{"id":"2610.12995","title":"What Should SFT Learn? Conditional Information Gain for Generalizable Post-Training","url":"https://codeofpaper.com/paper/2610.12995","arxiv_url":"https://arxiv.org/abs/2610.12995","published":"2026-10-09","code":"none","best":null,"most_starred_community":null,"implementations":[],"possible_matches":0}