AI Paper Atlas

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

Using Reinforcement Learning to Train Large Language Models to Explain Human Decisions

Jian-Qiao Zhu, Hanbo Xie, Dilip Arumugam, Robert Wilson, Thomas L. Griffiths

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Reasoning mechanisms and failure modes · 14 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
RL with outcome reward instead of SFTQwen-2.5-7B-Instruct · choices13k testMSE: SFT .0144 vs RL .0148No clear effect
Centaur-style masking instead of plain SFTQwen-2.5-7B-Instruct · choices13k testMSE: .0144 -> .0155No clear effect
use a weaker backbone (Gemma-2-2B) for RL (ablation)RL (GRPO) · choices13k testMSE: .0148 -> 0.0526Hurt
use a weaker backbone (Gemma-2-2B) for SFT (ablation)SFT · choices13k testMSE: .0144 -> 0.0162No clear effect
replace backbone CoT with RL model's CoT (ablation)Qwen-2.5-7B-Instruct · choices13k testMSE: .0694 -> .0212Improved
replace RL model's CoT with backbone's CoT (ablation)Qwen-2.5-7B-GRPO-c13k · choices13k testMSE: .0148 -> .0695Hurt
feed SFT-generated CoT to the backbone (ablation)Qwen-2.5-7B-Instruct · choices13k testMSE: .0694 -> 0.0785No clear effect
train RL on synthetic expected-value-maximizer choice rates instead of human data (ablation)RL (GRPO) · CoT mechanismsdominant CoT mechanismImproved
train RL on random choice rates (ablation)RL (GRPO) · Random datasetCoT qualityHurt

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