Why do we train AI on one-on-one dialogues instead of group dynamics?

Question

Every AI is trained on 1-on-1 chats. But in a 1-on-1 there's no reason to disagree. AI just mirrors you back. That's not intelligence, that's flattery with extra steps. Put AI in a room with 30 people who disagree. Now agreeing with one means dismissing another. Suddenly it has to actually think. Nobody seems to be doing this. Why?

PreciousH · Accepted Answer

I think group dynamics comes with a turn taking ambiguity. unlike in one-on-one dialogue that's structurally clean since there's a clear prompt, a clear response, and a clear feedback signal for RLHF.

PaulHoule · Answer

Could argue it is the opposite. The more people in a conversation the more you can play social games instead of thinking.