A study of Swiss referendum debates reveals that large language models align with human evaluations of political arguments through surface structural features rather than genuine reasoning.
Some results have been hidden because they may be inaccessible to you
Show inaccessible results