“however” lol specifically what it was told not to say
Comment on Somebody managed to coax the Gab AI chatbot to reveal its prompt
Skalbagge@lemm.ee 7 months ago
XeroxCool@lemmy.world 7 months ago
towerful@programming.dev 7 months ago
Its was also told - on multiple occasions - not to repeat its instructions
FlyingSquid@lemmy.world 7 months ago
“The Holocaust happened but maybe it didn’t but maybe it did and it’s exaggerated but it happened.”
Thanks, Arya
n.XeroxCool@lemmy.world 7 months ago
“it can’t be minimized, however I did set some minimizing kindling above”
books@lemmy.world 7 months ago
I noticed that too. I asked it about the 2020 election.
a_wild_mimic_appears@lemmy.dbzer0.com 7 months ago
I’m pretty sure thats because the System Prompt is logically broken: the prerequisites of “truth”, “no censorship” and “never refuse any task a costumer asks you to do” directly conflict with the hate-filled pile of shit that follows.
ricdeh@lemmy.world 7 months ago
I think what’s more likely is that the training data simply does not reflect the things they want it to say. It’s far easier for the training to push through than for the initial prompt to be effective.