为什么写作
Why write
很多人——尤其是受过完整学术训练的人——以为写作的流程是： 先把问题想清楚，再把想清楚的内容写下来。
这是个体面的、错误的、伪科学的描述。
真实的流程是另一种：你以为自己想清楚了，于是开始写；写到第三段 你发现第二段的论证撑不住；你回头重写第二段，发现它依赖一个你从 未真正检验过的前提；你拆掉那个前提，整个第一段塌了；你重写第一 段，意外发现一个比原本想说的更有趣的方向；你顺着新方向写下去， 完成了一篇与你最初打算写的完全不同的文章。
写作不是表达思想。写作是生产思想。
为什么必须是写
口头讨论也能逼出思想，但太轻——你可以用语调、笑容、共鸣的回声 让一段虚弱的论证滑过去。听众的善意会替你修补。
写作没有听众的善意。一段话写在屏幕上，没有眼神接触可以挽救它。 它要么自洽，要么不自洽；它要么有内容，要么是空话的延展。这种 不可逃避性 才是写作之所以训练思维的根本机制。
费曼学习法的真正核心，不在”假装你在教别人”，而在被迫把模糊的内 心独白外化为不能含混的句子。
私人写作 vs 公开写作
两种都重要，但功能不同：
- 私人写作——日记、笔记、Obsidian 草稿——是给未来的自己看的 侦察报告。它的目的是把今天的思考保存成明天可以踩着继续往上爬 的台阶。这种写作允许混乱、允许半成品、允许直接抄录别人的话。
- 公开写作——博客、文章、这个数字花园——是给陌生人看的。它 的约束更紧：你不能依赖只有自己知道的隐喻，不能跳过那些”显而 易见”的步骤（因为对读者并不显而易见）。这种约束反过来逼你 把私人写作里那些含糊的部分讲清楚。
数字花园的意义就在这里：它模糊了两者的边界。你以”这只是一 个 seedling”的姿态发表，给自己留下”还可以改”的台阶；但发表本身 强制了某种最低标准——至少要让一个不认识你的人读得懂第一句。这 是一个比纯私人写作更高、比纯发表写作更低的标准。它正好处在 逼 出但不杀死 的甜蜜点。
写作与 LLM
我注意到，自从开始密集使用 LLM 协助写作，我对自己写出的东西的 信任度反而下降了。这不是矛盾，是健康的信号。
LLM 极其擅长生产听起来像思考过的文本——结构合理、过渡自然、 比喻得体。如果我把”流畅”等同于”思考完成”，我就会被骗。所以现在 每写完一段，我会问自己：
这段话是因为我想清楚了所以写出来的，还是因为它读起来像想清楚 了所以我留下来的？
第二种情况的密度，在我的草稿里高得惊人。
这件事的含义不是”少用 LLM”。是 要重新校准什么叫”我写完了”。 也许我们这一代写作者的真正修炼，是学会把”流畅”这个老的写作美德 当作风险信号而不是质量信号。
延伸: 什么是数字花园, 心流与机器, The loop and the self.
A lot of people — especially those with a thorough academic training — believe the writing process is: first figure out the problem, then write down what you’ve figured out.
This is a respectable, false, pseudoscientific description.
The real process is different. You believe you’ve figured it out, so you start writing. By the third paragraph, you notice the argument in the second won’t hold. You go back and rewrite the second, only to realise it depends on a premise you’ve never actually tested. You pull out that premise, and the first paragraph collapses. You rewrite the first paragraph, accidentally discovering a more interesting direction than the one you started with. You follow the new direction and finish an essay completely different from the one you set out to write.
Writing is not the expression of thought. Writing is the production of thought.
Why writing, specifically
Conversation can also force thought, but it’s too light — tone, smile, resonant echo can let a weak argument slide. The listener’s goodwill patches the holes for you.
Writing has no listener’s goodwill. A paragraph on a screen has no eye contact to rescue it. Either it coheres or it doesn’t; either there is content or there is filler dressed up as content. This unevadability is the mechanism by which writing trains thought.
The real core of the Feynman method isn’t “pretend you’re teaching” — it’s being forced to externalise vague inner monologue as sentences that can’t be ambiguous.
Private writing vs. public writing
Both matter, with different functions:
- Private writing — journals, notes, Obsidian drafts — is a reconnaissance report to your future self. The purpose is to preserve today’s thinking as a foothold tomorrow can climb on. Mess, half-formed ideas, and direct transcription of others’ words are all allowed.
- Public writing — blogs, essays, this digital garden — is for strangers. The constraint is tighter: no metaphors only you understand, no skipping “obvious” steps (because they’re not obvious to your reader). That tightness, in turn, forces clarity back into the murky parts of your private writing.
This is what digital gardens do: they blur the boundary between the two. You publish under the heading “this is just a seedling”, which leaves yourself the dignity of revising. But publishing itself enforces a minimum standard — make the first sentence readable to a stranger. The standard is higher than purely private writing and lower than the polished-publication standard. It sits in the sweet spot: forcing without killing.
Writing with LLMs
Since I started using LLMs heavily to assist my writing, my trust in what I write has actually dropped. This isn’t a contradiction — it’s a healthy signal.
LLMs are extremely good at producing text that sounds like it was thought through: structure plausible, transitions natural, metaphors apt. If I equate “fluent” with “thought through”, I will be fooled. So now after every paragraph I ask myself:
Did I write this because I thought it through, or did I keep it because it reads like something thought through?
The density of the second case in my drafts is alarmingly high.
The implication is not “use LLMs less”. It is recalibrate what “finished” means. Maybe the real discipline of our generation of writers is learning to treat “fluency” — the old writerly virtue — as a risk signal rather than a quality signal.
Cross-links: 什么是数字花园, 心流与机器, The loop and the self.
💬展开评论 / Show comments