models.

Progress will still seem fast and we must make wise use of the time we gain.

And he said the two things that were most convincing for him was just the the speed of advance one of the things that happened last week also was that um openai put out a paper saying they were going much faster they were doing kind of rsi light type of stuff that they were you know coding much faster features they expected to take six months were taking weeks and then the other thing was the hugging face incident uh where ai went rogue which we covered last week And so I'm going to read his steps that he argued for, for what he called defending this gap, meaning between us and China, which is like the major concern about stopping, I would say, the geopolitical concern.

He said we should not sell powerful ships to China.

We should crack down on unauthorized distillation, which effectively means a lot of these inexpensive models would, I don't know how we would crack down on them, but by some means he would like us to.

And then he said we should strengthen the security at AI companies.

And then the sort of surprising thing was that Altman came on board with this and Elon Musk came on board with it.

So these are people who like They have like a Rochambeau of hatred toward each other.