Alyssa CoghlanSometimes I wonder how much of my AI skepticism is fuelled by the fact that a decent chunk of $dayjob is getting tools originally written for Linux to play nice with Windows dev environments, so a significant subset of the documented functionality doesn't work (with no written caveats), and plenty of the publicly available example projects don't work either (since they assume execution from a Linux environment, where WSL itself qualifies, but a Windows client to a WSL service doesn't)
Julia Evans@ancoghlan do you mean because the information the LLM is trained on doesn't apply so the LLM is very often wrong?
@b0rk Yeah, pretty much. There's an awful lot of documentation and code out there that doesn't describe its minimum version requirements or its platform or configuration limitations, so as soon as you deviate from "I'm running the most extensively used and/or documented version on the most common platform in the most common configuration", both LLMs and regular internet searches will feed you a lot of nonsense, but the LLM output often makes the problem less obvious.