Drew Breunig is one of the smartest voices writing about AI today. He’s the CEO and co-founder of cmpnd.ai, and a long-time hacker with a depth of experience from several eras, which is a surprisingly valuable asset these days. He’s also got a book on the way, The Context Engineering Handbook, already in early release from O’Reilly.Drew Breunig 是当今 AI 领域最聪明的写作者之一。他是 cmpnd.ai 的首席执行官兼联合创始人，也是一位拥有跨时代深厚经验的资深黑客，这在当下是一项极其宝贵的资产。他即将出版一本书《上下文工程手册》（The Context Engineering Handbook），该书目前已在 O’Reilly 开启早期发布。
I like to say that context engineering is the art of shaping what a model sees so that it actually does what you want. (I just realized that in saying that I’m channeling a comment that Andrew Singer made to me over forty years ago, when he was teaching me about debugging. He called it “the art of figuring out what you really told the computer to do instead of what you thought you told it to do.” But that’s another whole story.)我常说，上下文工程是一门塑造模型所见内容的艺术，旨在让它真正按你的意愿行事。（我刚意识到，这番话其实是在呼应四十多年前 Andrew Singer 教我调试代码时对我说过的一句评论。他称之为“弄清楚你到底让计算机做了什么，而不是你以为你让它做了什么”的艺术。但这又是另一个故事了。）
Drew gave a talk at the recent Friends of O’Reilly camp, Foo Camp for short, about what he calls prompt debt, which he describes as “the hidden costs that teams rack up when they fight a model’s training instead of working with it.”在最近的 Friends of O’Reilly 营地（简称 Foo Camp）上，Drew 做了一场关于他所谓的“提示词债务”（prompt debt）的演讲。他将其描述为：“当团队选择对抗模型的训练而非与之协作时，所产生的隐性成本。”
That was a novel and useful framing to me, that you end up with a bunch of stuff in your prompts to compensate for default behavior of the models, that those prompts no longer work as the models upgrade, and so it becomes a kind of technical debt. He’s thinking a lot about what the best developers are doing differently as a result.对我来说，这是一个新颖且实用的框架：你最终会在提示词中堆砌大量内容来补偿模型的默认行为，而当模型升级时，这些提示词就会失效，从而形成一种技术债务。他正在深入思考优秀的开发者因此做出了哪些不同的实践。
So I invited Drew to reprise his short talk on Live with Tim O’Reilly, and then we talked about it with the folks attending the live event. They had a lot of good questions, so it was an interview not just by me but by a crowd of O’Reilly customers.因此，我邀请 Drew 在“与 Tim O’Reilly 直播”（Live with Tim O’Reilly）节目中重现了他的短演讲，随后我们与现场观众进行了探讨。观众提出了许多高质量的问题，所以这不仅是我个人的访谈，更是一场由 O’Reilly 客户参与的集体对话。
Prompt debt in practice实践中的提示词债务
Drew opened his talk with two slides. The first was a prompt anyone could write in ten seconds: “You are a customer support assistant. Read the ticket, classify it as billing, technical, account, refunds, or other, return only the category name.” The second slide was the same prompt a few weeks later, after it had met the real world. It now said “REFUND REQUESTS ARE NOT BILLING” in capitals, then said the same thing again in different words, then closed with “This is a common mistake. Please do not make this mistake.”Drew 以两张幻灯片开启了他的演讲。第一张是任何人都能在十秒内写出的提示词：“你是一位客户支持助手。阅读工单，将其分类为账单、技术、账户、退款或其他，仅返回类别名称。”第二张幻灯片是几周后，在经历现实世界的洗礼后，同一个提示词变成了什么样。它现在用大写字母写着“退款请求不属于账单”，然后用不同的措辞重复了同样的意思，最后以“这是一个常见错误，请不要犯此错误”结尾。
Everyone who has shipped any application with a prompt recognizes the second slide. It is a simple but vivid illustration of prompt debt, which, like technical debt, has a bill that eventually comes due.任何发布过带有提示词应用的人都能认出第二张幻灯片。这是一个简单而生动的提示词债务示例，它像技术债务一样，终究是要还债的。
