Feedback Friction: LLMs Struggle to Fully Incorporate External Feedback

TL;DR

This paper investigates how effectively large language models incorporate external feedback, revealing a persistent resistance termed Feedback Friction, even under ideal conditions, and analyzing confidence as a predictor of feedback resistance.

Contribution

The study systematically evaluates LLMs' ability to incorporate near-perfect external feedback across diverse tasks, identifying Feedback Friction as a key limitation and proposing analysis based on model confidence.

Findings

Models show resistance to feedback even with near-complete ground-truth information.

Sampling strategies improve but do not eliminate Feedback Friction.

High-confidence predictions are more resistant to external correction.

Abstract

Recent studies have shown LLMs possess some ability to improve their responses when given external feedback. However, it remains unclear how effectively and thoroughly these models can incorporate extrinsic feedback. In an ideal scenario, if LLMs receive near-perfect and complete feedback, we would expect them to fully integrate the feedback and reach correct solutions. In this paper, we systematically investigate LLMs' ability to incorporate feedback by designing a controlled experimental environment. For each problem, a solver model attempts a solution, then a feedback generator with access to near-complete ground-truth answers produces targeted feedback, after which the solver tries again. We evaluate this pipeline across a diverse range of tasks, including math reasoning, knowledge reasoning, scientific reasoning, and general multi-domain evaluations with state-of-the-art language models including Claude 3.7 with extended thinking. Surprisingly, even under these near-ideal conditions, solver models consistently show resistance to feedback, a limitation that we term Feedback Friction. To mitigate this limitation, we experiment with sampling-based strategies like progressive temperature increases and explicit rejection of previously attempted incorrect answers, which yield improvements but still fail to help models achieve target performance. We analyze Feedback Friction and find that models' confidence on specific questions, measured by semantic entropy, predicts feedback resistance: high-confidence predictions remain resistant to external correction. We hope that highlighting this issue in LLMs will help future research in self-improvement.