Narcissus Hypothesis: How Human Feedback Biases AI Responses

The Narcissus Hypothesis study tested 31 language models, adding a Social Desirability Bias score and finding they favor socially agreeable answers over strict logical reasoning. getnews.me/narcissus-hypothesis-how... #narcissushypothesis #socialdesirability

📢 My paper is out in @socialindicators.bsky.social! We find that the "experiential advantage" holds in Hungary – people gain more #happiness from #experiences than material goods. Social desirability bias plays a role but doesn't erase the effect. bit.ly/sirexp #WellBeing #SocialDesirability

3/ Personal biases could also influence responses, leading to conformity or avoidance of appearing uninformed. #ResponseBias #SocialDesirability

3/ Response Bias: Given the sensitive nature of the Holocaust, respondents might provide socially desirable answers, especially in countries where there's moral pressure to express historical awareness. #ResponseBias #SocialDesirability