In my experience, AI seems like it’s helping debug problems, but it’s very hard to tell when fictional information starts being added. I’ve wasted a lot of time trying AI suggested solutions that I only realized were pointless when I started asking questions like ”I think this distro is missing a package, could that be the problem?” It would agree and tell me a specific package to download. I’d then ask “could iptables be the problem?” It would agree and give me a specific configuration to change.

LLMs can be useful, but I haven’t found a way to use them where I’d be confident in using it to solve technical problems I didn’t already deeply understand.