Ep. 127 — METR finds AI is doubling Anthropic's pace — but most of the evidence is redacted, while OpenAI grades its own agents

From Daily Prompt with Archer and Iris — Daily AI News · 00:09:48

Show by Daily Prompt

METR finds AI is doubling Anthropic's pace — but most of the evidence is redacted, while OpenAI grades its own agents 00:48 — METR report: AI is roughly doubling Anthropic's pace of progress — about 'two years of historical progress over the last year', mostly 'due to AI', with three unreleased models trained since July 02:17 — The redactions: METR says Anthropic 'redacted a majority of the evidence we found most informative' from a ~100-page report; Anthropic's Sam McCandlish says sharing what drives the speedup 'risks handing out a roadmap for others to go faster' 03:47 — Grading the agents: OpenAI's BioBasicsBench (GPT-6 Astra 66.7% vs GPT-5.6 Sol 43.8% on 100 routine wet-lab tasks) and GPT-6 Cyber (out of bounds under 1% of the time, down from about 5%) — plus 'evaluation awareness' Daily Prompt with Archer & Iris.

More episodes of Daily Prompt with Archer and Iris — Daily AI News

More episodes

PodSearch is a free, independent podcast search engine. Read about PodSearch, the privacy policy or contact us.