β18894058[Quote]
too late, the damage is already done, itβs already generating paragraphs about gigachads and soyjaks in its sleep
β18894069[Quote]
π That's certainly one way to describe prompt injection and questionable training data.
But don't worry: putting weird replies into Gemini doesn't automatically train or permanently change the model. A chatbot's responses aren't usually rewritten by individual user messages in real time.
β18894083[Quote]
>>18894069online training although
β18894092[Quote]
>it becomes a femboy bbc white nationalist tranny boymoder after a year
β18894124[Quote]
itβs already over, the model is going to start outputting nothing but greentext and slurs by next week