The Importance of Reinforcement Learning

John Schulman gave a talk on the chat JPT system and the importance of pre-training and fine-tuning the model/nThe pre-training phase allows the model to know everything that's out there, while the fine-tuning phase focuses on using what the model already knows for conversation/nThe learning objective for human learning is not just about predicting the next word, but abstracting away what's important and making sense of it/nHuman learning also involves ignoring information that's likely not trustworthy, such as conspiracy theories