The author treats blogging not as simple knowledge transfer but as a learning process for thinking more clearly, doing research, and changing your mind. Every post should make a clear, arguable claim, and the research required to defend that claim should teach the writer something new. The core argument is that honestly writing about topics you are still learning — rather than staying only within your existing expertise — and getting feedback on them is valuable for both writing and learning.

1. Every post contains at least two things learned

The author explains that every blog post he publishes is the result of at least two learnings. One is the problem or idea that made him want to write in the first place; the other is what he actually discovered while writing and researching. If he didn't learn something new while writing, he considers the post not interesting enough to publish.

"If I don't learn something new while writing a post, it isn't interesting enough to publish."

In practice he says he learns far more than two things. For instance, his post on the geo-guessing abilities of the o3 model started from the idea that "most AI prompts probably don't work well." After researching, he discovered that the latest OpenAI models had lost the geolocation-inference ability o3 once demonstrated.

His post on C2PA likewise began from the concern that the technology only works properly if it is adopted almost everywhere. But while writing, he ended up learning broadly about public key infrastructure (PKI), how private keys are managed on-device, and how C2PA actually works. With the Luddites, he originally wanted to cover the fact that the movement was essentially decentralized, but during research he became interested in the fact that the culture at the time was far more elitist, misogynistic, and violent than popular Luddite books portray.

In other words, for the author a blog is not a place to display already-settled answers, but a place to hold on to a question, explore it, and arrive at an unexpected answer. ✍️

2. You need a clear, arguable claim

The key reason this approach works, he says, is that every post makes some clear claim. He doesn't write posts that scatter fragmentary thoughts, or that read another piece and simply say "I agree." If a draft is so obvious that no reasonable person could disagree with it, he throws it out entirely.

"Every post on my blog makes an argument."

"If I've written a draft that no reasonable person could disagree with, I bin the draft."

The standard that a post must be at least somewhat contentious acts as a kind of forcing function for the author. It makes him think about which of his positions is most interesting, and it makes him research enough to answer the obvious counterarguments. This does not mean stating opinions forcefully; it means not settling for vague impressions and instead offering a claim and evidence that can be examined.

This contrasts with the common advice to treat a blog as a space for free self-expression. The author doesn't reject free self-expression itself. A blog is a space where you can write whatever you want, and if you want to write that way, he accepts that it's fine. He just thinks "write however you feel" is less helpful than people assume for writing consistently.

Before entering the tech industry he was a philosophy graduate student, and before that a poet. Writing poetry taught him that constraints actually make writing easier. A poet has to choose the next word at every moment, and formal structures like meter or rhyme dramatically shrink the range of possible words. Choosing within a fixed structure is easier than choosing from every word in English with no constraints at all.

The same is true of blogging. The constraint of having to write about one specific, potentially contentious point doesn't make writing harder — it actually makes it easier to pick topics and keep producing posts.

3. Writing exposes the gaps in your thinking

The author considers writing the best way to think clearly about a topic. When you only turn ideas over in your head, it's easy to believe you understand something well. But when you have to compress it into sentences, exactly how far your understanding goes and where it breaks down becomes visible.

"Writing is the best way to think clearly about a topic."

He says he often stops mid-post, looks at the sentence he just wrote, and thinks:

"Wait, this can't actually be right?"

"Is this really true?"

Those moments are precisely the value of writing. Turning something you vaguely assumed you knew into sentences exposes flawed premises or missing evidence. By the time you finish a post you usually understand the topic far more sharply than when you started, which is why the concluding paragraph ends up much better than the introduction.

For that reason the author has developed the habit of going back to the opening paragraph and rewriting it immediately while drafting. Once you write all the way through, your initial framing or explanation is going to change into a better form anyway. It's also an attitude that frees you from the pressure to perfect the introduction before moving on. Think it all the way through first, then fix the beginning with the understanding you've newly gained.

4. Changing your position while researching is a good sign

The author frequently changes his mind while writing. In fact, several posts — on cooling data centers in space, prediction markets and insider trading, so-called "engineers who just say no," giving personalities to LLMs, AI detection, Tempo, AI impact research, and AI interpretability — ended with the opposite conclusion to the position he started with.

He sees this not as a bad thing but as a good sign.

"I think this is a good sign, and I hope it never stops."

If you research and think in order to write, your existing views should often change because of what you newly learn. It's the opposite of deciding a conclusion first and then collecting only the evidence that fits — you have to be prepared to revise the conclusion based on what the research shows.

This is why the author deliberately chooses topics he does not yet fully understand but wants to know about. Examples include:

• Steering vectors in LLMs and controlling model behavior
• Stripe's Tempo blockchain
• C2PA and watermarking of AI content
• Cooling data centers in a space environment
• Interaction models between users and systems
• The internal workings of LLM inference

Writing about these topics is clearly good for the author himself, because he learns a lot. But an important question follows: is it also good for the reader when someone still learning explains a topic publicly?

