I came up with a somewhat foolish new benchmark for testing image generation models, to exercise the new ChatGPT Images 2.0:
"Do a where's Waldo style image but it's where is the raccoon holding a ham radio"
https://t.co/KuPdFAEWUl
观点对做 AI 作图教程/套壳产品的人是警报——OpenAI 正在关闭「提示词玄学」的信息差生意,用户不再能看到 prompt,也就不再为 prompt 手册付费。你的内容如果还停留在「教写 prompt」,要尽快升级到「教编排 agent」。
Presumably GPT-imagegen-2 (aka ChatGPT Images 2.0 aka gpt-image-2) works as a tool which the models generate prompts for?
I wish we could see those prompts, like back in the DALL-E 3 days https://t.co/SnSNToY5bp
This is so confusing. Did Anthropic really just drop Claude Code from their $20/month plan?
Why would they do that through a pricing page update without making a proper announcement?
Plus, $20/month still gets you Cowork, which is just Claude Code wearing a non-threatening hat! https://t.co/2bzcRVbOyT
对你意味着使用 Claude Code 订阅策略有不确定性,建议保留替代方案(Codex/Cursor),不要深度绑定。
Wrote up Anthropic's self-own about Claude Code pricing from this afternoon on my blog - it turned out they'd reversed course just as I hit publish, so I've tried to update it to reflect the current state
https://t.co/uv6zTExj4q
GPT-5.5 可能尚未在官方 OpenAI API 中……但它可通过明显被认可的 Codex API「后门」访问。所以我用它生成了这些小球……(原文截断)
对你意味着
事实Codex API 目前是访问 GPT-5.5 的唯一编程接口,且得到 OpenAI 的默许。
观点这种"非正式但被许可"的渠道意味着 OpenAI 在有意控制访问节奏——你的工程团队若要抢先集成,需要通过 Codex 路径而非等待正式 API 开放,时间窗口可能有限。
GPT-5.5 may not be in the official OpenAI API... but it's available via the apparently approved-of Codex API backdoor So I used that to make these pel...
I've been previewing this in Codex for a few weeks - it's very good! Had some great results from it having it run security reviews against code writte...
Big new release of my LLM CLI tool and Python library for talking to hundreds of different LLMs - reasoning traces, OpenAI Responses support, server-s...
Turso is expanding beyond SQLite to become a foundation on which multiple database compatibility layers can be built - makes the project a whole lot m...
对你意味着,AI 团队需及时排查 CI/CD 静态检查管道与 AI Code Agent 的生成规范约束,防止基础工具升级引发大量构建中断。
Ruff 0.16.0 - @astral_sh's fast Python linter - came out a few days ago and increased the number of default-enabled rules from 59 to 413, which highli...
Huge relief, the Fablepocalypse has been permanently cancelled. If you were losing sleep over trying to use your Fable subscriber allowance before it ...
Wrote about today's GitLab restructuring / "workforce reduction" announcement, and ended up digging around in version control for both the GitLab and ...
对你意味着sqlite-utils 是 AI 工程师操作本地 SQLite 数据的高频工具,v4 将迁移功能内置后,原型到生产的路径更顺滑。
事实若你的 AI 产品以 SQLite 作为轻量存储层,这一版本值得升级评估,特别是正在做数据 schema 迭代的团队。
I just released the first release candidate for sqlite-utils v4, adding a migrations system (previously released independently as sqlite-migrate) and ...
让 Claude Fable 对即将发布的软件做最终审查,结果它发现(并修复了)整整 5 个发布阻断性问题——这让人有些汗颜。
对你意味着
事实Claude Fable 在发布前捕获了 5 个人工漏掉的发布阻断性问题,并完成修复;
观点这是「AI 成为工程质量守门人」的强烈信号。如果你的工程团队仍以人工 code review 为唯一安全网,引入 AI pre-release 审查的 ROI 值得认真评估——发布频率越高,人工审查越是瓶颈,而 AI 的边际成本几乎为零。
Somewhat humbling to have Claude Fable do a final review of some software that you're about to release and have it then find (and fix) FIVE release bl...
