AI Paper Atlas

Research map / Large language models / Alignment and preferences

Discern Truth from Falsehood: Reducing Over-Refusal via Contrastive Refinement

Yuxiao Lu, Lin Xu, Yang Sun, Wenjun Li, Jie Shi

ICLR 2026 · Large language models › Alignment and preferences · approach: LLM-as-judge and feedback data · 5 citations (OpenAlex)

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 changedWhereMeasuredEffect
add a contrastive refinement stage before safety SFT (DCR)STL on Qwen2.5-1.5B · XSTestcompliance rate: 0.73 -> 0.98Improved
add a contrastive refinement stage before safety SFT (DCR)STL on Qwen2.5-7B · OR-Benchcompliance rate: 0.34 -> 0.71Improved
add a contrastive refinement stage before safety SFT (DCR)STL on LLaMA-3-8B · XSTestcompliance rate: 0.79 -> 0.93Improved
add a contrastive refinement stage before safety SFT (DCR)STL on Qwen2.5-1.5B · safety (defense success rate)defense success rate: 0.72 -> 0.81Improved
add a contrastive refinement stage before safety SFT (DCR)STL on LLaMA-3-8B · safety (defense success rate)defense success rate: 0.93 -> 0.91Hurt
add a contrastive refinement stage before safety SFT (DCR)STL on LLaMA-3-8B · general QA (ARC_c, OpenBookQA)accuracy: ARC_c 0.56 -> 0.51; OpQA 0.45 -> 0.39Hurt
add a contrastive refinement stage before safety SFT (DCR)STL on Qwen2.5-7B · ARC_e / ARC_caccuracy: ARC_e 0.77 -> 0.83; ARC_c 0.51 -> 0.59; OpQA 0.47 -> 0.44Mixed
use contrastive learning on seemingly toxic prompts instead of adding them to SFT dataSTL-aug · XSTestcompliance rate: Qwen2.5-1.5B: 0.75 -> 0.98Improved
compare DCR with SCANS on response quality (Qwen2.5-7B)SCANS · AlpacaEval win ratewin rate (%): 45.5 -> 45.8Improved
contrastive learning applied before alignmentLLM representations · kernel similarity between seemingly toxic and toxic promptsnormalized K Frobenius normImproved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy