Is AI Speeding Up Its Own Creation? An Outside Check Inside Anthropic Says Probably Yes
A big worry in AI safety is a feedback loop, where AI helps build better AI faster than people can check it. The nonprofit METR spent eleven weeks studying Anthropic, the maker of Claude, from the inside. It concludes that Anthropic is probably improving its AI about twice as fast as it used to, and that most of the speed-up likely comes from using AI to do the research. But much of the evidence was withheld from the public.
In this episode
What METR did, what data Anthropic shared, and the limits on METR's access
The findings: test scores rising about twice as fast, not explained by more…
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.