😭 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.