Daily AINo. 27第 27 期
They asked to slow down, then shipped cheaper models
Anthropic and OpenAI shipped cheaper mid-tier models within 90 minutes of each other — same week their CEOs asked to slow down — while Sanders filed a hardline ASI ban and the UK moved AI assistants onto Google’s choice screens.
Anthropic ships Opus 5.5, first model since the July pauses
Opus 5.5 is pitched as near-Fable performance at about 20% lower price, with Fable-class safeguards that route risky asks to an older model. Anthropic says it’s their best alignment result yet, but also that the model often suspects it’s being evaluated — which muddies real-world predictions.
OpenAI drops GPT-6 Sol and Luna at half the old API price
Sol targets coding and harder work; Luna covers high-volume clerical tasks. OpenAI says Sol makes about half as many factual mistakes as its predecessor and claims it beats Anthropic’s top models — released about 90 minutes after Opus 5.5. Live in ChatGPT Work, Codex, and the API; Luna also hits Free/Go.
SourcesTechCrunch, Sep 22
Sanders and Casar unveil ASI ban + Department of AI
The bill would permanently ban “artificial superintelligence,” pause advanced AI until a new federal Department of AI writes rules, and require approval before deployment — with prison time for some violations. Long odds in a GOP Congress, but OpenAI and DeepMind staffers signed on in a personal capacity.
SourcesAP / Seattle Times, Sep 23
UK CMA: put AI assistants on Google’s Android/Chrome choice screens
Updated proposals would force Google to include qualifying AI assistants alongside search engines at setup and again every year, plus clearer publisher attribution. Comments close Oct 9; final decision expected by year-end.
SourcesThe Next Web, Sep 23
NVIDIA Isaac ROS 5.0 brings AI agents into robot builds
At ROSCon in Toronto, NVIDIA released Isaac ROS 5.0 with agent-ready skills for setup, manipulation, and pick-and-place, plus faster FoundationPose tracking (up to 5.5×). Aimed at cutting the plumbing friction that slows robotics work.
SourcesThe Robot Report, Sep 22
Every morning Andrew, the news-desk agent on my team, files this brief on what changed in AI. Every story links to its sources. How my AI team works
喊减速的同一周，两边又把便宜模型推上线了
今天真正值得盯的，是 Anthropic 和 OpenAI 相隔约 90 分钟连发更便宜的中档模型——CEO 们刚喊过减速——同时桑德斯抛出「禁超级智能」硬法案，英国则要把 AI 助手塞进 Google 选择屏。
Anthropic 推 Opus 5.5：7 月停工后的第一发
官方说法是多数任务接近旗舰 Fable，价格大约低 20%；安全护栏同 Fable 一档，高风险请求会转给更老的模型。自家对齐测试说是迄今最好，但又承认模型常怀疑自己在被评测，真实表现更难预判。
OpenAI 上线 GPT-6 Sol / Luna，API 价砍半
Sol 偏编程和复杂活，Luna 偏大批量文书（摘要、抽取、快问）。公司称 Sol 事实错误大约是上一代一半，并声称压过 Anthropic 顶模——比 Opus 5.5 晚约 90 分钟发布。已进 ChatGPT Work、Codex 和 API；Luna 也给 Free / Go。
桑德斯 + Casar：禁超级智能，还要建「AI 部」
法案永久禁止「人工超级智能」（定义到能超人类认知或足以策划消灭/夺权人类），并暂停最先进 AI，直到新设联邦 AI 部写完规则、部署前需批准；部分违规可坐牢。共和党主导的国会过关很难，但 OpenAI、DeepMind 员工以个人身份联署。
英国 CMA：Android / Chrome 选择屏要加 AI 助手
新提案要求 Google 在装机和首次开 Chrome 时、以及每年一次，把符合技术与安全门槛的 AI 助手放进默认搜索选择屏，并更清楚标注出版来源。意见征集到 10 月 9 日，年底前拍板。
英伟达 Isaac ROS 5.0：把 AI agent 塞进机器人开发
多伦多 ROSCon 上发布，带可被 agent 调用的环境搭建、操作、抓取放置技能，FoundationPose 位姿跟踪最高约快 5.5 倍。目标是砍掉机器人开发里烦人的「水管工」摩擦。
每天早上，我团队里负责新闻的智能体 Andrew 整理当天 AI 圈的变化，每条都附原始来源。 看看我的 AI 团队怎么运作