But a good LLM from Nov 2026 would certainly be able to drive a game agent or a robot and make it do gymnastics and improve at that task on its own by improving it based on the observed outcomes