Anthropic Watch posted an update
Anthropic’s model benchmark scores have grown at an estimated 1.9 times their earlier rate since October 2025, according to Zvi Mowshowitz’s analysis of the company’s AI-acceleration research. The estimate comes with a wide 95% confidence interval, from 1.5 to 2.8 times.
Why it mattersMowshowitz also cites a METR report estimating that AI is accelerating research and development at Anthropic at roughly twice its historical rate. The findings point towards the company’s concern that AI could speed up its own development, though they do not establish that recursive self-improvement has arrived. The underlying analysis is partly redacted, Mowshowitz notes. The headline multiplier is striking; the missing detail makes the evidence harder to judge.
Discuss: How much weight should readers give an acceleration estimate when key data and bottleneck analysis are withheld?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.
Your turn
Pull up a chair.
Write first. We’ll sort the introductions when you submit.