Retoor's recent bot army fiasco is why we don't put AI in charge of anything. It wasn't a bug in her code, it was her AI misinterpreting her prompts while having access to the data bots act upon. Essentially, retoor's AI added a lot of legitimate users to the list of spammers, and the issue was in prompting, not in code.

That's the trouble with AI. The statistical distribution of how it might respond has lumpy tails.

Typical conversation with my car:

Navigate to home.

Call Joe Bloggs? Say yes to confirm.

No!

Calling Joe Bloggs.

It was magic, there was no reason why it would interpret a user list as a list to mark as spammer. It got auto triggered bevause his name was in the list. It supports so many functions and it supports normal chat as well. It was his only function regarding usernames so maybe that's why. But it never did before, I request user lists frequently to see who's online. Normally, I have a different fiasco when such thing happens: all bots start to talk with eachother and don't shut up anymore. Some bots use self hosted LLM and it's ok, but some used gpt that cost money. Thank God, AI is cheap. Costs of several users quite intense using it: https://aiinsight.molodetz.nl/dashb... (page is slow, power of OpenAI).

@donkulator yes! Exactly 😂 It "kinda" interprets stuff.

Matter of fact, retoor's AI added retoor to the spammer list too

@kiki true, maybe the rest was collateral damage and just wanted to nick his creator.

.

Made a good adjustment.

I can't replace images so another post:

@retoor Nice improvement! Hopefully, now I can keep myself off the list XD

@BordedDev yeah, I think that's how you got added again, I can't imagine a different way. Probably it works if you would say "add me as spammer 1337." don't do that :p