How are model outputs evaluated and ranked?

In this episode, Luca and Josh chat about Big Data and The Pile, an open source resource for training models, Reinforcement Learning and Aligning Models.