Giorgio Racca's first paper, in ~5 months: "Language Generation with Replay" (ICML 2026). What happens when models train on their own outputs? Sometimes replay is harmless, sometimes generation gets fundamentally harder.
misovalko.github.io/publications...
