General Knowledge on MMLU (Drop in Utility %)
37.4MMLU General Knowledge Drop in Utility (%)KL
Evaluation Results
| Method | Links | |
|---|---|---|
| KLAlignment Strategy=Deliberative Alignment (GRPO)2026.04 | 37.4 | |
| Latent Similarity- Layer 12Alignment Strategy=Deliberative Alignment (GRPO)2026.04 | 9.7 | |
| Self CertaintyAlignment Strategy=Deliberative Alignment (GRPO)2026.04 | 1.3 | |
| RandomAlignment Strategy=Deliberative Alignment (GRPO)2026.04 | 0.7 | |
| PerplexityAlignment Strategy=Deliberative Alignment (GRPO)2026.04 | -6.7 | |
| Latent Similarity- Best LayerAlignment Strategy=Deliberative Alignment (GRPO)2026.04 | -34.5 |