ConflictGuide improves machine learning by revealing competing behaviors
ConflictGuide: AutoResearch Improves When Competing Behaviors Are Made Visible
Artificial IntelligenceMachine Learning
Summary
When improving machine learning models, making progress on one goal can sometimes hurt another. The authors show that just looking at overall scores doesn't reveal these hidden conflicts well. They created ConflictGuide, which helps identify and measure competing behaviors so models can improve them together. This two-step process leads to better overall performance than using single-number feedback alone.
What this means in practice
- •For machine learning engineers: Improve training workflows by detecting and reducing conflicting behaviors in models for better balanced performance.
- •For software developers building code generation tools: Enhance automated code editing agents by integrating specific behavior probes to guide more effective code improvements.
Authors
Binqian Xu, Qiran Zou, Xiangbo Shu, Dianbo Liu
Abstract
When designing machine learning models, desirable properties are often in tension: improving one behavior can impair another, so task progress can depend on alleviating the conflict. LLM-based AutoResearch systems, which iteratively edit model code and retain edits based on scalar task-performance feedback, have largely ignored this trade-off. We find that scalar feedback supports broad exploration early in search, but it does not reveal how edits affect competing behaviors. In matched-budget experiments, introducing competing-behavior feedback as task gains diminish increases the share of proposals that improve both behaviors and sustains progress beyond scalar-only plateaus. Obtaining this feedback for a given model requires identifying its competing behaviors and designing probes to measure them. To make competing-behavior feedback actionable, we introduce ConflictGuide. Its reusable ConflictGuide-Skill combines a literature-grounded taxonomy with model-specific evidence to identify competing behaviors and specify probes for a code agent to implement as metrics. Evolution proceeds in two stages: Stage I explores with task feedback; Stage II uses probe feedback to steer proposals toward conflict alleviation and retains marginal-gain edits only when probes indicate sufficient alleviation. Across five diverse model families, ConflictGuide reduces task and conflict-related errors by up to 28% and 14%, respectively, relative to scalar-only AutoResearch, with gains extending to other code agents.