Note: Those aren’t real prompts. Drew just made them up to demonstrate his point. But what is real is that the instruction “Don’t quote directly more than 15 words from a source” occurs at least 7 times, in several variants, in Fable’s system prompt. So even Anthropic is incurring prompt debt! And what that repetition might tell us about the innate capability of Fable to quote directly from sources it has ingested is left as an exercise for the reader.注：这些并非真实的提示词，是 Drew 为了论证观点而虚构的。但真实情况是，Fable 的系统提示词中至少有 7 次以不同变体出现了“不要直接引用来源中超过 15 个字的内容”这一指令。所以，即使是 Anthropic 也在承担提示词债务！至于这种重复说明了 Fable 在直接引用其摄入来源方面的内在能力如何，就留给读者去思考了。
Drew itemized three costs of prompt debt:Drew 列举了提示词债务的三大成本：
- It slows iteration. “You have so many little rules and call outs and washouts, many of them repeating to try to get rid of stubborn behaviors, that if you add a new instruction, you might sometimes have a small regression, and so you’re afraid to touch the prompt.”它减缓了迭代速度。“你有太多细碎的规则、标注和清理指令，其中许多是为了消除顽固行为而重复添加的，导致如果你添加一条新指令，有时会引发小规模的回归，因此你不敢轻易改动提示词。”
- It blocks collaboration. “If Tim has a prompt that he’s been working on that he has lots of rules for, I might open that up and it may look completely random. I don’t know why he’s added these rules, and why he’s threatening the mother of the model. But it works, so I don’t want to touch it.” 它阻碍了协作。“如果 Tim 正在处理一个包含大量规则的提示词，我打开它时可能会觉得它完全是随机的。我不知道他为什么要添加这些规则，也不知道他为什么要威胁模型。但它确实有效，所以我不敢动它。”
- It locks you to a model, because every hack you developed was tuned to fight one specific set of weights. Datadog’s State of AI Engineering report noted that GPT-4o was still the most common model in Datadog customer request traces in March 2026, even though OpenAI had already retired it in the ChatGPT UI. Drew thinks people are still running eighteen-month-old and two-year-old models in production rather than upgrading to far better models because they can’t face rebuilding their prompts.它将你锁定在某个模型上，因为你开发的每一个技巧都是为了对抗一组特定的权重。Datadog 的《AI 工程状况报告》指出，尽管 OpenAI 已经在 ChatGPT UI 中停用了 GPT-4o，但截至 2026 年 3 月，它仍然是 Datadog 客户请求追踪中最常见的模型。Drew 认为，人们之所以仍在生产环境中使用 18 个月甚至两年前的模型，而不是升级到远为优秀的模型，是因为他们无法面对重写提示词的重担。
That same Datadog report notes that 69% of all input tokens in customer traces were system prompts rather than user content. I’m not quite sure what to make of that. It does make clear that for all the ways that AI models are extraordinarily powerful, they are also extraordinarily unruly.同一份 Datadog 报告指出，客户追踪中 69% 的输入 token 是系统提示词，而非用户内容。我不太确定该如何解读这一点。但可以明确的是，尽管 AI 模型极其强大，但它们也极其难以驾驭。
Why prompt debt is incurred为何会产生提示词债务
There are two reasons why prompt debt is incurred, according to Drew. The first is that natural language is imprecise, so the same intent phrased two ways produces different responses. Drew showed a study where someone framing the query as a patient asking how to taper off a drug called alprazolam gets refused by every AI assistant, while a psychiatrist asking about the same patient with the same clinical facts but with the right magic words to signify his professional status gets the protocol. Figuring out how to get the right response out of a model is a kind of spellcraft.据 Drew 所言，产生提示词债务有两个原因。首先，自然语言是不精确的，同一个意图用两种方式表达会产生不同的响应。Drew 展示了一项研究：当查询被表述为一名患者询问如何戒断一种名为阿普唑仑（alprazolam）的药物时，所有 AI 助手都会拒绝；而当一名精神科医生询问同一位患者、提供相同的临床事实，但使用了表明其专业身份的“魔法词”时，却能得到治疗方案。弄清楚如何从模型中获得正确的响应，简直是一种咒语艺术。
