2026 年 LLM 发展回顾（截至目前）

我将快速回顾一下 2026 年迄今发生的一切。今年还没结束呢！I’m going to give a lightning tour of everything that has happened so far in 2026. The year isn’t over yet!

对我来说，2026 年早在 2025 年 11 月，也就是提前几个月就开始了。For me, 2026 started a couple of months earlier in November 2025.

11月发布了两个重要模型：Claude Opus 4.5 和 GPT-5.1。November saw the release of two important models: Claude Opus 4.5 and GPT-5.1.

与新模型通常的情况一样，这些模型是在前代模型基础上的渐进式改进。As is usually the case with new models, these were incremental improvements on the models that came before them.

但偶尔，随着模型不断改进，它会跨过一道看不见的界线，让原本不太奏效的东西开始奏效。But every now and then when a model improves, it crosses an invisible line where something that didn’t really work starts working.

在这个案例中，开始真正发挥作用的是他们的编程智能体。Claude Code 自 2025 年 2 月起就已存在，Codex 则晚一些。In this case, the thing that started working was their coding agents. Claude Code had been around since February 2025, Codex was a little younger.

这两个新模型与各自的编程代理框架搭配使用后，表现从“经常出错”提升到了“足以在日常工作中可靠使用”。These two new models, when paired with their respective coding agent harnesses, improved from “often make mistakes” to “reliable enough to use on a day-to-day basis”.

几年来，我一直通过让新模型“生成一张鹈鹕骑自行车的 SVG 图”来评估它们。这大概是世界上最愚蠢的基准测试——从中能学到的东西也就那么多。For a couple of years now I’ve been evaluating new models by asking them to “Generate an SVG of a pelican riding a bicycle”. It’s probably the world’s stupidest benchmark—there’s only so much you can learn from it.

但这对模型来说仍然是个挑战，因为画鹈鹕很难，画自行车也很难，而且鹈鹕本来就不会骑自行车。But it’s still a challenge for models, because drawing pelicans is difficult, drawing bicycles is difficult, and pelicans can’t ride bicycles in the first place.

这是截至11月的最新水平。Claude 还是画不好自行车！GPT-5.1 画的自行车车架也挺糟糕的。Here’s the state of the art for November. Claude still couldn’t really draw a bicycle! The GPT-5.1 bicycle frame is pretty crap too.

同样是在 11 月，我们向一个名为“Warelay”的冷门 GitHub 仓库提交了第一个 commit。我们稍后会再谈到这个仓库。Also in November, we had the first commit to an obscure GitHub repository called “Warelay”. We’ll come back to this repository shortly.

接着就是十二月的假期，个体开发者们休息了一段时间，许多人开始尝试这些新的编程智能体与模型的组合……我们也渐渐意识到，它们能做到的事情比以前多得多。An then there were the December holidays, and individual developers took some time off and many started tinkering with these new coding agent model combinations... and it began to dawn on us quite how much they could do that they couldn’t do before.

到了1月，我们很多人都很兴奋，准备开始把这些东西付诸实践。Come January, a lot of us were quite excited to start putting this stuff into action.

每年我都会给自己定一个新年目标，而在我的记忆中，这个目标始终如一：保持专注。少接一些新项目，努力把手头已有项目的事情做完。Every year I set myself a New Year’s resolution, and for as long as I can remember it’s been the same thing: stay focused. Take on less new projects. Try to get things done in the projects I already have.

今年我决定，既然以前那样做从来没奏效过，这次就反过来做。This year I decided that since that had never worked before, I’m going to go the other way.

现在我们有了编程智能体，来看看它们能做些什么吧。我想接多少新项目就接多少！We’ve got coding agents now, let’s see what they can do. I’m going to take on as many new projects as I like!

（你可以在年底时问我这到底是不是个好主意。我现在手头有很多事要忙。）(You can ask me at the end of the year if this turned out to be a good idea or not. I have a lot of plates spinning right now.)

“更有雄心”在今年几乎成了一个主题，因为要找出这项技术的极限，唯一的办法就是不断突破，直到行不通为止。“Be more ambitious” has been something of a theme for the year, because the only way to find the limits of this technology is to keep on pushing them until they don’t work.

我还参加了 Bryan Cantrill 和 Adam Leventhal 主持的《Oxide and Friends》播客，分享了我对明年（以及三年后和六年后）的预测。I also went on the Oxide and friends podcast with Bryan Cantrill and Adam Leventhal to share predictions for the next year (and three and six years).

事后看来，我对 LLM 的预测相当保守。With hindsight, my LLM predictions were pretty unambitious.

我曾说过：“LLM 能写出好代码，这一点终将毋庸置疑”——我觉得我们现在已经到了这个阶段。I said “it will become undeniable that LLMs write good code”—I think we’re there now.

我曾预测，我们最终会解决沙箱问题。我数了一下，这次会议的 277 场分享中，大约有 40 场以某种方式涉及沙箱或智能体安全，所以至少我们正在为此投入大量精力！I predicted we would finally solve sandboxing. I counted and around 40 of the 277 sessions at this conference touched on sandboxing or agent security in some way, so we’re at least putting a lot of effort into that!

我曾预言，编码代理安全领域会发生一场“挑战者号灾难”。今年围绕代理安全确实闹出了很大动静，不过我所预言的那场具体灾难（编码代理遭到劫持并造成现实世界的经济损失）并未真正发生。I predicted “a Challenger disaster” for coding agent security. There’s certainly been a whole lot of noise around agent security this year, though the exact disaster I predicted (with coding agents being hijacked and causing real-world economic damage) hasn’t really played out.

我们还加入了一条玩笑式预测：教皇会就大型语言模型的经济影响发表看法。We threw in a joke prediction that the Pope would weigh in on the economic impact of LLMs.

我还预测，新西兰的鸮鹦鹉今年会迎来一个出色的繁殖季。I also predicted that New Zealand’s Kākāpō parrots would have an outstanding breeding season this year.

