Announcement
METR says frontier AI time horizons doubled roughly every 3.6 months in its tests
METR says capabilities relevant to AI research and development are improving rapidly—and getting harder to measure.
TLDR
Across its TH1.1 and TH2 tasks, METR reports frontier time horizons doubled roughly every 3.6 months, with a 95% confidence interval of 2.1–4.9 months. METR says agents do somewhat worse on costlier tasks of the same length, but doubling times are similar for cheaper and costlier tasks.
Combined views
27.9K
9 Sources, first seen ago

Comments