A study published in PNAS reveals that X's algorithm amplifies content that clashes with users' self-stated values, with a stronger effect among Democrats.