I'd nuke and start over (or known-good prior snapshot).
Fortunately Claude helped me identify the problem, and helped me fix it all on my linux server. I'm a bit out of my depth with anything difficult on Linux, and AI has really been helping a lot.
LLM-decoding of Linux's advanced underpinnings is what finally allowed me to switch-over, full-time (after three decades of failed distro attempts)...
----
I didn't have internet for the past few days, and a local 30b model helped me move from "absolute beginner" to "knows just enough to be dangerous" in the TrueNAS ZFS file-system-management process. Entirely offline learning aquisition (with plenty of "errors," but it helped me decipher what so much of the technically-obtuse underpinnings are (i.e: wtf does this sub-sub-sub tickbox mean with "bullshit option XYZ"?)!
My Windows system started having problems resuming from hibernate, it would just end up restarting when I woke it, losing all my application state, etc. Claude diagnosed it to being an old Nvidia driver, told me to just install the latest version, and now there are no more problems with hibernate since. I don't know if I could have figured that out on my own and I've been using Windows since v3.0. It would have taken me forever to narrow it down, and Claude did it in about 10 seconds.
I also recently switched to Linux on my server and a bunch of VMs, and it was a shaky start at first, but now that I have Claude to help, it's become pretty simple.
I'm only paying $20/mo for Claude, and I have to wonder - when they raise the price, will I follow? I think I have to at this point, but the question is how much will the raise the price, and how much am I willing to pay for it.
Essentially for free.
I think this is the biggest thing here. With p>0.05 ("95% of the time it's correct [for simple things]"), and an effective cost of nothing... this is why Anthropic can publish it's-so-obvious™ papers on "reducing junior humans" and they're still frontpage tech news. This is what is so scary: busywork is gone.
What do you do with a general populace that has nothing productive to do (and any attempts at make-work are easily defeated by lay-AI-vibecoders)?