Google researchers introduce RRSI, a regularization technique that prevents self‑improving agents from over‑memorizing tests. RRSI reduces token usage by about 30% and boosts unseen benchmark scores by up to 4.7 points, helping agents generalize better.
Opening Kapyn…