They have, through their efforts, earned the right for institutions to take the Internet seriously, and you should honor their (continuing or otherwise) memory and efforts by being taken seriously, including adopting low-cost cultural dress of corridors of power occasionally.

"You can get samizdat'ed into corridors of power regardless of how little time or intellectual effort you put into packaging! Look at our extensive history of this!". Yes and look how many years of grinding in the wilderness it took your forebears to earn a first crumb of cred.

Zohran Mamdani: He votes for activists, not you

Part of the Essay Meta is understanding that essays are underexplored space in the intersection of a) the production function for them is available to people with relatively little institutional power and b) institutions are allowed to consume essays, including legibly.

. your opening move should not be concession!.

That is not the only valid goal to have in the portion of life which one writes! I wrote essays to keep literate in English and have at least one place I could share what I was working on, at least back in 2006. But if you're *extremely aware* of an influence/status game. .

But a website where people will post brief essays from time to time *with the goal of influencing the corridors of power* is a) not "just" a website and b) undermines its own effectiveness by adopting the "blog" label.

(No, really.).

RT @ComplexSports: We pulled up to the @WSL’s 2025 Lexus US Open of Surfing to catch waves, vibes, and the culture. Powered by @Lexus htt….

“Would you want big tech doing that?!”. Could be Harvard medical running a population-wide sleep survey on a budget too low for tech to even meter, tracking against clinical endpoints and surveys for a subsample sized in six figures.

Meanwhile, some of the largest firms in capitalism could, with a DB query, trivially graph my sleep times for 99% of days going back to 2008. … But nobody actually does that. Not where our civilization put its competence points.

Also true with regards to passive data collection by tech companies. Someone one accidentally diagnosed me with a sleep disorder by drawing a histogram of HN post times over a 24 hour day then failing to identify hours where I was not posting.

It is interesting, in a sense alluded to in Inadequate Equilibria, that we put our competence points into remembering for all time which YouTube videos I liked in the late 2000s but that the only persistent medical records I have are those I kept on paper.

Later, after actually implementing the business logic, it removed the TKTK comment. And, above the old hallucinated code, added:. # Maintaining legacy logic for backwards compatibility.

And then in still another vignette, in stubbing out new functionality, it totally hallucinated some business logic. I went in, added a TKTK comment to replace with the real logic, and then had it implement an affordance script I like. (On running tests, count TKTK/etc + report.).

OK boss it is done and I also took the liberty of implementing all the numeric comparators and boolean operators. Me: … I did not expect to appreciate the initiative today.

And then on another occasion, when I asked for a particular data file to be able to specify a consequence if another data file matched or did not match a tag or numeric, I said “Let’s be generous to the user and accept !, ~ or NOT {tag} to define a tag.”. *whirs*.

RT @PardonMyTake: Tuesday night max woke Big Cat up with a flashlight at 2am because he thought we were going to get sued. @forthepeople ht….

And that it missed one instance of it. Me: You missed one in … .Claude: Oh sorry about that let me read that file. Me: In the future, when you are refactoring in this fashion, recursively grepping for…. *sudden flashback to when I was 24 and getting a polite talking to*.

Some are surprisingly human, like the time it successfully refactored blah_id, which was not an auto incrementing Rails ID but rather an alphanumeric string (and should have been blah_code for that reason) in 36 places, including successfully adjusting spec, tests, and call sites.

One of the emerging intuitions for LLM-assisted development is which consequences you’ll have to spell out to the agent, which you can just allude to, and which it will figure out on its own. I have had many surprises this week.

(Unfortunately, it seems evident based on engagement numbers that far fewer people see tweets about the podcast than I'd expect given e.g. number of followers and reach for other tweets. Grumble grumble, Twitter product decision, grumble grumble.).

In the meanwhile, did you know that I'm much better at publishing this podcast than tweeting about it? Yep, 50+ weekly episodes at this point. If you want to reliably know when the new ones come live, get on a feed at your service of choice versus hoping a tweet reaches you.