Dual-Axis Policy Optimization for LLM Agents: Bayesian Feedback Attribution and Trajectory Mass Normalization
A new training method for LLM agents that improves policy optimization using Bayesian feedback attribution and trajectory mass normalization.
A new training method for LLM agents that improves policy optimization using Bayesian feedback attribution and trajectory mass normalization.
Auto-scored: This is a technical capability improvement for LLM agents with no direct risk or mitigation implications.