模型发布/更新
-
Meta Superintelligence Labs launches Muse Image and Muse Video
Meta Superintelligence Labs has released its first media generation models, Muse Image and Muse Video. Muse Image is…
产品发布/更新
-
Rowboat: an open-source, local-first desktop AI assistant
Rowboat is an open-source, local-first desktop AI assistant that indexes data from email, meetings, Slack, and more…
-
Grok Imagine update: 15-second videos supported
Grok Imagine update. Update your Grok app! 15-second Imagine videos are now available, with incredible quality.
-
Claude Cowork opens to mobile and web
Claude Cowork is opening to mobile and web, letting conversations and files sync across devices. The beta will roll…
-
Gemini API Managed Agents add background execution, remote MCP, and custom functions
Google has added background execution, remote MCP server integration, custom function calling, and credential refresh…
-
NotebookLM short video overviews officially launch
Short video overviews are now fully available on mobile and web for all English-language users! As always, your…
-
Hugging Face Storage becomes a first-class SkyPilot backend: zero egress cross-cloud storage
Hugging Face Storage is now a first-class backend for SkyPilot. Using `hf://` URLs and an existing HFTOKEN, users can…
-
Claude Code v2.1.203 发布
本次更新新增登录过期警告和手动权限模式标记,并将附加工作目录加入 MCP roots/list。修复了 macOS 下因内存检测误报导致后台会话卡顿 15–20 秒(回归自 2.1.196)、后台会话因 token 过期永久不可用(现自动…
行业动态
-
China plans to restrict foreign access to its most powerful AI models
China plans to restrict foreign access to its most powerful AI models, having recently met with companies including…
-
Microsoft replaces OpenAI and Anthropic models with in-house MAI models in Copilot to cut costs
Microsoft is replacing OpenAI and Anthropic models in Copilot products with its in-house MAI models to reduce…
-
First U.S. autonomous ground vehicles enter combat in Ukraine
U.S. autonomous vehicle company Forterra announced that over the past nine months it has deployed more than 100 Lancer…
论文研究
-
Apple study: a single neuron can bypass large language model safety alignment
Apple researchers found that safety alignment is regulated by two types of neurons: refusal neurons control whether…
-
Weblica: a scalable and reproducible training environment for visual web agents
Apple researchers proposed the Weblica framework, which uses HTTP-level caching to preserve stable visual states of…
-
DynaMiCS: dynamic mixed fine-tuning of large language models with performance constraints
DynaMiCS is a dynamic mixture optimizer that formulates multi-domain fine-tuning as an optimization problem with…
技巧与观点
-
FDE boom: AI companies commit $9.75B to deployment engineering in 12 months
AI companies have collectively committed $9.75 billion over 12 months to build forward-deployed engineering (FDE)…
-
Liquid AI open-sources Antidoom: a final-token preference optimization fix for reasoning model doom loops
Liquid AI has open-sourced Antidoom, a targeted fix based on Final Token Preference Optimization (FTPO) to reduce doom…
-
Krea 2 identity preservation feature launches
Krea 2's identity preservation feature has been released, and the accompanying model and ComfyUI nodes are also live. 🔥
-
人们如何使用Claude Cowork
基于2026年5月11-31日120万次匿名会话样本,Claude Cowork最大用途为业务流程与运营(33.4%),如整理报告、核对表格;其次为内容创作与文案(16.4%),如起草稿件、制作幻灯片;软件开发仅占8.7%。用户多用它处理…
-
在 Claude Code 中选择 Claude 模型与努力级别
Claude Code 提供模型(model)和努力级别(effort)两种设置。模型选择决定能力范围,更大模型(如 Claude Fable 5)在基准上优于 Claude Sonnet。努力级别不单是思考时间,还控制读取文件数、验证步…