CLM lets language models autonomously manage their context
InfoQ 中文 AI&大模型 reports that researchers from Meta, MIT, and the University of Washington proposed Context Language Model (CLM), which treats context as an editable file and can rewrite messages, preserve important facts, remove irrelevant information, and maintain progress notes. The researchers evaluated zero-shot, in-context, and reinforcement-learning approaches; reported results included an 11.4% accuracy gain and 21.5% fewer FLOPs on BrowseComp-Plus, while the reinforcement-learning variant raised Qwen3.5-9B accuracy from 28.8% to 42.5% with 12% fewer FLOPs. The report also notes risks including lost information, persistent prompt injection or self-generated instructions, cache invalidation, and incorrect context-management decisions.
CLM offers a research-backed alternative to predefined summarization, compression, and retrieval-based context management, with reported gains in task performance and compute efficiency on several benchmarks.
Meta、麻省理工学院与华盛顿大学的研究人员提出上下文语言模型（Context Language Model，CLM），一种使语言模型能够自主管理和编辑自身上下文的新方法"。该方法不再依赖预定义的摘要、压缩和信息检索机制，在性能和计算效率方面均取得显著提升。
研究人员实现了一种让 LLM 自主管理上下文的机制，将上下文视为一个可以无限制更新的文件。这种方法为传统的上下文管理策略提供了一种替代方案，传统策略（如压缩、外存卸载和检索）用于防止上下文无限增长，但各有局限。摘要可能会丢失关键细节或引入错误，压缩通常使用预定义的规则集，而外部存储则需要智能体决定哪些信息需要重新载入上下文。
#Meta#MIT#University of Washington#Context Language Model (CLM)#Qwen3.5-9B#BrowseComp-Plus#EdgeBench#ContextBench