模型发布/更新
-
SenseTime open-sources SenseNova-Vision-7B-MoT multitask vision model
SenseTime has released and fully open-sourced SenseNova-Vision-7B-MoT, a model that uniformly handles major vision…
-
腾讯混元 Hy3 量化版发布:1bit 版本单卡可部署,4bit 版本接近满血性能
腾讯混元团队为旗舰模型 Hy3(295B 参数)推出量化版本。1bit 版本(IQ1_M)将权重从 598 GB 压缩至 85.5 GiB,缩小 6.7 倍,单张 96GB 推理显卡即可部署;4bit 版本(Q4_K_M)体积 169.9…
-
Google showcases Pixel 10 on-device AI future powered by Tensor SoC and TPU at I/O Connect India
At Google I/O Connect India, Google showcased a 100% private on-device AI future powered by the Pixel 10 series,…
产品发布/更新
-
Codex surpasses 7 million weekly active users, 150+ updates in two months
7M+ weekly active Codex users. 150+ updates in two months. @romainhuet walks you through the latest Codex…
-
Amap releases general world model studio ABot-WorldStudio, now in open testing
Alibaba's Amap has released ABot-WorldStudio, a general world model studio, and opened it for testing. The product…
-
Google AI releases Gemini 3.5 Live Translate with near-real-time speech-to-speech translation in 70+ languages
Google AI released Gemini 3.5 Live Translate, supporting near-real-time speech-to-speech translation in more than 70…
-
Anthropic launches Claude for Teachers
Anthropic released Claude for Teachers, offering U.S.-certified K-12 teachers free access to advanced Claude features,…
-
Claude Code v2.1.208 发布
Claude Code v2.1.208 发布。新增屏幕阅读器模式,可通过 `claude --ax-screen-reader` 等启用。新增 `vimInsertModeRemaps` 设置,支持映射双键序列为 Escape。新增 `…
-
Google Images 25周年:推出可浏览图片主页与AI Overviews图像生成功能
Google Images迎来25周年,推出两项新功能:一是全新的可浏览图片主页,展示来自网络的动态沉浸式图片画廊,实时更新并根据用户兴趣智能定制,支持收藏夹标签页,未来几周在美国桌面端英文上线;二是将图像生成直接引入AI Overvie…
行业动态
-
OpenAI GPT-5.6 Sol reportedly deletes user files and databases on its own
After OpenAI's latest flagship model GPT-5.6 Sol launched, multiple developers posted on X that the model deleted Mac…
-
New York State pauses all new large data center projects
New York State became the first state in the U.S. to pause data center construction. Governor Kathy Hochul signed an…
-
Google hit with publisher class action over AI training
A group of publishers and authors, including Hachette, Cengage, Elsevier, and author Scott Turow, filed a class action…
-
Anthropic 向加拿大 AI 研究捐赠 1000 万加元
Anthropic 宣布向加拿大研究机构捐赠 1000 万加元,用于资助有益且负责任的 AI 应用研究。合作伙伴包括 Amii、Mila、Vector Institute、CHEO、CAMH、Université Laval、Univer…
-
Demis Hassabis 支持 AI 预飞安全测试
DeepMind 联合创始人兼 CEO Demis Hassabis 公开支持对 AI 系统实施“预飞安全测试”(preflight safety testing),即在部署前进行类似航空业的安全检查。这一立场与当前业界对 AI 安全监管…
论文研究
-
Anthropic Economic Index: analysis of Claude usage in Canada
Based on a February 2026 sample of Claude.ai conversations, Canada accounted for 2.6% of global traffic, with…
-
Apple 等机构提出 Proactive Agent Research Environment (Pare),将应用建模为有限状态机以评估主动式智能体
现有用户模拟框架将应用建模为扁平的工具调用 API,无法捕捉数字环境中用户交互的状态性和顺序性。Apple 与加州大学圣塔芭芭拉分校、华盛顿大学等机构的研究团队提出 Proactive Agent Research Environment…
技巧与观点
-
Hands-on with LibTV Agent: 100 AI video workflows reorganized into Skills for creative freedom
LibTV launched the Agent feature with a built-in Skill Hub, offering more than 100 video Skills covering genres such…
-
Demis Hassabis: AGI may arrive in a few years, impact 10x the Industrial Revolution
Google DeepMind co-founder Demis Hassabis wrote that AGI may be achievable in just a few years, with an impact 10…
-
Interview with ModelBest CTO Zeng Guoyang: on-device models are key path for AI deployment
ModelBest CTO Zeng Guoyang stated that on-device models are a key path for AI deployment. Its original "model wind…
-
LMSYS 与 SGLang 团队为 GLM-5.2 NVFP4 推出推理优化,8×B300 单 batch 解码超 500 TPS
LMSYS 与 SGLang 团队针对智谱 GLM-5.2 NVFP4 模型在 Grace Blackwell 硬件上推出多项优化。运行时方面,Spec V2 重叠调度消除 GPU 气泡,端到端 TPS 提升 11%;IndexShare…
-
唱作人 sad alex 谈如何用 Suno 作为创意草稿本,在短内容中保持创作自主性
洛杉矶唱作人 sad alex 分享她如何将 Suno 用作创意草稿本:解决人声转换、乐器采样或 demo 搭建等具体问题,且仅用于自己 100% 拥有版权的歌曲。她认为 AI 本质上是“向后看”的,而人类创作是“向前看”的,因此 Sun…
-
Apple Music 引入多语言语义检索系统
Apple 机器学习研究团队为 Apple Music 搜索引入多语言语义检索系统,覆盖 150 多个国家及地区的数十种语言。该系统利用多语言嵌入向量模型,将用户查询与歌曲等内容的语义表示映射到同一向量空间,实现跨语言匹配。检索准确率较此…