1
消息称《GTA 6》大幅强化偷车系统：部分载具游戏初期无法窃取、汽车配有追踪器
IT之家 8 月 29 日消息，R 星开发负责人 Rob Nelson 在接受 IGN 采访时透露，《GTA 6》大幅强化偷车系统，过去玩家只需要按下按键即可轻松开走汽车，但新作中部分载具在游戏初期无法直接盗取，玩家需要随着剧情推进获得相应工具后才能顺利劫车。 Rob Nelson 表示，在游戏之初…
2
《GTA 6》游戏强化角色状态反馈：饮食、运动和睡眠都可能影响外观
IT之家 8 月 27 日消息，在接受《Dazed》采访时， Rockstar North 开发负责人罗布 · 纳尔逊（Rob Nelson）透露《侠盗猎车手 VI》（GTA 6）游戏机制，并谈到了两名游戏主角的角色设计。 纳尔逊透露杰森（Jason）和露西娅（Lucia）两位主角的身体会根据玩家选…
3
Learning Generalizable Behaviors for Terminal Agents
用强化学习让大模型在终端环境中学会通用行为，打开日常自动化新可能。
arXiv:2608.22631v1 Announce Type: new Abstract: Terminal agents are a compelling application of large language models (LLMs), with the potential to in…
4
RetroDFM-R: Reasoning-Driven Retrosynthesis Prediction with Large Language Models via Reinforcement Learning
用强化学习让大模型学会「讲道理」，逆合成预测从黑箱猜测走向可解释推理——AI驱动分子设计又进一步。
arXiv:2507.17448v2 Announce Type: replace-cross Abstract: Retrosynthetic planning is a cornerstone of organic synthesis and drug discovery. Yet existi…
5
Reinforcing Multi-Turn Reasoning in LLM Agents via Fine-Grained Reward Structure and Credit Assignment
多轮推理强化学习新方案，细粒度奖励结构直击信用分配难题，提升LLM代理决策可靠性。
arXiv:2505.11821v3 Announce Type: replace Abstract: Reinforcement Learning (RL) approaches have been wildly used to enhance the reasoning capabilities…
6
OpenAI 一改此前立场，呼吁加州强化 AI 安全法案
OpenAI改口支持加州AI安全法案，监管态度大转弯背后有何考量？
IT之家 8 月 23 日消息，OpenAI 呼吁加州进一步加强一项具有里程碑意义的 AI 安全法案。该法案已于去年获得通过。 OpenAI 全球事务团队在 LinkedIn 上发表的一篇文章中表示，加州 SB 53 法案“应该进一步扩大安全保障措施”。例如，可以要求对正在训练或评估的前沿 AI 模…
7
Credit Without Ground Truth: Auditing Step-Level Credit Assignment in LLM Agents Against Executed Replay
一步错步步错？实测三类信用信号在LLM智能体训练中均无法锁定关键步骤，揭示因果归因盲区。
arXiv:2608.19760v1 Announce Type: new Abstract: Audited against causal ground truth from executed replay in a single-agent tool environment (ALFWorld)…
8
Future Policy Approximation for Offline Reinforcement Learning in LLM Reasoning
离线强化学习遇上LLM推理，未来策略近似方法带来新突破，AI研究者不容错过。
arXiv:2509.19893v3 Announce Type: replace Abstract: Reinforcement learning (RL) has emerged as a key driver of post-training for complex reasoning in …
9
Agentic ESOpt: Fine-Tuning Long-Horizon LLM Agents with Minimal GPU Requirements
只用极少GPU就能微调长时程LLM智能体，破解长程推理训练算力困局的新方案。
arXiv:2608.17310v1 Announce Type: new Abstract: Reinforcement Learning (RL) has been promising in single-turn LLM fine-tuning. However, long-horizon a…
10
Reinforcing Consistency in Video MLLMs with Structured Rewards
视频多模态模型常“看走眼”？这项研究用结构化奖励强化推理一致性，为视频理解立新标尺。
arXiv:2604.01460v2 Announce Type: replace Abstract: Multimodal large language models (MLLMs) have achieved remarkable progress in video understanding.…
11
Effective Personalized AI Tutors via LLM-Guided Reinforcement Learning
