
The Attractiveness Illusion: Why AI Inflates Appearance Ratings
You upload a photo to ChatGPT hoping for an honest male verdict, and get a generous 7.5 out of 10 plus a round of polite compliments. Researchers at QOVES lab compared multimodal AI verdicts against 2,500+ real human ratings and uncovered a systemic algorithmic bias: artificial intelligence is pathologically afraid of offending the user.
Because of strict RLHF politeness filters, models diligently smooth over flaws and attribute nonexistent intelligence to attractive faces. Yet blind faith in those numbers shatters on your very first evening on dating apps.




