We got used to it before we understood it
We got used to it before we understood it
我们在理解它之前，就已经习惯了它
Swipe and there is food at the door. Swipe and there is a date on Thursday. Swipe and there is a stranger’s car outside taking you home. Ask for a function and there is code. Ask for a design and there is a mockup. Ask for a summary of a forty page document and it is done before you have finished reading the first page yourself. None of this felt like a decision. It arrived, and we adapted to the interface without ever adapting to the change. 滑动一下，食物就送到了门口。滑动一下，周四就有了约会。滑动一下，陌生人的车就在外面等着带你回家。要求一个功能，代码就写好了。要求一个设计，样稿就出来了。要求总结一份四十页的文档，在你读完第一页之前，它就已经完成了。这一切感觉都不像是一个经过深思熟虑的决定。它就这样发生了，我们适应了界面，却从未真正适应这种变化。
That is the part I keep thinking about. Not whether the technology is good. Whether the gap between how fast it moves and how fast we absorb it is still a survivable gap. The race is worth running I want to be clear about where I stand, because the easy version of this article is a complaint, and I do not want to write that one. I like this. I have built things in the last two years that I could not have built in the previous ten. I read release notes for fun. The speed is genuinely exciting and I am not nostalgic for a slower industry. The problem is not the speed. It is that the speed now exceeds my ability to form an opinion about what is moving. 这就是我一直在思考的部分。不是技术本身好不好，而是技术发展的速度与我们吸收它的速度之间的差距，是否还在我们可承受的范围内。这场竞赛值得参与，我想明确我的立场，因为这篇文章最简单的写法就是抱怨，而我不想那样写。我喜欢这一切。过去两年里，我构建的东西比之前十年还要多。我甚至会为了乐趣去阅读发布说明。这种速度确实令人兴奋，我并不怀念那个节奏缓慢的行业。问题不在于速度，而在于这种速度现在已经超出了我对事物发展形成判断的能力。
Where I notice it in my own work The clearest version for me is tooling. I wrote a while back about Rust taking over the JavaScript toolchain. Vite, Rolldown, the whole rewrite of the layer underneath everything we ship. It was a good story to write. It was also a story where, by the time I understood the tradeoffs well enough to have a real position, the position was already outdated. That used to be an annoyance. Now it is the normal condition. I do not reach competence with a tool anymore. I reach a working familiarity, and then the tool changes, and I am back near the start with a slightly better instinct than last time. 我在自己的工作中注意到了这一点，最明显的例子就是工具链。前段时间我写过关于 Rust 接管 JavaScript 工具链的文章。Vite、Rolldown，以及我们发布的所有内容底层的全面重写。那是一个很好的选题，但也是一个这样的故事：当我足够理解其中的权衡取舍并形成观点时，这个观点已经过时了。过去这令人烦恼，现在这成了常态。我再也无法精通一个工具了。我只能达到一种“工作熟悉度”，然后工具就变了，我又回到了起点，只是比上次多了一点直觉。
Evaluation has become the bottleneck. Not learning. Deciding what is worth learning. Multiply that by every developer on a team, and you get an organisation where nobody is deeply expert in anything because expertise takes longer to build than the thing takes to be replaced. That has architectural consequences. It is why so many codebases are now three eras of tooling stacked on top of each other, with nobody left who understands the bottom layer. 评估成了瓶颈，而不是学习本身。难的是决定什么值得学习。把这一点乘以团队里的每一个开发者，你就会得到一个组织：没有人是任何领域的深度专家，因为建立专业知识所需的时间比工具被替换的时间还要长。这会带来架构上的后果。这就是为什么现在许多代码库都是三代工具堆叠在一起，却没有人能理解最底层的原因。
The honest version of “our brains cannot keep up” The line I first wanted to write was that our brains do not finish processing the change. It feels true. It is how the tiredness actually feels from the inside. But I do not think I can defend it as stated, and there is a lot of confident pop neuroscience about attention spans and dopamine that does not survive contact with the actual research. So let me say the thing I can defend instead. What lags is not biology. It is everything built on top of biology. Norms take years. Nobody knew what phone etiquette was for a decade. Laws take longer. Professional identity takes a career, which is why a developer who trained on one paradigm spends their whole working life partly translating. 关于“我们的大脑跟不上”的诚实版本：我最初想写的是，我们的大脑无法完成对变化的加工。这感觉很真实，也是从内部感受到的那种疲惫感。但我认为我无法为这种说法辩护，而且关于注意力跨度和多巴胺的流行神经科学观点，很多都经不起实际研究的推敲。所以，让我说一个我能辩护的观点。滞后的不是生物学，而是建立在生物学之上的所有东西。规范需要数年时间才能形成。十年来，没人知道什么是电话礼仪。法律需要更久。职业认同需要整个职业生涯来建立，这就是为什么一个在某种范式下受训的开发者，整个职业生涯都在进行某种程度的“翻译”。