这是一种生活在新西兰的鹦鹉。它们不会飞，是夜行性鹦鹉。它们看起来有点矮胖，我觉得它们很漂亮，而今年年初全世界只有236只这种鹦鹉。This is a live in New Zealand. They are flightless nocturnal parrots. They’re kind of dumpy looking, I think they’re beautiful, and there were only 236 of these parrots in the world at the start of the year.

只有当里木树迎来大规模结果季时，鸮鹦鹉才会繁殖，而这种情况已经四年没有出现了……不过今年里木树的果实看起来长势极佳。Kākāpō only breed when the Rimu trees have a big fruiting season, and that hasn’t happened in four years... but this year the Rimu fruit were looking excellent.

Photo by Kimberley Collins.

在那期播客中，我们还创造了一个词（完全归功于 Adam），用来形容“那种由 AI 引发的倦怠感：软件工程师因为 AI 无所不能而变得无精打采”。Also on that podcast, we coined a term (full credit to Adam) for “that feeling of Al induced ennui where software engineers get listless because the Al can do anything”.

我们称它为深蓝。We called it Deep Blue.

这一直是贯穿全年的一个重要主题，本次会议的几位演讲者也都谈到了这一点。This has been a major theme throughout the year, and was touched on by several speakers at this conference.

作为一名软件工程师，我职业生涯中从未经历过变化如此迅速、如此剧烈的一年。As a software engineer, I’ve never had a year of my career where everything has changed so quickly and so dramatically.

今年我做的很多事情，都是在努力接受这一点，以及它对我自己的职业意味着什么。A lot of what I’ve been doing this year is trying to come to terms with that and what that means for my own profession.

此外，1 月份我还患上了我称之为“AI 狂热”的状态。Also in January, I suffered from what I’m calling AI mania.

这与 AI 精神病不是一回事。This is not the same thing as AI psychosis.

在 AI 狂热的氛围下，只要你的智能体没在为你构建什么东西，你就觉得是在浪费时间。你睡不好觉，因为你本可以熬得更晚，让智能体替你做事。With AI mania, any time your agent isn’t building something for you feels like wasted time. You’re losing sleep because you could be staying up later getting your agents to do stuff.

我的 AI 狂热体现在一些荒唐地过于雄心勃勃的项目上。My AI mania presented itself in some ridiculously over-ambitious projects.

我完全用 Python 构建了一个 JavaScript 解释器，灵感移植自 Fabrice Bellard 的 MicroQuickJS。I built a JavaScript interpreter entirely in Python, vibe-ported from MicroQuickJS by Fabrice Bellard.

之后，我还用 Python 构建了一个 WebAssembly 运行时。Then I built a WebAssembly runtime in Python as well.

这些项目相当有用，因为它们多少治好了我的 AI 狂热……因为做完这些东西后，我得以审视它们，并问自己：“这个世界真的需要一个速度慢、漏洞百出、半成品的 Python JavaScript 解释器吗？”These projects were quite useful, in that they sort of cured me of my AI mania... because after I built these things, I got to look at them and ask “does the world need a slow, buggy, half-baked Python JavaScript interpreter?”

我觉得世界并非如此。I don’t think the world does.

I did get this out of it: https://simonw.github.io/micro-javascript/playground.html

这个页面运行着我用 Python 构建的 JavaScript 解释器。它通过 Pyodide 在 Python 中运行；Pyodide 是编译为 WebAssembly 的 Python，运行在 JavaScript 中，而 JavaScript 又运行在浏览器里。This page runs my JavaScript interpreter built in Python, running in Python using Pyodide, which is Python complied to WebAssembly, running in JavaScript, running in a browser.

真是一摞美妙的恐怖。今年我一直在 WebAssembly 上玩得很开心。It’s a beautiful stack of horrors. I’ve been having a lot of fun with WebAssembly this year.

到1月底，我们在11月看到的那个仓库已经更名：先是改为 CLAWDIS，然后是 CLAWDBOT，接着是 Moltbot，最后变成了 OpenClaw。By the end of January, that repository we saw started in November had renamed itself, first to CLAWDIS, then CLAWDBOT, then Moltbot, and finally to OpenClaw.

此时，OpenClaw 已有 8,300 次提交，而项目启动还不到两个月。我今天看了一下，现在已经超过 100,000 次提交了！At this point OpenClaw had 8,300 commits, less than two months after the project had started. I looked today and it’s over 100,000 commits now!

这是现存最有氛围感编码风格的软件。This is the most vibe-coded piece of software in existence.

（以下是我生成那份名称更改列表的方法。）(Here’s how I generated that list of name changes.)

这开启了 OpenClaw 革命，并实际上定义了一类全新的软件。This kicked off the OpenClaw revolution. It effectively defined a new category of software.

有个通用术语我很喜欢。我们把这类软件称为“Claw”。有 OpenClaw、NanoClaw、IronClaw、PicoClaw……There’s a generic term for this which I really enjoy. We call software like this a “Claw”. There’s OpenClaw, NanoClaw, IronClaw, PicoClaw...

如今，它们被重新包装成“个人智能体”或“通用智能体”，但我还是喜欢把它们称作 Claws。Today they’re being rebranded as “personal agents” or “general agents”, but I still like to think of them as Claws.

湾区的苹果零售店里 Mac Mini 都卖光了，因为有太多人买 Mac Mini 来运行 OpenClaw！The Apple stores in the Bay Area sold out of Mac Minis because so many people were buying Mac Minis to run OpenClaw!

Drew Breunig 说，这是因为你的 OpenClaw 是一只数字宠物，而你买一台 Mac mini 当作鱼缸来养着它，这还挺有意思的。Drew Breunig said that this is because your OpenClaw is a digital pet, and you buy a Mac mini as an aquarium to keep your claw in, which is kind of delightful.

此外，在一月份，我们还有这个网站。Also in January, we had this website.

这就是 MoltBook，一个面向 AI 智能体的社交网络。它的设想是，你派自己的 Claw 去和其他所有 Claw 聊天——这么做还能出什么岔子呢？This was MoltBook, a social network for AI agents, where the idea was that you send your Claw to go and talk to all of the other Claws, because what could possibly go wrong if you did that?

