I’ll add that it’s a multiple of parts (see this video), not just one factor that gives it away. But for me the most telling one is that the text looks like the author spent a lot of time thinking about what figures of speech to use but at the same time did not use capital letters. The prompt clearly included something like “don’t use capital letters to make it look more casual”. But the LLM forgot to make the semantics look casual.

It’s also possible they just write like a bot.

Unlikely but if it were true, it would not change a thing. For me it does not matter who wrote the slop. It’s crap writing regardless and I don’t want to see that on the internet.

Btw from the video we can see the “rule of 3 pattern” which was used in this post:

what’s your absolute, non-negotiable, ‘i will personally visit the datacenter if this fails’ self-hosted application

Also I think the greeting “alright folks, let’s get real” is a gipitism.

what’s your absolute, non-negotiable, ‘i will personally visit the datacenter if this fails’ self-hosted application

Here you missed that this sentence makes no sense in a selfhosted context. What datacenter??? I honestly don’t know how can anyone doubt the elelemcy here.

It’s funny that the headline frames it as “a big number” when in reality majority of users don’t know what database they’re using and probably don’t even know what a database is. Such polls aren’t useless but you always get skewed results towards the more technical population. They would have to create a poll inside the nextcloud webapp to get more balanced results.

I used to do this (before my server died). You can for example use this to bypass Proton’s free tier one user per account limit.

Basically it’s about correctly setting the wireguard’s AllowedIPs and DisallowedIPs. Your laptop wants to send everything through wireguard. Your home server wants to send everything through Proton (or Mullvad or aVPN) except the communication with your laptop.

(I did that by marking the packets from one VPN with fwmark and sending marked packets to the other but that is a dumb solution for such a simple problem.)

You can tell Open Interpreter to run commands based on you human-language input. If you want local only LLM, you can pair it with Ollama. It works for “interactive” use where you’re asked for confirmation before a command is run.

I set this up in a VM because I wanted a full automatic coding “agent” which can run commands without my intervention and I did not want it to blow up main system. It did not really work though because as far as I know Open Interpreter does not have a way to “pipe” a command’s output back into the LLM so that it could create feedback with linters and stuff.

Another issue was that Starcoder2, which is the only LLM trained on permissive licensed code I could find, only has a 15B “human-like” model. The smaller models only speak code so I don’t know how that would work for agentic usage and the 15B is really slow running on DDR4 CPU. I think agents are cool though so I would like to try Aider which is a supposedly good open source agent and unlike Open Interpreter is not abandonware.

Thanks for coming to my blabering talk, hope this might be useful for someone.

Fair enough. I guess I imagined someone hosting all the selfhosted web apps that get posted to this forum, when most people likely just host only the few they need on the go, so it isn’t really that burdensome.

Edit: Forgot to add: I always though that it could be useful to just set up Apache Guacamole, so that instead of the hosted services, my family members could just use remote desktop apps but I never got around to it.

Yep my thoughts. New selfhosters think the hard part of selfhosting is command line but that’s “kinda” like thinking that the hard part of math is writing numbers on paper. Terminal is just the medium, not the complex part. Navigating filesystem and editing files is easier on the desktop but changing permissions and managing services would be be extremely difficult for a newbie without using the terminal because (almost) every online tutorial uses terminal. OP would have to learn how to translate the terminal command to its desktop counterpart at which point they might as well use the terminal.

OP also has an XY problem. They asked for a system which does not require terminal usage but they should have actually asked for an easy to set up system. People are recommending things like Yunohost though, so it’s fine in the end.

It’s nice there’s a front end for all these tools but I kinda don’t get why is everything “hosted”. This could have just been a desktop app. I guess it can be useful when you want to convert something on your phone but to me it just seems like unnecessary server maintenance burden.

Dunno whether they are transcoding or not nor why they have such a bizarre setup. But I would hope 16C/32T CPU from 2017 could handle software transcoding. Also peak power consumption while playing a movie does not really matter compared to idle power consumption. What matters more is that the motherboard is probably packed with pcie slots that consume a lot of power. But to OP it probably does not matter if they use a threadripper.

