Open-set assistance on Overcooked Defects (held-out)
77.8Coaching (LLM-as-judge)Open-Set Corrective Assistance
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Open-Set Corrective AssistanceModel Scale=8B2026.03 | 77.8 | 54.6 | |
| Open-Set Corrective AssistanceModel Scale=1B2026.03 | 76.6 | 55.7 | |
| Behavior Critic + SummariesSummaries usage=true2026.03 | 55.7 | 19.8 | |
| Behavior Critic2026.03 | 21 | 20.4 |