AI 日报 /2026年9月8日

AI 日报 · 2026年9月8日

今天几条值得看的:微软Phi-4系列开源多模态指令版,图像、音频、文本一把抓,主打小参数、跑得动;颜水成团队的VoiceMem用「流式双脑架构」专攻实时语音AI;法国Mistral刚融了30亿欧元,估值210亿,主打主权AI生意。另外Hugging Face在聊AI安全的新思路——只拒危险那一小撮,别一竿子打死整个话题。

打开交互阅读器 →

实用技巧 · 1 条

  • 实用技巧 TechCrunch

    Opaque recurrence, and other AI terms that you should probably know

    The rise of AI has brought an avalanche of new terms and slang. Here is a glossary with definitions of some of the most…

    来源:TechCrunch

行业资讯 · 7 条

  • 行业资讯 机器之心

    刚刚,OpenAl最新内部数据公开!2028年实现全自动A研究

    刚刚,OpenAl最新内部数据公开!2028年实现全自动A研究

    来源:机器之心

  • 行业资讯 新智元

    仅成立4个月,这家中国公司的全原子生物基础大模型登顶全球

    仅成立4个月,这家中国公司的全原子生物基础大模型登顶全球

    来源:新智元

  • 行业资讯 新智元

    虚幻引擎CEO围观最新世界模型:代码驱动,视频模型生成画面

    虚幻引擎CEO围观最新世界模型:代码驱动,视频模型生成画面

    来源:新智元

  • 行业资讯 机器之心

    颜水成团队重磅发布VoiceMem,提出「流式双脑架构」打造下一代实时...

    颜水成团队重磅发布VoiceMem,提出「流式双脑架构」打造下一代实时...

    来源:机器之心

  • 行业资讯 HuggingFace Trending Models

    微软Phi-4多模态指令模型开源

    微软Phi-4系列新增多模态指令版本,支持图像、音频和文本联合理解与生成,参数效率高,适合边缘设备和本地推理场景。

    来源:HuggingFace Trending Models

  • 行业资讯 TechCrunch

    Mistral raises €3B as sovereign AI becomes big business

    The French AI lab has raised €3 billion at a €21 billion valuation in a Series D round led by Samsung, Scaleup Europe an…

    来源:TechCrunch

  • 行业资讯 The Verge

    谷歌DeepMind发布AlphaGenome Atlas基因组AI工具

    Google DeepMind推出名为AlphaGenome Atlas的AI平台,包含人类基因组中所有可能的DNA碱基变化预测图谱,有望加速生物学研究并推动新疗法开发。

    来源:The Verge

项目推荐 · 10 条

  • 项目推荐 GitHub TypeScript

    AdventDevInc/kudu

    Free Windows, Mac and Linux cleaner, scanner, and more.https://usekudu.com Kudu Free, open-source system cleaner & s…

    来源:GitHub TypeScript

  • 项目推荐 GitHub TypeScript

    aipoch/open-science

    Open Science by AIPOCH is an open-source, local-first, model-agnostic AI research workbench for macOS, Windows, and Linu…

    来源:GitHub TypeScript

  • 项目推荐 GitHub TypeScript

    EveryInc/compound-engineering-plugin

    Official Compound Engineering plugin for Claude Code, Codex, Cursor, and morehttps://every.to/guides/compound-engineerin…

    来源:GitHub TypeScript

  • 项目推荐 GitHub TypeScript

    Nutlope/logocreator

    A free + OSS logo generator powered by Flux on Together AIhttps://www.logo-creator.io/ LogoCreator An open source AI log…

    来源:GitHub TypeScript

  • 项目推荐 GitHub TypeScript

    ruvnet/ruflo

    🌊 The original agent meta-harness. Deploy intelligent multi-player swarms, coordinate autonomous workflows, and build co…

    来源:GitHub TypeScript

  • 项目推荐 GitHub TypeScript

    heygen-com/hyperframes

    Write HTML. Render video. Built for agents. Write HTML. Render video. Built for agents. Quickstart | Showcase | Playgrou…

    来源:GitHub TypeScript

  • 项目推荐 GitHub TypeScript

    BraveOPotato/FckSignups

    A list of tools that are open-source, in-browser, and require no-signups!https://nosignups.netNoSignups (formerly FckSig…

    来源:GitHub TypeScript

  • 项目推荐 GitHub TypeScript

    pascalorg/editor

    Create and share 3D architectural projects.https://editor.pascal.appPascal Editor A 3D building editor built with React…

    来源:GitHub TypeScript

  • 项目推荐 GitHub TypeScript

    fluxerapp/fluxer

    A free and open source instant messaging and VoIP chat app built for friends, groups, and communities.https://fluxer.app…

    来源:GitHub TypeScript

  • 项目推荐 GitHub TypeScript

    streetwriters/notesnook

    A fully open source & end-to-end encrypted note taking alternative to Evernote.https://notesnook.com/ Notesnook An e…

    来源:GitHub TypeScript

高质量文章 · 3 条

  • 高质量文章 Towards AI

    MCP Just Quietly Became AI’s Most Important Protocol.

    Model Context Protocol went from a niche Anthropic experiment to infrastructure inside Bloomberg, PayPal, Cisco, and Mic…

    来源:Towards AI

  • 高质量文章 Towards AI

    OpenAI Says Astra Saturates the Benchmarks. I Gave It 19,656 Cycles Instead.

    Codex vs. the Commodore 64: a benchmark nobody funded, on a machine nobody can game — and what a 1 MHz CPU says about an…

    来源:Towards AI

  • 高质量文章 Hugging Face Blog

    AI安全新思路:拒绝主题的特定子集而非整个主题

    Hugging Face博客探讨AI安全中的细粒度拒绝策略,主张模型应学会拒绝有害主题中的特定危险子集,而非对整个主题一概拒绝,以平衡安全性与可用性。

    来源:Hugging Face Blog