Allison Kaptur

If you've noticed a distinct dialect emerging in frontier AI models and struggled to understand what they're talking about, you're not crazy! We turned to psycholinguistics to investigate what makes the writing style so weird, and why this might be happening 🚨New preprint! 🧵

i spent all of my free time for a month on this. it's done. so i wrote about it.

I’m happy to see more people hopping mad about the democrats‘ text campaigns. Hopping mad is the correct amount of mad to be data4democracy.substack.com/p/hook-and-s...

Compression and LLMs are trying to solve the exact same problem: predicting what comes next. @anniesexton.com's interactive essay walks us through the basics of compression to reveal its surprising overlap with every language model you've ever used.

This is grim but I think it’s all correct

I’ve been noticing this also with Opus 5. It’s not just jargon I don’t know but jargon it invented, when I already have perfectly good jargon I was already using.

Moishe is so thoughtful and that really comes through in this piece (thereby proving his point!) tk.xyz/notes/fe9ecc...

Engineering management after the cost of code collapsed: “Investing in machine-checkable correctness is now among the highest-leverage infrastructure work an org can fund” karimjedda.com/engineering-...

“Legibility of effort“ is a great, useful framing from Nolen: eieio.games/blog//blog/l...

This is a really good writeup about keeping a mental model of your code in the age of LLMs: yes, it’s wonderful to have great mental model of your entire system, but in practice almost everyone with a coding job is working in a system where that’s not possible. www.seangoedecke.com/in-defense-o...

here's the other half of my post: why AI token cost discussions are basically just noise and have little relevance to whether businesses should adopt AI tools for software development

Great read surfingcomplexity.blog/2026/02/08/n...

Here’s a very similar article about astrophysics: if you do science that no one understands, what have you accomplished? (Serious question, not rhetorical, that all these scientists are wrestling.)

The mental model problem has been an interesting one for me too. With coding tools, when and how do I refresh my mental model of the existing system and build my mental model of how my changes fit into it? www.scattered-thoughts.net/writing/arti...

Very thankful for this letter to Becky Pepper-Jackson (the 16 year-old trans girl SCOTUS ruled against yesterday) from James Dale, a gay man whose expulsion from the Boy Scouts was upheld by the Supreme Court in 2000.

Drawing the line on AI writing, literally notes.archie-hall.com/p/drawing-the-… #AI #writing

it has come to my attention that busytown is anti pig

Boring but functional post on something I think more leaders need to address: guidelines for respectful use of AI medium.com/@skamille/gu...

This is a smart take - there’s a way to use AI to turbocharge your growth, and a way to use it so you never learn anything. I’ve also seen “problem decomposition” remain a huge part of the job and one not really accelerated by AI.

"Absent LLMs, it is presumed that of the reader and the writer, it is the writer that has undertaken the greater intellectual exertion." This is a pithy explanation of why I dislike LLM "slop"

The engineer-manager pendulum model no longer accurately reflects how engineering leadership operates. Most of us aren’t just an IC or a manager. Technology has forced us to live somewhere in the middle. I wrote about what this actually looks like now: www.modernleader.is/p/pendulum-r...

And just for clarification, "learn to write well" doesn't mean "learn to write with big words and complex grammar" It means: learn to... -articulate an argument and support it with sound evidence and logical reasoning -organize your writing so the reader can easily understand complex ideas

This resonates: blog.val.town/vibe-code

We spent two months talking with RC alums and thinking deeply about how LLMs are changing programming and learning. Here’s what we learned, and how we're currently thinking about AI at RC: t.co/ddFj486ch8

This is shocking. Pure totalitarianism from DHS, and also blatantly unconstitutional. www.newyorker.com/news/the-led...

“how many politicians have you seen doing photo ops in nurse scrubs?”

Holy crap this article

One thing Democratic leadership could do right now is to name an alternate HHS secretary--someone to provide ongoing public updates and health information. And do it for other departments too. Start showing voters what a Democratic government would look like--press conferences, speeches, all of it.

I keep thinking about this www.liberalcurrents.com/once-and-fut...