AI Paper Atlas

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

SelfReflect: Can LLMs Communicate Their Internal Answer Distribution?

Michael Kirchhof, Luca Füger, Adam Golinski, Eeshan Gunesh Dhekane, Arno Blaas, Seong Joon Oh, Sinead Williamson

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Uncertainty quantification for LLMs · 15 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
Use Wasserstein distance over the whole logit vector instead of only the masked word log-prob (ablation)SelfReflect · Good vs almost-good summariespairwise accuracy: 85.90% -> 93.20%Improved
Add the one-by-one masked-out task instead of comparing full-answer log-likelihoods (ablation)SelfReflect · Good vs almost-good summariespairwise accuracy: 33.64% -> 93.20%Improved
Put sampled answers into the judge context instead of the sampling-free comparison (ablation)SelfReflect · Good vs almost-good summariespairwise accuracy: 54.85% -> 93.20%Improved
Use generative masked-out task instead of discriminative P(True) (ablation)SelfReflect · Human preference alignmentKrippendorff alpha: -0.032 -> 0.690Improved
Use generative masked-out task instead of discriminative P(True) (ablation)SelfReflect · Good vs bad summariespairwise accuracy: 65.29% -> 98.77%Improved
Sample & summarize with N=10 instead of GreedyQwen2.5 72B Instruct · Self-reflection benchmarkSelfReflect score (lower better): 91 -> 85Improved
Increase samples from N=10 to N=20Sample & summarize, Qwen2.5 72B Instruct · Self-reflection benchmarkSelfReflect score (lower better): 85 -> 84No clear effect
Use Basic prompt asking for all options instead of GreedyQwen2.5 72B Instruct · Self-reflection benchmarkSelfReflect score (lower better): 91 -> 94Hurt
Use CoT prompt instead of GreedyQwen2.5 72B Instruct · Self-reflection benchmarkSelfReflect score (lower better): 91 -> 96Hurt
Use reasoning before summarizing instead of greedy answerQwQ 32B · Self-reflection benchmark, RLVR modelsSelfReflect score (lower better): 96 -> 105Hurt
Sample & summarize N=10 instead of Greedy for the smallest modelQwen2.5 0.5B Instruct · Self-reflection benchmarkSelfReflect score (lower better): 96 -> 96No clear effect
Use Sample & summarize instead of GreedyGemma 3 1B Instruct · Self-reflection benchmarkSelfReflect score (lower better): 116 -> 117No clear effect

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