自验证智能体工具：分离长时程智能体中的承诺漂移与绑定漂移
原标题：The LLM Proposes, the Executive Disposes: A Self-Verifying Agent Instrument that Dissociates Commitment Drift from Binding Drift in Long-Horizon Agents
AI 摘要
该研究提出一种自验证的智能体工具，通过确定性执行器控制信念、语言模型仅提交类型化提案，并利用预注册预测与代码观测匹配来验证长时程智能体。实验表明，移除承诺机制会导致目标放弃率从0.00升至1.00，而绑定错误保持为0.00，验证了漂移分解的有效性。尽管任务效能为零（52次门控运行中无一次完成ARC-AGI-3），但该方法为智能体开发提供了结构性验证方法论。
正文节选
Computer Science > Artificial Intelligence Title:The LLM Proposes, the Executive Disposes: A Self-Verifying Agent Instrument that Dissociates Commitment Drift from Binding Drift in Long-Horizon Agents View PDF HTML (experimental) Abstract:How do you verify a long-horizon agent when its own state and self-reports are exactly what you cannot trust? We present an agent instrument built so that verification is structural rather than post-hoc. A deterministic Executive owns all belief; a