Stanford researchers release HumanLM, an RL framework that aligns LLM states with human beliefs and emotions to simulate users · Digg