Probably depends a lot on what programming language you're generating, though. Python is notorious for needing tons of asserts and unit tests to ensure correctness, and is frequently used by programming beginners as well as for prototyping, where code quality isn't looked after as thoroughly.

I also have to say, though, whether a code snippet is accepted or not always felt like a bad metric. I frequently see folks accepting a code snippet and then fixing it up inline. Or worse, checking in that code snippet and then a few weeks later when someone has to read that code, then it has to be cleaned up.

I would have guessed that my mother that said to me that stealing doesn't pay off in the long term but I don't know.

worse?

they've always been fucking terrible.

coding assistants in general are annoying and get in the way more than they help.

For people who hate LLM, do you hate it because it performs badly and failed to help you? Or because of other reason?

if it has failed to help you, what have you tried to use it for? And what is your expectation?

Who would have thought...Oh wait, any sane person would. Don't move, you might burst it.

I mean it's kind of obvious... they are giving their LLMs simulators. access to test etc..., IE chat gpt can run code in a python environment and detect errors. but obviously it can't know what the intention is, so it's inevitably going to stop when it gets it's first "working" result.

of course I'm sure further issues will come from incestuous code... IE AIs train on all publicly listed github code.

Vibe coders begin working on a lot of "projects" that they upload to github. now new AI can pick up all the mistakes of it's predicesors on top of making it's new ones.

A task that might have taken five hours assisted by AI, and perhaps ten hours without it, is now more commonly taking seven or eight hours, or even longer.

What kind of work do they do?

in my role as CEO of Carrington Labs, a provider of predictive-analytics risk models for lenders. My team has a sandbox where we create, deploy, and run AI-generated code without a human in the loop. We use them to extract useful features for model construction, a natural-selection approach to feature development.

I wonder what I have to imagine this is doing and how. How do they interface with the loop-without-a-human?

Either way, they do seem to have a (small, narrow) systematic test case and the product variance to be useful at least anecdotally/for a sample case.

As opposed to the older ones working perfectly well?

/s

Sorry for the self-reference but... https://medium.com/@nandofm/ai-the-danger-of-endogamic-programming-8114b90dffb2

I use LLM-generated code extensively in my role as CEO of Carrington Labs, a provider of predictive-analytics risk models for lenders.

Well you know where not to buy now...

Paint huffer surprised when other paint huffers are happy to accept any old solvent.

GiGo.

Correction: Newer versions of ChatGPT (GPT-5.x) are failing in insidious ways. The article has no mention of the other popular services or the dozens of open source coding assist AI models (e.g. Qwen, gpt-oss, etc).

The open source stuff is amazing and gets better just as quickly as the big AI options. Yet they're boring so they don't make the news.

Not failing, just Skynet doing its work.

is it just me or does LLMS feel like its slowly collapsing