[bslanej]

Have you tried using code generated by an LLM? It rarely runs at all until you fix it yourself.

It reminds me of how ezines would publish source codes to run exploits with slight errors so script kiddies couldn’t run them if they didn’t know how to fix them.

[nneonneo]

We’re talking about coding competitions here. These models have been fed the entirety of ACM, SPOJ, LeetCode, and whatever other competitions they can get their hands on. They are good at constrained coding competition tasks, and certainly strong enough to place highly in a competition meant for high school students.

[ToucanLoucan]

> Have you tried using code generated by an LLM? It rarely runs at all until you fix it yourself.

In College Humor's Brennan's exasperated tone:

THEN WHAT ARE WE DOING!?

Like Jesus fuck. I feel insane. We scraped the internet, broke tons of trust in our community, fed tons of code we didn't ask for into an industrial shredder, and worked out nonsense generators that can make awful code that barely, and apparently sometimes just doesn't work, burning enough electricity to power several small countries in the process and lighting billions of dollars on fire.

What, and I can't stress this enough, the fuck are we doing anymore. I swear to God the entire valley needs to be pushed into the ocean and humanity will lurch forward 200 years.

[bslanej]

That’s hyperbolic. An LLM gets you like… 90% of the way there. That’s much better than the previous baseline of 0%

[Deukhoofd]

To be entirely fair, that's the same experience I've had with some coworkers.