[almosthere]

Death threats mainly. Personally I think it would be easier if they just made it so that platforms ran a tiny LLM against the content that will be posted - determined if it is a death threat, then require them to be identified before it's posted, then it would solve a lot of these problems.

TLDR: Evil people be doxxed internally not everyone.

[numpad0]

That turns jokes into contracts that nobody wants. Bad idea.

[Filligree]

Maybe just don’t make “jokes” like that.

[numpad0]

I don't make such "jokes". Idiots do.

And when the idiots do, the proposed system locks the fire door for them. That's just dangerous. We'd want them with bunch of confusing options and better illuminated de-escalation paths.

[bigfishrunning]

a "tiny large language model"? lol

[reverius42]

See https://tinyllm.org

These days the name "LLM" refers more to the architecture & usage patterns than it does to the size of model (though to be fair, even the "tiny" LLMs are huge compared to any models from 10+ years ago, so it's all relative).

[almosthere]

Yeah, a small one that is cheaper because they'll be processing billions of messages per year.

[lazide]

Good thing all the kind people doing death threats won’t just bypass it?

[almosthere]

I'm totally lost here. If you don't identify, you don't post.

[lazide]

Good thing no one ever breaks any rules!

[BenjiWiebe]

If a platform decides to require an account to post, or requires your message to pass an LLM sniff test before publishing it, you can break all the rules you want but your message won't be visible to others on said platform.