Old RL Ideas Should Combine With Large Models

Many ideas from earlier deep reinforcement learning work like DQN and AlphaGo are coming back into fashion and should be recombined with the new advances in large multimodal models, with significant potential in merging older and newer ideas.