Well there are guardrails from what I understood, including :

• executing commands (off by default)
• executing commands without user confirmation (off by default)

which are IMHO reasonable but if the person this happened to is right, there is no filesystem sandbox, e.g. limited solely to the project repository.

Can’t talk about AMD but I’m on NVIDIA and I always followed https://wiki.debian.org/NvidiaGraphicsDrivers and never had issues others seem to be having. I typically hear good things about AMD GPU support, on Debian and elsewhere so I’m surprised.

Now in practice IMHO GPU support doesn’t matter much for NAS, as you’re probably going headless (no monitor, mouse or keyboard). You probably though do want GPU instruction set support for transcoding but here again can’t advise for this brand of GPU. It should just be relying on e.g. https://trac.ffmpeg.org/wiki/Hardware/AMF

Finally I’m a Debian user and I’m quite familiar with setting it up, locally on remotely. I also made ISOs for RPi based on Raspbian so this post made me realize I never (at least I don’t remember) installed Debian headlessly, by that I mean booting on a computer with no OS all the way to getting a working ssh connection established on LAN or WiFi. I relied on Imager for RPi configuration or making my own ISO via a microSD card (using dd) but it made me curious about preseeding wiki.debian.org/DebianInstaller/Preseed so I might tinker with it via QEMU. Advices welcomed.

PS: based on few other comments, consider minidlna over more complex setups. Consider Wireguard over tailscale (or at least headscale for a version relying solely on your infrastructure) with e.g. wg-easy if you want to manage everything without 3rd parties.

As mentioned on another Lemmy server IMHO and as the vibe coder mentions in his video the main problem isn’t that LLMs suck in general (hallucinations, ecological costs, lack of openness for the most popular ones, performance, etc) but rather that this specific tool made by Google does not sandbox anything by default.

Copilote

Is it the Mexican version? /s

IMHO fix whatever you can, donate it all locally (HackerSpace, RepairCafe, Linux non-profit, etc) as there are quite a few people dedicated to refurbishing computers for schools, people who need a computer to find work, etc.

Then for the tinkering aspect, keep one, that’s enough.

Honestly even 1 isn’t really required. Pretty much everything listed here can be done more efficiently without an actual physical computer :

• your current computer can be a server, just turn off the screen or even accept (which I’d argue is a fair assumption) that at night it will be off. If you want external access put WireGuard or another VPN on it.
• Want to test distributions or anything else? QEMU or containers, no need for actual hardware

Sounds absolutely stupid… and yet my (gaming) desktop (model CORSAIR ONE i180) remains untouched after nearly 6 years. I still play indies to AAA to VR with it. I still work with it, specifically VR prototyping, so dev.

If I were to give it away or use as a self-hosted server with GPU used on e.g Immich or video transcoding it would still do pretty well.

So…IMHO it’s not a bad take but damn I remembered I paid a LOT of money back then. As other pointed out if you can afford it, sure. If you are not a professional then probably not.