STEER: Inference-Time Risk Control via Constrained Quality-Diversity Search

2026-02-02

AI Summary

This work addresses the rigidity of large language models (LLMs) in ordinal decision-making tasks—such as clinical triage—that require balancing sensitivity and specificity, noting their lack of tunable risk-control mechanisms. To overcome this limitation, the authors propose STEER, a framework that constructs a diverse ensemble of natural language personas via offline constrained quality-diversity search and dynamically selects a persona at inference time using a single interpretable parameter corresponding to a user-specified risk percentile. This enables monotonic control over decision conservativeness without additional training. Evaluated on two clinical triage benchmarks, STEER achieves a broader behavioral spectrum than temperature sampling or static role ensembles, maintaining high accuracy on critical cases while offering enhanced controllability for ambiguous decisions.

Technology Category

• Natural Language Processing: Safety and Robustness
• Machine Learning: Large Multimodal Models (LMMs)
• Reasoning under Uncertainty: Sequential Decision Making

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work

Abstract

Large Language Models (LLMs) trained for average correctness often exhibit mode collapse, producing narrow decision behaviors on tasks where multiple responses may be reasonable. This limitation is particularly problematic in ordinal decision settings such as clinical triage, where standard alignment removes the ability to trade off specificity and sensitivity (the ROC operating point) based on contextual constraints. We propose STEER (Steerable Tuning via Evolutionary Ensemble Refinement), a training-free framework that reintroduces this tunable control. STEER constructs a population of natural-language personas through an offline, constrained quality-diversity search that promotes behavioral coverage while enforcing minimum safety, reasoning, and stability thresholds. At inference time, STEER exposes a single, interpretable control parameter that maps a user-specified risk percentile to a selected persona, yielding a monotonic adjustment of decision conservativeness. On two clinical triage benchmarks, STEER achieves broader behavioral coverage compared to temperature-based sampling and static persona ensembles. Compared to a representative post-training method, STEER maintains substantially higher accuracy on unambiguous urgent cases while providing comparable control over ambiguous decisions. These results demonstrate STEER as a safety-preserving paradigm for risk control, capable of steering behavior without compromising domain competence.

Problem

• mode collapse
• risk control
• ordinal decision
• behavioral diversity
• LLM alignment

Innovation

• STEER
• quality-diversity search
• risk control
• persona ensemble
• inference-time steering