1
Value-Gradient Hypothesis of RL for LLMs
从价值梯度假说揭秘PPO和GRPO为何有效，为LLM后训练提供新理论框架。
arXiv:2605.21654v1 Announce Type: cross Abstract: Reinforcement learning substantially improves pretrained language models, but it remains understudie…
2
Reinforcement Learning for LLM Post-Training: A Survey
一篇系统梳理LLM后训练中强化学习的综述，涵盖RLHF、DPO、RLVR等前沿方法
arXiv:2407.16216v4 Announce Type: replace Abstract: Large language models (LLMs) trained via pretraining and supervised fine-tuning (SFT) can still pr…
3
CheckSupport: A Local LLM-Powered Tool for Automated Manuscript Submission Checklist Selection and Completion
本地LLM驱动，自动完成学术手稿提交检查清单，提升投稿效率。
arXiv:2605.16377v1 Announce Type: cross Abstract: Transparent and standardized reporting is essential for reproducible scientific research, yet adhere…
4
What Is Preference Optimization Doing, and Why?
揭示DPO与PPO本质差异，挑战“监督学习vs强化学习”传统认知的深度技术论文。
arXiv:2512.00778v2 Announce Type: replace Abstract: Preference optimization (PO) is indispensable for large language models (LLMs), with methods such …
5
Proximal Policy Optimization
OpenAI发布新强化学习算法PPO，简单易调优且性能卓越，已成为默认算法。
We’re releasing a new class of reinforcement learning algorithms, Proximal Policy Optimization (PPO), which perform comparably or better than state-of…
6
Learning Montezuma’s Revenge from a single demonstration
单个人类演示教会AI狂揽74,500分，刷新《蒙提祖马的复仇》最高分记录。
We’ve trained an agent to achieve a high score of 74,500 on Montezuma’s Revenge from a single human demonstration, better than any previously publishe…