SafetyOpenAI
OpenAI said its coding agents now finish 46% of two-to-four hour research tasks alone, up from 23%
46%of OpenAI's two-to-four hour research tasks done by agents alone in October, from 23% in March
OpenAI published a research acceleration report on 18 November. It measures how coding agents speed up OpenAI's own AI research. From March to October, agents' solo success on two-to-four hour tasks rose from 23% to 46%. On four-to-eight hour tasks it rose from 13% to 35%. Success was 79% on monitoring training runs but 25% on inference and serving code. Median researcher agent use reached $2,400 a day at list prices in October. OpenAI says labs should have to report such self-improvement indicators.
OpenAIfrom the source itself