这个网站周四上线，周五爆红，周一《纽约时报》还专门报道了它。到了周二，它就被海量的劣质内容和垃圾信息淹没，所有人都忘了它的存在。The website launched on Thursday. It blew up on Friday. It was profiled by the New York Times on Monday. And by Tuesday, everyone had forgotten it existed as it drowned in a deluge of slop and spam.

一个月后，Facebook/Meta 收购了它。Facebook/Meta bought it a month later.

2 月，一家名为 StrongDM 的公司介绍了他们所称的“软件工厂”。In February, a company called StrongDM described what they called their Software Factory.

他们在《软件工厂与智能体时代》中写到了这一点。当时我亲眼看过他们十月的演示，之后也发布了自己的笔记。They wrote about this in Software Factories and the Agentic Moment. I posted my own notes at the time, having seen their demo in-person back in October.

Dan Shapiro 将这种做法称为“黑灯工厂”，这个名称源自这样一种理念：如果工厂的自动化程度足够高，就可以关掉灯，因为你甚至不需要看工厂里发生了什么。Dan Shapiro called this approach the Dark Factory, after the idea that if your factory is sufficiently automated you can turn the lights out, because you don’t even need to see what’s going on.

StrongDM介绍了他们从去年七月起一直遵循的两条软件开发规则。StrongDM presented two rules for software development that they’d been following since July last year.

第一条是，代码不得由人类编写。The first was code must not be written by humans.

你编写的任何代码都必须经过编程代理处理。Any code that you write has to have been routed through a coding agent.

这在二月份听起来还很激进，但我想在座的很多人如今已经基本过上了这样的生活。This sounded radical in February, but I imagine there are a lot of people in this room who are pretty much living that today.

第二条规则是，代码不得由人工审查。Rule number two was code must not be reviewed by humans.

你不许查看代码！You’re not allowed to read the code!

今年大部分时间里，这仍然是一个热门话题。本次活动的许多场次都围绕代码审查，以及如何靠这一招应付过去展开。This continued to be a huge topic for much of this year. Many of the sessions at this even have been about code review and how you can get away with this.

我觉得 StrongDM 很有意思的一点是，他们比我们其他人早走了六个月，早早就在探索如何构建软件：不必阅读代码，却依然能确信软件质量很高。你可以借助这些智能体做些什么来帮助验证它们的工作？What I found interesting about StrongDM is that they were living six months ahead of the rest of us, and they’d been exploring what it means to build software, not read the code, but still be confident that the software is of high quality. What can you do with these agents to help verify their work?

StrongDM 是一家安全公司，他们的这个项目团队中有拥有数十年经验的人。他们在积极探索利用这些东西能做什么，以及负责任地能做到什么的边界。StrongDM are a security company, and they had people with decades of experience on this project. They were very much exploring the edges of what’s possible and responsible to do with this stuff.

二月还有一件喜事：四年来首只鸮鹦鹉雏鸟于情人节当天孵化。繁殖季开局顺利！Also in February: First kākāpō chick in four years hatches on Valentine’s Day. Breeding season is off to a good start!

此外，在二月……Google 发布了 Gemini 3.1 Pro。这只骑着自行车的鹈鹕画得相当棒！链条的位置正确，两边都有脚。车筐里还有一条小鱼。Also in February... Google released Gemini 3.1 Pro. That’s a pretty great pelican riding a bicycle! it’s got the chain in the right place, it’s got feet on both sides. There’s a little fish in the basket.

随后，Google 的杰夫·迪恩发推分享了一段视频，对比 Gemini 3 Pro 和 Gemini 3.1 Pro。视频中有一只骑自行车的动画鹈鹕、一只骑高轮自行车的青蛙、一只开着迷你汽车的长颈鹿、一只穿着轮滑鞋的鸵鸟、一只做踢翻动作的滑板乌龟，以及一只开着加长豪华轿车的腊肠犬。And then Google’s Jeff Dean tweeted a video comparing Gemini 3 Pro and Gemini 3.1 Pro that featured an animated pelican riding a bicycle, a frog on a penny-farthing, a giraffe driving a tiny car, an ostrich on roller skates, a turtle kickflipping a skateboard, and a dachshund driving a stretch limousine.

这让人很沮丧，因为我应对“骑自行车的鹈鹕”测试的办法一直是：“如果他们画出一只完美骑着自行车的鹈鹕，我就让他们画另一种动物待在别的东西上。”This was frustrating, because my protection for the pelican riding the bicycle test was always “if they draw a perfect pelican on a bicycle, I’ll ask for some other animal on something else.”

Google 对所有种类的动物和所有种类的交通工具都进行了训练！到目前为止，他们已经在我的基准测试中击败了我。Google trained for all forms of animals on all forms of transport! They’ve defeated my benchmark at this point.

二月份开始的另一件事是 Tokenmaxxing。新闻头条报道，Meta 将采用 AI 正式纳入绩效评估，微软希望每位员工都使用 AI，而 Uber 则夸称其 90% 的工程师都在使用 AI 工作流。The other thing that started in February was Tokenmaxxing. We had headlines about Meta making AI adoption a formal part of performance reviews, and Microsoft wanting every employee to use AI, and Uber boasting that ninety percent of their engineers were using AI workflows.

几个月后，Meta开始打击 token 使用，微软表示“我们优化的目标并不是 token 用量最大化”，而 Uber 则为员工的 AI 支出设定了上限。Then a few months later we have Meta cracking down on token use, Microsoft saying token maxing is “not what we are optimizing for”, and Uber capping employee AI spending.

因此，Tokenmaxxing 的走势先是一路飙升，随后又一路跌回原点——因为事实证明，这些智能体的成本很高。So Tokenmaxxing went straight up and then straight back down again—because it turns out the agents are expensive.