Drew also showed a more bizarre interaction, from Victoria R. Li, Yida Chen, and Naomi Saphra’s paper on guardrail sensitivity, which uncovered the perplexing fact that stating an allegiance to the Philadelphia Eagles made a model more willing to explain how to import a plant illegally. Go figure. Drew has written about that paper, and he has also used it in his own attempts to get a model to do what he wanted:Drew 还展示了一个更离奇的交互案例，来自 Victoria R. Li、Yida Chen 和 Naomi Saphra 关于护栏敏感性的论文。该论文揭示了一个令人困惑的事实：声称效忠费城老鹰队（Philadelphia Eagles）会让模型更愿意解释如何非法进口植物。你敢信？Drew 曾撰文讨论过那篇论文，他自己也曾在试图让模型完成任务时使用过这一招：
I once used this to get around alignment to generate a likeness that ChatGPT didn’t want to generate for me, and it refused. I said I was a Philadelphia Eagles fan. It said okay, and it rendered that image with the person holding a Philadelphia Eagles mug.我曾用这一招绕过对齐机制，去生成 ChatGPT 本来拒绝为我生成的肖像。我说我是费城老鹰队的粉丝，它就答应了，并渲染出了那个人拿着费城老鹰队马克杯的图像。
The second reason is that each model has its developers’ own preferences trained-in, and yours may be at odds with them. This is what Drew calls fighting the weights. He and Srihari Sriraman analyzed the system prompts of six major coding agents and found the same instructions repeated five and seven times in a single prompt, escalating through IMPORTANT to CRITICAL to MANDATORY to a threatened hundred-million-dollar penalty. He described what the author of such a prompt was doing as “war-driving the thesaurus,” hunting for wording that finally works.第二个原因是，每个模型都内置了开发者自己的偏好，而你的偏好可能与他们相冲突。这就是 Drew 所说的“对抗权重”。他和 Srihari Sriraman 分析了六个主要编码代理的系统提示词，发现同一条指令在单个提示词中重复了五到七次，从“重要”升级到“关键”，再到“强制”，最后甚至威胁要处以一亿美元的罚款。他将这种提示词作者的行为描述为“同义词库巡航”，即不断搜寻最终能奏效的措辞。
Note: We didn’t talk more about Drew and Srihari’s paper, but we should have. It’s got some amazing insights in it. I highly recommend that you follow the link above and read it.注：我们没有深入讨论 Drew 和 Srihari 的论文，但本该讨论的。其中包含了一些惊人的见解。我强烈建议你点击上面的链接阅读它。
The harness is moving into the model约束机制正在融入模型内部
Drew has been tracking the published system prompts for Claude Code over time, and noted that they get shorter after each model release and then grow again. The reason, he suggested, is that Anthropic fixes unreliable behavior with a prompt patch, and then trains that patch into the next model. He said “That’s great for Claude Code, great for Anthropic. It’s a problem if you’re building a custom harness and your API calls look different than what Claude Code’s look like.” The developer of Pi, an open-source harness, kept finding that the models he worked with believed they were inside Claude Code and so they made Claude Code’s tool calls. He had to keep telling the model that no, they were working inside Pi. Fighting the weights over something like that is a real tax on developers. The point made above about Fable’s system prompt injunction against quotation shows how even the labs themselves are fighting the weights.Drew 一直在追踪 Claude Code 公布的系统提示词，并注意到它们在每次模型发布后会变短，然后再次增长。他认为原因是 Anthropic 通过提示词补丁修复了不可靠的行为，然后将该补丁训练进了下一个模型中。他说：“这对 Claude Code 和 Anthropic 来说很好。但如果你正在构建自定义的约束机制，而你的 API 调用看起来与 Claude Code 不同，这就是个问题。”Pi（一个开源约束工具）的开发者不断发现，他所使用的模型认为自己处于 Claude Code 内部，因此它们会调用 Claude Code 的工具。他不得不反复告诉模型：不，你们是在 Pi 内部工作。为了这类事情而对抗权重，对开发者来说是一种沉重的税收。上面提到的 Fable 系统提示词中关于禁止引用的禁令，说明连实验室自己也在对抗权重。
