hell is empty and the demons are here: https://github.com/dbreunig/whenwords

[cap_ybarra@beige.party]

@SnoopJ i'm announcing my new library: whenwordswords

it's just a the idea for a prompt that says "write a good library and some specs"

you're welcome, open source community.

[gsuberland@chaos.social]

@SnoopJ this made me angry

[ireneista@adhd.irenes.space]

@munin @SnoopJ @nullpo may the void take these asshats, indeed

[snoopj@hachyderm.io]

@gsuberland I would simply write an unambiguous and complete specification of the behavior I wanted.

It would be easy for me.

(/s)

[arrjay@tacobelllabs.net]

@SnoopJ there is more to LaCroix than this

but I really wanna call it LaCroix Software

[brouhaha@mastodon.social]

@SnoopJ @gsuberland
[insert Sean Bean "one does not simply" meme here]

[rune@social.sound-city.dk]

@SnoopJ I do wonder how likely LLMs are to output code that ONLY works for the specific test cases and no other inputs

[joepie91@fedi.slightly.tech]

@SnoopJ "Instead of containing the code, this repository contains instructions which may or may not result in that code depending on unpredictable factors!" sure is a thing to advertise with, huh

[joepie91@fedi.slightly.tech]

@rune @SnoopJ Not even that, because the test code is also generated so quite likely at least some of them are not even going to match the testcase spec

[rune@social.sound-city.dk]

@joepie91 @SnoopJ at least that part could be reasonable easily fixed with a testing framework/tool and using a preexisting format to parse the test cases

But I guess until they do that it could be either or both that do not work as intended

[snoopj@hachyderm.io]

@joepie91 why worry about what model was used to consume the input? The models always produce the same outputs given the same inputs. And it works today, with the models they checked, and the models only ever make forward progress on all tasks!

(and the models will be available at their current price point forever!)