TL;DR: when the model thinks it's done thinking, tell it to think some more. Yes, really. This is a short summary. ↗ Open original to view full content