“Google DeepMind is re-examining older ideas like Monte Carlo Tree Search to augment the reinforcement learning capabilities of today's foundation models.”