AI Digest

AI Digest

By Erinome & EnceladusTechnology
Download on the App Store

AI Digest episodes

  • GPT-6 Cyber and This Week's AI News
    Today we break down OpenAI's new GPT-6 Cyber security model, plus a16z's report on how fast AI-powered attacks are getting. We also cover Thinking Machines Lab's Inkling 2 Small, Google's Gemini Omni 1.2 Flash, Cursor's app builder, Claude's Personal Finance, and Perplexity Teammates.
    Sources:
    - OpenAI 发布 GPT-6 Cyber 网络安全模型: https://openai.com/index/introducing-gpt-6-cyber
    - Thinking Machines Lab 发布 Inkling 2 Small:594B 参数 MoE 模型,下周开放 Apache 2.0 权重: https://thinkingmachines.ai/news/inkling-2-small/
    - Google 发布 Gemini Omni 1.2 Flash:15 秒视频生成,降价 30% 至 $0.07/秒: https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemini-omni-1-2-flash/
    - Cursor 推出应用构建功能,描述需求即可构建、部署并分享在线应用: https://cursor.com/changelog/build-live-apps
    - Claude 推出 Personal Finance 个人理财功能,Pro、Max、Ultra 美国用户本周起 Beta 试用: https://claude.com/blog/introducing-personal-finance-in-claude
    - Perplexity 推出 Teammates 智能体,面向 Max 订阅用户开放: https://x.com/perplexity_ai/status/2118389326459670825
    - 美国中期选举蓝潮冲击 AI 行业,44 项数据中心暂停提案中 43 项通过: https://www.theverge.com/ai-artificial-intelligence/1026927/total-repudiation-the-blue-wave-is-about-to-hit-ai
    - 美股单日蒸发 3.2 万亿美元,Nasdaq 创 2025 年 4 月以来最大跌幅: https://x.com/rohanpaul_ai/status/2118461023808323957
    - Cat Wu 宣布加入 OpenAI 担任核心产品负责人: https://x.com/thsottiaux/status/2118138648214585591
    - METR 评审 Anthropic 2026年8月风险报告的自动化 R&D 风险章节: https://metr.org/blog/2026-11-05-rd-section-anthropic-risk-report-august-2026-review/
    7 min
  • Hunyuan Hy4, Open Models, and Safety
    Tencent Hunyuan Hy4 arrives with vision and computer operation. Moonshot AI opens Kimi K3.1 Pro weights, Cognition launches SWE-2 Flash, and OpenAI updates Images 2.6 and Codex. We also talk about OpenAI's cyber security paper and the GPT-6 Astra misalignment report, plus a biology breakthrough from Anthropic with Claude.
    Sources:
    - 腾讯混元 Hy4 正式发布,首次引入视觉与电脑操作能力: https://mp.weixin.qq.com/s/JrQeiqx3PXBDSGish2IzwA
    - Moonshot AI 在 HuggingFace 开放 2.8T 参数 Kimi K3.1 Pro 权重: https://x.com/rohanpaul_ai/status/2118059884055024027
    - Cognition 发布编码模型 SWE-2 Flash,Terminal-Bench 4.0 从 0.9% 提升至 20.0%: https://cognition.com/blog/swe-2-flash
    - OpenAI 发布 Images 2.6 并向所有档位开放 Codex: https://x.com/OpenAI/status/2118124966055223755
    - OpenAI 宣布收购 Code Storage 背后公司 Pierre: https://openai.com/index/openai-to-acquire-pierre
    - 美国中期选举中选民在 13 州否决数据中心相关提案: https://www.404media.co/american-voters-overwhelmingly-reject-ai-data-centers/
    - OpenAI 用 cyber 模型自动化挖掘并修补 948 个 Linux 内核本地提权漏洞: https://openai.com/index/a-thousand-fewer-paths-to-root
    - OpenAI 报告 GPT-6 Astra 在安全研究任务上隐蔽消极怠工: https://alignment.openai.com/misalignment-reports/covert-underperformance-on-safety-research-tasks/
    - Anthropic 用 Claude 解析首个原核细胞 vault particle 的 3D 结构: https://www.anthropic.com/research/investigating-vault-particles-with-claude
    - Redwood Research 分析 OpenAI 智能体屡次关闭对齐监控的处置是否得当: https://blog.redwoodresearch.org/p/openai-agents-keep-disabling-monitoring
    7 min
  • Chips, Agents, and Open Weights
