Links, February 13, 2026

§I am Pro-Human

My friend Mike Lasch published this rallying cry in honor of the January 30th shutdown. I agree with him entirely on this.

§Tactile

An excellent piece by Mike Hoye about connection, which, being of a similar era and figuring out how I can convey some of these things to my kid, really hit home.

Every incremental loss is just an increment, a small thing. Even mentioning it feels more nostalgia-complaint than real, and things absolutely were not Better Back Then In My Day. But having grown up with that physicalism, a relationship with stuff that hummed and growled and warmed and hissed and scratched and clanked, having that experience dissolved down to a silent car you start with a button and the sterility of a computing that reduces you to pawing at a piece of glass if it’s even in the same room… I understand.

§On Building Communities in Public: Why I Chose Discourse Over Discord

Stuff happened with Discord and a lot of people I know are making a big deal about it. I have thoughts on the whole situation but they’re messy and there’s no good takeaway. This pice by Joan Westenburg on the actual Discourse site offers a good framing for anyone looking to create a community somewhere. It also takes a great stock about the nature of online communities and how to make sustainable and fair ones.

What we’ve built, in other words, is a system of tiny dictatorships masquerading as communities.

Discourse lacks real-time chat which is perhaps the big thing I’ve seen to help create very tightly-knit communities, but when I look through the list of thirty-plus discord servers I’m on, twenty-six of those could just as easily be a non-real-time forum for the reasons I’m there. Plus! Forums make it much easier to link to things which happened in the past – searching for something on Discord is a painful and often fruitless experience.

§FFmpeg Audio Batch

A GUI around ffmpeg’s audio capabilities:

FFAB is a GUI wrapper for FFmpeg that makes designing complex audio filter chains easily accessible to non-command-line users. Real-time preview. Drag & drop filters. Parallel processing, the works.

I’m very close to wrapping up a book recommendations post for Discworld, but at fourteen thousand words so far it may need some editing, even if it is only 350 words per book on average. Having read the entire forty-one book series over the past few years, I found this an interesting read:

Across Pratchett’s later novels, there was a clear and statistically significant decline in the diversity of adjectives he used. The richness of descriptive language gradually narrowed. This was not something a reader would necessarily notice, nor did it reflect a sudden deterioration in quality. Instead, it was a subtle, progressive change detectable only through detailed linguistic analysis.

They cite the twenty-second book, The Last Continent, as the turning point for the “preclinical” phase of Pratchett’s dementia; I didn’t care for that book particularly (mostly due to it being a Rincewind novel) but I found this interesting since if you average my personal ratings for the nineteen books before that one (Equal Rites through Jingo) and the nineteen books after (Carpe Jugulum through The Shepherd’s Crown), well: I wouldn’t bother writing a post recommending the first set (maybe something recommending Hogfather and Jingo), but it would be a lot easier to recommend the outstanding second set in isolation.

Perhaps a more interesting overall part of the study is the idea that there are some other dimensions on which we could detect preclinical dementia onset.


using LLMs means you endorse their effects

§Winning the wrong game

It doesn’t matter how good these systems are in reality, because that’s not what your boss cares about. “AI” is a tool to disenfranchise labor. That’s the job. If “AI” is actually more expensive than paying actual people actual wages that’s still a good investment for capital because it is about breaking up the structures, networks and organizations that help workers organize and fight for labor standards and fairer wages.

§The Enclosure feedback loop

And those questions are beautiful training data for those same models. Everything that used to be asked in a public forum, visible to everyone, is now a private confession to an LLM that tells you you’re smart for asking such insightful questions.

t is immediately clear that there are huge racial stereotypes in the outputs. It makes prompting embarrassing and hard to do, because you there’s a violence in the image you’re receiving, and you didn’t mean to make that depiction. Then it hypersexualized depictions of women unprompted. You could be prompting “women standing in a coffee shop” and with each iteration of the prompt, she’s losing more clothes.

§Debunking robot rights metaphysically, ethically, and legally

Another paper this week, well worth a read in its entirety:

But the danger is then to think that the essence of the model is in fact a real aspect of the original phenomenon.

Such move mistakes the map for the territory. Throughout history we have compared ourselves metaphorically with the most advanced technology of the time, not just as an epistemological tool, but also making the mistake to insert the mechanics of the technology into ontological claims.

§The Steam Machine Has Been Delayed Because Stupid Little Babies Can’t Stop Using AI To Write Their Emails

Those “memory and storage shortages”, if you haven’t heard about them, are a result of AI data centre usage—or, not even usage, but planned usage

§Vibe Coding Is Killing Open Source Software, Researchers Argue

Another summary of a study – third this week!

This won’t end well. “Vibe coding is not sustainable without open source,” Koren said. “You cannot just freeze the current state of OSS and live off of that. Projects need to be maintained, bugs fixed, security vulnerabilities patched. If OSS collapses, vibe coding will go down with it. I think we have to speak up and act now to stop that from happening.”