AI · Studies · 2026-09-09 09:13:00
Large language models are increasingly deployed as agents that plan over long horizons and act through external tools. Most agents select actions through unconstrained generation over an accumulating history, leaving procedural knowledge implicit. The authors introduce the Procedural Graph, which organizes procedural knowledge into (procedure, relation, procedure) triplets. At each step, a guidance model turns the surrounding subgraph into step-level guidance that biases but does not dictate the solver's next action. The graph self-evolves by contrasting failed and successful trajectories, committing only edits that preserve or improve held-out validation performance. Across datasets, task types, and LLMs, it delivers consistent gains over memory-based baselines.
Auto-generated summary for reference only. RSS