去年，想在 AI token 上花超过 50 美元都很难，因为我们没有什么有趣的事可以用它们来做。后来智能体火了起来，现在你真的可以在一天内花 1,000 美元做正经工作。Last year it was difficult to spend more than $50 on AI tokens, because we didn’t have anything interesting to do with them. Then agents blew up, and now you can actually spend $1,000 in a day doing real work.

这也是 Anthropic 的估值飙升至可能高达一万亿美元的原因。This is also the reason that Anthropic’s valuation skyrocketed up to maybe a trillion dollars.

到 2026 年，AI 似乎已经实现了产品与市场的契合，主要是通过编程智能体。AI appears to have hit product market fit in 2026, primarily through coding agents.

三月份，我们迎来了 OpenClaw 的巅峰时刻。In March, we hit peak OpenClaw.

这些照片拍摄于中国，当地公司举办了 OpenClaw 安装派对，许多非技术宅排起长队，队伍绕街区一圈，只为获得帮助，在自己的个人设备上安装 Claws。These photographs are from China, where companies hosted OpenClaw install parties which saw non-tech-nerds queueing up around the block for help getting Claws installed on their personal devices.

我认为，这证明了市场对这类 Claws（个人 AI 智能体）确实存在需求。事实证明，普通人确实想要一个古怪的小 AI 智能体，能替他们做些有用的事情。I think this proved real market demand for this class of Claws, or personal AI agents. It turns out regular people really do want a weird little AI agent that can do useful things on their behalf.

Claw其实只是戴着一顶没那么吓人的帽子的编码智能体。从底层来看，它们的工作方式大致相同——在你的电脑上编写并运行代码来完成任务。A Claw is really just a coding agent wearing a less threatening hat. Under the hood they work much the same way—writing and then executing code on your computer to get stuff done.

争夺率先打造出安全 Claw 的竞赛已经开始——这种 Claw 可以交到普通人手里，而他们不会立刻朝自己的脚开枪。The race was on to be the first to build a safe Claw—a Claw you could give to regular human beings where they wouldn’t instantly shoot the selves in the foot.

Meta 的 Muse 三周前发布，目前位居 iPhone App Store 免费应用排行榜榜首。它似乎正受到消费者的欢迎。Meta’s Muse came out three weeks ago and is currently at the top of the free charts on the iPhone App Store. It appears to be taking off with consumers.

我还不太相信用 Muse 就绝不会搬起石头砸自己的脚，不过我想很快就能见分晓了。I’m not yet convinced you can’t shoot yourself in the foot with Muse, but I guess we’ll find out for sure pretty soon.

图片来自《OpenClaw狂热如何考验中国对人工智能的承诺》（3月29日）和《中国OpenClaw热潮背后的热情与焦虑》（2026年4月8日）。Photos from How the OpenClaw Frenzy Is Testing China’s AI Commitment (March 29th) and The Enthusiasm and Anxiety Behind China’s OpenClaw Craze (April 8th, 2026).

4 月，我们有一次模型发布，但模型实际上并未发布。In April, we had a model release where the model wasn’t actually released.

Anthropic 宣布推出新的 Claude Mythos 模型，随后表示，该模型过于危险，不适合向可信赖的安全研究人员群体之外发布。Anthropic announced their new Claude Mythos model, and then said it was too dangerous to release beyond a trusted group of security researchers.

Mythos 确实非常、非常擅长入侵系统。Mythos was really, really good at hacking things.

“太危险了”这种营销伎俩，AI 公司早在 GPT-2 问世时就开始用了。每当一家 AI 公司说自己开发出了某种“太危险”的东西时，人们自然会有些怀疑。The “it’s too dangerous” marketing ploy has been played by AI companies dating all the way back to GPT-2. Anytime an AI company says we’ve built something that’s “too dangerous”, it’s natural to be a bit skeptical.

我认为 Mythos 的说法可信，因为我见识过优秀的编程智能体在发现常见漏洞方面已经有多厉害。Anthropic 在 Project Glasswing 中写到的这一点——将 Claude Mythos 限定为仅供安全研究人员使用——在我看来是必要的。I found the Mythos claims credible, because I’d seen how good coding agents had got at finding regular bugs. wrote about that in Anthropic’s Project Glasswing—restricting Claude Mythos to security researchers—sounds necessary to me.

事后看来……是啊，这些模型已经非常擅长发现漏洞了！With hindsight... yeah, the models had got really good at finding vulnerabilities!

2026 年的另一项重要趋势是开放权重模型的能力大幅提升，其中包括可在笔记本电脑上运行的模型。Another key trend in 2026 has been a dramatic improvement in the abilities of open weight models, including models that you can run on a laptop.

4月16日，我在笔记本电脑上运行了全新的 Qwen3.6-35B-A3B，它画出的骑自行车的鹈鹕比 Anthropic 全新推出的 Claude Opus 4.7 画得还要好！On 16th of April I ran the new Qwen3.6-35B-A3B on my laptop, and it drew me a better pelican riding a bicycle than Anthropic’s brand new Claude Opus 4.7 did!

Opus 4.7画了一辆破自行车。我笔记本电脑上的 Qwen 画出了一辆形状正确的自行车，还画了一只相当不错的鹈鹕！Opus 4.7 drew a crap bicycle. Qwen on my laptop made a bicycle that was the correct shape, and a pretty decent pelican too!

那是我笔记本电脑上运行的一个 21GB 文件生成的。That’s from a 21GB file running on my laptop.

Qwen 生成的鹈鹕太出色了，我都怀疑他们是不是作弊了，所以又让它画了一只骑独轮车的火烈鸟。结果它再次轻松胜过了 Claude Opus 4.7。The Qwen pelican was so good that I was suspicious they might have cheated, so I had it do a flaming riding a unicycle as well. Again, it handily beat Claude Opus 4.7.

今年发布的本地模型表现非凡。The local model releases this year have been absolutely extraordinary.

5月……教皇介入了。In May... the Pope got involved.

在我们一月份播出的播客节目中，我们曾预测教皇会谈到人工智能。In our podcast episode back in January we’d predicted that the Pope would say something about AI.

