Karpathy: ask LLMs for diagrams, HTML pages, and explainer videos
Karpathy argues that as models improve, more of our work shifts upward into oversight and understanding rather than production, so we should optimize the format of model output. He suggests a ladder of increasing readability: have the model rewrite explanations in ASD-STE100, the controlled-language spec built for aerospace maintenance docs (or ask for "80% of the way" there when the constraints bite); ask for diagrams; ask for HTML and get an interactive page with animation; and, his most bullish bet, custom 3b1b-style explainer videos with ElevenLabs narration or a free local substitute. Because intelligence and code are increasingly abundant, you can request large, discardable software artifacts that never made economic sense before.
We'll be spending a lot more time trying to understand the outputs of language models. A few thoughts, tips & tricks:
我们会花多得多的时间去理解语言模型的输出。下面是一些想法、技巧和窍门：
Writing. Something I've had success with: Ask your LLM to explain something in ASD-STE100, it's a controlled language specification originally developed for aerospace maintenance documentation. LLMs well-versed in this language and it comes with heavy constraints on clean writing style that I often find a lot more readable. Sometimes I've tried to soften it a bit e.g. ask for "80% of the way to ASD-STE100" because the spec is quite stringent.
写作。有个办法我试过挺有效：让 LLM 用 ASD-STE100 来解释某个东西。它是一套受控语言规范，最初为航空航天维修文档开发。LLM 很熟悉这套语言，而它对干净的写作风格有很强的约束，我常常觉得这样写出来的内容可读性高得多。有时我会试着放松一点，比如要求“达到 ASD-STE100 的 80%”，因为这套规范相当严格。
But even better: Diagrams / images. Instead of writing, ask your LLM to create a diagram. These can be a lot easier to process, parse, and understand.
但还有更好的：图表 / 图像。与其写文字，不如让 LLM 画一张图。图往往更容易处理、解析和理解。
But even better: Web pages. Ask for output "in HTML" to get a beautiful, interactive webpage. LLMs are getting really good at frontend and can create beautiful experiences, animations, etc.
但还有更好的：网页。要求模型“用 HTML 输出”，就能得到一个漂亮、可交互的网页。LLM 的前端能力正变得越来越强，可以做出漂亮的交互体验、动画等。
But even better: Explainer videos. The output format I am most bullish on is fully custom / bespoke explainer videos generated on any arbitrary topic. Experiment with things like "Create a 3b1b style video explainer on X. Use my ElevenLabs API key for audio narration". (you'd need an API key for the latter or you can ask your LLM to find you decent free alternatives that use your local compute). This is actually starting to work!
但还有更好的：讲解视频。我最看好的输出形式，是针对任意主题生成的完全定制/量身打造的讲解视频。可以试试这样的提示：“用 3b1b 风格做一个关于 X 的视频讲解。用我的 ElevenLabs API key 做音频旁白。”（后者需要 API key，或者你也可以让 LLM 找一些利用本地算力的靠谱免费替代方案。）这事儿现在真的开始能成了！
In summary:
- As LLMs get better, they will do more and more of the legwork autonomously, and a lot more of our work will rise up the abstractions into oversight and understanding.
总结：
- 随着 LLM 变得更强，它们会越来越自主地承担繁重的基础工作，我们更多的工作会沿着抽象层级上移，转向监督与理解。
Luckily, LLMs can help here too because as intelligence and code are increasingly abundant, you can ask for large, custom, discardable software artifacts (e.g. web apps, video explainers) that would have never made sense to create before. Push the boundaries here and you'll be surprised.
幸运的是，LLM 在这方面也能帮忙：智能和代码正变得越来越充裕，你可以让模型生成大型、定制、用完即弃的软件制品（比如网页应用、视频讲解），这类东西以前根本不值得做。在这里大胆推进边界，你会感到惊讶。
