Neglected Free Lunch from Post-training: Progress Advantage for LLM Agents
- 类型：arxiv
- 标识：2606.26080
- 链接：https://arxiv.org/abs/2606.26080
- 主分类：engineering
- 形态：benchmark
- 被引：1
- 被引来源：Semantic Scholar
- S2被引：1
- OpenAlex被引：0
- 影响力被引：0
- TLDR：This work shows that reinforcement learning (RL) post-training already provides the ingredients for effective step-level scoring, eliminating the need for dedicated reward model training altogether, and derives an implicit advantage under a general stochastic Markov decision process, which is term progress advantage.
- OpenAlex ID：W7165849379
- OpenAlex DOI：10.48550/arxiv.2606.26080
- DOI：10.48550/arxiv.2606.26080
- DOI来源：OpenAlex
- 开放获取：green
- 开放获取链接：https://doi.org/10.48550/arxiv.2606.26080
- OpenAlex更新：2026-07-19
- 副分类：agent
- 待LLM分类：否
- 标题中文：Neglected Free Lunch from Post-training: Progress Advantage for LLM Agents
- TLDR中文：本文表明强化学习（RL）后训练已具备实现有效 step-level 评分所需的要素，从而完全无需额外的奖励模型训练，并在通用随机 Markov 决策过程下推导出一种隐式 advantage，称为 progress advantage。
- 来源文件：
- /inbox/tom/_candidates/2026-06-29-agent-memory-tool-use-candidates.json
- /inbox/tom/_candidates/2026-06-28-agent-rag-longcontext-candidates.json
- [S2 enrich]
- [OpenAlex backfill]