They had originally planned on june and my part is in so…. Me too.

Rust is fine if you need it but most things don’t and ocaml has more convenient abstractions for “regular” work. If you need rust you need rust but ocaml isn’t that far off in perf.

Haskell type system is too expressive, it itself becomes a place for the agent to make mistakes and get bogged down.

* Other “pure” hindley-milner languages are tied but among them ocaml has some particular strengths that I’m sure others will discuss.

Neither are anything like a solo effort in the way these examples are, and it's hard to say what they'd be like without the long individual involvement.

The fact that his writings are still so good and useful and interesting speaks incredibly highly of his powers of perception and skill as a writer. But it makes them nearly useless in a "how ought we to be" conversation, which is one of the main things his works engage with. So it really is a huge problem with his entire project.

They love htmx! Claude will independently suggest using it instead of more general js frameworks for smaller projects. A gpt model told me, and I wrote this down because it was striking, "htmx won the argument it was making."

It's the same thing with rust. I'm an ocaml expert and ocaml genuinely fits a lot of the problems I work with better regardless of that. But because of when the models were trained and what they were trained on, a certain sort of alt-consensus opinion is frozen into them, and it believes rust is a better choice for nearly everything!

I think you're good for the next few years at least. It's an open question what happens after that dead internet theory recursive loop of llms trained on llm output etc so who knows you might be good forever.

Is it not "legitimate" in the same way to kill terrorists?

This is already an extreme perversion of justice all on its own. Damaging military equipment is absolutely outside any useful definition of terrorism. This is simply armed conflict.

But the idea they’d be better than a bigger model is cope, you’re pretty much always better off running the biggest one you can bring to bear within your constraints.

The way vision and language models converge into the same geometric space should be extremely alarming for the "you don't need global knowledge for local tasks" type dreams.

And to be clear I'm not saying that smaller models don't or can't work well, or that we shouldn't be heading in this direction. And it's not quite the case that broad knowledge is strictly necessary. But it never seems to be negative! And so far it is the best way we've found to do... everything. Small models are good to the extent they are like big models, not to the extent that they are small.

I don't believe this very strongly. Again it's a contract issue not a moral one. But the metaphor still holds I think.

It reminds me of matt levine's reframing of insider trading where it's not about fairness it's about theft. You're supposed to get secret insights and use them to get an edge. What you can't do is get an edge for yourself with secret insights that your employer got.

So it roughly comes down to "that data is valuable and rightfully belongs to the originating company." Which then makes this a contract diligence type situation.