RE-Bench (METR): on 7 ML research engineering tasks, AI agents with a 2-hour budget score higher than human experts on average, while with 32 hours the average human score is almost twice that of the best agent
translated by AI, reviewed by the editors
· Science · Software development
- technology
- Recursive self-improvement of AI
- source
- metr.org ↗
¶ csrkaiiv2