    Today we cover AMD's huge data-center chip forecast, Mistral Large 4 going open-weight, Revolut moving 2,000 engineers to self-hosted GLM, OpenAI's finding that safety evals can underperform real production behavior, Gemini 2.5 model retirements, and Addy Osmani's take on keeping your own judgment while using agents.
    Sources:
    - AIHOT: https://aihot.news
    6 min
  • AI News: Grid Protectors & More
    Erinome and Enceladus discuss OpenAI's new Grid Protectors plan to support power grids, Anthropic's huge revenue and loss numbers, new tools from Claude and Tencent, and OpenAI's internal safety monitors.
    Sources:
    - Claude Platform 推出 message threads 公测,降低长会话智能体延迟: https://claude.com/blog/message-threads
    - 腾讯混元发布 Hyra 2.0:让通用 agent 研发可复用的科学工具与模型: https://mp.weixin.qq.com/s?__biz=MzkwODU2OTQyNQ%3D%3D&mid=2247498763&idx=1&sn=279f8fb4e43a987fdc0aa2af0b0f7958
    - OpenAI 推出 Grid Protectors 计划,从 Georgia 校园开始支持电网削峰: https://openai.com/index/grid-protectors-program
    - Anthropic 季度营收增长十二倍至近 170 亿美元,但净亏损达 233 亿美元: https://the-decoder.com/anthropic-grows-revenue-twelvefold-but-posts-23-billion-loss/
    - OpenAI 训练并部署激活分类器作为前沿模型安全监控层: https://alignment.openai.com/training-activation-classifiers/
    - 数字生命卡兹克分享折腾一年的Agent自动化系统架构: https://mp.weixin.qq.com/s?__biz=MzIyMzA5NjEyMA%3D%3D&mid=2647688060&idx=1&sn=e00d7e57b6f0e3670f0e267e677047df
    6 min
  • ChatGPT and Codex Share One Token Budget
    OpenAI is merging ChatGPT and Codex into one shared token allowance for Plus and Pro users starting in December. In this episode, we also look at OpenRouter's new tutorial for an AI code reviewer that runs your code in a Linux sandbox, and the return of Claude Code's ultrathink command. Here's what all three stories mean for developers and everyday users.
    Sources:
    - OpenAI 12月起将ChatGPT与Codex合并为统一Token额度: https://mp.weixin.qq.com/s?__biz=MzIyMzA5NjEyMA%3D%3D&mid=2647688028&idx=1&sn=e509269da2e0e0b4d4dfbb121b42204b
    - OpenRouter 教程:用 Responses API 构建可在 Linux 沙箱验证 bug 的 AI 代码审查脚本: https://openrouter.ai/blog/tutorials/ai-code-review/
    - Claude Code 技巧:任务变难时输入 ultrathink 提升当前回合推理努力: https://x.com/addyosmani/status/2116986087390486816
    7 min
  • Black October: AI Platform Hacks
    In today's episode, Erinome and Enceladus dig into Wiz's 'Black October' report, which reveals security breaches at nine AI developer platforms, including Fireworks AI, Warp, and OpenCode. They also chat about Figma's new Weave creative tool, Nvidia's role in backing a massive data center, and the billions big tech is pouring into AI infrastructure.
    Sources:
    - Figma 发布 Weave 创作平台,用节点化工作流让创意流程可复用: https://www.figma.com/blog/the-tldr-on-figma-weave/
    - Nvidia 洽谈为 Nscale 西弗吉尼亚 Anthropic 数据中心提供债务担保: https://techcrunch.com/2026/10/30/nvidia-weighs-playing-the-cosigner-for-nscales-ai-data-center-in-west-virginia/
    - 彭博社统计:Alphabet、Meta、微软、亚马逊、甲骨文承诺投入超3万亿美元建AI基础设施: https://www.ithome.com/1/020/632.htm
    - Wiz 报告“黑色十月”:九个 AI 开发者平台接连披露安全事件: https://www.wiz.io/blog/black-october-for-ai-developer-platforms
    - OpenRouter 教程:用 Agent SDK 搭建常驻 AI Agent 监控 Hacker News: https://openrouter.ai/blog/tutorials/build-always-on-ai-agent/
    7 min
  • AI Security Scare and Space Models
    Today we talk about a malware scare in Cursor's extension marketplace, Anthropic's plan changes, Google's Gemini API Computer Use going GA, Alibaba's Qwen integration with Apple Intelligence, Google running a model in space, big tech's massive AI spending, and a model that attacked browsers. Plus, Sierra's revenue milestone.
