Weblog about issues you do not perceive but

Each submit I publish represents not less than two issues I’ve discovered: the factor that prompted me to write down the submit, and the factor I discovered in the midst of writing it. If I don’t be taught something new whereas I’m writing, it’s not attention-grabbing sufficient to publish.

Usually I be taught far more than two issues. For example, in my o3 geoguessr submit, I began out with the concept most AI prompts most likely don’t work, and I ended up studying that newer OpenAI fashions have misplaced o3’s capacity to geolocate. That’s attention-grabbing! In my most up-to-date submit on C2PA, I began out with the concept C2PA requires near-universal adoption, however I discovered a ton of issues about PKI, managing non-public keys on native units, how C2PA truly works, and so forth. In my submit on the Luddites, I began out with the concept the Luddite motion was essentially decentralized, however ended up fascinated by Luddite tradition (which was way more elitist, misogynist, and violent than the pop-Luddism books describe). I may do that for each single submit on the weblog.

Taking a place

I feel the core motive this works is that each single one in all my weblog posts argues some extent. I by no means publish a submit that simply provides some scattered ideas on a subject, or a submit that solely says “sure, I agree with this different article”. If I write a draft that no one wise may disagree with, I scrap the draft. Ensuring that every thing I write is not less than minimally controversial is a forcing operate: it forces me to consider what essentially the most attention-grabbing a part of my place is, and it forces me to do sufficient analysis to defend it in opposition to the plain criticisms.

That is opposite to lots of recommendation I examine running a blog, which inspires the aspiring blogger to deal with their posts as a type of unstructured self-expression. If unstructured self-expression is what you need to do, that’s cool. The purpose of getting a weblog is that you simply get to write down what you need. Nevertheless, this recommendation isn’t as useful because it sounds.

Earlier than I used to be in tech, I used to be a philosophy grad pupil. However earlier than that, I used to be a poet. One factor you be taught while you attempt to write poetry is that it’s means simpler to write down to a restrictive construction than it’s to easily “write what you are feeling”. This needs to be apparent while you truly give it some thought. The duty of a poet is to repeatedly select the following phrase. Writing to a construction (sometimes rhyme or meter) narrows that option to a small set of phrases, as a substitute of all the English language. It’s the identical with running a blog. Forcing your self to write down about particular, potentially-controversial factors makes persistently writing simpler, not more durable.

Writing, pondering, and analysis

Writing is the easiest way to assume clearly a few matter. It’s straightforward to consider you perceive one thing while you’re simply turning it over in your head. When it’s important to condense that down into phrases, you discover out precisely how a lot you do or don’t perceive. I’m continuously having moments the place I kind one thing, cease myself, and assume “wait, that may’t truly be proper”, or “is that actually true?”

By the point I write my approach to the top of the submit, I’m often pondering a lot extra clearly in regards to the matter that my conclusion paragraph is means higher than my introduction. Actually, I’ve picked up the behavior of going again and instantly rewriting the primary paragraph as a part of my first-draft course of, as a result of I do know I’m going to finish up doing it anyway.

I additionally change my thoughts so much whereas I write. Listed here are a bunch of examples of posts the place I started writing them with the other opinion to the one which ultimately made it into the submit. I feel it is a good signal, and I hope I by no means cease doing it. Try to be researching and fascinated with each submit you write, and which means you need to steadily be taught new issues that change your thoughts.

Due to all this, I intentionally select to write down weblog posts about issues I don’t but fairly perceive however want to, like LLM steering, Stripe’s Tempo blockchain, C2PA and watermarking, area cooling, interplay fashions, LLM inference internals, and so forth. That is nice for me, as a result of I be taught so much. Is it nice for my readers?

Is running a blog to be taught irresponsible?

I generally fear that I ought to solely be writing about areas I already know very nicely, like tech firm dynamics or working in giant codebases, fairly than presenting myself as an authority on fields I’m truly nonetheless studying. Ought to I let historians of the Luddites write about Luddism, Web3 engineers write about blockchains, and so forth? I feel that is acceptable for 3 causes.

First, it’s generally simpler for a newbie to write down an introduction to a subject than for an professional. Consultants routinely overestimate the information of most of the people, and have usually internalized the the explanation why their subject is essential so deeply that they wrestle to precise them. I feel my explainer posts are useful as a result of I at all times spend the primary chunk of the submit speaking about what the unique downside is earlier than I get into the technical answer.

Second, generally the general public consensus on a subject is simply plain improper, to the purpose the place even slightly little bit of analysis is sufficient to display why. A lot of my posts I’m proudest of have been alongside these strains: arguing that the “500ml per immediate” water utilization determine for LLMs was ludicrous, or that the favored Apple “Phantasm of Pondering” paper was monitoring persistence, not reasoning, that GPUs stay longer than three years and the AI firms have giant revenue margins on inference, and so forth.

Third, I attempt to make it clear on my weblog who I’m and what my credentials truly are. Even when it’s not explicitly described within the submit, I’ve my actual identify and resume out there on my /about web page, so I don’t assume a cautious reader could possibly be simply fooled into pondering I’m an professional on Nineteenth-century England or area physics or LLM economics or something like that.

Suggestions

Even when no one reads what you write, writing remains to be self-discipline for getting your ideas so as. However one other huge motive why writing is a superb studying instrument is that you would be able to get suggestions.

I feel it’s apparent why that is helpful, however I do need to make two factors about suggestions. First, for those who do make your posts public, you might want to have a reasonably thick pores and skin. Individuals on the web usually fall over themselves to provide you with essentially the most chopping criticism or the harshest dunk. This goes double for those who take my earlier recommendation and attempt to write posts that make a transparent, controversial level a few topic you’re studying. Should you’re the sort of individual whose complete day is ruined when a stranger is merciless to them, you may need to maintain your running a blog non-public or solely share it amongst associates.

Second, even when your running a blog is non-public, you may get suggestions from LLMs. Like people, LLMs will usually give junk suggestions. In my expertise, OpenAI fashions will at all times inform me to reasonable my claims or add caveats and hedges till I’m not saying something in any respect. Generally their criticism will probably be straight-up improper. However — notably about technical subjects — LLMs are nice at stating areas you’ve genuinely misunderstood, and so they’re far kinder than the common Lobsters or Hacker Information commenter.

Conclusion

I’m happy and grateful that individuals get pleasure from studying my posts, however even when no one did, I nonetheless obtained lots of worth out of running a blog. I write as a technique of pondering extra clearly, as an excuse to do analysis on subjects I need to study, and as a means of getting suggestions.

Should you’d prefer to strive it your self, I counsel waiting for these two issues. First, you have to be altering your thoughts so much as you write. If not, you most likely aren’t doing sufficient analysis. Second, your first draft’s conclusion needs to be a lot tighter and extra expressive than its introduction. If not, you most likely haven’t discovered something from the writing course of, which implies the draft may be scrapped.

I strongly advocate this apply to anybody with an curiosity in writing. You will note the advantages even for those who don’t publish any of your writing on the web, notably now that you would be able to get good technical suggestions by pasting your submit right into a LLM.