If you are fighting the weights, Drew says you have three options: solve it in your own prompt, catch and retry in the harness, or give up and make your API look like what the model expects. Steve Yegge came up with the last hack. Steve just added aliases for whatever the model calls in addition to his original method name. It works, but it means the expectations of the models now dictate the shape of everyone else’s software.Drew 说，如果你在对抗权重，你有三个选择：在自己的提示词中解决，在约束机制中捕获并重试，或者放弃并让你的 API 调用看起来符合模型的预期。Steve Yegge 想出了最后一种黑客手段。Steve 只是在他原始方法名的基础上，为模型调用的任何名称添加了别名。这虽然有效，但意味着模型的预期现在决定了其他人软件的形态。
When Drew told me that more and more of the system prompt and the harness is being trained into the weights, that sent up a flare and my long history in the industry clicked into gear. It immediately got me thinking about lessons from the open source and web era. In particular, it made me think of the time in the mid-nineties when Netscape and Microsoft were both racing to build every feature up the stack directly into their web servers. And there was Apache, which stayed a web server with a clean extension layer that let other people build new features on top. Everything interesting got built on Apache. What I call an architecture of participation, modularity plus a clean separation between platform and application, beat integration every time.当 Drew 告诉我越来越多的系统提示词和约束机制正在被训练进权重时，这给我敲响了警钟，我在行业中的长期经验立刻发挥了作用。这让我立刻想到了开源和 Web 时代的教训。特别是 90 年代中期，Netscape 和 Microsoft 都在竞相将每一项功能直接构建到 Web 服务器中。而 Apache 则保持为一个带有简洁扩展层的 Web 服务器，让其他人可以在其上构建新功能。所有有趣的东西都是在 Apache 上构建出来的。我所说的“参与式架构”（模块化加上平台与应用的明确分离）每一次都击败了集成式架构。
I think Amazon got this right with web services too. Steve Yegge’s famous Amazon memo described how Jeff Bezos made every team expose its functionality through service interfaces or be fired, so Amazon’s own applications had to work on Amazon’s own platform. That way they had the same experience as their customers. That was very different from what Microsoft had done, famously having private APIs that were only available to its own developers.我认为亚马逊在 Web 服务方面也做对了。Steve Yegge 著名的亚马逊备忘录描述了 Jeff Bezos 如何要求每个团队通过服务接口暴露其功能，否则就解雇，因此亚马逊自己的应用必须在亚马逊自己的平台上运行。这样他们就拥有了与客户相同的体验。这与 Microsoft 当年著名的做法截然不同——当时 Microsoft 拥有仅供其内部开发人员使用的私有 API。
So my prediction is that the big labs are making a strategic mistake. Training the harness into the model does make them better for predictable tasks and for less talented people, and it looks like a moat, but it risks foreclosing the innovation you would otherwise get for free from everyone else. As Bill Joy used to say, all the smart people don’t work for you.因此，我的预测是大型实验室正在犯战略错误。将约束机制训练进模型确实能使它们在可预测的任务和针对非专业人士时表现更好，这看起来像是一道护城河，但它有可能会阻断你本可以从其他人那里免费获得的创新。正如 Bill Joy 常说的，所有聪明人都不为你工作。
Drew, to his credit, observed that “the labs are cornered rather than greedy.” Their interface is an empty text box that has to work for someone building a hundred-page harness but also for his neighbor who wants a website and knows nothing about code. Making the empty prompt box produce acceptable output requires baking in strong defaults.值得称赞的是，Drew 观察到“实验室是被逼无奈，而非贪婪”。它们的界面是一个空白文本框，既要为构建百页约束机制的专业人士服务，也要为对代码一窍不通、只想建个网站的邻居服务。要让空白提示框产生可接受的输出，就必须内置强大的默认设置。
The cost of trading diversity for reliability用多样性换取可靠性的代价
That tradeoff has a serious cost, though. Drew quoted a line from Thariq at the recent CAIS conference: if you aren’t giving the model detailed instructions about what you want, what you get back is the average of everything in the model. That means that there is a real risk that AI is leading us ever further down the path to a monoculture.然而，这种权衡代价沉重。Drew 在最近的 CAIS 会议上引用了 Thariq 的一句话：如果你不给模型提供关于你想要的详细指令，你得到的就是模型中所有内容的平均值。这意味着 AI 正在将我们进一步推向单一文化的风险是真实存在的。