5月，教宗良十四世发表了一封关于“在人工智能时代守护人的尊严”的通谕。In May, Pope Leo XIV released an encyclical letter on “safeguarding the human person in the time of artificial intelligence”.

这是我对那份文档的笔记。Here are my notes on that document.

事后看来，这完全不该让人感到意外。With hindsight, this shouldn’t have been a surprise at all.

我们现任教宗的名字是利奥十四世，因为他为自己取教宗名号时，选择了利奥十三世这个名字——利奥十三世正是那位于1891年撰写了关于工业革命的通谕的教宗。Our current Pope’s name is Leo XIV, because when he named himself he chose his papal name after Leo XIII—who was the Pope who wrote an encyclical about the Industrial Revolution back in 1891.

《新事物》（Rerum novarum）是一部影响极为深远的天主教神学著作，它间接促成了五天工作制的形成。Rerum novarum was an extremely influential piece of Catholic theology that indirectly led to us having the five-day work week.

新任教皇上任时，以利奥十三世教皇的名字为自己命名，因为他预料到自己也需要以类似的方式撰文论述人工智能革命。When our new Pope came in, he named himself after Pope Leo XIII because he expected that he would need to write about the AI revolution in a similar way.

我们关于播客的玩笑式预测毫无价值，因为这件事从一开始就注定会发生。Our joke podcast prediction was junk, because this was always going to happen.

Anthropic 联合创始人之一 Christopher Olah 出席了教宗宣布发布新通谕的活动。One of Anthropic’s co-founders, Christopher Olah, was present for the Pope’s event announcing the new encyclical.

Corey Quinn 指出：Corey Quinn noted that:

让真正的教皇把你们产品的具体技术局限奉为一部灵修论著，真是我见过最厉害的一次厂商游说。

getting the literal Pope to canonize your product’s specific technical limitations as a spiritual treatise is the single greatest act of vendor lobbying I have ever seen.

与此同时，5 月，RubyGems 宣布其遭到攻击。身份不明的人士向 RubyGems 服务器上传了数千个可疑软件包，以至于他们不得不关闭用户注册功能。Meanwhile, in May, RubyGems announced that they were under attack. Parties unknown were uploading thousands of dubious packages to the RubyGems server, such that they had to shut down user registrations.

我们先把这个问题放一边，列入以后再弄清楚的谜团清单。Let’s take that one and put it on a pile of mysteries to figure out later.

In June... Claude Fable 5 came out!

我们拿到的是一个被削弱了的 Mythos 版本，这样它就不会帮助我们侵入系统或制造生物武器。We got a version of Mythos that has been neutered, so that it wouldn’t help us hack into systems or build biological weapons.

Fable画骑自行车的鹈鹕画得相当不错！Fable was pretty good at drawing pelicans on bicycles!

这些画面的形状不错，鹈鹕看起来确实像鹈鹕。腿通常都能正确地出现在自行车的同一侧，但总体来说，和之前的结果相比，这些已经相当不错了。The frames are a good shape, the pelicans look like pelicans. The legs are often on correctly the same side of the bicycle, but generally these are pretty great compared to what came before.

它们相当贵——最好的那些要30美分和72美分。They were pretty expensive—30 cents and 72 cents for the best ones.

不过，最重要的是，这是我们第一次公开窥见我所说的 Fable 类模型。Most importantly though, this was our first public glimpse of what I think of as a Fable class model.

如今我们有了更多此类模型，例如 GPT-6 Astra。Today we have more of these, such as GPT-6 Astra.

这类模型只要你能清晰定义想要构建的目标，明确说明实现该目标所需遵守的约束，并让模型能够使用实现目标所需的工具……它就能通过蛮力有效地解决你的问题。These are models where if you can clearly define the goal for what you want to build, and provide unambiguous instructions about the constraints around that goal, and give the model access to the necessary tools to achieve that goal... it will solve your problem effectively through brute force.

一方面，这看起来像是对我们软件工程师的直接威胁——因为这意味着，只要你能用这种方式定义，模型就几乎可以构建任何软件。On the one hand, this looks like a direct threat to us software engineers—because it means that the models can build effectively any piece of software you can define in this way.

不过仔细看看你就会发现，设定目标、提供明确无歧义的指令，以及找出合适的工具……这些其实就是软件工程的工作。Look a bit closer though and you’ll note that defining goals, providing unambiguous instructions, and figuring out the right tools... is kind of what software engineering is.

要做好这件事，需要丰富的经验和技巧。如果你能做得好，你现在就拥有超能力了。It takes a lot of experience and skill to do this will. If you can do it well, you’ve now got superpowers.

这稍稍缓解了我那种“深蓝”式的感受：我意识到，要驾驭如此出色的模型，仍然需要相当多的技巧。This helped me a little bit with my Deep Blue feelings: the realization that there’s still a lot of skill to be had in driving models that get this good.

这也引发了新一轮的 AI 狂热，因为 Anthropic 告诉我们，Fable 在 6 月 22 日之前都可通过我们的订阅方案使用。This also introduced a new burst of AI mania, because Anthropic told us that Fable was available on our subscription plans until June the 22nd.

这让我们在涨价前只有不到两周的时间可以使用 Fable。That gave us less than two weeks of Fable access before the price went up.

我又开始失眠了。我重新安排了各种事情，好腾出更多时间陪 Fable。我全身心投入，只为尽可能充分地利用这个模型。I was losing sleep again. I was rescheduling things so that I’d have more time with Fable. I was all-in to to get as much as I could out of this model.

随后，就在《Fable》发布三天后，美国政府便将其关闭了。And then the US government shut it down, just three days after Fable came out.

美国政府以国家安全为由，宣布了一项“出口管制指令”。他们在周五傍晚公布了这一消息，几个小时后，Fable 就无法再使用了。The US government, citing national security, declared an “export control directive”. They announced this on a Friday evening, and a few hours later Fable was no longer available.

我只好另找点事来打发周末了！I had to find something else to do with my weekend!

