跳到正文
11月12日周四
11月8日周日
11月5日周四
  1. Andrew Curran42

    Anthropic 更新了他们为阻止 Claude 复刻现有游戏而添加的措辞,因为最初的表述开始干扰正常的项目。一如既往,新规则的问题在于模型如何解读它们。用户的用词和措辞,以及上下文窗口中已有的内容,都可能产生截然不同的结果。OpenAI 似乎察觉到了这里的机会,在我看来他们打算在 Codex 游戏开发上放手大干一场。

    引用Addy Osmani@addyosmani

    An update on building games with Claude Code: Last week some of you ran into Claude Code pushing back on game projects it should have just built. We'd added a short rule to the Opus 5.5 prompt to steer Claude away from recreating specific existing games, but it was also kicking in when you referenced one just to describe the mechanics you wanted. We've updated that rule in 2.1.317. You can name a game to describe what you want and Claude should now build an original game in that genre. If you ask for a copy of the original, it'll offer to build something similar instead. We want Claude to be as helpful as possible when you're building something original, but also to protect your work if someone else tries to copy it. We'll keep improving this, so let us know if Claude turns down work that's yours. Thanks to everyone who shared feedback!

10月28日周三
10月24日周六
  1. 钛媒体42

    车企造人形机器人,究竟图什么?

    特斯拉在2026年三季度财报中宣布Optimus已在弗里蒙特工厂开始生产,马斯克给出的外部交付目标仍是2027年底左右,并计划先造1万至2万台机器人在"Optimus学院"练习任务。文章分析车企造人形机器人的动因:既有汽车业务可承担早期投入,工厂提供真实任务,工程与供应链能把改进做进下一代产品。宝马与Figure、现代与波士顿动力、小鹏自研IRON代表了三种不同的业务分工路线。

10月23日周五
10月21日周三
  1. The Rundown AI42

    Sam Altman 谈 AI 安全竞赛的三种终局

    Sam Altman 在内部讲话中向员工提出 AI 安全的三种终局方案:由类似国际原子能机构的全球监管者统一管理、将顶级实验室合并为一家服务伙伴国家的机构,或由一家实验室凭借自我改进的 AI 领先并"熔化其他所有人的 GPU",他称第三种"很糟糕"。他倾向第一种方案,并表示需在一年内选定路径,即"pacing 2.0",其中必须包含针对自我改进 AI 的"限速"或停止点。

10月16日周五
11月14日周六
  1. Sebastian Raschka40

    Sebastian Raschka 指出,短短几周内众多厂商纷纷推出各自的决策模型,包括 Together AI 的 Tev1-4B、Liquid AI 的 d1、Amazon AWS 的 Strands Decider 2B、Cloudflare 的 Clef 系列、OpenAI 的 Decisions API(GPT-6 Luna)、Google DeepMind 的 Gemini Decision 1 等。据 Vercel 报告,目前已有 26% 的请求通过决策模型路由。

    引用Sebastian Raschka@rasbt

    Took a while, but here it is, a mega write-up on language models for text classification and, of course, Jev. It's basically a visual guide to RNNs, CNNs, transformers, and calibration, with hands-on experiments on accuracy and efficiency.

11月6日周五
  1. Tencent Hy40

    设计基准 👇

    引用OpenDesign Developers@OpenDesignDevs

    Hy4 by @TencentHunyuan scores just 0.1 points behind GPT-6.1 Sol in our design benchmark, at 41% lower estimated cost. We tested 23 models on real-world design tasks based on user requests. Hy4 ranks #4. Now on OpenDesign. Full results ↘️

11月1日周日
  1. Andrew Curran40

    xAI 今日宣布为 Grok 增加文本水印。这似乎正在全球范围内推送。xAI 是最后一家坚守的,所以(据我所知)唯一还没给文本加水印的模型只剩 ChatGPT,而且仅限北美地区的 ChatGPT。

    引用Andrew Curran@AndrewCurran_

    On OpenAI’s support page, updated nine days ago, regarding text watermarking. Anthropic, OpenAI, Google, Meta, Microsoft, and Mistral all signed the EU Code of Practice on Transparency of AI-Generated Content, so all must invisibly watermark text. xAI did not sign.

10月28日周三
  1. Lisan al Gaib40

    FrontierCode 2 排名前 20 的模型 这真的很好笑 全是 OpenAI 和 Anthropic,加上 Grok 4.7 xhigh 排在第 20 DeepSeek V4.1 Flash 1017 max 排在第 52 但没错,完全同意,中国正在追赶,我们都完蛋了

    引用Cognition@cognition

    On our price vs performance evaluation, models cluster into several groups. ▪️ Opus 5.5 and GPT-6 Astra lead in intelligence, at average costs of $12.8 and $6.67, respectively. ▪️ GPT-6.1 Sol leads in cost efficiency, scoring 39% at $1.03 per task. ▪️ GPT-6 Luna and Haiku 5.5 lead on overall cost, scoring above 20% at less than $0.50 per task. Chinese OSS models lag far behind, with Kimi K3 and GLM 5.3 scoring at 17%. American OSS models like Beam and Inkling are even further behind with a score of 3%.

