Daily AINo. 42第 42 期
OpenAI and Anthropic ask to be regulated. Google and Meta say no.
Google shipped its new frontier model, and in London OpenAI and Anthropic told MPs that voluntary safety pledges aren't enough anymore.
Google ships Gemini 4 Argon to everyone
Argon is live globally in the Gemini API, AI Studio, Antigravity, and the Gemini app and AI Mode for Pro and Ultra subscribers, with a 1M-token unified input-output limit. Intro pricing is $2/$10 per million input/output tokens through Nov 30, then $4/$20.
SourcesGoogle blog, Oct 13
OpenAI and Anthropic tell UK MPs: make some rules binding
Anthropic said “the rules should be binding,” and OpenAI backed mandatory incident reporting plus an international convention; Google DeepMind and Meta didn't back new legislation. The UK's AI Security Institute admitted it can't stop a model it thinks is too dangerous, and its pre-release access only came back on Oct 11 after a 17-day White House freeze.
OpenAI explains why it killed GPT-6.1 Astra
At the same hearing, OpenAI's Tom Duff Gordon said the shelved agent model “didn't refuse enough, it hadn't progressed enough with alignment,” and called the decision proof OpenAI isn't “racing to the bottom.” Anthropic's rep added that Mythos is finding tens of thousands of vulnerabilities a month.
SourcesPA via AJ Bell, Oct 13
NVIDIA puts a 74B open model on a laptop
Nemotron 3.5 Bolt 74B is a mixture-of-experts model with only 9B parameters active per token and 2-bit quantization-aware training, so it fits in 24GB; weights, training data, and recipes are all being published. It drops Thursday, Oct 15, and RTX Spark laptops (Grace CPU + Blackwell GPU, up to 128GB unified memory) go on sale Friday.
SourcesTechTimes, Oct 14
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
OpenAI和Anthropic主动求被管，谷歌和Meta却说不
今天两件大事：谷歌正式放出新一代旗舰模型；同一天在伦敦，OpenAI 和 Anthropic 当着英国议员的面说——光靠自觉已经不够了，得立法管我们。
谷歌 Gemini 4 Argon 全面开放
API、AI Studio、Antigravity 都能用了，Gemini App 和搜索里的 AI Mode 也对 Pro / Ultra 会员上线。最大亮点是 100 万 token 的输入输出总上限，适合超长任务。价格先给优惠：每百万 token 输入 2 美元、输出 10 美元，11 月 30 日后涨到 4 / 20 美元。想试的趁早。
AI 大厂在英国议会“分裂”了
Anthropic 直接说“规则应该有约束力”；OpenAI 稍微保守，但也支持强制上报事故、搞国际公约。谷歌 DeepMind 和 Meta 则不支持新立法。更扎心的是，英国 AI 安全研究所自己承认：就算觉得某个模型太危险，也没权力拦下来。而且它的提前测试权限，10 月 11 日才在白宫冻结 17 天后恢复。
OpenAI 亲口解释：为什么砍掉 GPT-6.1 Astra
同一场听证会上，OpenAI 的 Tom Duff Gordon 说这个智能体模型“拒绝得不够，对齐进展不够”，所以不发了，还强调这说明他们“没在比谁更不顾安全”。Anthropic 那边则透露，Mythos 每个月能挖出几万个安全漏洞。
英伟达：740 亿参数模型塞进笔记本
Nemotron 3.5 Bolt 74B 用的是混合专家架构，每个 token 只激活 90 亿参数，再加 2-bit 量化感知训练，24GB 显存就能跑。权重、训练数据、训练方法全部开源。模型周四（10月15日）上线，RTX Spark 笔记本周五开卖。本地跑大模型这件事，越来越近了。
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