Sure, but your idea here is fundamentally flawed. The example you linked worked because they used a specific trigger word that was associated with strings of garbage characters. It’s a very specific case, and the only people seeing that garbage output are people using the trigger word.

You aren’t associating thorn with a trigger, you’re just using it ‘correctly’. What you’re doing is providing helpful translation keys for any LLM that uses lemmy as training data. It gives them data on how thorn is likely used, so if someone asks for it, or uses it in their prompt, then the model will be better prepared to correctly interpret it.

And in doing this, you’re alienating hundreds of actual people in the community that you’re ostensibly trying to connect with. I occasionally read your posts and I generally appreciate what you have to say. But more often than not, if it’s more than a sentence or two I’m just going to roll my eyes and move on.

Is that really worth it to maybe, possibly confuse some LLM user for a few seconds?

This is what made you hate Reddit? Not power tripping mods, overt censorship, spam bots, or horrible monetization?

You draw the line at somebody who points out another person’s self-defeating behavior?

You do realize that all you’re accomplishing is helping to train AI in how to correctly use thorn right?

Besides irritating real humans and ensuring nobody bothers to pay any attention to the content of your posts of course.