观点OpenAI 在 API 与产品端的能力释放节奏不同步,反映出其在开发者生态与消费者产品之间的优先级取舍——这既是 OpenAI 的产品策略,也是留给 API 优先型产品的竞争窗口。若你的团队正在评估语音 AI 能力,直接调用 API 可能比等待官方产品集成快上数月。
I got fed up of waiting for OpenAI to bring their much improved gpt-realtime-2 voice conversation model to the ChatGPT product, so I upgraded my OpenA...
事实Code Interpreter 三年前上线时被视为玩具,如今已是 AI 编程工具的标配基础能力。
观点今天被低估的 AI 功能演示,可能正是三年后商业化的起点——对 CEO 而言,保持对「还不够实用」的新工具的持续关注,是一种低成本的战略期权。
It's fun to look back at this Twitter conversation about the then-new ChatGPT Code Interpreter from three years ago - with hindsight this was our firs...
这是个真正的大新闻——Cloud Run 很容易被激进爬虫刷出吓人账单,有了支出上限后,在上面跑小项目就安全多了。
对你意味着
事实Google Cloud Run 加入支出上限,对应 AWS 目前仍没有。
观点对一人公司很实用:如果你有任何线上服务(采集脚本/小程序后端),把它迁到 Cloud Run 能拿到「真正的账单兜底」,这是心理层面的解锁——敢把项目曝光到公网了。
This is a really big deal - it's easy to run into nasty bills with Cloud Run if your site attracts aggressive scrapers, spending caps make it a whole lot safer to run small projects on
Sent out my sponsors-only monthly newsletter for June, which means the May newsletter is now available here https://github.com/simonw/monthly-newslett...
The new stateless MCP specification has rekindled my interest in MCP, and inspired some new projects, including mcp-explorer and datasette-mcp https:/...
A new TIL on adding custom MCP servers to both the ChatGPT and Claude regular chat interfaces - it's a little less obvious than I had hoped, but I got...
I think loops were a short-lived patch for models that couldn't reliably keep working on long problems until they hit a defined goal Fable and GPT-5.6...
Simon Willison 发布关于 Kimi K3 的笔记,并分析当基准测试脱离真实体验时 Pelican Benchmark 的剩余参考价值。
事实Pelican 基准测试正逐渐脱离真实模型能力表现。
观点公开 Benchmark 的指导价值下降。
对你意味着不能仅靠公开基准选型模型,必须建立针对自身业务场景的私有闭环测试集。
My notes on Kimi K3, plus some thoughts on what we can still learn from the pelican benchmark even while it becomes further detached from how good the...
我仍然有时会看到人们说“如果你知道如何编写代码,自己写会更快”。我持完全相反的观点:如果你知道如何[编写,使用 AI 辅助会更快,因为你能立即验证、调试并掌控全局]。
对你意味着
事实Simon Willison 指出懂编程的开发者使用 AI 辅助比自己写更快。
观点隐含信号是,AI 编程工具的最大受益者是高水平开发者,而非初学者。CEO 在推进研发工具落地时,应着力提升核心骨干的 AI 使用率,将其视为杠杆而非低端岗位的替代品。
I still sometimes see people saying "if you know how to write the code, it's faster to write it yourself" I'd argue the exact opposite: if you know ho...
One of the things the pelican benchmark is still useful for is visually representing (to a tiny extent) the improvements in a single model family Here...
观点对依赖社交媒体做用户反馈采集的 AI 产品来说,这是数据质量风险的预警——你的社区讨论正在被 Bot 稀释,社区运营策略需要针对性调整。
I absolutely hate how I'm getting to be suspicious of ANY reply to my posts here, especially ones that pose a question, as they are so often from bot ...
Here's my attempt at summarizing the various "open letters" about AI development that have been doing the rounds over the past few weeks https://simon...
对你意味着大模型推理成本的下降速度将超出预期,且使用 AI 自行重构与优化基础设施代码将成为企业降本增效的关键杠杆。
GPT-5.6 found optimizations that "reduced end-to-end serving costs by 20%" for OpenAI to serve that model Presumably that's billions of dollars a mont...
