Nathan Lambert's RLHF Monograph: Data Over Algorithm
Nathan Lambert's RLHF monograph covers instruction tuning, reward models, and reinforcement learning, arguing that data is far more important than algorithm. The book is available online with open-source code.