Probably more importantly, the current generation of LLM chatbots are specifically modified to do this: OpenAIs model includes filters to require the answers Ch

Probably more importantly, the current generation of LLM chatbots are specifically modified to do this: OpenAIs model includes filters to require the answers ChatGPT gives to be pleasant and friendly. The model is specifically instructed to be agreeable. It would be just as easy to invert those: tell it to voraciously argue against whatever you tell it (in a non-political context this would actually be interesting: i.e. "tell me why my code sucks" is actually a pretty useful capability).