Forge Decision Context：AI 智能体的持久行为学习
原标题：Decision Context: Persistent Behavioral Learning for AI Agents
AI 摘要
该论文提出 Forge Decision Context（forge_dc）系统，从人类纠正和工具调用失败中捕获行为信号，编译为规则指令并在决策点注入，以解决 LLM 智能体跨会话不学习纠正的问题。作者同时发布 DC-Bench 基准（71 个场景、9 个行为维度），forge_dc 达到 95% 合规率，而朴素检索仅 55%，且每次查询注入的 token 约少 10 倍。研究指出表示、激活与放置是持久行为学习的核心设计选择。
正文节选
Personalized, Aligned, Long-Term Memory for AI Systems Decision Context: Persistent Behavioral Learning for AI Agents Abstract LLM agents do not learn from corrections across sessions. A user fixes the same mistake at the start of every conversation. Existing agent memory systems handle facts (preferences, entities, conversational history [1, 2, 3]) but not procedural knowledge [4]. We present Forge Decision Context (forge_dc), a system that captures behavioral signals from human corrections and