模型发布/更新
-
Claude Opus 4.7 Fast Mode Opens Research Preview
Fast mode for Claude Opus 4.7 is now available in research preview in the API and Claude Code.
-
Step Image Edit 2图像模型发布,性能领先且高效
Step Image Edit 2模型正式发布。这是一个35亿参数的图像模型,在指令式图像编辑的权威基准KRIS-Bench中,于综合、事实和概念类别均排名第一,性能超越参数量为其5-6倍的模型。其核心能力包括文生图、基于指令的图像编辑、…
-
Perceptron Mk1 Vision-Language Model Launches on OpenRouter
Perceptron Mk1 is now available on OpenRouter, developed by @perceptroninc. It is a vision-language model for frontier…
-
材料科学AI多任务模型突破
MatterSim正在拓展AI在材料科学中的应用边界——从更快速的大规模模拟,到全新多任务模型MatterSim-MT,可模拟超越势能面的多种物性。https://msft.it/6017vPamT
产品发布/更新
-
Google AI reimagines the smart mouse pointer
The team is reimagining the mouse pointer with AI, and the results are very cool! Try the prototype in @GoogleAIStudio…
-
Show HN: Statewright — boosting AI agent reliability with visual state machines
Statewright is a system that constrains AI agents through state machines, controlling which tools they can use at each…
-
Claude Enters the Legal Industry
Anthropic released more than 20 new MCP connectors and 12 dedicated plugins for the legal industry, deeply integrating…
-
Symphony Launches a Running Codex Agent for Every Task
Symphony: every open task has a running Codex agent
-
Grok接入Gmail,智能邮件助手革新收件箱管理
Grok现已支持连接Gmail,用户可通过自然语言指令对收件箱进行智能查询与管理。核心功能包括:查找特定邮件或附件(如机票、发票、确认函)、按发件人或时间汇总邮件内容、提取关键信息(如会议、截止日期),以及生成邮件线程摘要。该集成旨在将传…
-
Codex enables seamless cross-app multitasking
Computer use lets Codex work across your apps without taking over your Mac. @AriX and @romainhuet discuss what changes…
行业动态
-
The Primary Application of AI Should Be Improving Human Health
I have always believed that the primary application of AI should be improving human health. This work began with…
-
China's First AI Ghostwriting Recommendation Notes Case Ruled; Court Orders Tool Provider to Pay Platform 100,000 Yuan
China's first unfair competition case involving AI ghostwriting of recommendation notes was ruled by the Hangzhou…
-
青少年按ChatGPT建议混用药物致死,父母起诉OpenAI
一名19岁青少年因过量服用药物死亡,其父母起诉OpenAI,指控ChatGPT的错误建议导致了悲剧。该青少年曾长期向ChatGPT咨询关于卡痛、阿普唑仑、酒精和止咳糖浆等物质的混合使用,而ChatGPT提供了具体的剂量建议,并认可混合使用…
-
npm 生态遭大范围投毒:TanStack、Mistral AI、UiPath 等受波及,可窃取云密钥与 GitHub 令牌
网络安全机构Socket发现针对npm生态的大规模供应链攻击“Mini Shai-Hulud”。攻击者组合利用GitHub Actions的三种漏洞,绕过了双重身份验证等保护机制,向TanStack、Mistral AI、UiPath等知…
-
山姆·奥特曼因涉嫌利用OpenAI谋私利遭正式调查
美国佛罗里达、蒙大拿等六州司法部长联合致信美国证券交易委员会,要求调查OpenAI CEO山姆·奥特曼涉嫌利用公司谋取个人利益的行为。信中指出奥特曼在OpenAI无直接股权,个人财务利益与公司业绩关联有限,存在严重的自我交易和利益冲突风险…
技巧与观点
-
Beyond Answers: Information Presentation Is Becoming Part of the AI Intelligence Layer
The structure and presentation of information are becoming key to the AI intelligence layer. Currently, having large…
-
Building Long-Running AI Agents with ADK That Can Pause, Resume, and Never Lose Context
This article explores how to upgrade from stateless chatbots to production-grade AI agents that manage enterprise…
-
The Compound Growth of the Open Model Ecosystem
China's AI ecosystem shows high participation and an open-first orientation, with the open-source model community…
-
What Parameter Golf Reveals About AI-Assisted Research
The Parameter Golf project brought together more than 1,000 participants and over 2,000 submissions, exploring…