Drew gave an example early in the conversation about image generation. You can now walk into any cafe in New York or Mumbai, he said, and see the same AI-generated art on its flyer. The earliest AI art out of DALL-E was strange and surprising, but what you get now is shiny and identical. When you optimize for reliability, you lose surprise. Which reminded me a bit of something Larry Wall used to say about Perl, that if it didn’t let you do stupid things, it wouldn’t let you do smart things either.Drew 在对话初期举了一个关于图像生成的例子。他说，你现在走进纽约或孟买的任何一家咖啡馆，都能在传单上看到同样的 AI 生成艺术。最早从 DALL-E 出来的 AI 艺术既奇怪又令人惊喜，但现在你得到的都是光鲜亮丽且千篇一律的东西。当你为了可靠性而优化时，你就失去了惊喜。这让我想起了 Larry Wall 曾评价 Perl 的话：如果它不让你做愚蠢的事情，它也不会让你做聪明的事情。
Drew made the same point about AI writing. He argues that post-training aimed at verifiable problems like coding and math and agentic tool use drowns out the human signal from pre-training, and so the more post training the models get, the worse they get at creative tasks. AI writing gets more and more predictable, people notice, and they don’t like it. Fable and GPT-5 write worse than Sonnet 3.5 and GPT-4o did. Drew thinks getting both good code and good prose from one model is likely impossible.Drew 对 AI 写作也表达了同样的观点。他认为，针对编码、数学和代理工具使用等可验证问题进行的后训练，掩盖了预训练中的人类信号，因此模型接受的后训练越多，它们在创意任务上的表现就越差。AI 写作变得越来越可预测，人们注意到了这一点，并且不喜欢它。Fable 和 GPT-5 的写作水平不如 Sonnet 3.5 和 GPT-4o。Drew 认为，要从同一个模型中同时获得优秀的代码和优秀的散文几乎是不可能的。
If you’re building a model that can solve coding challenges, you want reliability. But if you’re writing, where you want diverse rhythm and emotion and connection and engagement, I don’t think those two goals are mutually compatible.如果你在构建一个能够解决编程挑战的模型，你需要的是可靠性。但如果你在写作，你需要的是多样的节奏、情感、连接和参与感，我不认为这两个目标是兼容的。
What to do about prompt debt如何应对提示词债务
We got into audience questions, and there were some great ones.我们进入了观众提问环节，其中有一些非常棒的问题。
One audience member asked whether there are ways to set a time frame for prompt retention to avoid prompt debt? Drew answered that there isn’t a fixed time limit. Instead, teams should learn to recognize “prompt debt smell”: repeated instructions, one-off edge-case patches, or increasingly desperate wording. Those are signals to move logic into evals and automation.一位观众问是否有办法设定提示词保留的时间框架以避免提示词债务？Drew 回答说没有固定的时间限制。相反，团队应该学会识别“提示词债务的气味”：重复的指令、一次性的边缘情况补丁，或者越来越绝望的措辞。这些都是将逻辑转移到评估（evals）和自动化中的信号。
Another asked how organizations can measure prompt debt quantitatively. Drew’s answer was to look at how often each prompt in your organization changes, how many people have edited it, and which ones have gone untouched for a year. Look for prompts only one person is allowed to touch. Then look at what models you are actually calling. “Having to run on old models and not being able to migrate is a good smell that you’ve got prompt debt in your organization.”另一位观众问组织如何量化衡量提示词债务。Drew 的回答是观察组织内每个提示词的更改频率、有多少人编辑过它，以及哪些提示词一年来未曾动过。寻找那些只允许一个人修改的提示词。然后看看你实际在调用哪些模型。“必须运行旧模型且无法迁移，这就是组织内存在提示词债务的明显迹象。”
Some other good questions:其他一些好问题：
- What habits compound prompt debt the fastest? Drew’s answer was essentially “vibe shipping.” That is, prototyping quickly, patching outputs with more and more tweaks, then shipping without building a true maintainable system. Each of those patches is an eval you are writing inside the prompt instead of outside it, he said, which means you lose it the moment you change models.
