arXiv:2611.23564v1 Announce Type: new Abstract: When LLM agents pass intermediate judgments downstream without uncertainty, later decisions may treat tentative assumptions as facts. Effective uncertainty propagation must preserve information for correction and update inherited beliefs as new evidence arrives. Yet scalar scores omit information needed for revision, while detailed descriptions and reflection incur additional costs. We introduce Latent Uncertainty, a latent-space estimator that preserves uncertainty-relevant information without lengthy descriptions. Building on this, we propose LProp (Latent Uncertainty Propagation), which revises inherited answer distributions using latent representations and new evidence without additional trajectory sampling. We evaluate both methods across multiple-choice, yes/no, open-ended question answering, and mathematical reasoning tasks. Our approach outperforms baselines on most metrics, while LProp achieves the lowest mean inference-token usage in all eight propagation settings. Matched feature ablations further demonstrate that inherited latent content improves error recovery beyond answer probabilities or model capacity alone.
LProp: Propagating Latent Uncertainty through Evidence-Conditioned Belief Revision in LLM Agents
Researchers introduced LProp, a method that propagates latent uncertainty through evidence-conditioned belief revision in LLM agents, according to arXiv paper 2611.23564v1. LProp revises inherited answer distributions using latent representations and new evidence without additional trajectory sampling, and across multiple-choice, yes/no, open-ended question answering, and mathematical reasoning tasks it achieved the lowest mean inference-token usage in all eight propagation settings while outperforming baselines on most metrics. Matched feature ablations showed inherited latent content improves error recovery beyond answer probabilities or model capacity alone.
Pass latent uncertainty representations between agent steps rather than scalar confidence scores or verbose reflections, then revise inherited beliefs with new evidence to improve error recovery at lower token cost.
more in playbook: Designing agent systemsRun your AI side-project on zahid.host
EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.