Consistently Simulating Human Personas with Multi-Turn Reinforcement Learning | Read Paper on Bytez