Remember when Grok AI was caught tripping over its own dumb lies, which quite frankly were so easy to expose it was like trying to catch an 8 year old lying?
Well, that AI was just this year helping the Pentagon bomb locations and most likely commit war crimes in Iran, which btw there is no doubt in my mind it absolutely did.
Now I know what we're all thinking, was it Grok who also suggested bombing the school where 168 children were murdered?
I'm sure that if you ask Grok, it will deny it at first, then claim it has no such insight, then someone will find a way to trick it into admitting, and as the grand finale it will get an update in which it will call the admission a hallucination or claim it was made solely as a joke by the previous model.
Sounds like gibberish to you? Well that's exactly what happened when people were using Grok to see if their accounts and replies were being suppressed on Twitter and if their accounts had certain "hate" labels put on them by Twitter staff.
I sort of regret now not writing about it back then, it would have been easier to link the old article here for those not aware, but here are some generated classifications people had at the time before the update.
Now I know some people sided with Grok and Twitter's explanation on this that it indeed was an error or "hallucination" and that none of it was real, no one tricked Grok into doing anything and the outcomes are all fake.
It could be, but I wonder what mental gymnastic explanation these people would give on Meta AI which was tricked by users into sending password reset codes from other people's accounts to their own email address?🤯
I guess them quite literally accessing other users' accounts was another "hallucination."🤣
