Beyond Memory: Harnessing Long-Horizon Agents with Explicit Belief States
PoS keeps explicit belief states so long-horizon LLM agents avoid Belief Trapping and beat memory baselines.
The paper introduces PoS, an inference-time framework that keeps explicit belief states as an LLM agent's decision context instead of relying only on interaction history. Each belief pairs an estimate of the current world state with unresolved task requirements, and the system checks consistency and progress to detect Belief Trapping. Recovery is chosen according to the trapping pattern and the type of unfinished requirement. On four execution and diagnosis benchmarks, PoS was the top method with every one of three LLM backbones.
- PoS builds explicit belief states from world estimates and unresolved task requirements.
- Belief Trapping is detected when actions do not advance the goal, then recovery is tailored.
- PoS ranked highest on four execution and diagnosis benchmarks across three LLM backbones.
- Ablations credit consistency validation and recovery; context-scaling stayed resilient as context grew.
From the original article
1 min read at huggingface.co
Large language model (LLM) agents can now undertake increasingly complex tasks, but the way they organize interaction history into memory does not ensure a coherent understanding of the current world. We introduce PoS, an inference-time framework that constructs and continually maintains explicit belief states as the agent's decision context.
By Yu Luo, Jiamin Jiang, Yimin Zuo · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.