Different evidence types affect AI prompt improvement results
What Should the Reflector See? An Empirical Study of Evidence in Reflective Prompt Optimization
Computation and LanguageMachine Learning
Summary
When improving instructions given to AI models, it’s unclear which examples the model should see to do better. The authors tested different ways of showing past results to the AI and found some strategies help more than others. For instance, showing only the failures improves test scores a lot, but sometimes success nearby doesn’t mean better final results. They recommend evaluating improvement methods by looking separately at final scores, local improvements, and overfitting signs.
What this means in practice
- •For ai model trainers: Choose which previous examples to provide when improving prompt instructions to maximize final model task accuracy.
- •For data pipeline engineers: Design evaluation systems that separately track performance gains, local reflection success, and calibration gaps to detect overfitting in AI prompt tuning.
Authors
Xiaofan Zhou, Lu Cheng
Abstract
Reflective prompt optimization revises instructions using examples of a model's behavior, but which evidence the reflector should receive remains unclear. We study evidence composition, visibility of examples, candidate selection and domain-knowledge policy within a single-parent Pareto-guided search. Using Qwen3.5-9B as both task model and reflector, we evaluate nine reflection strategies on five datasets. From these experiments, we find three distinct patterns. For performance improvement, Failures-only produces the largest mean test gain (+8.0 percentage points), while Balanced-mix and No-examples+Val share the best mean performance rank. For reflection effectiveness, No-examples achieves the best rank for improving sampled parents, yet yields only a 1.4-point mean test gain: local reflection success does not necessarily produce a stronger final prompt. For overfitting assessment, Failures-only and Balanced-mix share the lowest mean calibration-gap rank, while larger gaps on GPQA and IFBench show that calibration gains can overstate held-out improvement. This gap is a descriptive indicator, not a direct measure of overfitting. Together, these results show why reflection strategies should be assessed separately on final performance, parent improvement and calibration-to-test transfer.