Recursive self-improvement (RSI) remains hard in long-horizon tasks, where growing histori…
Group-relative reinforcement learning waits for sibling rollouts of the same prompt, which…
Web agents that act from rendered pixels avoid the fragility and heavy token cost of readi…
Real-world data for knowledge graph question answering is often distributed across differe…
Large language models (LLMs) are increasingly deployed as AI analysts to process financial…
The rapid expansion of large-scale assessments and the growing adoption of automatic item …
Large language models are increasingly used for knowledge graph question answering (KGQA),…
Recent work has applied Mamba style state space models (SSMs) to video anomaly detection, …
We present Crase, a bounded and inspectable alternative to deep research agents for schola…
We present StarHarness, a framework for evolving environment-specific agent harnesses whil…
Outcome-supervised search agents learn when and how to retrieve evidence, but terminal rew…
Clinicians read chain-of-thought (CoT) rationales as evidence of medical reasoning, but wh…
Distinguishing machine-generated text (MGT) from human-written text (HWT) becomes increasi…
LLM-based agents can interact with external environments through tool invocation, but this…
Large language model agents are moving beyond conventional retrieval-augmented generation …
Text-to-CAD aims to generate executable CAD programs from natural-language descriptions. H…
Discovering stable neuron behavior across entire domains remains a challenge in mechanisti…
Self-improving LLM agents refine answers, not the process that produces those answers. Sys…
Lifted inference algorithms enable scalable probabilistic inference even for large object …
Agentic systems increasingly gate actions on a model's own stated confidence, which assume…
Large language models (LLMs) are commonly evaluated under the assumption that their observ…
Scaling test-time reasoning has substantially improved the problem-solving ability of larg…
Accurate assessment of student competencies is essential for enabling educators to identif…
Automated high-density storage systems (warehouses, robotic parking, plant logistics, etc.…
Video multimodal large language models support language guided video segmentation, but the…
Open-source AI-assisted per-application routing for Windows with OpenAI/Ollama drafts and …
NodeSeek 用户 AI 画像 Userscript：快速画像、深度交易分析、多用户并发、管理记录、多 AI Provider、Token 展示与图片分享。
A lightweight Windows companion for switching Codex pets and viewing ChatGPT-backed usage …
A recap of all the incidents involving LLMs made by Anthropic, Meta, and OpenAI, which wen…
NVIDIA Vice President of Hyperscale and HPC Ian Buck hand-delivers Vera CPU systems across…
Plaud's new 'agentic' earbuds are priced at $249.
8月26日晚，智谱正式上线并开源GLM-5.3-Flash(320B-A18B），这是GLM-5系列的首个原生多模态模型。其总参数320B，能力超过GLM-5.2，在全球权威的Ar…
好能增长，好能商业化啊！
OpenAI has more than 100 million weekly active ChatGPT users in India, a huge chunk of who…
8月27日，“搭子进行时——百度搭子产品发布会”在上海举行。百度搭子围绕“交付即惊艳”宣布个人版、企业版、专业套件及工作台能力一揽子升级，重点面向自媒体、金融等专业场景，覆盖选题策…
<img data-src="https://static.leiphone.com/uploads/new/images/20260827/6a90182c80098.jpg" …
2026具身智能黑马玩家昆仑行机器人，全面领先全球权威榜单 WorldArena。 <span st
<img data-src="https://static.leiphone.com/6a6c4981e5bcb.jpg" alt="图片" class="rich_pages w…
今天，距离 ECCV 2026 在瑞典马尔默开幕只剩 12 天，会前预热迎来最重磅的一天：官方完整议程首次揭晓，图灵奖得主 Yann LeCun 携手自动驾驶与世界模型领域的两位顶…
<img class="rich_pages wxw-img" data-aistatus="1" data-croporisrc="https://mmbiz.qpic.cn/m…
8月27日，阿里发布全新Qoder，这是一款以Coding为核心能力，面向所有人的智能体工作台。全新Qoder采用更加简洁、以任务为中心的用户界面。用户不必从代码开始，只需用自然语…
<img class="rich_pages wxw-img" data-aistatus="1" data-croporisrc="https://mmbiz.qpic.cn/s…
Nvidia has reportedly agreed to buy Hugging Face, the popular open-source AI hub, for $12.…
当不少企业还在讨论AI能做什么时，中国飞鹤已经把问题往前推进了一步——“飞鹤生产端已经不是要不要投入AI的阶段，而是进入哪些场景值得规模化的阶段。”中国飞鹤副总裁、生产事业部负责人…
近日，AI基础设施公司基元律动（TokenRhythm）宣布完成新一轮融资，融资额达数千万美元。
线下实体网点正经历一场静默的效率困局。一方面，人力成本刚性攀升，单厅月均运维成本达8至15万元；另一方面，客流波谷时段的人力空转与高频重复咨询造成的隐性浪费，每年可达20至40万元…
OpenAI is expanding its presence in Brazil, deepening engagement with developers, business…
西门子Xcelerator与普通软件货架最本质的区别。货架解决的是「把产品卖出去」，西门子Xcelerator要解决的是「让产品在真实工业场景中持续生长」。
8月26日晚，千问办公首发上线刚刚发布的Qwen3.8-Flash模型，同时推出标准模式。
The startup is only a year old but it has already generated a massive amount of hype (and …
Amazon is adding another 2 million Nvidia GPU chips to its data centers over the next two …
The new deal with the infrastructure provider is the latest example of Anthropic's white-h…
The next wave of AI is placing new demands on infrastructure. As AI agents and trillion-pa…
Consumer AI apps need to stop making users learn their product architecture.
Was Greg Brockman the right executive all along?
The report, which spans several discrete cybersecurity compromises, is the most complete a…
GLM-5.3 Flash
Particle’s new podcast intelligence platform transcribes and analyzes more than 130,000 po…
Perceptron offers an AI model that it says can help machines navigate the world while also…
Gates is mostly in the Responsible AI camp, but there are a few ideas in here we hadn't he…
Z.ai confirms it is behind Ox Alpha, the mysterious open AI model topping benchmarks and l…
8月26日晚，阿里发布并同步开源千问最新模型 Qwen3.8-Flash。该模型采用全新下一代架构，千亿总参数仅激活60亿（6B）即可获得超越Claude Opus4.6的前沿性能…
商汤董事长兼首席执行官徐立表示：“商汤构建的‘一套模型、一座 Token 工厂、一套智能体管控系统’核心能力体系， 用一套模型实现持续 提升智能上限
OpenAI’s new report explores how students and educators use ChatGPT to make learning more …
ChatGPT for Teachers is expanding to 55 U.S. school systems, bringing secure AI tools, tra…
OpenAI shares findings from the Hugging Face security incident and the steps we’re taking …
Introducing Intelligence Age, a new OpenAI blog exploring how transformative AI could resh…
YouTube · NVIDIA
YouTube · Google for Developers
YouTube · Lex Fridman
YouTube · Google for Developers