TheOneCurly

I have not seen quadlets before, that’s really neat.

I have the exact same setup. It works perfectly and integrates really well into home assistant if that’s your thing. Getting a coral TPU also makes object detection really easy even on low power hardware.

Yep Lemmy uses SMTP and in my experience most self-hostable platforms do as well. You can see in the Lemmy config documents how it gets set up: https://join-lemmy.org/docs/administration/configuration.html.

Made with Gtk4, WebKitGTK, libadwaita and Flatpak.

WebKit based, which is interesting. I don’t have much experience with WebKit on Linux.

As I understand it, NAT is a firewall with only a very basic configuration: allow all outbound and accept only established inbound. If you don’t expect to have any incoming connections and completely trust all your internal devices then its good enough.

However, if you start wanting to port forward for servers (SSH, FTP, video games) you need to poke holes in the NAT firewall and it has no additional configuration options to help you. The same goes for if you have internal (ex. IoT) devices that you don’t necessarily trust, there are no rules to block outbound traffic.

I wonder what the risks are to including deleted and pre-edited content in training data. Most of the edits are going to be typos and formatting, do you want 2-3 copies of the same message with typos in them for training data? Similarly, deleted comments are mostly nonsense, unhelpful, duplicate, or highly controversial things.

If someone wants to dig through and find individual users to restore that’s one thing, but I don’t think I’d immediately choose to train off of that other data unless I had to.

That’s what finally did in my 10 year old Corsair. I was technically within specs on wattage with my new 4070 but certain loads would cause it to trip the over current protection anyway.