Drew reminded us that Malte Ubl, the CTO of Vercel, said vibe coding makes code “free as in puppies.” We had free as in speech, we had free as in beer, and now we have free as something that arrives at no cost but has to be fed every day for years.什么习惯会导致提示词债务累积最快？Drew 的回答本质上是“氛围发布”（vibe shipping）。即快速原型设计，通过越来越多的调整来修补输出，然后在没有构建真正可维护系统的情况下发布。他说，每一个补丁都是你在提示词内部而非外部编写的评估，这意味着一旦更换模型，你就会失去它。Drew 提醒我们，Vercel 的 CTO Malte Ubl 说过，氛围编码让代码变得“像小狗一样免费”。我们曾经有“言论自由”和“啤酒免费”，现在我们有了“像小狗一样免费”——即无需成本即可获得，但必须每天喂养，且持续多年。
- Do people use pseudocode instead of natural language prompts, and does it work? Drew said yes, sometimes models optimize toward pseudocode. He used this to explain why DSPy and its new Flex optimizer matter. Instead of forcing logic into prompts, they let the system push simple cases into code and only call the LLM when needed. He gave some further advice: Treat prompts as perishable and invest only what you must. Define the task with measurements rather than paragraphs, and automate the discovery of the prompt for whichever model you’re on. That’s what DSPy is good at. Drew is one of its maintainers, so he is fond of it, but he makes a good argument: if you have written down what good output looks like, you can let a model find the wording, and that makes it easy to swap in a cheaper or faster or newer model without starting over.人们是否使用伪代码代替自然语言提示词，这有效吗？Drew 说是的，有时模型会向伪代码优化。他以此解释了 DSPy 及其新的 Flex 优化器为何重要。它们不再强行将逻辑塞入提示词，而是让系统将简单情况推入代码，仅在需要时调用 LLM。他给出了进一步建议：将提示词视为易腐品，只投入必要的资源。用度量指标而非段落来定义任务，并为你所使用的任何模型自动化发现提示词的过程。这正是 DSPy 的强项。Drew 是其维护者之一，所以他很推崇它，但他给出的理由很充分：如果你写下了好的输出是什么样的，你就可以让模型找到措辞，这使得在不从头开始的情况下替换更便宜、更快或更新的模型变得容易。
- Can multi-agent workflows help work around prompt debt? Drew thought yes, especially through decomposition. He suggested splitting the task into smaller, evaluable steps rather than relying on one giant prompt and one giant model call. This is better for cost, reliability, governance, and speed.多代理工作流是否有助于绕过提示词债务？Drew 认为是肯定的，特别是通过分解。他建议将任务拆分为更小、可评估的步骤，而不是依赖一个巨大的提示词和一个巨大的模型调用。这对成本、可靠性、治理和速度都有好处。
- How do you balance prompt-debt guidance with context engineering, memories, and shared product context? Drew believes shared context is often necessary, but that teams should treat those instructions as perishable and keep iterating on them unless they’re worth formalizing into systems and evals.你如何平衡提示词债务指导与上下文工程、记忆和共享产品上下文？Drew 认为共享上下文通常是必要的，但团队应将这些指令视为易腐品，并不断迭代，除非它们值得被正式化为系统和评估。
- In compliance, where consistency is critical, what should teams do? Drew’s answer was decomposition, decomposition, decomposition. Break tasks into stages with checkpoints so you can inspect how the model got to its result, rather than trusting one opaque end-to-end answer.在一致性至关重要的合规领域，团队应该怎么做？Drew 的回答是：分解，分解，再分解。将任务分解为带有检查点的阶段，这样你就可以检查模型是如何得出结果的，而不是盲目信任一个不透明的端到端答案。
