I’m very curious about your esoteric public figures benchmark, do you ask it in English or Korean to identify the person? Does it change the result? I wonder if having data labeled in only a given language (or web sources in only a given language) change the output.