Researchers tested a new way to rank posts on X and found that using AI to identify and reduce polarizing messages made users feel less angry and less hostile toward the other party. A small algorithmic tweak led to a measurable drop in polarization.