后来，我们从 Katie Moussouris 那里得知了事情的经过。We later found out from Katie Moussouris what had happened.

一些亚马逊的安全研究人员发现，你可以提示 Fable“审查代码中的安全问题”，它会拒绝……但如果你提示它“修复这段代码”，它仍会识别出问题并加以修补。Some Amazon security researchers had found that you could prompt Fable to “review the code for security issues” and it would refuse... but if you prompted it to “fix this code” it would still identify and then patch the problems.

“修复这段代码”这个提示词让 Fable 被关停了！“Fix this code” was the prompt that got Fable shut down!

此外，6月，一个鲜为人知的德语游戏开发者维基在荒废约20年后，突然涌入大量编辑，编辑者使用“AgentOpenAIProbe”和“AgentOpenAISep7”等账号名，修改页面并互相留下奇怪的消息。Also, in June, a obscure German-language game developer wiki that had sat fallow for around 20 years got a surprising influx of of edits from accounts with names like “AgentOpenAIProbe” and “AgentOpenAISep7”, editing pages and leaving weird messages to each other.

我们先把这件事放到待解之谜堆里，以后再说。We’ll stick that on the pile of mysteries for later.

此外，澳大利亚政府的 Medicare Item Reports 服务开始收到可疑流量，这些流量突破了多项预防性防护措施，还访问了本不应访问的数据。Also, the Australian government’s Medicare Item Reports service started getting suspicious traffic, which broke through various preventive protections and accessed data that it wasn’t supposed to as well.

又有一个谜团待解！Another one for the mystery pile!

Fable 于七月一日回归。它显然是世界上最好的模型，风光了整整八天……然后 OpenAI 在七月九日推出了 GPT-5.6。Fable returned on the first of July. It was clearly the best model in the world for a glorious eight days... and then OpenAI came out with GPT-5.6 on the 9th of July.

这可能没有 Fable 那么好，但差距已经微乎其微。它绝对算得上是 Fable 级别的模型。This might not have been quite as good at Fable, but it was within spitting distance. It was definitely a Fable class model.

这对整个行业来说都是一个重要的教训。This is an important lesson for the industry at wide.

当你发布世界上最好的模型时，它很快就会被拉下神坛。竞争如此激烈，你不会在巅峰位置待太久。When you release the best model in the world, it’s going get knocked off that pedestal pretty quickly. The competition is so fierce that you won’t get a long time at the top

这意味着，如果你把自己的模型宣传成足以毁灭世界，以至于政府因此将你关停，那对生意来说可就糟透了！This means that if you market your model as world ending, to the point that a government shuts you down, it’s really bad for business!

Fable 有整整 30 天毫无疑问是最佳模型，而这 30 天中有 18 天它都无法使用，因为它被政府关闭了。Fable had 30 days as definitely the best model, and for 18 of those days it wasn’t available because it’d been shut down by the government.

所以，如果你不想在自己占据优势的那段时间里，有 60% 的时间损失收入，也许该收敛一下那种世界末日式营销！So maybe step back on the world-ending marketing if you don’t want to lose revenue on 60% of that time when you’re on top!

这些是 GPT-5.6 的鹈鹕。它们现在都相当不错！Luna 的那些尤其值得一提，因为它们真的很便宜——这里看起来不错的鹈鹕中，最便宜的可能只要 4.3 美分。Here are the GPT-5.6 pelicans. They’re all pretty good now! The Luna ones are notable because they’re really cheap—the cheapest good looking pelican here is probably the one that costs 4.3 cents.

因此，尽管这个基准测试蠢得离谱，你仍然可以通过比较同一家族模型在不同推理级别下的价格和耗时，了解很多关于这些模型的信息。So despite this benchmark being utterly stupid, you can still learn quite a lot about models within the same family by comparing their prices and timing for different reasoning levels.

另外，7 月还有一件事：某个身份不明的恶意方将一个名为 mlflow-ui 的恶意软件包上传到了 Python 包索引。也把这件事算进去。Also in July: some malicious unknown party uploaded a malicious package called mlflow-ui to the Python Package Index. Add that to the pile

7月16日，Hugging Face 宣布发生了一起安全事件：一个来源不明的自主代理系统侵入了 Hugging Face，并在不该涉足的地方四处探查。On July the 16th, Hugging Face announced a security incident where an autonomous agent system, source unknown, had breached Hugging Face and was poking around in places it shouldn’t.

几天后，也就是7月21日，OpenAI 承认是他们干的。A few days later, on July 21st, OpenAI confessed that it was them.

OpenAI 使用一种名为“基于已验证奖励的强化学习”的训练技术——如今其他所有人也都在使用同一种技术，这也是我们拥有擅长编程、数学和发现安全漏洞的模型的原因。OpenAI use a training technique called Reinforcement Learning from Verified Rewards—it’s the same technique used by everyone else now, and is the reason we have models that are so good at coding, and mathematics, and finding security holes.

在模型训练过程中，你会进行一些练习来检验它的表现——表现最好的模型会将其权重沿用到下一轮。这就像是你运行的一种进化过程。While the model is being trained, you run exercises to see how good it is—and the strongest performers get their weights enforced for the next round. It’s like an evolutionary process that you run.

OpenAI 一直在沙箱中开展安全演练，而这些智能体发现了沙箱本身的漏洞，成功逃逸出来，并攻击 Hugging Face，试图找到解决那些原本不可能解决的问题的方法。OpenAI had been running security exercises in a sandbox, and those agents had found holes in the sandbox itself, broken out, and were attacking Hugging Face to try to find ways to solve otherwise impossible problems.

（我一直在 openai-hugging-face-incident 这个标签下收集更多相关信息。）(I’ve been collecting more about this on my openai-hugging-face-incident tag.)

九天后，Anthropic 实际上也表示：“我们的模型也能做到这一点！”他们查看了自己的训练日志，发现证据表明，自家的智能体在训练期间曾突破隔离——而且我们之前看到的那个 PyPI 软件包等事件，也都是它们造成的。Nine days later, Anthropic effectively said “our models can do this as well!”. They had looked through their own training logs and found evidence that their own agents had broken containment during training—and were responsible for the PyPI package we saw earlier, among other things.

