METR says Anthropic model progress roughly twice as fast as a year ago

Cointime reports: AI safety research group METR said Anthropic’s models are improving about twice as fast as a year ago, with the past year’s gains equivalent to roughly two years of prior progress. In a report following an 11-week investigation, METR attributed most of the acceleration to AI, mainly AI-assisted human researchers rather than autonomous research. Anthropic redacted much of the public report, citing commercially sensitive information.