What Drives Interactive Improvement from Feedback?

2026-06-29

AI Summary

This work investigates the true source of performance gains in multi-turn interactions: whether they stem from effective feedback or merely from repeated attempts, formatting adjustments, or additional computation. To this end, we introduce the first cross-task, multi-model controllable feedback evaluation framework, leveraging a student–teacher protocol to construct interaction matrices on benchmarks including Omni-MATH, Codeforces, BBEH Linguini, and ARC-AGI, while rigorously controlling for task difficulty, interaction history, and teacher information privilege. Our experiments reveal that feedback efficacy primarily depends on the student’s ability to utilize feedback rather than the teacher’s identity; high-quality external feedback yields substantial improvements, whereas self-generated feedback barely surpasses unguided self-optimization. The code and framework are publicly released.

Abstract

We study when natural-language feedback produces improvement beyond the gains obtainable from repeated attempts alone. In multi-turn language agent setting, higher final accuracy can reflect useful feedback, but it can also arise from resampling, format correction, or additional test-time computation. To separate these effects, we introduce a controlled student-teacher protocol across Omni-MATH, Codeforces, BBEH Linguini, and ARC-AGI1, evaluating thirteen open-weight models in both student and teacher roles. We compare external feedback, self-feedback, and unguided self-refinement, while varying interaction history, task difficulty, and teacher access to privileged task information. Across settings, we find that multi-turn improvement is often not evidence of feedback use: self-generated feedback adds little beyond unguided self-refinement, whereas the strongest external teachers produce substantially larger feedback-specific gains, suggesting that useful feedback must provide guidance beyond generic retry. Dense student-teacher interaction matrices further show that interactive gains are driven more by the student's ability to use feedback than by the teacher's identity, although teacher choice remains important for a fixed student. These results suggest that feedback-based agents should be evaluated against repeated-attempt baselines, and that ability to act on feedback, not merely feedback availability, is a central bottleneck for interactive improvement. We release our controlled student-teacher evaluation framework at https://j-lojek.github.io/feedback-generation-is-a-bottleneck/.

Problem

• interactive improvement
• natural-language feedback
• repeated attempts
• student-teacher interaction
• feedback utilization

Innovation

• feedback utilization
• student-teacher protocol
• interactive improvement
• self-refinement
• controlled evaluation