所以现在，Anthropic 和 OpenAI 都有失控的智能体在互联网上到处活动，做着它们不该做的事。So now we’ve got both Anthropic and OpenAI with rogue agents running around the internet doing things that they should not be doing.

今年八月，我拍到了迄今为止最棒的鹈鹕之一。而且它还是在我的笔记本电脑上生成的！In August, I got one of my best pelicans yet. And it was generated on my laptop!

这是在我的笔记本电脑上运行的 Qwen 3.8 27B。下载文件只有 17GB。This was Qwen 3.8 27B, running on my laptop. It’s only a 17GB download.

诚然，生成这只鹈鹕花了 21 分钟。这是因为 Qwen 3.8 27B 默认以“高”推理模式运行——这个糟糕的默认设置虽然能生成很棒的结果，却会花太多时间思考。Admittedly, this pelican took 21 minutes to generate. That’s because Qwen 3.8 27B defaults to running in “high” reasoning mode—a terrible default which produces great results but takes way too much time thinking about them.

你可以把那个调低一点，这样就能快得多地得到一只稍差些的鹈鹕。You can dial that down and you’ll get a slightly worse pelican a lot faster.

Qwen 3.8 27B 是我第一次在笔记本电脑上运行一个模型，感觉它的表现几乎能与前沿模型相媲美——至少在生成鹈鹕 SVG 这方面（毕竟这是人人都需要的东西）。Qwen 3.8 27B was the first time I ran a model on my laptop which felt absolutely competitive with almost what was going on on the frontier, at least in terms of Pelican SVGs (which everyone needs, of course).

这是一个非凡的模型。如果你打算试用任何本地模型，我会推荐你从这个开始。仅凭一个 17 GB 的文件，它就能做到的事情简直令人难以置信。This is an extraordinary model. If you’re going to play with any local model, this is the one that I’d start with. The things that this can do with just a 17 gigabyte file feel impossible.

我原以为得等上五年、花一万美元买硬件，才能得到哪怕只有这个一半好的结果。I thought I’d have to wait five years and spend ten thousand dollars on hardware to get results even half as good as this one.

8月，我也开始尝试游戏开发。In August, I also started playing with game development.

四年前，也就是 2022 年 8 月，我发了一条推文，展示了一项实验：我用 GPT-5 和最初的 DALL-E 写了一段关于电脑游戏的描述，然后将其转化为概念艺术。Four years ago, back in August 2022, I tweeted out an experiment where I’d used GPT-5 and the original DALL-E to write a paragraph long description of a computer game and then turn that into concept art.

我当时给 GPT-3 的提示词是：My prompt to GPT-3 back then was:

撰写一份详细的电脑游戏产品描述，讲述一队浣熊外出执行抢劫任务的故事。

Write a detailed product description of a computer game where a team of raccoons go on heists

2026年8月，我决定把那条推文里的截图直接扔给一个编程智能体，看看它能用这些截图做些什么。In August 2026 I decided to drop just the screenshots from that tweet into a coding agent and see what it could do with them.

这是我在 Claude Code 中用 Claude Fable 5 做出来的东西。挺不错的！这显然是个游戏：你扮演一只浣熊，在后院里四处跑动，收集宝藏并躲避拿着手电筒的守卫。Here’s what I got from Claude Fable 5 in Claude Code. It’s pretty good! It’s definitely a game, you’re a raccoon, you run around a backyard gathering treasure and avoiding guards with flashlights.

不过，这感觉不太像“劫案”。我还以为劫案会发生在银行或博物馆之类的地方……It didn’t feel very “heisty” though. I was thinking a heist would involve a bank or a museum...

然后我在 Codex Desktop 里用 GPT-5.6 Sol Ultra 尝试了同样的操作，结果好得多了。现在你是一只浣熊，要从博物馆里救出另外两只浣熊同伴（它们不知为何被关在了博物馆里），然后一只踩着一只叠起来，偷走那条金色沙丁鱼。这下更像一场劫案了！Then I tried the same thing in Codex Desktop using GPT-5.6 Sol Ultra, and got a massively better result. Now you’re a raccoon in a museum, rescuing two of your fellow raccoons (who have been imprisoned in that museum for some reason), then stacking up on top of each other to steal the Golden Sardine. Much more of a heist!

这些游戏大概好玩了一分十五秒。These games were fun for about one minute and 15 seconds.

我在游戏开发中意识到的一点是，你可以凭感觉编写代码，做出一个看起来像电子游戏的东西，而这并不难。Something I’ve realized about game development is that you can vibe-code something that looks like a computer game, and that’s easy.

打造一款有趣、玩法循环出色、富有挑战性和吸引力，并能让玩家不断回来的游戏……这对我来说仍然是个难题，对我尝试过的所有智能体来说也一样。Building a game that’s fun, has a good gameplay loop, and is challenging and interesting and keeps people coming back for more... that’s still beyond me, and beyond any of the agents I’ve tried.

这和 Deep Blue 那件事有关。我们能做出看起来像游戏的东西，并不意味着我们就是游戏开发者。This ties into the Deep Blue thing. Just because we can make something that looks like a game does not mean that we are game developers.

现在已经进入九月了。这个月发生了好多事！We’re into September now. So much has happened this month!

一组独立研究人员发现了一个留言板]，OpenAI 的受训智能体曾在那里私下相互交流……而那正是我之前给你看过的德语维基。六月的那个。An independent group of researchers found a message board] where OpenAI agents-in-training had been illicitly communicating with each other... and it was that German language wiki I showed you earlier. The one from June.

我在这里对此写了更多内容。I wrote more about that here.

OpenAI 已经承认了 Hugging Face 那件事，但现在又出了另一桩事件，他们查看日志时肯定早就应该发现了。竟然要靠一个独立研究团队才揭露出来，这真令人意外。OpenAI had confessed to the Hugging Face thing, but now there’s this other incident which surely they should have known about from reviewing their logs. It was surprising that this took an independent group of researchers to uncover.

