METR: Anthropic model progress rate is about twice that of one year ago

深潮 TechFlow reported on November 7 that, according to The Verge, after an 11-week investigation of Anthropic, AI safety research organization METR said the company's model progress rate is about twice that of one year ago, and progress made over the past year is equivalent to about two years of previous progress.

METR believes that as of October, AI accounts for most of the acceleration in R&D, and its contribution is growing. The acceleration currently comes mainly from AI-assisted human researchers, rather than autonomous work. Employees using Anthropic's strongest internal model self-reported in an informal survey that their work efficiency was on average about 12 times higher than without AI assistance.

The report noted that Anthropic made extensive redactions to the public report. Co-founder Sam McCandlish said the relevant information involves commercially sensitive content and could also help other companies accelerate R&D.

Source: 深潮 TechFlow