5. Is writing to learn irresponsible?

The author wonders whether he should only write about the organizational dynamics of tech companies or how to work in large codebases — areas he already knows well. Shouldn't the Luddites be left to historians, and blockchain to Web3 engineers?

But he offers three reasons why he thinks it's fine to cover topics he's still learning.

Beginners can give better introductory explanations

First, a beginner is sometimes better than an expert at giving an introduction to a field. Experts tend to overestimate what the general public knows, and because they have so deeply internalized why their field matters, they may struggle to explain it in accessible terms.

The author cites as one reason his explanatory posts are valuable that he always explains "what problem was this originally trying to solve" before explaining a technical solution. It's a way of helping the reader understand why something became necessary before they learn its name or structural details.

If the conventional wisdom is wrong, even a little research is worthwhile

Second, on some topics the popular consensus can be plainly wrong. In those cases even a modest amount of research is enough to show why it's mistaken.

The posts he is particularly proud of include claims such as:

• That the widely circulated figure of 500 ml of water per LLM prompt is absurd
• That Apple's "Illusion of Thinking" paper was measuring persistence rather than reasoning ability as such
• That the real lifespan of GPUs is longer than the commonly cited three years
• That AI companies can achieve substantial margins on inference services

The value of these posts is not that "you can say anything even without being an expert." It's that even widely accepted numbers or interpretations need their evidence rechecked and publicly examined.

Be transparent about who you are and where your limits are

Third, the author tries to make clear on his blog who he is and what his background is. Even if he doesn't spell out the limits of his expertise in every post, his real name and résumé are available on his about page. So he thinks a careful reader is unlikely to mistake him for an expert in 19th-century British history, space physics, or LLM economics.

The core of this principle is the honesty of not pretending to be an expert. It's fine for a learner to write, but you must not blur your level of knowledge and background in a way that misleads readers.

6. Feedback is the powerful reward of writing in public

Even a post nobody reads is valuable enough as an exercise in organizing your thinking. But another reason writing is a great learning tool is that you can get feedback.

"The other big reason writing is a great learning tool is that you can get feedback on it."

That said, writing publicly requires a certain amount of thick skin — an attitude that doesn't collapse under hostile responses. On the internet people sometimes behave as if competing to produce the sharpest criticism or the harshest mockery. Making a clear, contentious claim about a topic you're still learning can draw an especially strong reaction.

"If having a stranger be cruel to you would ruin your whole day, you might be better off keeping your blog private or sharing it only with friends."

But writing privately doesn't mean you get no feedback at all. The author suggests getting feedback from an LLM. LLMs, like people, can give strange or useless feedback. He points out that OpenAI models in particular repeatedly recommend softening claims, adding caveats, and toning down phrasing, until you end up saying nothing at all. Sometimes the criticism itself is simply wrong.

Even so, for technical topics he finds LLMs useful for locating the parts he has genuinely misunderstood, and far kinder than the average Lobsters or Hacker News commenter. He adds that there's no need to obsess over crafting an elaborate review prompt. In his experience a simple request works well enough:

Writing "Please review:" and pasting the post gets you results nearly as good as a carefully engineered "review prompt."

7. How to tell whether good learning is happening

The author is glad and grateful that people enjoy reading his posts. But even back when he had no readers at all, the blog gave him a great deal. Writing was a way to think more clearly, an excuse to research topics he wanted to learn, and a channel for getting feedback from other people and from LLMs.

He recommends that anyone trying this approach check two things.

1. Does your thinking change often while you write?
   If it never changes, your research may not be sufficient. Research isn't just finding material that supports what you already believed — it's testing your own premises.

2. Is the conclusion of your first draft more accurate and better expressed than the introduction?
   If the conclusion hasn't become firmer and sharper than the start, it's possible you didn't actually learn anything in the process of writing. Such a draft can be discarded without hesitation.

"You should be changing your mind a lot as you write. If not, you're probably not doing enough research."

"The conclusion of your first draft should be much firmer and more expressive than the introduction."

8. Closing: from writing what you understand to writing in order to understand

This piece suggests you shouldn't see a blog only as a place to display authoritative answers. Instead you can use it as a space to honestly dig into what you don't yet know, make an arguable claim, and revise your own thinking through research and writing. 📝

The point is not to write while blithely pretending to know. It's to expose your limits, check your evidence, accept counterarguments and feedback, and reach a better understanding by the end of the post than you had at the start. The exercise is valuable even if you never publish, and in 2026 in particular you can also get kind, fast feedback on technical writing from an LLM.

The post ends with a preview of a related piece arguing that saying the obvious is surprisingly useful too. The idea is that we carry a lot of knowledge without being conscious of it, and a sentence that seems obvious can sharpen your sense of why you dislike something or which values matter to you.