My example: I installed Debian 13 recently. I installed on the second SSD of my laptop, so I can dual boot and keep working with Debian 11 on the first SSD.

I encrypt my disks. Debian 13 can use the hardware encryption of my Samsung SSD, 11 didn't. The installer offered me the option and I accepted it. That nearly bricked the SSD because of (I'm not totally sure) a mismatch between the block size of the file system and the block size required by the SSD encryption. The installer should have made a check and at least warned me. It did nothing of that and the laptop didn't boot. I couldn't even change the partitions on that disk. It enforced its encryption and refused to do anything. I appreciate that but it left me without my disk. I asked questions to either chatgtp or Claude, found the problem and after a few attempts I got the right sequence of commands to unblock the SSD and get an empty one. I reverted to the standard OS based encryption and all is well now. I would have had to dig deeply into forums and learn the meaning of those commands. AI saved me a lot of time. Is this a Linux only thing or a Windows installer would have made the same mistake? No idea.

LLMs have probably trawled through ArchWiki+StackOverflow and can enough content to help you debug your system. That plus a few “are you sure” responses to LLM hallucinations have gotten me far.

So much Linux advice on the web is woefully outdated. Like answers for Ubuntu 8 still sometimes come up high in my search results. True that some things are still the same, but not many, especially pre-systemd.

I write in the prompt the distro and its version and so far no much problem with old answers. If anything, it made me realise that some knowledge I had is outdated.

I made a Gemini Gem with some of my preferences (vi not nano, Ubuntu LTS, my hardware) and it's been quite efficient and on-point. A few times it forgets to check ancillary contexts and I have to feather things back on course.

Enough things have changed in Linuxland in recent years that some of those diversions, as you pointed out, helped me work from old knowledge to new knowledge.

I had to fix a kiosk style linux desktop that had suddenly changed its touch input behaviour from touch to cursor today. ChatGPT gave me the steps immediately for troubleshooting, wrote me a udev rule and explained the potential reasons it could have happened and offered to walk me through the process of isolation. I suspect 99% of user problems in Linux can be solved this way.

There's a catch-22 in that the average Linux user is pretty loudly opposed to AI and so it appears hard for pro-AI software to gain footing in the space. Granted, most AI tools are Linux-friendly ATM, but my uneducated guess is that a larger % of Apple/Windows users use AI daily than Linux users

Agreed. We, the Linux crowd, are pretty big on understanding what our machines are doing, which is hard with LLMs. But I'd wager that a distribution that brings a local open source LLM for system administration tasks would find interest. Train it on man-pages and off you go ;)

Yes, though the idea would be to make Linux more viable for regular folks, and a sysad on their side should help.