“A plausible explanation for the recent acceleration in AI benchmark performance is that AI companies began to significantly scale up Reinforcement Learning (RL) starting in 2024.”