Link Dump: Origins

March 31, 2026 at 1:34 AM (UTC)

I have not been writing much lately, it's true. It's almost been a full month since I talked about dumping e-Reader cards!

Truth is, I haven't had a lot to say, and/or the thoughts aren't coming together right.

But I have been reading. So let me share some stuff with you!

Here's a juicy one to start out: Epistemological Fault Lines Between Human and Artificial Intelligence. If you take away nothing else from this post, take away that "epistemia" is a word you can use to describe that thing where people read what large language models generate and believe it instead of critically applying judgment, because that's hard work.

Also, A.I. Isn't People. Hilarious takedown of a pervasive idea that LLMs constitute some sort of new kind of being, and a reminder of how much smarter a toddler is than computers.

AI hallucinates because it’s trained to fake answers it doesn’t know should be on the other end of any link you write that says something to the effect of "LLMs are literally made to bullshit." It almost was in a piece I was writing but urgh, it is still not coming together.

A sufficiently detailed spec is code is a wonderful counterpoint to the idea that somehow, endlessly refining a Markdown file until an LLM can generate something useful from it, is better than, you know, very specifically specifying what you want in code.

I never thought I'd be sharing a LinkedIn post but Christine Haskell, Ph.D.’s post on the Grammarly incident is beautiful just for how much truth it crams into four measly paragraphs.

I apologize to everyone for the horrifying header image of Sam Altman looking very altmanly and vaguely into the distance in "CEO Said A Thing!" Journalism, but it makes such a good point about what's wrong with all these headlines that keep telling us what people who have nothing to say just said.

While we're talking about our ridiculous tendencies to even care about what people like Altman say, I want to recommend Ghost in the Machine, the still-pretty-brand-new documentary about the really kind of bad people and thinking behind all the worst shit about "AI". Amnesty International hosted a Q&A with the cast which is also great watching.

Just today, Taggart has a really heartfelt, honest, and real look at his experiences with 2026's Claude Code in I used AI. It worked. I hated it.. The infosec.exchange thread where I discovered it is also beautiful, except for the replybomber. You can just mute them.

Finally, Mario Zechner, who made the Pi coding agent that backs the blight on society named OpenClaw, has some good shit to say in his Thoughts on slowing the fuck down.

I'll have more to write sometime soon, I hope. In the meantime, thanks for reading.