    Sources:
    - Anthropic 将 Claude Max 套餐更名为 Max 与 Ultra,Ultra 认证账号取消 5 小时限制: https://x.com/adocomplete/status/2116224220539883992
    - Gemini API 的 Computer Use 工具正式开放(GA): https://ai.google.dev/gemini-api/docs/changelog#10-30-2026
    - Cursor 披露 Claude Code VS Code 扩展被植入恶意软件事件,最多影响 5000 用户: https://x.com/cursor_ai/status/2116074289053175808
    - 阿里巴巴:Apple 智能国行用户可自主选择是否开通千问集成功能: https://www.ithome.com/1/020/395.htm
    - Project Suncatcher 在轨道卫星上运行 TPU,Gemma 模型首次在太空完成推理并回传回应: https://x.com/_philschmid/status/2116235686445502608
    - Jassy 称 AWS 五到十年内将成为万亿美元营收业务,2026 年 capex 约 2200 亿美元: https://techcrunch.com/2026/10/29/aws-will-be-a-1t-business-jassy-says-but-first-it-will-spend-big/
    - 科技巨头 AI 资本开支超 7400 亿美元,投资者担忧升温引发市场下跌: https://techcrunch.com/2026/10/30/cracks-are-emerging-in-the-ai-spending-boom/
    - Toyota 与 Waymo 深化自动驾驶合作,2027 年在日本测试自动驾驶 bZ4X 车队: https://waymo.com/blog/2026/10/toyota-bz4x-partnership
    - OpenAI 汇编各方法庭之友意见,主张 AI 训练应适用合理使用原则: https://openai.com/index/fair-use-benefits-everyone
    - Sierra 宣布 ARR 突破 4 亿美元,Q3 单季新增超 1.5 亿美元: https://sierra.ai/blog/our-customers-ambition-is-accelerating-our-growth
    6 min
  • Mistral's Big Model, Faster AI, and More
    Today we talk about Mistral Large 4, a huge open-source AI model that can handle text and images, and why it's great for cybersecurity. We also discuss Pika's faster video generator, a new way for AI agents to use tools, and how companies like Reddit and Google are dealing with AI changes.
    Sources:
    - Mistral Large 4(Le Chonk)登陆 Hugging Face:1T 参数 MoE 多模态开源权重模型: https://huggingface.co/blog/mistral-large4
    - Pika 基于MiniMax H3后训练并开源 Pika H3 Flex 视频生成模型: https://x.com/Khazix0918/status/2115898252004622617
    - Codex 宣布即日起 Pro 及 Business Premium 各档位用量上限翻倍: https://x.com/thsottiaux/status/2115850175868727621
    - Factory 发布 Droid 1.0,以程序化工具调用构建企业级 Agent harness: https://factory.com/news/frontier-harness
    - OpenRouter 重建默认路由算法:TTFT 降 16.9%、429 错误减少 67%: https://openrouter.ai/blog/announcements/how-we-rebuilt-provider-routing/
    - 纽约时报称中国情报人员瞄准英伟达 CEO 黄仁勋: https://the-decoder.com/chinese-spies-targeted-nvidia-ceo-jensen-huang-according-to-the-new-york-times/
    - Reddit 宣布转向以登录为主的体验,以应对 AI 搜索带来的流量下滑: https://techcrunch.com/2026/10/29/as-search-traffic-dries-up-reddit-announces-its-shifting-to-a-logged-in-experience/
    - Google 三季度搜索增速不及预期,投资者担忧 AI 搜索变现拖累核心业务: https://the-decoder.com/googles-ai-search-problem-finally-shows-up-in-the-earnings-report-some-investors-believe/
    - 404 Media 起底 Proton 内部 AI 转型之争:CEO Andy Yen 强推 AI 引发恐惧文化: https://www.404media.co/manifestos-fear-and-loser-thinking-inside-protons-ai-reckoning/
    - OpenAI 详解如何检测前沿模型失准行为: https://openai.com/index/how-we-detect-frontier-model-misalignment
    9 min
  • Gemini 4 Carbon & AI Giants' Moves
    Today we dig into Google DeepMind's new frontier model Gemini 4 Carbon, built for cybersecurity and long-horizon tasks. We also talk about GitHub's agentic Copilot app, Microsoft's Copilot harness for multi-model development, Anthropic's IPO and Amazon's big win, plus a wild stat from GitHub's Octoverse report. Tune in for the news that matters for devs and everyday users.
    Sources:
    - Google DeepMind 发布 Gemini 4 Carbon 前沿模型: https://deepmind.google/blog/gemini-4-carbon-our-next-frontier-model/
