The worst thing is that I know there's something interesting in there but I'm too arsed to take the time and decode what data the author must have used to generate that text.
Exactly, it would be so much better if they shared the prompts instead of the output, at least then we'd be able to make some sense of what they were thinking about.
Unfortunately the prompt may have been something like "investigate this bug and file an issue when you think you understand what's going on". The "better just to share the prompt" heuristic falls down when the actual underlying intellectual work was also done by the machine.
(Feel free to pretend that I used some phrase other than "intellectual work" if you dislike seeing it used to describe something done by AI.)
Yeah in that case just closing it and moving on would be better IMO. A sufficiently large fraction of the time the issues with vibeslopped code and/or analysis are bad enough once you start digging into it that it's not worth spending time on. My heuristic would be to simply ignore contributions derived from lazy prompts. I don't think there are very many healthy babies in that bath water.
So having the prompt available would be useful even in the degenerate case.
Maybe it's not going to get any better. LLMs were trained to do two things:
- Mimic human language (and logic, since that's part of what we express with language). This includes code written by humans.
- Write code to solve problems. The figure of merit here is solving the problem, not reproducing anything human.
I'm not an expert on LLMs, but it's not obvious that the human reasoning they are fitting in the first case is going to be anything like the problem solving required in the second case. It was trained to get itself out of a problem, not to do it in a way that a human would relate to.
And we're already running out of human data to train on, while synthetic data has no limit. Bug reports in the future might amount to "fix this because then I'll get a cookie".
A lot of it is that AI can talk with different "voice" but that requires giving it cues, and that costs money and expands context, so people that don't know, don't, and those that do have no incentive to
> Headline. The crash is real and reproducible.