模型发布/更新
-
Microsoft releases MAI-Thinking-1, its first advanced reasoning AI model
At Build 2026, Microsoft released MAI-Thinking-1, its first advanced reasoning AI model. Positioned as "medium-sized,"…
-
Holo3.1: a fast local computer-use agent
Holo3.1 is a family of computer-use agents built on the Qwen model family, designed to improve robustness in desktop,…
-
阶跃星辰Step 3.7 Flash发布,专为高效推理设计
阶跃星辰发布其推理优化型模型Step 3.7 Flash。该模型为196B MoE架构,从设计之初就专注于推理效率。其采用多矩阵分解注意力机制,使KV-cache成本仅为DeepSeek模型的约22%;同时通过注意力与FFN解耦技术,实现…
产品发布/更新
-
Claude Code adds dynamic workflows
Claude Code has added dynamic workflows, allowing the model to improvise and coordinate multi-agent frameworks at…
-
Claude Platform adds a CLI tool
We have added a CLI to Claude Platform so every API endpoint can be run from your terminal. Call the Messages API,…
-
Google DeepMind open-sources science agent toolkit
Building autonomous agents for scientific discovery? 🧬🤖 @GoogleDeepMind Science Skills is now available on GitHub. We…
-
微软发布开源框架 Adaptive Spec-driven Scoring:支持用文本描述创建 AI 评估测试
微软开源发布 Adaptive Spec-driven Scoring for Evaluation and Regression Testing 框架。开发者可通过文本描述快速生成 AI 行为测试,用于模型评估与回归测试。
-
GitHub Copilot app: an agent-native desktop experience
At Microsoft Build 2026, GitHub released new tools and updates and positioned the Copilot app as an "agent-native…
行业动态
-
Alphabet拟融资800亿美元 Anthropic提交IPO申请
Alphabet宣布拟通过股权融资800亿美元,用于扩展AI基础设施。Anthropic已秘密提交IPO申请,在上市竞赛中领先于竞争对手OpenAI。此外,SpaceX正与华尔街机构协商其IPO的承销费用,HPE则因AI基础设施需求旺盛,…
-
Anthropic expands Project Glasswing program
Anthropic is expanding its Project Glasswing program to about 150 new organizations, following an initial cohort of…
-
SK Chairman Chey Tae-won: SK hynix plans to double wafer capacity within five years
SK hynix Chairman Chey Tae-won announced plans to double overall wafer capacity over the next five years to address…
-
Anthropic supports implementation of US AI executive order
This executive order is an important step in strengthening US AI leadership. We look forward to working with the White…
-
OpenAI呼吁通过全球领导力推进青年AI安全与机遇
OpenAI呼吁通过设立专门的AI安全研究所,在全球范围内采取行动,以保障青少年在使用AI时的安全,并创造更多发展机遇。
-
Travelers借助OpenAI在全国部署AI理赔助手
美国保险公司Travelers与OpenAI合作,构建了一款AI驱动的Claim Assistant。该工具旨在引导客户完成理赔流程,并提供全天候支持,以在业务高峰期扩展运营规模。
论文研究
-
Microsoft Research: Aurora forecasts weather thousands of times faster than traditional supercomputers
Weather forecasting thousands of times faster than traditional supercomputers. Hear Kenji Takeda share about Aurora at…
-
Anthropic interpretability research: distinguishing features with similar causal effects
Anthropic's interpretability team presented new progress in its Circuits research. To distinguish model features with…
-
Codex正在成为每个人的生产力工具
The Next Era of Knowledge Work报告指出,Codex正通过AI增强的研究、数据分析、工作流自动化与内容创作,变革知识工作的生产力。
技巧与观点
-
为了不花那120刀,我把电脑清理软件做成了开源skill
作者受一条推文启发,使用Codex对自己的MacBook进行了只读存储分析,发现了B站缓存视频等大量可清理空间(激进方案超140G)。为替代收费软件CleanMyMac,作者制作并开源了一个跨平台(支持Mac/Windows)的AI清理s…
-
SenseTime open-sources SenseNova-Skills AI office skills suite
SenseTime has open-sourced SenseNova-Skills, an AI office skills suite. It is an open-source skill collection designed…
-
Karpathy shares learning methodology
🎯 Andrej Karpathy on how to learn.
-
Claude Code team practice: how agentic coding reshapes engineering organizations and processes
At Code w/ Claude SF 2026, the Claude Code engineering team shared the process and structural changes that followed…
-
Tips for Claude Code self-checks and feedback loops
How do you get Claude Code to check its own work before handing it back? See how to encode your manual checks so…
-
Gemini Spark:最令人印象深刻也最可怕的AI体验
Google DeepMind的AI模型Gemini Spark提供了一次极为深刻但同时令人感到不安的用户体验。该模型展现的强大能力令人印象深刻,但其带来的影响和潜力也引发了深刻的恐惧感。
-
The thriving ecosystem of open models
According to OpenRouter platform data, open model usage has grown significantly since 2025. The latest data shows…
-
Gary Marcus: why things will eventually fall apart
Well-known AI critic Gary Marcus explores the fundamental challenges facing AI development in his column on…