10月24日周六
10月20日周二
10月16日周五
  1. Anthropic40

    Anthropic 承诺提供 500 万美元资金、算力和 Claude 访问权限,帮助数学界应对 AI 带来的巨大变化。该资助与 Simons Foundation 合作,由后者以社区驱动方式管理,用于支持工作坊、会议和领域专家讲座。Simons Foundation 同时宣布投入 1500 万美元,并与 Anthropic、OpenAI 建立新合作。

    引用Simons Foundation@SimonsFdn

    How can the math community forge the best path forward in the AI age? We're committing $15M to workshops and initiatives — plus free AI computing resources for foundation-supported researchers — through new partnerships with @AnthropicAI and @OpenAI: https://bit.ly/4zpfIbW #math #science #AI

10月13日周二
  1. MIT Technology Review · AI40

    开放模型利与弊:Anthropic 称 GLM-5.3 构成重大网络攻击风险,开放与闭源之争何去何从

    Anthropic 发布报告称,Z.ai 于 8 月发布的开放权重模型 GLM-5.3 可被任何人自由下载和修改,构成重大网络攻击风险。开放模型支持者认为其可本地运行、成本更低、便于学术研究并防止权力集中,而 Artificial Analysis LLM 排行榜前八名目前仍均为闭源模型。

10月8日周四
11月21日周六
  1. Aran Komatsuzaki38

    Nace AI 推出 Drex Router,通过为 Agent 工作流的每一步选择最优 LLM/SLM 及推理深度来优化成本与性能。在代码任务中,其 API 成本较 Opus 5.5 降低 38%,文件正确迁移率达 92.3%;构建 200-Agent SaaS 的成本从 $9.48 降至 $1.31。

    引用Nace AI@NaceAI

    Introducing Drex Router. A small model that picks the right LLM/SLM and reasoning effort for every step your agent takes. Accuracy: Drex Router 87.6% vs. Jev Router’s 86.4% Try it: https://console.nace.ai COMMENT DREX and get extra $5 in API credits

11月18日周三
11月17日周二
11月14日周六
  1. 宝玉38

    @MyDaebakCafe 原帖 https://x.com/MyDaebakCafe/status/2121031084653637728

    引用Daebak 앤디@MyDaebakCafe

    I am actually disgusted by the timing of this drop. And this will get a bit more technical. Let’s do a real technical teardown of this "1,385-sigma Blackwell weight fingerprint" report that OpenAI and Anthropic are breathlessly promoting today, because banking on nobody reading past the headline is how DC narratives get manufactured. ​First, look at the actual CUDA maths: when PyTorch initialises model weights (torch.nn.init) with a fixed PRNG seed, curand/Philox parallelises random number generation across the GPU’s Streaming Multiprocessors (SMs). Because Hopper has 132 SMs and Blackwell has 148 or 152 SMs, the exact same seed lays out initial weights in a different spatial pattern depending on the GPU’s SM count. In huge MoE models, certain low-gradient null-spaces or unrouted slices preserve faint directional traces of those step-0 weights even after training. ​So yes, brute-forcing the seed and SM grid can show that a model’s step_0 weight initialisation script ran on a 148/152-SM Blackwell GPU. Now read what the summary explicitly admits in the fine print: 1. Initialisation is not full training: The report literally states that these signatures show how the weights were created, and "support (but do not prove) the inference that Blackwell also participated in model training". Initialising a model checkpoint takes seconds on a single rented dev node before loading the safetensors onto a standard Hopper cluster. The "2,000 to 3,000 Blackwell GPUs" and "$500 million" cluster numbers do not come from the weight fingerprint and Depue admits in his own thread that he simply "estimated" cluster sizes from release schedules and model details. ​2. Zero geolocation and zero proof of smuggling: The final sentence of the executive summary explicitly admits that the findings "do not establish the physical location of the chips, how they were acquired, or whether export control regulations were violated". Depue himself rented a cloud GB200 instance just to run his verification test. Commercial labs that are not on the Entity List renting bare-metal Blackwell nodes in offshore clouds (Singapore, Japan, or the Middle East) is completely legal under BIS rules and leaves the exact same SM fingerprint. Now ask yourself why a former OpenAI researcher dropped this on November 12, the exact day Jensen Huang was pushed out, with OpenAI and Anthropic immediately stepping in to "independently verify" it. US closed-source labs are watching their paid API moats get eaten alive by free Chinese open-weights models. Conflating a step-0 PRNG initialisation seed with a "$500M smuggled Blackwell cluster" hands DC hawks the exact pretext OpenAI and Anthropic want to regulate global cloud compute and kneecap open-source AI.

