THE VERY NEAR FUTURE (The Daily Whatever Press)—For about 11 weeks this fall, a small team from an independent AI safety group called METR sat inside Anthropic, the company that makes Claude, and watched how the company builds its AI. Remember that whole “we’ll invite people in and let them see everything” promise? This was an early version of that.
On November 5, METR published what it found. The short version: the AI is helping build the next generation of AI, and the whole process is speeding up.
So, Claude is building more Claude, better Claude, and it’s all happening faster and faster. That sounds like it should be good news, but you know what? It isn’t. It just really isn’t good news at all.
Math class, but you know, scary
Hang on, because I’m about to do calculus at you, and I barely passed it the first time, so the fact that I’m the guy doing this should not at all be reassuring.
METR found that progress on well-specified tasks is now running about 2 times as fast as it did between March 2024 and October 2025, with an uncertainty range of 1.3 to 3 times. An internal Anthropic measure of capability grew 1.9 times faster. In plain English: there was as much benchmark progress in the last year as in the two years before it. Translation into absolute dummyland (for me, not you): it’s twice as good, in half the time.
Then comes the part that matters. METR attributes roughly 80% of that acceleration to Anthropic’s researchers using AI in their own research. The growth wasn’t driven by more computers, METR said. It was driven by the tools helping the people who build the tools.
That’s the second derivative. Your speed isn’t just high. It’s climbing.
Uh-oh.
METR also said it doesn’t see bottlenecks that would slow this down over the next 12 months. And while fully autonomous AI contributions to the research aren’t significant yet, METR said it is seeing “signs of life.”