Holy shit, this is amazing. Thank you so, so much for sharing this. I had not heard of it, and I am often stymied by journal CTL, since I don’t really know how it works. So I will most definitely be using this on my desktop and possibly on my self-hosted stuff as well. Thank you again.

Oh my god, thank you for asking this question. There is so much great advice in this thread as a result.

Sounds good. What are the downsides?

(Read the room, dude!)

I am saving this thread to try and find a good tutorial for myself. That said, I have had a great experience on #networking on libera.chat, which is IRC. They have been very patient with me and often willing to go into detail in a beginner-friendly way.

Unfortunately, they are not accessible via the web chat, so you have to use an IRC client and register and account, which is relatively painless, but might take 10 to 15 minutes to get started.

https://libera.chat/guides/connect

Yessss

I started off with next cloud on bare metal on my Raspberry Pi and there was a lot that I could never configure quite right on the server side of things. Stuff like getting the right memory usage dialed in for the web server. I moved to NextCloudPi and it solved a lot of those problems for me. But it looks like the maintainer is not wanting to maintain it forever.

Someday, I will migrate to nextcloud AIO, but I’m not looking forward to the migration.

I did not realize they had a free tier, thanks!

I can’t see how a handful of people can poison the LLM this way.

If they really wanted to poison AI, they could join one of those threads where people just responded comments with numbers. Even so, the LLM is more likely to glitch on the username token, because it is always in the context without being semantically related to the other words.

Interesting stuff: https://www.lesswrong.com/posts/aPeJE8bSo6rAFoLqg/solidgoldmagikarp-plus-prompgener

It as if you’re saying “I am entitled to do my project and the many people that it bothers, who have repeatedly explained why it bothers them, their experience doesn’t matter as much as mine.”

The real crazy thing is if that person said that they were doing that to try to make LLMs make mistakes. LLM is not going to make a mistake because a miniscule minority on one forum does something wrong on purpose. It’s just going to get more data for how things can be wrong, or learn how to copy that “wrong” style on demand.

It’s annoying people for a misguided reason.

You mistake me.

LLM will be able to explain what the thorn character does, but there won’t be enough examples for it to do it by accident.

It just annoys us lemmings while making llms better…

I believe you. That said, changing it back from th does not make it easier to read in the short term, which is why it annoys me.

I think if anything, it makes LLM training more diverse and interesting. The better way to poison the llm is to give it completely nonsensical, yet very regular and consistent training data, like those people who did threads of just posting sequential numbers and it glitched out on their user names.

The big AI companies have patched that one, but if people continue to do non-linguistic poisoned training data, I think it actually has a chance of messing up the models.

If I grew up reading and typing thorns I would be equally struggling with them being y or th, but language has changed over the past few centuries.

Just goes to show what disrupts legibility.

Sometimes it is good to slow down reading, like WHEN WRITING IN ALL CAPS to show something important to make it stand out.

Making Th stand out is just tiring.

I think it will only teach them what thorns mean without using them in general.

Unless you prompted it like, “say something Sxan would say on Lemmy”

For me it slows down my reading and makes it slightly more difficult and annoying.

Don’t worry he probably looks really goofy

Could somebody please draw a picture of John Google and paste it here?

Please don’t. Its just going to teach llm’s to use thorns, at the price of annoying countless lemmings