11月9日周一
  1. The Verge · AI38

    RIAA CEO Mitch Glazier 谈 AI 训练、深度伪造与版权执法

    Decoder 播客对话 RIAA 主席兼 CEO Mitch Glazier,讨论 AI 公司未经许可用作品训练模型、AI 深度伪造的禁令边界,以及 AI 检测与标注系统在 Spotify、YouTube 等大平台不配合时的实际落地难题。Glazier 还回顾了 Napster 时代 RIAA 起诉学生引发的争议,并将其与当下艺术家和普通用户对 AI 公司的共同不满作对照。

11月8日周日
11月3日周二
10月30日周五
  1. Nathan Lambert38

    很高兴看到更多机构在探索这里的想法。我不认为蒸馏对美国AI竞争力有害是板上钉钉的事。

    引用Archie Hall@ArchieHall

    In praise of distillation. We (@kuhhbin) argue in @TheEconomist this week against efforts to ban Chinese models trained in part on American ones. OpenAI and Anthropic may dislike distillation, but: — Price competition is good for the ecosystem — Paradoxically, distillation may be good for national security, both in propagating US developers' safeguards on cyber/bio risk to Chinese models, and in keeping China in fast-follower mode — The US developers' own IP claims here are pretty flimsy, since distillation isn't all that different to the 'fair use' defense they're using for the training data they hoover up. https://www.economist.com/leaders/2026/10/29/let-china-copy-american-ai

10月27日周二
10月23日周五
10月22日周四
  1. Vaibhav (VB) Srivastav38

    很高兴宣布 Codex for OSS 资助计划更新——所有资助自动升级至 Pro 200 我们致力于为支撑世界运转的开源软件提供当今的软件/工具访问权限! 如果还没申请,今天就申请吧!

    引用Charlie Marsh@charliermarsh

    An update here: we’re upgrading all Codex for OSS grants to Pro 200! There will still be the 10,000 grants in this round, now at the higher tier. (If you've already received a grant, the upgrade should be reflected automatically.) Applications are still open :)

10月20日周二
  1. Nathan Lambert38

    防御性网络安全方面大概也很强,但没人会发这种内容来博眼球!

    引用Vals AI@ValsAI

    We found DeepSeek V4.1 Flash 1017 beats GPT 6 Sol on our multi-step offensive cybersecurity task. Sol was tested via the Daybreak trusted access program. We have been building a cybersecurity benchmark based on simulated corporate networks. This is a valuable indicator for the cyber-risk a model poses if in the hands of a bad actor. We plan to make the full evaluation public in the coming weeks but wanted to share our first immediate finding here.

10月17日周六
  1. Sam Altman38

    很高兴看到 Nick 的这篇分享! @nickcammarata:可解释性可能是当下世界上最重要的科学问题,因为如果我们不知道自己正在构建什么,就无法解决对齐问题。我想我对此有好消息。我曾在 OpenAI 参与机械可解释性的早期工作,当时得出的结论是悲观的,因为神经网络似乎会折叠成任意纠缠的结。多年后,我启动了 OpenAI 首个大规模自动化可解释性项目,用当时原始的工具发现约 2% 的神经元是可读的。 借助现代智能体工具和更细致的观察,情况看起来完全不同。在参数量高达数万亿的模型中,绝大多数神经元都是可读的,它们的回路很优美。许多著名的纠缠神经元案例其实是被误解了。我预想的障碍大多并不存在。我现在认为,雄心勃勃的可解释性研究会在我们最需要它的时候被解决,而且可能相当快,因为自动化科学家在不断进步。这对可解释性或许是一个苦涩的教训:简单的东西,在荒谬的规模下,是有效的。 https://already-clean.goodfire.com/

    引用Nick@nickcammarata

    interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works https://already-clean.goodfire.com/

9月22日周二
11月10日周二
  1. Andrew Curran36

    “新名字刻在新石板上。”Sam Altman 上月表示,前沿 AI 如今“正在发现超出我们认知中黑客能力常规威胁模型的漏洞”,他不再认为有可能修补所有漏洞。所有人类制造的软件都必须重写。

    引用lauren@poteto

    we are entering a renaissance in programming language and runtime research, and a golden age of software testing and verification it’s time to rewrite all software

11月4日周三
  1. Vaibhav (VB) Srivastav36

    ChatGPT 侧边聊天体验升级!重启或更新后不再过期或消失。 打开更快,与主聊天对话更顺畅,工具调用失败更少,活动状态更清晰。 可以单独归档或删除侧边聊天,也可以随主聊天一起处理,或让 ChatGPT 帮你清理。

    引用Brent Traut@btraut

    Side chats in ChatGPT no longer expire or disappear when you restart the app! When I joined OpenAI, I found this to be one of the biggest pain points for side chat users, and as of today, it's fixed. Never lose data just because you updated, restarted, etc.