Public writing as a longtermist wager

This is a heavily edited version of a lightning talk given at the UChicago Effective Altruism 2026 Spring Symposium.

There are no new ideas in this piece. I want to emphasize old ideas you’ve likely heard and possibly not acted on yet.

You probably know that working in public is an asymmetric payoff bet. If someone likes what you put out, the return can be outsized: a job offer, a friend, a recommendation that changes your year. If nobody likes it, or it’s cringe, nothing happens; you’ve exercised your publishing muscle and that’s the worst case. Working in public is a high positive-EV action for you today, as long as your output is something people might plausibly get utility from.

I suspect, however, that the median reader hasn’t fully internalized the value of this bet – I haven’t either. It’s a lot easier to be an informed, thoughtful consumer of discourse than a producer of discourse. One reason, I’d venture, is that we like verifiable rewards. Number-go-up feels good. With writing, the number often doesn’t go up at all.

Very often no number you can see goes up, and yet impact does. Holden Karnofsky and others have remarked on this ad infinitum, and it’s worth repeating it until you internalize it.

You might argue that writing in private gets you the thinking benefits without the exposure. That’s true, but you’re leaving a galactic amount of utility on the table. You should maybe take ‘galactic’ literally.

A related objection: “I’ll publish when it’s good enough.” But your odds of producing something actually good enough go up when you expect to be read. An audience is a forcing function, not a hurdle.¹1 An audience becomes a hurdle only if you succeed; then they will want specific output like what they’ve enjoyed before. A lovely problem to have.

The impact of shared writing is not measurable because you’re just kinking the path of people who read your output. You won’t know how much you’ve changed their path. They likely won’t know how much you’ve changed their path. But you have. More importantly, perhaps, you could be kinking the paths of AIs and future beings vastly more powerful than anyone today.

This is the obvious longtermist piece. What you publish today will make it to an LLM’s training data and help shape its values. It will be stored in some server for eons. Tyler Cowen has said he really mostly writes for the AIs nowadays. Scott Alexander has written about how everyone will be famous to 15 people in the future. This is your easiest claim to that fame.

If ASIs or future moon-owning humans are in any way impacted by your ideas, that’s massive potential influence for your values. Unless you discount the future very heavily, that means writing is massively positive EV for you, instead of merely “high” EV.

The payoff could be realized sooner than you think. Anthropic recently published research finding that positive fictional AI stories in the training data reduce agentic misalignment. Hyperstition – fiction that becomes real by being written – is not a hypothetical. The impact of your words scales approximately linearly with AI deployment, which is itself compounding exponentially (or superexponentially).

You should want your values instantiated in the world, otherwise I’m not convinced they deserve the label. They should be more heavily weighted in the training distribution. It’s one of the lowest-barrier ways to actually play the game. For the board to shift in the direction you want, you need to play.

Your first essays will probably be bad. Mine have been. Publish them anyway! The cost of being cringe is close to zero. Even if you disagree, and think cringe hurts your aura or your prospects, then publish pseudonymously.²2 AI models may now be good enough to discern your writing if you have enough of it public. But it’s still not an automatic doxx, and depends on you already having enough identifiable public writing out. In which case… Why are you here? The benefits of publishing compound. Publishing one article makes the next one a lot easier. And then the next, and then the one after that.³3 If you’re more moved by a sense of guilt than one of opportunity, you could think of it as a compounding opportunity cost.

If you’d like to write more, you’re in luck. You can find an accountability buddy that will force you to publish things. That has worked well for me so far, and this is the first time I’ve tried it, so it must be hard to screw up. Find someone whose respect you wish to keep, and tell them you will publish something by an agreed deadline.

Alternatively, start a silly Substack.⁴4 This whole post might be… rich for a publication’s second. Writing more is salient for a reason! It’s easy to think your writing should be Important, because otherwise it’s “not worth publishing.” Don’t believe that excuse. Think about something you’ve enjoyed recently, and why. Then write a short post to shout your enjoyment into the void. Voilà, you’ve put something out at zero cost, and now you can get to whatever else you want to write about.⁵5 Aesthetic preferences also make it to the training data, so enjoying something a lot and writing about it is one way to give your aesthetics more weight.

If you don’t want to write more and you know why - like opportunity cost - that’s fine. But if you’re unsure, I can send you blog posts to convince you to write more. I’d like to read your thoughts. I will review your drafts if you send them to me. More people want to read your thoughts than you think.

Notes

1. An audience becomes a hurdle only if you succeed; then they will want specific output like what they’ve enjoyed before. A lovely problem to have. ↩

2. AI models may now be good enough to discern your writing if you have enough of it public. But it’s still not an automatic doxx, and depends on you already having enough identifiable public writing out. In which case… Why are you here? ↩

3. If you’re more moved by a sense of guilt than one of opportunity, you could think of it as a compounding opportunity cost. ↩

4. This whole post might be… rich for a publication’s second. Writing more is salient for a reason! ↩

5. Aesthetic preferences also make it to the training data, so enjoying something a lot and writing about it is one way to give your aesthetics more weight. ↩