arXiv 论文将长时程智能体的世界模型崩塌刻画为相变
World-Model Collapse as a Phase Transition
AI 导读
arXiv 论文（arXiv:2606.31399）研究长时程语言智能体维持世界状态表征的能力，发现其类似相变：低压力区域持续成功，临界边界附近任务参数的微小变化导致成功率骤降，高压力区域持续失败。逐步轨迹显示智能体在动作失效前就已丢失真实世界状态，是在基于被破坏的世界模型行动；更强模型更晚到达临界边界但无法消除突变。代码即将公开。
正文 · 原文
Abstract:Water looks unchanged as it warms, then at a critical point it boils. We ask whether long-horizon language agents show an analogous transition in their ability to maintain an accurate representation of the evolving world state. In some parameter settings, a small change in a task parameter, or adding a single step to the task, leaves behavior nearly unchanged; near a critical boundary, the same small change causes a sudden collapse. We study this effect in a deterministic task family where the true world state is known at every step. A large grid search over state size, state dependency, horizon, number of available actions, observation mode, and frequency of state changes reveals a phase diagram: a low-stress region where the agent consistently succeeds, a narrow region where success drops sharply, and a high-stress region where the agent consistently fails. Step-by-step traces show the mechanism: the agent loses track of the true world before its actions become invalid, so the agent is not merely choosing a bad action; it is acting from a corrupted world. Stronger models reach the critical boundary later but do not eliminate the sharp transition. These results make world-model collapse a measurable bottleneck for long-horizon agents. Code will be made publicly available soon.
| Comments: | Under Review |
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2606.31399 [cs.AI] |
| (or arXiv:2606.31399v3 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2606.31399 arXiv-issued DOI via DataCite |
Submission history
From: Zekun Cai [view email]
[v1]
Tue, 30 Jun 2026 09:28:01 UTC (8,056 KB)
[v2]
Sat, 4 Jul 2026 03:39:38 UTC (6,547 KB)
[v3]
Thu, 12 Nov 2026 09:17:23 UTC (4,454 KB)
来源：arXiv cs.AI · arxiv.org