chpatrick

Okay, what's the great compiled alternative then?

I don't think writing your frontend in Rust instead of TypeScript or your computer vision pipeline in Java would be better at all. We rewrote our frontend in GHCJS (Haskell) at one point and it was a colossal waste of time.

Happily using both in production here, guess I'm just hallucinating.

What job are we talking about and why is TypeScript or typed Python actually bad at it?

As far as I know Watson was actually smoke and mirrors, relying on a ton of human hand written rules and didn't do any reasoning at all. It was a giant program written in PROLOG, not machine learning.

So you'll only be impressed if it's superintelligent and solves a problem no human has? That's a ridiculously high bar.

For the kind of work I'm using Python for (computer vision, ML), not really. The ecosystem isn't there and even when it's possible it would be much less productive for very little gain. Typed Python actually works quite well in my experience. We do use C++ for some hand-written things that need to be fast or use libraries like CGAL, but it has a lot of disadvantages like the lack of a REPL, slow compile times and bad error messages.

I disagree and I've been using Haskell professionally for ten years so I know what I'm talking about when it comes to types. Typed Python isn't perfect but it's totally workable with medium sized projects and gives you access to a great ecosystem.

None of those things are to do with typing. Python is slow because it's interpreted, bad at parallelism because of the GIL (which is going away), and bad at deployment because of the messy ecosystem and build tools (although it's pretty good with Nix). Conversely other languages aren't good at those because of static typing.

It is a great choice though for many problems where performance isn't critical (or you can hand the hard work off to a non-Python library like Numpy or Torch). Typing just makes it even better.

That was more because of Java than static typing.

You generally don't write Python if you want it to be really fast anyway (non-Python parts like numpy notwithstanding).

To be pedantic compiled languages only check types at compile time as well. If you have a C library that takes void* then it can easily go wrong at runtime.

Would you? Why?

Python has a great experience for a bunch of tasks and with typing you get the developer experience and reliability as well.

Because Python has a lot of things it's great at (numeric stuff, ML, computer vision, scripting), and with types you can actually rely on it to work. It's the best of both worlds.

I went from mypy to pyright to basedpyright and just started checking out pyrefly (the OP), and it's very promising. It's written in Rust so it's very efficient.

Is it? If you're the only user then you care about latency more than throughput.

Poorly optimized apps need big batteries.

Whatever you say.

It's not new scientific ground but a machine beating a challenging computer science problem unassisted is a big deal. If they can do that then there are a lot of other challenging things they can do.

It doesn't say anywhere that Gemini used any of those things at ICPC, or that it used more real-world time than the humans.

Also, who cares? It's a self contained non-human system that could solve an ICPC problem it hasn't seen before on its own, which hasn't been achieved before.

If there was a savant human contestant with photographic memory who could remember every previous ICPC problem verbatim and can think really fast you wouldn't say they're cheating, just that they're really smart. Same here.

If there was a man behind the curtain that was somehow making this not an AI achievement then you would have a point, but there isn't.

Humans are allowed to look up and learn from as many previous problems as they want before the competition. The AI is also trained on many previous problems before the competition. What's the difference?

I mean strong human contestants would also know a lot of similar problems, I'm not seeing how it's fundamentally different or not a meaningful achievement.

Where is this magic ICPC competition answers database that they're using?

Don't worry, they're just stochastic parrots copying answers from Stack Overflow. ;)

To be fair Tesla was in a very primitive state when he took over.

Why wouldn't they just take a carriage pulled by a horse?

Hence "allegedly", but I think it's pretty telling that the people who work at Google and Meta are okay with this.

The Orbán government here in Hungary is one of the biggest ad spenders in the EU. You literally can't open a YouTube video without seeing propaganda with just plain lies, increasingly with AI-generated video. I find it really hypocritical that these allegedly progressive companies are willing to sell millions of dollars of brainwashing to the most hateful toxic regime in the EU.

This article doesn't mention "slop" at all.

They're linked but they're very different. Speaking from personal experience, It's a whole different task to solve an engineering problem that's been assigned to you where you need to break it down and reason your way to a solution, vs. coming up with something brand new like a song or a piece of art where there's no guidance. It's just a very different use of your brain.