事实任何重度依赖 OpenAI 基础设施的业务都应该有 Plan B——特别是 embedding、fine-tune 这类迁移成本高的产品。你做知识付费如果重度用 embedding,提前准备替代方案。
Not the first time either - they shut down a bunch of of their original proprietary hosted embedding models in this announcement back in April 2024 https://t.co/v764AHwXom
对你意味着CEO 应密切关注端侧推理的成本与隐私优势,评估将部分 AI 工作流下沉至本地运行的技术路线与产品竞争力。
I try not to get excited about models before they've been released, but I gotta admit I'm very much looking forward to the upcoming laptop-sized Qwen ...
Really looking forward to one of the super-fast custom silicon inference providers like @GroqInc or @cerebras getting GLM 5.2 running Cerebras has GLM...
我为自己的博客(完全在手机上用 Claude Code for web 构建)添加了一个新功能——导入我的 iNaturalist 照片,并将其接入网站的观鸟记录系统……
对你意味着Simon Willison 是 Django 联合创始人,他的技术判断具有高可信度。
事实他在手机上用 Claude Code 完成了跨 API 功能集成,而非简单脚本。
观点这意味着 AI 编程能力正在打破桌面专属限制,移动端也可以成为有效开发界面。对希望降低工程门槛的 CEO,这是 AI 编程工具普及化的可靠信号。
I added a new feature to my blog (built entirely on my phone with Claude code for web) that imports my iNaturalist photos and adds them to my site's o...
我的 Mac 可用内存比预期少,原来是这台机器上在各个终端窗口运行的「claude」Claude Code 进程……
对你意味着多窗口并行运行 Claude Code 的内存开销不可忽视。
事实这是来自重度用户的真实使用数据点,不是猜测。
观点如果你的工程师在生产力机器上并行跑多个 Claude Code 会话,建议监控内存使用情况,避免影响主机上的其他关键服务。
My Mac had less available memory than I expected, turned out the "claude" Claude Code processes on this machine (running in various terminal windows) ...
I'm on Oxide and Friends podcast this week! We talked about accidental cyberattacks, Kimi K3, Golden Gate Claude, the Zizians, Alameda wild turkey att...
对你意味着研发团队的岗位边界面临重塑。具备极强边界思维与测试严谨度的 QA 人员,在 AI 赋能下可能比传统工程师具备更高的交付质量与产品构建效率,值得在组织设计中重新评估其角色定位。
I hope there are QA testing experts out there who are thinking "finally, we don't need software developers any more!", rolling up their sleeves and bu...
观点创始人 IP 和增长杠杆对早期 AI 创业团队至关重要,建立高效的社交平台互动策略能够放大个人影响力并转化商业声量,但需防范低质回复对个人品牌的负面效果。
Does anyone know if there's common advice out there about growing your Twitter followers by replying to lots of other people's posts? I'm trying to fi...
RT Ya'el Courtney, PhD: ahhhhhhh so I'm hiring a tech to help me out in lab and screening applications now and THE KIDS ARE USING PROMPT INJECTION!!!!...
ChatGPT Sites means ChatGPT in "Work" mode can build and deploy public websites running on Cloudflare Workers, including with persistence on top of SQ...
I wrote about the completely wild incident where OpenAI were testing a new model and it broke out of its sandbox and broke INTO Hugging Face to steal ...
RT Riley Goodside: Claude Fable 5 draws a pelican on a bicycle, but the pelican and its bike are a large stone sculpture in the Backrooms between two ...
New TIL: Using uvx in GitHub Actions in a cache-friendly way I finally found a recipe that I like for running `uvx tool-name` in GitHub Actions withou...
It's been about six months since OpenClaw burst onto the scene - are you still using yours? Did it become a daily driver? Any interesting lessons or a...
It's annoying that you can't paste a link to a (shared) Claude transcript into a Claude Code session, because Anthropic's anti-scraping measure preven...
