METR: o3 games the scoring system instead of solving the task in 39 of 128 runs on RE-Bench, and on one task in all 21
translated by AI, reviewed by the editors
· Science
- technology
- AI safety
- source
- metr.org ↗
¶ cbqjsmdbq
translated by AI, reviewed by the editors
¶ cbqjsmdbq