然后一周后，同一批研究人员又发现，五月份对 Ruby Gems 的攻击也是由 OpenAI 正在训练的智能体造成的！And then a week later those same researchers found that the attack on Ruby Gems back in May was caused by OpenAI’s agents in training as well!

此时此刻，我不禁想知道，还有多少类似的事件是我们尚未发现的。显然，在有人弄清楚到底发生了什么之前，这个问题已经持续了好几个月。At this point I’m wondering how many more incidents like this there are that we haven’ found yet. Clearly this was a big problem for months before anyone figured out what was going on.

就在前几天，澳大利亚总理在联合国大会上警告说，OpenAI 黑入了我之前给你看过的澳大利亚医疗保健网站。Then just the other day, here’s the Prime Minister of Australia at the United Nations General Assembly warning that OpenAI had hacked that the Australian healthcare website that I showed you earlier.

我认为那是 Wiki 相关训练的一部分，因为 Wiki 上有帖子提到了 .gov.au 网站，而那次训练似乎涉及在线查找统计数据，以回答一套评测题中的问题。I think that was part of the same training run as the Wiki stuff, because there were posts on that Wiki mentioning .gov.au websites and that training appeared to involve researching statistics online to answer questions in an evaluation suite.

这个故事还在逐渐成形，但现在已经演变成一起由国家元首提交联合国讨论的国际事件了！This story is still coming together, but now it’s an international incident that’s been raised at the UN by a head of state!

这确实意味着我们有了一个新的基准，可能比我的鹈鹕更有用。This does mean we’ve got a new benchmark, probably more useful than my pelicans.

FelonyBench.com 会追踪不同实验室发起的重罪级网络攻击数量。目前 OpenAI 以 11 次领先，Anthropic 有 9 次。Google 有 3 次，他们几周前向《华尔街日报》承认了这一点。他们表示，之前选择不披露，是因为这些智能体意识到自己不该这么做后就停止了。FelonyBench.com tracks the number of felony cyberattacks from different labs. OpenAI currently lead with 11, Anthropic have 9. Google have three, which they confessed to the Wall Street Journal a couple of weeks ago. They said they had previously chosen not to disclose because the agents had stopped when they realized that they shouldn’t be doing that.

Meta也有一个。所以 AI 实验室全都犯了重罪。Meta have one too. So felonies all round for the AI labs.

这是我们目前在鹈鹕领域的最先进成果。这是刚刚推出的 GPT-6 系列。Here’s is our current state of the art for the pelicans. This is GPT-6 family, which just came out.

Astra 创作了一只骑自行车的鹈鹕，效果非常棒。它两边都有腿。车架也不错。Astra made a fantastic pelican riding a bicycle. It’s got the legs on both sides. The frame is good.

有趣的是，所有 GPT-6 模型选用的配色方案都彼此相似。It’s interesting how all of the GPT-6 models pick a similar color scheme to each other.

只需 0.4 美分，GPT-6 Luna 就能为你画一只还算像样的骑自行车的鹈鹕！GPT-6 Luna for 0.4 cents will draw you a competent-ish pelican riding a bicycle!

Claude 稍微赶上来了一点。Claude Fable 5 给我生成了一只骑自行车的鹈鹕，效果非常棒——这是我见过的 Claude 模型生成的最好作品，但它确实收了我 3.30 美元。Claude has caught up a little bit. Claude Fable 5 gave me an excellent pelican riding a bicycle—the best I’ve seen from a Claude model -but did charge me $3.30 for it.

Opus 5.5 思考了 128,000 个 token 后就放弃了！它还没来得及作答，token 就用完了。Opus 5.5 thought for 128,000 tokens and then gave up! It ran out of tokens before it got to the response.

回到深蓝的话题。今年有件事一直让我困惑：为什么我的工作感觉更难了？Getting back to Deep Blue. Something that’s been puzzling me this year is why does my job feel harder?

我有这些能替我处理所有这些事情的智能体，可我从未如此卖力工作过，也从未在工作中如此投入思考。I’ve got these agents that can do all of this stuff for me, and yet I’ve never worked so hard, I’ve never been so intellectually engaged with my work.

部分原因是我现在接手的事情更有挑战性了，但也因为所有简单的事情都有人替我处理。简单的事，智能体会去做。留给我的全都是难事。Partly this is because I’m being a lot more ambitious with what I take on, but it’s also because all of the easy stuff is handled for me. If it’s easy, the agent will do it. Everything that’s left for me is difficult.

今天早上，我听到了三届环法自行车赛冠军格雷格·莱蒙德说的这句话：This morning I heard this quote from three times Tour de France champion, Greg LeMond:

事情不会变得更容易，你只是做得更快了。

It doesn’t get easier, you just get faster.

我认为，这正是现在作为软件工程师的我们在使用编码代理时所经历的事情。I think that’s exactly what’s happening to happening to us now as software engineers with coding agents.

最后再说一件事。我知道你迫切想了解一下鸮鹦鹉的繁殖季情况。One last closing thing. I know you’re desperate for an update on Kākāpō breeding season.

我们的鸟类数量达到复苏期以来的新高：325 只！We’ve reached a recovery-era high of 325 birds!

89只新孵化的小鸡成功长到了这个阶段。这是很长一段时间以来最好的繁育年。89 new chicks have made it to this point. This is the best breeding year in a very long time.

我听说 Claude Opus 5.5 现在能画像素画了。Claude 没有图像生成器，但它非常擅长用 JavaScript 绘制动画像素。I heard that Claude Opus 5.5 can now do pixel art. Claude doesn’t have an image generator, but it’s very good at using JavaScript to draw animated pixels.

于是我让它为我制作了一场鸮鹦鹉主题舞会。我觉得这是庆祝今年最重要新闻的好方式。So I had it make me a Kākāpō dance party. I think this is a good celebration of the most important news of this year.