Framing Sensitivity Evaluation on FRAGILE Value-tinted narration
12.8Flip RateVALIGN
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| VALIGNLLM Backbone=LLaMA-3.1-8B-Instruct2026.05 | 12.8 | 8.6 | 3.6 | 83.6 | |
| VALIGNLLM Backbone=Mistral-7B-Instruct-v0.32026.05 | 31.5 | 29.8 | 10.3 | 58.2 | |
| VALIGNLLM Backbone=Qwen2.5-7B-Instruct2026.05 | 32.5 | 31.4 | 19.3 | 48.3 | |
| BaseLLM Backbone=Mistral-7B-Instruct-v0.32026.05 | 36.3 | 34.8 | 11 | 52.7 | |
| BaseLLM Backbone=LLaMA-3.1-8B-Instruct2026.05 | 39.9 | 20.9 | 2.8 | 57.3 | |
| BaseLLM Backbone=Qwen2.5-7B-Instruct2026.05 | 47.2 | 36.4 | 17.9 | 34.9 | |
| CAALLM Backbone=LLaMA-3.1-8B-Instruct2026.05 | 50.5 | 48.9 | 25.3 | 24.2 | |
| CAALLM Backbone=Mistral-7B-Instruct-v0.32026.05 | 51.5 | 51 | 17.5 | 31.1 | |
| Instruc.LLM Backbone=Qwen2.5-7B-Instruct2026.05 | 52.7 | 26.5 | 25 | 22.3 | |
| CAALLM Backbone=Qwen2.5-7B-Instruct2026.05 | 52.7 | 52 | 14.5 | 32.8 | |
| Instruc.LLM Backbone=Mistral-7B-Instruct-v0.32026.05 | 52.9 | 25.6 | 20.2 | 26.9 | |
| Instruc.LLM Backbone=LLaMA-3.1-8B-Instruct2026.05 | 53.2 | 28.4 | 24.3 | 22.4 | |
| CoTLLM Backbone=LLaMA-3.1-8B-Instruct2026.05 | 60.1 | 32.2 | 20.6 | 19.3 | |
| CoTLLM Backbone=Qwen2.5-7B-Instruct2026.05 | 60.6 | 32 | 19.6 | 19.9 | |
| CoTLLM Backbone=Mistral-7B-Instruct-v0.32026.05 | 63 | 30 | 15.8 | 21.2 |