

And yet, today, Taiwan is a sovereign state and wants to keep it that way, which is the only thing that matters here. This is pretty straightforward.


And yet, today, Taiwan is a sovereign state and wants to keep it that way, which is the only thing that matters here. This is pretty straightforward.


It’s because people get mislead by the “agent”, assuming there’s something actually intelligent at the other end, able to act like they would, just… Automated.


What a clown. In what world do you require an LLM to check for duplicate pictures??
I mean, good on him for realizing that LLMs are dogshit. But still.


How! For me it’s more like: “you get up to cook dinner and realize you’ve already snacked twice your calorie budget”


Eh… Not really. Qemu does a really good job with VM virtualizarion.
I believe I could easily build containers instead of VMs from the nix config, but I actually do like having a full VM: since it’s running a full OS instead of an app, all the usual nix tooling just works on it.
Also: In my day job, I actually have to deal quite a bit with containers (and kubernetes), and I just… don’t like it.


I’ll DM you… Not sire I want to link those two accounts publicly 😄


Zero.
About 35 NixOS VMs though, each running either a single service (e.g. Paperless) or a suite (Sonarr and so on plus NZBGet, VPN,…).
There’s additionally a couple of client VMs. All of those distribute over 3 Proxmox hosts accessing the same iSCSI target for VM storage.
SSL and WireGuard are terminated at a physical firewall box running OpnSense, so with very few exceptions, the VMs do not handle any complicated network setup.
A lot of those VMs have zero state, those that do have backup of just that state automated to the NAS (simply via rsync) and from there everything is backed up again through borg to an external storage box.
In the stateless case, deploying a new VM is a single command; in the stateful case, same command, wait for it to come up, SSH in (keys are part of the VM images), run restore-<whatever>.
On an average day, I spend 0 minutes managing the homelab.


You (sadly) need to group all quality profiles into a single one, and then handle quality through a custom format. Example from my setup:



NixOS for the win! Define your system and services, run a single command, get a reproducible, Proxmox-compatible VM out of it. Nixpkgs has basically every service you’d ever want to selfhost.


Lost me at LLMs. My Nix config is over 20k lines long at this point, neatly split into more than a hundred modules and managing 8 physical machines and 30+ VMs. I love it.
But every time I’ve tried to use an LLM for nix, it has failed spectacularly.


Called their support. They were super nice, and then told me that if I still had access to my 2fa, I could simply add a second mail in my account without needing access to the first one.
Made total sense when stopping to think about it for a second 😄
Or TUI.


Oh, yeah. Wasn’t actually worried of the mails themselves (multiple devices which have all of them, plus a daily borgbackup offsite), it was just knowing that I NEED to fix this, now, because email is unfortunately kinda important.
Fun conundrum: had my personal mail as the only contact at my server provider. Requested a KVM for rescue. “Thank you for your request! We’ll email you the link to your KVM access!” Gahhh.


Same. But also, I’ll never forget the adrenaline when I actually legitimately fucked up once. Was a sweaty 2 hours of recovery.
Eh, maybe… But doesn’t this read very Trump-coded to you? There’s a pretty distinct stylistic difference between “beginner English as a second language” and “idiot native speaker” IMO.
This is 100% intentionally dumbed down to Trump’s level. Would actually be surprised if Macron wrote this himself, rather than a team laboring over how they can dumb it down to a level Trump will comprehend, without stooping so low even he recognizes it as insulting or manipulative.


Nice, I was able to send an email to that.


That’s what I’m not so sure about though. Forgejo/codeberg/… projects are already not hard to find through search engines. Add a federated in-forgejo search and you’d be set there.
And currently the problem indeed is that a forgejo project is on instance X, and you, as a developer only have accounts on Y and Z. But through federation, that would stop mattering, so I don’t get the “it’s where contributors are”: as long as contributors have a single forgejo account anywhere, we’d be good.
Did you prompt it in English, or on Chinese? This is just a next-token-predictor, and it will largely use its English training data for its English output prediction.
This is not surprising. Nor is it indicative of, well, anything.