大模型结合强化学习，打造精准适配每位学生的AI导师，个性化教育迎来新突破。
arXiv:2608.16907v1 Announce Type: cross Abstract: Generative AI (GenAI) is rapidly reshaping education by unlocking the potential for personalized tut…
12
Agent Lightning v1.0: Towards Harnessed Agentic RL
突破性智能体强化学习框架，解锁自主决策新范式，AI研究者必读前沿成果。
arXiv:2608.17528v1 Announce Type: new Abstract: Modern agents operate inside agent harnesses that manage tools, context, and control flow, making the …
13
Belayer: Efficient Fault Tolerance for LLM Agentic RL Training
应对大模型智能体强化学习训练中的故障难题，为长时训练提供高效容错保障，值得关注。
arXiv:2608.14635v1 Announce Type: cross Abstract: Large language model (LLM) agents are increasingly trained with reinforcement learning in long-horiz…
14
Le Critique: Privileged Value Functions for LLM Reinforcement Learning
给大模型强化学习装上“特权价值函数”，让模型更懂自己表现，训练效率有望大幅提升。
arXiv:2608.16739v1 Announce Type: new Abstract: Reinforcement learning algorithms for Large Language Models (LLMs) are largely distinguished by their …
15
TRCA: Transition-wise Rubric Credit Assignment for Long-horizon LLM Agents
面向长时程智能体，TRCA用过渡级评分信用分配破解稀疏奖励难题，值得关注。
arXiv:2608.16156v1 Announce Type: new Abstract: Long-horizon large language model (LLM) agents are typically optimized with sparse terminal outcomes, …
16
微软 Copilot Notebooks 扩展支持 Markdown，强化 AI 资料分析
集中管理Word、PPT、Excel等资料，AI精准分析摘要，支持Markdown让笔记更灵活。
IT之家 8 月 15 日消息，科技媒体 Neowin 昨日（8 月 14 日）发布博文，报道称微软更新 Copilot Notebooks 应用， 新增支持 Markdown、纯文本和富文本 3 种文本格式。 IT之家注：Copilot Notebooks 是微软 Microsoft 365 Co…
17
GCPO: Diagnosing and Constraining Subspace Geometry in Rollout RL for LLMs
揭秘大模型rollout强化学习中的子空间几何，提出GCPO约束方法，助你理解训练背后的数学本质。
arXiv:2608.11674v1 Announce Type: cross Abstract: On-policy rollout methods such as GRPO are central to post-training of large language models, yet th…
18
Cutting AI Datacenter Energy with Reinforcement Learning: Measured Power Control of LLM Training from One GPU to the Fleet
用强化学习给AI数据中心“降耗”，实测从单卡到集群的LLM训练功率控制，兼顾性能与能效。
arXiv:2608.11226v1 Announce Type: new Abstract: Reinforcement-learning post-training dominates modern language-model development, yet its power behavi…
19
A-3PO: Accelerating Asynchronous LLM Training with Staleness-aware Proximal Policy Approximation
异步LLM训练如何对抗数据陈旧？这项研究提出陈旧感知的近端策略近似，让PPO在异步场景下更稳更快。
arXiv:2512.06547v4 Announce Type: replace-cross Abstract: Decoupled PPO has been a successful reinforcement learning (RL) algorithm to deal with the h…
20
Reinforcing Step-level Reasoning for Effective Self-Correction in LLMs
从步骤级推理入手，强化大模型自我纠错能力，为提升推理可靠性提供新思路。
arXiv:2608.11573v1 Announce Type: cross Abstract: Achieving effective self-correction, where models verify and correct their own mistakes, remains a f…