Ignorance or Incompetence? Constructing Knowledge-Gated, Verifiable Tasks for LLM Agents

  • 类型:arxiv
  • 标识:2608.30322
  • 链接:https://arxiv.org/abs/2608.30322
  • 主分类:agent
  • 形态:benchmark
  • 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:A knowledge-gated task-construction protocol is introduced that separates a task instruction from a compact artefact containing private conventions, reference tables, and utility operators, and it is shown that the retained tasks improve post-training.
  • OpenAlex ID:W7204969630
  • OpenAlex DOI:10.48550/arxiv.2608.30322
  • DOI:10.48550/arxiv.2608.30322
  • DOI来源:OpenAlex
  • 开放获取:green
  • 开放获取链接:https://arxiv.org/pdf/2608.30322
  • OpenAlex更新:2026-09-06
  • 待LLM分类:否
  • 标题中文:无知还是无能?为 LLM 智能体构建知识门控的可验证任务
  • TLDR中文:本文提出一种知识门控的任务构建协议,将任务指令与一个包含私有约定、参考表与效用算子的紧凑工件分离,并证明被保留的任务能够改善 post-training。
  • 来源文件:
  • /inbox/tom/_candidates/2026-09-03-agent-rag-longcontext-candidates.json
  • [OpenAlex backfill]
  • [S2 enrich]