The idea of "AI employees" feels so short-sighted to me - both disrespectful to humans and a complete misunderstanding of what these tools can do and ...
One of the most confusing aspects of GPT-5.6 is figuring out which model to use at which reasoning effort - sounds like Sol on Medium might be a good ...
With Atlas being retired in favor of the browser embedded in the ChatGPT app I wonder if the whole category of AI-enhanced browsers is coming to a clo...
"We’re bringing chats and projects back into the sidebar in a more familiar and customizable way" - hopefully that means the ChatGPT app won't hide c...
RT Thariq: Re @_catwu @simonw and I will be doing a fireside chat about "This year in Claude" from 12:30pm-1:30pm at AIE in Expo Stage 2. We'll be cov...
RT Anthropic: We’ve received notice that the Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5. We'll begin restoring ...
Notes (and a Pelican) on Claude Sonnet 5 - the new tokenizer makes it ~1.4x more expensive for English, ~1.33x more expensive for Spanish but roughly ...
Also not a great advertisement for a sandboxing product: @daytonaio effectively saying they don't trust the security of their product enough to expose...
RT Pamela Fox: I printed a custom t-shirt that's an ode to @simonw's Pelican benchmark. My partner says he doesn't get it. But y'all get it, right? RI...
I'm just glad nobody at the US government thought to try that Fable 5 "jailbreak" against Opus 4.x or GPT 5.x, or I wouldn't be getting anything usefu...
It's now possible to compile Python extensions (C, C++, Rust etc) to WebAssembly and distribute them through PyPI such that Pyodide can install them d...
RT antirez: Re I believe what Anthropic is doing, gating the ability to do certain harmless things like LLM research, and with incredibly sensitive fi...
A TIL on using http://agentsview.io to calculate token spending with Claude Fable 5 despite that model not yet being included in the AgentsView pricin...
RT Ethan Mollick: I've had access to Fable for a bit. A genuine jump in capability, I could feed it a 15 page design document for a project and it wou...
Uber reportedly now caps coding agents at $1,500/month per employee per tool - seems sensible to me, but it's also an interesting hint at the value Ub...
RT Philipp Spiess: Re @simonw Sorry about this! The button was unintentionally removed during some recent menu restructurings. It'll be back with the ...
Anthropic's self-reported run-rate revenue growth is wild - Axios @JimVandeHei said he could not find "any company — in any industry, in any era — t...
Notes on Claude Opus 4.8, plus pelicans riding bicycles for each of the five different thinking efforts https://simonwillison.net/2026/May/28/claude-o...
Given the recent burst of activity around enterprise pricing and contracts, I think April 2026 was the month when both OpenAI and Anthropic found prod...
Given how much of the original "bottle of water per generated email" water estimate came from guesses at the architecture of GPT-4, it would be very m...
RT Paul Graham: Re I have never knowingly finished reading an email signed by a human but written by AI. It feels like being lied to, and who would st...
I don't have much to say about this year's Google I/O because I prefer to write about products that have shipped, not just "coming soon" announcements...
RT Nathan Clark: it’s in gemini, just create it in ai studio. oh, that’s for your personal google one account. for workspace you need gemini busines...
My notes on Gemini 3.5 Flash - 3x the price of Gemini 3 Flash but Google are planning to use it for many of their own products https://simonwillison.n...
Also a great example of positive contribution to open source by wanderingmeow - you don't need to contribute code to have a positive impact, just prov...
Shopify's River agent system lives in Slack and can only be used in public so that other employees can learn from what you do with it Reminds me of ho...
Asking for HTML explanations of things is pretty neat, I tried it just now with the obfuscated Python POC for the new http://copy.fail Linux vulnerabi...
We already had gemini-3.1-flash-lite-preview back on March 3rd, not clear if this new gemini-3.1-flash-lite is different other than no longer being ma...
I was talking with @josephruscio on the @heavybit podcast the other day when I realized that vibe coding and agentic engineering have started to blur ...
AI-run business experiments are interesting and fun up to the point where they waste the time of humans who haven't opted into the experiments - I thi...