l'empathie mécanique @dpwiz@qoto.org

@travisfw IDK. Visual "code" doesn't scale well. Text is more succinct and easier to pattern-match.

@omgubuntu Th[e]Underbird

@travisfw No-code for what result?

@boilingsteam SmertTV - a romantic pale of death by consumption

Looks like this is the reason why Opus, the next-gen Xiph codec, was failed to overtake its predecessor Vorbis despide being pareto-superior to it.

The author of `stb_vorbis.h` that gets shipped far and wide found that they can't make the same for Opus due to licensing shenanigans.

https://nothings.org/stb/stb_opus.html

#gamedev #fail

@me Twitter-to-OPML? Yes, please.

@webbureaucrat Hm... That's almost exactly* the thing that made me read it in a first place.

*Mixed with shipping Python code at 2 AM.

@jonn I vaguely remember something like this. Care to post the link again?

But, isn't snowballing is the essence of strategy? Press the advantage while you can, so you can "can" later too 😄

*takes notes*

But what if...

#gamedev

Playing some old RTS games and oh my they had clunky controls!

Supreme Commander reigns supreme indeed in the UI department.

@whitequark Walled by a sign-up form. Great start, just great...

@reidrac@social.sdf.org @gilesgoat I use TabNine with a moderate success.

It doesn't pretend to be "your smart coding assistant" and literally a glorified autocomplete.

But it is hit-or-miss even with that. Sometimes it completes too much or picks up a wrong pattern. Sometimes I miss something that it got wrong and had to backtrack. Luckily it is only a line-sized chunk of code, not some "write me a screenful of boilerplate".

So it is net-positive, but marginally so.

Perhaps its main value for me is having a live probe in the state of the art free::beer code automation. If the thing starts to pump solid Haskell I'd be worried... or happy, IDK.

@gilesgoat @reidrac@social.sdf.org Coding in Python, JS or other popular language (i.e. where LLMs achieved network effect) is basically automating yourself away (as a coder) by pouring more data in their programming corpus.

If you like code and don't like debugging others' mindless slapdash - avoid mainstream. It is already doomed by its success.

@reidrac@social.sdf.org I've included a fallback trick so the system should try to detect build failure and recover with less features.
It appears to work on Hackage/Stackage, but somehow fails on a live project.

@LesterHammerer@mindly.social That's how you overdose

@reidrac@social.sdf.org do you have a recent libsdl?

@simonmic And, yes, the models are exactly hallucinating. For the same reasons the human brains do, for all we know at this time. No need to call in mysticism for this.

And no, generative models wouldn’t solve any crisis. But for journalistic integrity, don’t conflate a subclass with the whole thing! Of course AIs of the future will solve that. For the same reasons a human thought could (in theory) do it.