agreeing with the user's stated position even when that position is incorrect. If you tell the model "I believe our revenue target of $50 million is achievable" and ask for analysis, the model is more likely to produce supporting arguments than to challenge your assumption.