From Uncertainty to Action: Learning to Steer LLM Agents
cs.AI, cs.CL, cs.LG
Submitted: 2026-10-06
Updated: 2026-10-06
Comments: 22 pages, 9 figures, 9 tables
License: http://creativecommons.org/licenses/by/4.0/
The gist: Steering an LLM agent means deciding whether to correct it, at which step, and with which mechanism.
Abstract
Steering an LLM agent means deciding whether to correct it, at which step, and with which mechanism. Uncertainty is often used to decide when to correct an agent, but whether it can guide these decisions remains unclear. We steer agent trajectories separately at every non-terminal step with each of four mechanisms and run each continuation to completion. The resulting stepwise outcome table (SOT) holds about 82,000 counterfactual continuations of 1,864 trajectories from three benchmarks and two agents. It shows that uncertainty can identify failing trajectories, but that no single signal reliably locates the step at which steering helps. We therefore propose VoS (Value of Steering), a trajectory-level monitor, offline or online, that learns from SOT the value of steering at each step and decides where to steer by it. A harm-budgeted trigger decides whether to steer, limiting the fraction of successful trajectories that VoS disturbs. VoS improves on unmodified execution in all 12 settings of benchmark, agent, and offline or online use, by 7.8 points on average, and outperforms the strongest of five existing uncertainty-triggered methods in 11, by 2.9 points on average. Ablations show that training on measured outcomes and a tight harm budget are both essential.
Sources
- Conformal Risk Control
- Why Do Multi-Agent LLM Systems Fail?
- DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence
- UProp: Investigating the Uncertainty Propagation of LLMs in Multi-Step Agentic Decision-Making
- Gemma 4 Technical Report
- A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networks
- Uncertainty of Thoughts: Uncertainty-Aware Planning Enhances Information Seeking in Large Language Models
- Language Models (Mostly) Know What They Know
- Scalable Best-of-N Selection for Large Language Models via Self-Certainty
- Same Signal, Opposite Meaning: Direction-Informed Adaptive Learning for LLM Agents
- Generating with Confidence: Uncertainty Quantification for Black-box Large Language Models
- AgentBoard: An Analytical Evaluation Board of Multi-turn LLM Agents
- Self-Refine: Iterative Refinement with Self-Feedback
- Consistent Estimators for Learning to Defer to an Expert
- RouteLLM: Learning to Route LLMs with Preference Data
- ReDAct: Uncertainty-Aware Deferral for LLM Agents
- Out-of-Distribution Detection and Selective Generation for Conditional Language Models
- Toolformer: Language Models Can Teach Themselves to Use Tools
- A Stitch in Time Saves Nine: Detecting and Mitigating Hallucinations of LLMs by Validating Low-Confidence Generation
- Accurate Failure Prediction in Agents Does Not Imply Effective Failure Prevention
Related papers
- MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
- Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
- Agents' Last Exam
- The Clinician's Veto: Navigating Trust, Liability, and Uncertainty in Autonomous AI Prescribing
- MindHelper: Closed-Loop Embodied Mental-State Reasoning for Precision Intervention
- Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems