Research map / Large language models / Alignment and preferences
Conflict-Aware Adaptive Alignment for LLM Hallucination Mitigation
Ruohan Zong, Yang Zhang, Dong Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add conflict-aware margin (apply helpfulness regulation only to conflicting pairs) (ablation) | CAMP · TruthfulQA | accuracy (%): 83.48 -> 85.92 | Improved |
| add conflict-aware margin (ablation) | CAMP · TruthfulQA | hallucination rate (%): 11.26 -> 9.79 | Improved |
| add adaptive margin strength (per-pair margin instead of batch-wise constant) (ablation) | CAMP · TruthfulQA | accuracy (%): 84.70 -> 85.92 | Improved |
| add adaptive margin strength (ablation) | CAMP · TruthfulQA | truthfulness score (1-5): 4.36 -> 4.41 | Improved |
| add CAMP margin term over DPO on truthfulness pairs (ablation) | DPO on truthfulness pairs · TruthfulQA | hallucination rate (%): 11.26 -> 9.79 | Improved |
| increase helpfulness regulation lambda from 0 to 0.25-0.75 (ablation) | CAMP · TruthfulQA | helpfulness score | Improved |
| increase lambda to 1.0 (ablation) | CAMP · TruthfulQA | helpfulness score | Hurt |
| increase lambda from 0 to 1.0 (ablation) | CAMP · TruthfulQA | truthfulness and accuracy | Mixed |
| use CAMP instead of MODPO on Qwen2.5-7B-Instruct | CAMP · TruthfulQA | hallucination rate (%): 7.22 -> 5.88 | Improved |
| replace truthfulness score with honesty score as the preference signal (ablation) | CAMP · TruthfulQA | hallucination rate (%): 12.48 (MODPO) -> 9.79 (CAMP) | Improved |
| CAMP instead of CPO baseline | Llama-3.1-8B-Instruct · TruthfulQA | hallucination rate (%): 15.18 -> 9.79 | Improved |
| increase training pairs from 2.5k to 25k (ablation) | CAMP · TruthfulQA | helpfulness, truthfulness, accuracy, hallucination | Improved |
Closest papers by meaning
- BALTO: Balanced Token-Level Policy Optimization for Hallucination Mitigation (NeurIPS 2026 · similarity 0.91)
- FLAME : Factuality-Aware Alignment for Large Language Models (NeurIPS 2024 · similarity 0.91)
- Intervene-All-Paths: Unified Mitigation of LVLM Hallucinations across Alignment Formats (NeurIPS 2025 · similarity 0.90)
- IRIS: Implicit Reward-Guided Internal Sifting for Mitigating Multimodal Hallucination (ICML 2026 · similarity 0.90)
- Mitigating Hallucination Through Theory-Consistent Symmetric Multimodal Preference Optimization (NeurIPS 2025 · similarity 0.89)
- Mitigating Hallucinations in Large Vision-Language Models via DPO: On-Policy Data Hold the Key (CVPR 2025 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
