scaling paradigm shift

RL agentic loops

Both point to reinforcement learning/reasoning as the emerging source of capability and reliability, moving compute away from pure pre-training.

1 source videos