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.
True for humans. AI has the opposite issue — in 1-on-1 there's only one person to impress. Easy to just agree. Add more people and you cant make everyone happy. Thats the whole point