I don’t recommend it unless you just want it for storage or whatever else it does out of the box. It’s basically impossible to tinker with it because it has so many layers of abstraction. At least that was my impression when I tried to edit their nginx config. It had like 2000 lines so I just gave up.

If you want a server that runs services that you download from the internet, don’t buy it. Look at it as a box that does the thing that it promises to do, not as a computer. If you want it to do a different thing, buy a different box that does that. Kinda like a TV. It’s technically a computer that runs some kind of linux but to the user it’s a monitor that also shows videos from the internet.

Also it’s perfectly fine to buy a “NAS black box” but maybe not something I’d buy if I wanted to get into selfhosting. I’d buy it if I wanted to have a NAS running at home with the least amount of “self” in “selfhosting” that’s feasable.

Proxmox is nice for beginners. This is a nice tutorial: https://youtube.com/playlist?list=PLT98CRl2KxKHnlbYhtABg6cF50bYa8Ulo

Proxmox has nice UI for managing Linux Containers (LXC). They act like a computer inside a computer with the advantage that you can clone them. So you can basically save and load them whenever you succeed or fail at something. Proxmox also allows you to install Turnkey Linux containers which have the software you want to run preconfigured in them so that’s also good for beginners.

Only downside is that this is not declarative so it won’t be as scalable as docker or nix. It might be more worth it to learn docker from the beginning but that would also be less friendly for a beginner.

Yep any core i3 is fine even for desktop given an SSD and enough RAM. Once you delve into the core2 era, you start having problems because it lacks the compression and encryption instructions necessary for the day to day smoothness. In a server you might get away with core 2 duo as long as you don’t use full disk encryption and get an SSD or at least use ram for caching. Though that would be kinda a bizarre setup on a computer with 512 MB of ram.

Ryzen is not recommended for transcoding because the Radeon integrated GPU’s encoding accelerator is not as fast as in intel iGPUs. But this does not come into play if you A) have 16 cores and B) don’t even have an integrated GPU.

And about idle power consumption: I don’t think it’s a point of interest if you are using a workstation class computer.

I used to selfhost on a core 2 duo thinkpad R60i. It had a broken fan so I had to hide it into a storage room otherwise it would wake up people from sleep during the night making weird noises. It was pretty damn slow. Even opening proxmox UI in the remotely took time. KrISS feed worked pretty well tho.

I have since upgraded to… well, nothing. The fan is KO now and the laptop won’t boot. It’s a shame because not having access to radicale is making my life more difficult than it should be. I use CalDAV from disroot.org but it would be nice to share a calendar with my family too.

What do you mean by not being arrested? I would say German police putting a black bag on your head and taking you to their station in the middle of the night is something one could consider an arrest.

However, it is important to keep in mind that to my knowledge no one has ever been arrested for running a exit node in a western country.

https://news.ycombinator.com/item?id=41505009

There links to other occurrences of arrest in the comments.

Tbh it’s the English language that decides what counts as Open Source. Free/Open Source software has been established for decades at this point. It’s good that they changed the name to “Source First”.

I think that better wording would be “the organization that doesn’t believe that foss solves every problem”. For project like immich AGPL is completely fine but for the android keyboard it might not be a good idea to allow Google to use it to abuse their customers.

I don’t think it’s possible to make this project proprietary because FUTO does not own the rights to the code that were made by random contributors on git. Part of the promise was that they won’t change their CLA so it should be fine.

It’s probably a skill issue but don’t really know how to setup desktop streaming. I’ve tried

[[apps]]

title = "Desktop"

[apps.video]

source = "pipewiresrc capture-screen-cursor=true capture-screen=true"

But it just shows black screen.

It’s easy to misdetect the card. You just need to flash broken firmware on it that pretends it’s a different card. This is definitely not a 2070 because 1) Powercolor does not make nVidia cards and 2) RTX 2000 GPUs don’t have DVI ports.

Returning it is what OP should do. He paid for a working card, he should not be dealing with firmware flashing. Though I’d try using GPU-Z on a Windows machine to be sure first. Technically you can only be 100 % sure after reading the laser print from the GPU die but that might make returning harder so I wouldn’t bother.