Institutions are the slowest of all, and they are the layer that decides what is permitted rather than what is possible. Those lags are measurable. You do not need a claim about neurons. The technology iterates in months and the scaffolding around it iterates in decades, and every year that ratio gets worse. 机构是最缓慢的，它们决定了什么是“被允许的”，而不是什么是“可能的”。这些滞后是可以衡量的。你不需要关于神经元的论断。技术以月为单位迭代，而围绕它的社会架构以十年为单位迭代，而且这种比例每年都在恶化。
Every generation said this I have to include this part or the article is dishonest. People said it about the printing press. The telegraph. Radio. Television. The internet. Every one of those came with a serious argument that human beings were about to be overwhelmed, and in every case we absorbed it and the panic reads as quaint now. “This time is different” is the most overused sentence in technology. I do not get to skip that objection just because I feel tired. So here is the only version of the difference I can actually argue for, and I hold it loosely. Every previous case had a plateau. The technology arrived, there was a shock, then a long flat stretch while people, laws and habits caught up, and then the next thing. The plateau is what made it survivable. Adaptation happened during the pause. The proposal on the table now is to remove the pause. 每一代人都这么说过：我必须包含这一部分，否则这篇文章就不诚实。人们在印刷机、电报、广播、电视和互联网出现时都这么说过。每一次都有严肃的论点认为人类即将被淹没，但每一次我们都吸收了它，现在的恐慌看起来显得有些古怪。“这次不一样”是科技界最被滥用的一句话。我不能仅仅因为我感到疲惫就跳过这个反驳。所以，这是我唯一能论证的“不同之处”，而且我对此持保留态度。以往的每一个案例都有一个“平台期”。技术到来，产生冲击，然后是一段漫长的平稳期，人们、法律和习惯在此期间跟上，然后才是下一波浪潮。这个平台期是让我们得以生存的关键。适应是在暂停期间发生的。而现在摆在桌面上的提议是：取消这个暂停。
Which brings me to the letter On July 28, 2026, more than a thousand employees of the frontier AI labs signed a statement called Pacing the Frontier. The site listed 1,178 signatories when I checked and the form is still open. Anthropic, OpenAI, Google and Meta are all represented, including several CEOs and chief scientists. They are not asking for a pause. The single request is that the U.S. government support an international effort to build the technical and governance tools needed to “deliberately pace the frontier of automated AI development.” 这让我想到了那封信。2026年7月28日，超过一千名顶尖 AI 实验室的员工签署了一份名为《跟上前沿》（Pacing the Frontier）的声明。我查看时，网站上列出了1,178名签署者，且表格仍在开放中。Anthropic、OpenAI、Google 和 Meta 的员工都在其中，包括几位首席执行官和首席科学家。他们并没有要求暂停。他们唯一的诉求是，美国政府支持一项国际努力，建立必要的技术和治理工具，以“审慎地把控自动化 AI 开发的前沿节奏”。
Automated AI development is the phrase that matters. Models used to build better models, which are then used to build better models again. That is a description of a process with no plateau in it. The gap between capability and comprehension stops being constant and starts compounding. What struck me most was not the request. It was the personal comments on the signatory page. Several of them are people who work on capability evaluations for a living saying the recent pace has been a shock to them too. That is the same feeling I have on a much smaller scale about build tools. It goes all the way up. Nobody is standing somewhere calm. “自动化 AI 开发”是关键所在。模型被用来构建更好的模型，然后再被用来构建更好的模型。这描述的是一个没有平台期的过程。能力与理解力之间的差距不再是恒定的，而是开始呈指数级增长。最让我震惊的不是他们的诉求，而是签署页面上的个人评论。其中几位是专门从事能力评估的专业人士，他们表示最近的进展速度让他们也感到震惊。这和我对构建工具的感受如出一辙，只是规模不同。这种感觉贯穿了整个行业。没有人站在一个平静的地方。
So what do we actually do I do not have a policy answer. I build websites and internal tools. The governance question is above my pay grade and I am suspicious of developers who pretend otherwise. But I have a working answer for myself, and it is not “keep up.” Keeping up is no longer a coherent goal. There is more happening per week than any person can evaluate, and treating that as a personal failure is a category error. The skill is not consumption. It is choosing what to deliberately not keep up with, and being at peace with that choice. For my own work that means a short list of things I go deep on and a much longer list I let pass. It means I do not adopt a tool until it has survived… 那么我们到底该怎么办？我没有政策层面的答案。我只是构建网站和内部工具。治理问题超出了我的能力范围，我怀疑那些假装自己能解决的开发者。但我为自己找到了一个可行的答案，那不是“跟上”。“跟上”不再是一个连贯的目标。每周发生的事情比任何人都无法评估，将其视为个人失败是一种范畴错误。现在的技能不是消费信息，而是选择“故意不去跟上”什么，并与这种选择达成和解。对于我自己的工作来说，这意味着我只深入研究少数几件事，而让更多的事情随它去。这意味着在工具经过考验之前，我不会采用它……