AI Paper Atlas

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

Fantastic Reasoning Behaviors and Where to Find Them: Unsupervised Discovery of the Reasoning Process

Zhenyu Zhang, Shujian Zhang, John Wheatley Lambert, Wenxuan Zhou, Zhangyang Wang, Mingqing Chen, Andrew Hard, Rajiv Mathews, Lun Wang

ICML 2026 · Large language models › Language, safety and interpretability · approach: Reasoning mechanisms and failure modes · 0 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
inject positive reflection vectorR1-1.5B · GPQA-Diamondreflection steering metric: 53.23 -> 62.77Improved
project out reflection vector (negative)R1-1.5B · GPQA-Diamondreflection steering metric: 53.23 -> 45.42Hurt
inject positive reflection vectorR1-1.5B · KnowLogicreflection steering metric: 35.56 -> 51.00Improved
inject positive reflection vectorR1-1.5B · GPQA-Diamondbacktracking steering metric: 11.83 -> 20.31Improved
negative reflection vectorR1-1.5B · KnowLogicbacktracking steering metric: 5.42 -> 2.33Hurt
increase reflection intervention strength alpha from -1.5 to 1.5 (ablation)R1-1.5B · AIME25number of reflection steps: 58.6 -> 166.9Improved
inject confidence vectorR1-1.5B · AIME25reflection steps: 90.53 -> 33.77Improved
inject confidence vectorR1-1.5B · AIME25backtracking steps: 35.50 -> 5.93Improved
inject confidence vectorR1-1.5B · AIME25accuracy (%): 23.33 -> 20.00No clear effect
negative confidence vectorR1-1.5B · GPQAreflection steering metric: 53.23 -> 37.25Hurt
positive confidence vectorR1-1.5B · KnowLogicbacktracking steering metric: 5.42 -> 5.98No clear effect
test-time learned combination of top-3 confidence vectors vs SEAL and TIPR1-1.5B · reasoning benchmarksaccuracy: up to 4.66 points; 13.69% token reductionImproved

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