HIPIF: Hierarchical Planning and Information Folding for Long-Horizon LLM Agent Learning

In three linesHIPIF introduces hierarchical reinforcement learning for long-horizon LLM agents. It decomposes tasks into explicit subgoals and folds completed histories to reduce long-context interference. Validated on three public benchmarks without costly auxiliary models.

Summary generated by Claude — human-verified