“Misha Laskin believes the final paradigm needed to reach Artificial Super Intelligence (ASI) is scaling reinforcement learning on top of large language models.”