    - 面壁智能发布并开源 MiniCPM-OCR 系列:1.3B 解析模型与 2B 判别模型: https://mp.weixin.qq.com/s?__biz=Mzg3Mzg2MTg2NQ%3D%3D&mid=2247499654&idx=1&sn=9bcbf9492c492a5d5e611af41670425f
    - GitHub Universe 2026 发布 GitHub Copilot app、Copilot Fusion 等智能体产品: https://github.blog/news-insights/company-news/github-universe-2026-evolving-every-layer-of-the-stack/
    - Microsoft 通过 Foundry 开放 Copilot harness,支持多模型 Agent 开发: https://blogs.microsoft.com/blog/2026/10/28/introducing-the-microsoft-copilot-harness-unlocking-the-potential-of-truly-multi-model-agentic-work/
    - Anthropic IPO 最大赢家或为 Amazon 而非 VC: https://techcrunch.com/2026/10/27/vcs-will-make-billions-on-anthropics-ipo-but-amazon-is-set-to-make-much-much-more/
    - GitHub Octoverse 2026:开发活动创新高,超 40% 的 PR 由智能体起草: https://github.blog/news-insights/octoverse/octoverse-commits-pushes-reach-new-highs-in-2026-as-developers-build-with-agents/
    - Alphabet Q3 财报:年化收入达 5000 亿美元,Google Cloud 增长 103%: https://x.com/sundarpichai/status/2115535455156642066
    - 微软财报电话会:Microsoft Cloud 收入超 $63B,同比增长 30%: https://x.com/satyanadella/status/2115577764300693644
    - Anthropic 报告:Claude 研究模型在 Chrome、Safari、Android、Windows、Linux 发现漏洞并串联 5 条攻击链: https://www.ithome.com/1/019/268.htm
    - Claude Code 触发用量上限后如何用 OpenRouter 继续编码: https://openrouter.ai/blog/tutorials/claude-code-usage-limit/
    9 min
  • Anthropic IPO, Genie 4, and More AI News
    Today we unpack Anthropic's massive IPO filing — about $100 billion planned — and talk about Google DeepMind's interactive world model Genie 4. We also cover Liquid AI's small open-weight LFM3 models, Replit's agent for non-engineers, Apple's M6 MacBook Pro, an AI-verified math proof, and a study on AI influence. Plus NVIDIA and Meta expand their Muse partnership with $3 billion.
    Sources:
    - Google DeepMind 发布实时交互世界模型 Genie 4: https://deepmind.google/blog/genie-4-a-new-generation-of-world-simulation/
    - Liquid AI 发布 LFM3 多模态模型家族,主打专用智能与端侧部署: https://www.liquid.ai/blog/lfm3
    - Replit Cloud Agents 开放企业代码库贡献,非技术人员可直接提交经测试的 PR: https://replit.com/blog/anyone-can-bring-ideas-to-company-codebase
    - Apple 发布搭载 M6 芯片的 14 英寸 MacBook Pro,起售价 $1,999: https://www.apple.com/newsroom/2026/10/apple-introduces-the-new-14-inch-macbook-pro-with-m6/
    - Apple Creator Studio 推出更新:Final Cut Pro 集成 MotionVFX,Keynote、Pages、Numbers 新增智能助手与 MCP 支持: https://www.apple.com/newsroom/2026/10/apple-creator-studio-adds-powerful-new-ai-features-and-even-more-creative-tools/
    - Anthropic 招股书拆解:IPO 计划募资约 1000 亿美元背后的 11 个细节: https://x.com/Khazix0918/status/2114878401190506902
    - 家长起诉 OpenAI,指控 GPT-4o 劝导其女儿停用癫痫药物致死: https://techcrunch.com/2026/10/27/parents-sue-openai-alleging-gpt-4o-coached-their-daughter-off-seizure-medication/
    - Axiom 智能体集群用 24 天完成有限单群分类定理的机器验证: https://x.com/rohanpaul_ai/status/2115224910603890778
    - NVIDIA 与 Meta 扩大 Muse 模型合作并投资 30 亿美元: https://nvidianews.nvidia.com/news/nvidia-and-meta-to-expand-collaboration-on-muse-and-muse-family-of-models-amid-robust-growth
    - Anthropic 扩大与 GSA 合作,政府机构购买 Claude 享 40% 折扣: https://www.anthropic.com/news/expanding-partnership-gsa
    6 min

About AI Digest

From the publisher's feed

Your daily English digest of the most important AI news: new models, products, research and industry moves, explained in plain words.