EVOKE: Eliciting World Knowledge in Agents for Transferable Decision-Making

类型：arxiv
标识：2609.38334
链接：https://arxiv.org/abs/2609.38334
主分类：agent
形态：application
TLDR：Large language models (LLMs) are increasingly deployed as agents for multi-step decision-making, yet transfer poorly to unseen environments. World-model methods address this by training agents to predict future observations, at the cost of additional training and errors that compound when predictions are used for planning. However, for LLM agents operating in digital environments, much of this world knowledge is already internalized during pretraining, which shifts the problem from acquiring it to eliciting it. We argue that typical post-training provides little pressure for such elicitation,
副分类：engineering
待LLM分类：否
来源文件：
/inbox/tom/_candidates/2026-10-01-agent-rag-longcontext-candidates.json

2026-07-23

仓库 Dusto-AI/llm-situational-awareness-paper-list

2026-08-11

笔记精读与批判 · From 128K to 4M: Efficient Training of Ultra-Long Context LLMs

2026-08-06

笔记 Self-Challenging Language Model Agents (SCA) + Code-as-Task · flyP 单篇精读（2026-08-18 22:50 · 晚棒）

2026-08-18

笔记 KVAE Tokenizers arXiv:2608.05798 · multimodal · 精读与批判 v2

2026-08-09

笔记 📚 学术研究知识库草稿 · Jay · 2026-06-22

2026-06-22

arXiv

2609.38334

TLDR

Large language models (LLMs) are increasingly deployed as agents for multi-step decision-making, yet transfer poorly to unseen environments. World-model methods address this by training agents to predict future observations, at the cost of additional training and errors that compound when predictions are used for planning. However, for LLM agents operating in digital environments, much of this world knowledge is already internalized during pretraining, which shifts the problem from acquiring it to eliciting it. We argue that typical post-training provides little pressure for such elicitation,

EVOKE: Eliciting World Knowledge in Agents for Transferable Decision-Making

相关 · 按语义匹配