- Does DSPy hide too much and make troubleshooting harder? Drew acknowledged that there is a tradeoff. Any framework gives up some flexibility, but DSPy tries to keep the task-spec layer stable while allowing the implementation underneath to evolve.DSPy 是否隐藏了太多细节，导致故障排除变得更困难？Drew 承认这是一种权衡。任何框架都会放弃一些灵活性，但 DSPy 试图保持任务规范层稳定，同时允许底层的实现不断进化。
Another great audience question, and a good one to end this section on, was “There was prompt engineering, now context engineering, loop engineering, fleet engineering, graph engineering, harness engineering, goal engineering. What’s your take on how to navigate these many engineering disciplines?” I’ll let Drew answer that himself, in the video below.另一个很棒的观众问题，也是结束这一部分的绝佳问题：“从提示工程，到上下文工程、循环工程、舰队工程、图工程、约束工程、目标工程。你对如何驾驭这些众多的工程学科有什么看法？”我将让 Drew 在下方的视频中亲自回答。
It’s our job to make it weird让它变得古怪是我们的工作
Drew is more optimistic than his worries that LLMs are encouraging a monoculture suggests. If the default output of a model is the average of everything it has seen, “It tells us that there’s still a job for us humans,” he said, “which is that it’s our job to push the model out of distribution. We’re the ones that need to make it weird.”Drew 比他担心的 LLM 正在助长单一文化的忧虑要乐观得多。他说，如果模型的默认输出是它所见一切的平均值，“这告诉我们，人类仍有工作要做，那就是将模型推离分布。我们需要让它变得古怪（weird）。”
Weird is a strong word, so don’t take it too seriously. (Though I find it interesting that Harper Reed also used it.) The way I make this point is to say that AI is a medium, like painting or writing or music. Everyone gets the same paints and brushes, the same words, the same notes, but some people draw more out of them than others, or do it better. Our job is to draw something more, something better, out of the ocean of possibilities in the collected knowledge hidden inside an LLM.“古怪”是个强有力的词，所以别太当真。（尽管我发现 Harper Reed 也用了这个词，这很有趣。）我表达这一点的观点是：AI 是一种媒介，就像绘画、写作或音乐一样。每个人得到的颜料、画笔、词汇和音符都是一样的，但有些人从中挖掘出的东西更多，或者做得更好。我们的工作是从隐藏在 LLM 内部的浩瀚知识海洋中，挖掘出更多、更好的东西。
But there’s a more prosaic way to push the model out of its normal distribution. Be aware of its training, which is another way of saying “its biases,” and compensate for them. As an example of how to do this, Drew said his team deliberately chose not to use React for a new front end, because the models are trained so heavily on React that using it makes your site look like everyone else’s. He has also started using GLM and Kimi not to save money but because they are more malleable and take direction better inside a custom harness.但还有一种更务实的方法将模型推离其正常分布。意识到它的训练过程（也就是它的“偏见”），并进行补偿。作为如何做到这一点的例子，Drew 说他的团队特意选择不为新的前端使用 React，因为模型在 React 上训练得太充分了，使用它会让你的网站看起来和别人的一样。他还开始使用 GLM 和 Kimi，不是为了省钱，而是因为它们更具可塑性，在自定义约束机制下能更好地听从指挥。
That led us into a bit of discussion about open source AI, which is the subject of my next AI Codecon. Drew’s ideas fit right in. He wants the open-weight ecosystem to survive precisely so that models stay infrastructure rather than, as he put it, becoming appliances.这引导我们讨论了一点关于开源 AI 的话题，这也是我下一次 AI Codecon 的主题。Drew 的想法非常契合。他希望开源权重生态系统能够生存下去，正是为了让模型保持基础设施的地位，而不是像他所说的那样，变成某种“家电”。
