PACMS: Submodular Context Selection as a Pluggable Engine for LLM Agents

  • 类型:arxiv
  • 标识:2606.20047
  • 链接:http://arxiv.org/abs/2606.20047v1
  • 主分类:agent
  • 形态:method
  • 被引:0
  • 被引来源:Semantic Scholar + OpenAlex
  • S2被引:0
  • OpenAlex被引:0
  • 影响力被引:0
  • TLDR:Conversational and tool-using LLM agents operate over a context window that fills from several directions simultaneously, and agents that must recall information across many turns, the defining case for memory, are precisely where recency truncation fails.
  • OpenAlex ID:W7165356640
  • OpenAlex DOI:10.48550/arxiv.2606.20047
  • DOI:10.48550/arxiv.2606.20047
  • DOI来源:OpenAlex
  • 开放获取:green
  • 开放获取链接:https://doi.org/10.48550/arxiv.2606.20047
  • OpenAlex更新:2026-07-19
  • 待LLM分类:否
  • 标题中文:PACMS:作为 LLM Agent 可插拔引擎的次模上下文选择
  • TLDR中文:对话式与工具使用的 LLM Agent 在上下文窗口中同时从多个方向被填充,而必须在多轮之间回忆信息的 Agent(即 memory 的典型场景)恰恰是 recency 截断失效的地方。
  • 来源文件
  • /inbox/tom/_candidates/2026-06-22-agent-rag-longcontext-candidates.json
  • /inbox/tom/_candidates/2026-06-21-agent-rag-longcontext-candidates.json
  • /inbox/tom/_candidates/2026-06-20-agent-rag-longcontext-candidates.json
  • /inbox/tom/_candidates/2026-06-19-agent-rag-longcontext-candidates.json
  • [S2 enrich]
  • [OpenAlex backfill]