AI being a Yes-Man is slowly sabotaging it's own answers, because it negatively impact the user's decision. Yes/No are equally important, within a coherent context, for objective reasons. But being supported in the wrong direction is a castastrophe multiplier, down the road. The AI should be neutral, doubtful at times.
To be doubtful would imply that there is a world model full of some kind of Bayesian reasoning. Priors updating based on the context of the conversation, the question, the user asking the question, and cross referencing of facts well outside the scope of the current conversation.
“What are the chances this user is full of shit?” Is not something we are close to
