1
One Token to Fool LLM-as-a-Judge
只需一个token就能轻松骗过LLM评判者，揭示AI评估体系的安全软肋。
arXiv:2507.08794v3 Announce Type: replace-cross Abstract: Large language models (LLMs) are increasingly trusted as automated judges, assisting evaluat…
2
ReSum: Synergizing LLM Reasoning and Summarization with Reinforcement Learning
新论文ReSum用强化学习协同LLM推理与摘要，解决长推理链低效问题，干货满满。
arXiv:2606.13316v1 Announce Type: new Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) is a central technique for improving long-horizo…
3
Certifiable Safe RLHF: Semantic Grounding and Fixed Penalty Constraint Optimization for Safer LLM Alignment
语义基础+固定惩罚约束优化，让大模型对齐过程获得可认证的安全保障
arXiv:2510.03520v2 Announce Type: replace-cross Abstract: Ensuring safety is a foundational requirement for large language models (LLMs). Achieving an…
4
TruthRL: Incentivizing Truthful LLMs via Reinforcement Learning
用强化学习让大模型更诚实，TruthRL方法提升LLM回答真实性，含代码开源
arXiv:2509.25760v2 Announce Type: replace-cross Abstract: While large language models (LLMs) have demonstrated strong performance on factoid question …
5
Beyond Uniform Token-Level Trust Region in LLM Reinforcement Learning
非均匀令牌级信任区域优化，突破传统限制提升大模型强化学习训练稳定性。
arXiv:2606.10968v1 Announce Type: cross Abstract: Reinforcement learning with verifiable rewards (RLVR) has become standard for improving LLM reasonin…
6
TD-Grokking: Learning from Zero-Reward Problems by Training-Time Decomposition
LLM推理训练新突破：通过训练时分解攻克零奖励难题，让模型从失败轨迹中学习！
arXiv:2606.09883v1 Announce Type: cross Abstract: Large language models (LLMs) have made remarkable progress in reasoning tasks, largely driven by pos…
7
ComplexConstraints and Beyond: Expert Rubrics for RLVR
提出专家评分标准解决RLVR中复杂约束问题，为强化学习奖励设计提供新范式
arXiv:2606.09118v1 Announce Type: new Abstract: As LLM capabilities advance rapidly, the evaluation methods used to assess them increasingly lag behin…
8
CATPO: Critique-Augmented Tree Policy Optimization
CATPO方法通过批评增强的树策略优化，显著提升大语言模型推理中的密集奖励获取效率。
arXiv:2606.08346v1 Announce Type: cross Abstract: Reinforcement learning with verifiable rewards (RLVR) has become a dominant paradigm for improving t…
9
OpenAI 申请上市，奥尔特曼估值 25 亿美元的眼球扫描公司却被曝裁员
OpenAI冲刺上市，奥尔特曼旗下估值25亿美元的眼球扫描公司却陷入裁员与各国监管围堵。
IT之家 6 月 9 日消息，OpenAI 于当地时间周一宣布已秘密提交首次公开募股（IPO）申请，这或将成为近十年最具标志性的上市事件之一。另据 Business Insider 报道，OpenAI 首席执行官山姆 · 奥尔特曼旗下的另一家公司 Tools for Humanity 正进行裁员。 …
10
As OpenAI files for IPO, Sam Altman’s eye-scanning company is doing layoffs, report says
OpenAI冲刺IPO之际，Sam Altman的虹膜扫描公司World却因监管与商业困境进行裁员，反差揭示科技巨头生态的复杂性。
Tools for Humanity, Sam Altman's identity verification company, is reportedly struggling to generate revenue and will downsize its staff.
11
What Do People Actually Want From AI? Mapping Preference Plurality
顶会论文揭示RLHF聚合偏好的根本缺陷，系统绘制人类对AI的真实多元需求图谱
arXiv:2606.06674v1 Announce Type: new Abstract: Large Language Models (LLMs) are often fine-tuned through Reinforcement Learning from Human Feedback (…
12
微软让步，撤回对白帽黑客“梦魇日蚀”的法律威胁
多引擎恶意文件检测平台，支持URL和文件扫描，社区共享威胁情报，助力白帽安全研究
IT之家 6 月 7 日消息，据科技媒体 Notebookcheck 今天报道，在全球网络安全行业人士的强烈反对之下，微软已正式收回此前针对白帽黑客“梦魇日蚀（Nightmare Eclipse）”的强硬法律威胁。 据报道，“梦魇日蚀”曾在此前绕过微软的传统漏洞提交流程，直接公开了多个 Window…
13
Show HN: Lathe – Use LLMs to learn a new domain, not skip past it
用LLM重构学习方式：Lathe让你深入新领域，而非走捷径，每个教程都透明记录来源、模型和提示。
Hey HN! Lathe is an experiment in using LLMs to teach me something new, instead of doing the work for me. It generates a hands-on, source-backed tutor…
14
Show HN: Summarize YT Video by pasting url into AI chat[video]
粘贴YouTube视频链接到AI聊天框，即可自动生成视频摘要，无需手动观看，集成在AI工具中操作更便捷。
We added tooling to our chat to make it agentic. It can control our 40+ apps suite. One of the tools is url fetching with pagination. Comments URL: ht…
15
Show HN: Clarity, See what concepts your LLM uses and trace it to training data
首个可解释AI平台Clarity，让你看到大模型使用了哪些概念并能追溯至训练数据。
Article URL: https://www.guidelabs.ai/post/meet-clarity/ Comments URL: https://news.ycombinator.com/item?id=48401606 Points: 3 # Comments: 1
16
A Systematic Investigation of RL-Jailbreaking in LLMs
系统研究强化学习对LLM的越狱攻击，揭示AI安全新风险，值得关注
arXiv:2605.07032v2 Announce Type: replace-cross Abstract: The evolution of generative models from next-token predictors to autonomous engines of compl…
17
AgentJet: A Flexible Swarm Training Framework for Agentic Reinforcement Learning
一种灵活的群体训练框架，让智能体强化学习更高效协同。
arXiv:2606.04484v1 Announce Type: new Abstract: We present AgentJet, a distributed swarm training framework for large language model (LLM) agent reinf…
18
Libra: Efficient Resource Management for Agentic RL Post-Training
高效管理Agentic RL后训练资源的新方案Libra，降低训练成本、提升性能。
arXiv:2606.03077v1 Announce Type: cross Abstract: Reinforcement learning (RL) has become a standard post-training paradigm for large language models (…
19
AUGUSTE: Online-Learning dApp for Predictive URLLC Scheduling
AUGUSTE是面向5G URLLC的在线学习dApp，用AI预测调度实现1毫秒级超可靠低延迟通信
arXiv:2606.03664v1 Announce Type: cross Abstract: Ultra Reliable and Low Latency Communications (URLLC) was one of the main motivations behind 5G, wit…
20
URL Encoding Explained: Special Characters and How to Handle Them
快速将URL中的特殊字符进行编码或解码，确保链接安全无错误，支持多种编码格式
URL Encoding Explained: Special Characters and How to Handle Them 📅 May 25, 2026⏱️ 7 min read🔗 Network Tools Every character in a URL has a meaning. S…