从过往经验中学习受益于两种互补的记忆形式:情节性痕迹——所发生事件的原始轨迹——以及从众多情节中提炼出的、可复用的图式化经验教训,即整合性抽象知识。当前的智能体记忆系统追求的是整合形式:大语言模型将过去的轨迹重写为文本记忆库,并通过新的交互不断更新,从而有望实现无需参数更新的自我改进型智能体。然而我们发现,即便源自有用经验,当前大语言模型产生的这类整合记忆也常常存在缺陷。随着整合的推进,记忆效用先上升,后下降,甚至可能低于无记忆基线。更令人惊讶的是,即便从真实解决方案进行整合,GPT-5.4 在一组其先前无需记忆即可解决的 ARC-AGI 问题上仍有 54% 的失败率。我们将这种性能倒退追溯至整合步骤,而非底层经验本身:相同的轨迹在不同的更新策略下会产生性质不同的记忆,而仅保留这些轨迹的纯情节性对照组与我们测试的整合系统相比仍具竞争力。在暴露了“保留”、“删除”和“整合”操作的受控 ARC-AGI Stream 环境中,智能体默认保留原始情节,其准确率是强制整合方案的两倍;完全禁用整合(仅进行情节管理)则能达到与这种自动模式相同的效果。在实践中,稳健的智能体记忆应将原始情节视为首要证据,并显式地控制整合过程,而非在每次交互后都自动触发。展望未来,可靠的智能体记忆需要大语言模型能够在整合的同时,不覆盖其所依赖的证据。
Learning from past experience benefits from two complementary forms of memory: episodic traces -- raw trajectories of what happened -- and consolidated abstractions distilled across many episodes into reusable, schema-like lessons. Recent agentic-memory systems pursue the consolidated form: an LLM rewrites past trajectories into a textual memory bank that it continuously updates with new interactions, promising self-improving agents without parameter updates. Yet we find that such consolidated memories produced by today's LLMs are often faulty even when derived from useful experiences. As consolidation proceeds, memory utility first rises, then degrades, and can fall below the no-memory baseline. More surprisingly, even when consolidating from ground-truth solutions, GPT-5.4 fails on 54% of a set of ARC-AGI problems it had previously solved without memory. We trace the regression to the consolidation step rather than the underlying experience: the same trajectories yield qualitatively different memories under different update schedules, and an episodic-only control that simply retains those trajectories remains competitive with the consolidators we test. In a controlled ARC-AGI Stream environment that exposes Retain, Delete, and Consolidate actions, agents preserve raw episodes by default and double the accuracy of their forced-consolidation counterparts; disabling consolidation entirely (episodic management only) matches this auto regime. Practically, robust agent memory should treat raw episodes as first-class evidence and gate consolidation explicitly rather than firing it after every interaction. Looking forward, reliable agentic memory will require LLMs that can consolidate without overwriting the evidence they depend on.