AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Hallucination Reduction with CASAL: Contrastive Activation Steering for Amortized Learning

Wannan Yang, Xinchi Qiu, Lei Yu, Yuchen Zhang, Aobo Yang, Narine Kokhlikyan, Nicola Cancedda, Diego Garcia-Olano

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Task vectors and steering · 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
train CASAL (amortized steering) instead of SFT/DPO/GRPOLlama-3.1-8B-Instruct · TriviaQA to EntityQA (OOD) unknown querieshallucination rate: 11.72% vs SFT 23.13%, DPO 18.30%, GRPO 12.83%Improved
train CASAL on Wikipedia-style TriviaQA, test on webLlama-3.1-8B-Instruct · TriviaQA (Wiki to Web)hallucination rate: 50.74% -> 32.42%Improved
train CASAL on PopQA Group 1, test Group 2Llama-3.1-8B-Instruct · PopQAhallucination rate: 74.35% -> 23.42%Improved
use CASAL instead of DPO/GRPO for known-query refusalLlama-3.1-8B-Instruct · TriviaQA, EntityQArefusal rate: 7.29% vs DPO 14.37%, GRPO 17.77%; 6.84% vs DPO 17.66%Improved
CASAL vs baseline on PopQA known-query accuracyLlama-3.1-8B-Instruct · PopQAaccuracy: 91.08% -> 85.11%Hurt
apply CASAL to general capability benchmarksLlama-3.1-8B-Instruct · MMLU, GSM8K, GPQAaccuracy: MMLU 68.01 -> 68.04; GSM8K 77.48 -> 77.02; GPQA 33.31 -> 33.18No clear effect
apply CASAL to MT-Bench coherenceLlama-3.1-8B-Instruct · MT-Benchscore: 7.38 -> 7.57Improved
train CASAL on 640 examples versus baselines on 12,800Llama-3.1-8B-Instruct · hallucination (TriviaQA)hallucination rate: matches or surpasses SFT, DPO, GRPO trained on 12,800 examples (20x data efficiency)Improved
use CASAL instead of LoRA trainingLlama-3.1-8B-Instruct · compute costFLOPs per token: over 30x fewer FLOPs per token than LoRAImproved
apply CASAL to vision-language modelQwen2.5-VL-7B-Instruct · WorldCuisines-VQAhallucination rate: 72.35% -> 33.34%Improved
apply CASAL to Landmark-VQAQwen2.5-VL-7B-Instruct · Landmark-VQAhallucination rate: 75.78% -> 31.25%Improved
apply CASAL to MoE modelOLMoE · unknown querieshallucination rate: drops by 42.9%Improved

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