Claude loves seams, GPT 5.6 loves arms and slices.
Michael Hart
@hichaelmart
Joined March 2008
- Another study showing that LLMs are better at producing more "correct" code with JS than TS. Not by a huge margin, but still more fuel for the idea that types aren't the slam-dunk guardrails for LLMs that many ppl assert they are.What programming languages are coding agents best at? danluu.com/pl-tokens/
- Firstly, they don’t *have* this patent. They just filed it a few months ago. Secondly, to them: Good luck patenting code mode bro!
- We almost had a Gas Town 2.0 on our hands ppl, be thankful.Look, I wanted to call it Lairs. Like your personal AI lair. The agents were "minions". You built "gadgets". See "AI" is right there in the name: "l-ai-r" I even registered lairs[.]net. And you're mad about it being called an OS? Be thankful they didn't let me name it.
- The bottleneck's not gonna be the model. It's gonna be your tests. Your linting, formatting, typechecking. All your CI pipelines. I expect to see a whole bunch of iteration in this space (Still! I know. We've been doing this forever. But it's still too slow after all this time)
