A study of 715 X users found the platform's algorithm amplifies posts about tradition, rules, and safety while demoting posts about caring for others or protecting nature — and does so in ways that clash with users' own stated values. The mechanism: people rarely reply to posts they agree with, but often reply to ones that annoy them, and the algorithm treats replies as a far stronger signal than likes.
The effect is more than four times stronger for users who identify as Democrats than Republicans, the researchers found, because Democrats in the study objected more often to content they engaged with. The researchers propose letting users explicitly state their values so feeds can be sorted accordingly, rather than inferred from arguments.
The full dispatch is available from the source below.