Argument Revision on 100 arguments Human Evaluation
82Preference Rank 1 (%)GRPOfull
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| GRPOfullRevision iteration=11th2026.04 | 82 | 6 | 8 | 4 | 1.34 | 0.903 | |
| GRPOfullRevision iteration=1st2026.04 | 9 | 55 | 26 | 10 | 2.37 | 0.515 | |
| LLaMA2026.04 | 8 | 33 | 45 | 14 | 2.65 | 0.405 | |
| PPOapp=sim2026.04 | 1 | 6 | 21 | 72 | 3.64 | 0.07 |