Mitigating the Uncanny Valley Effect in Hyper-Realistic Robots: A Student-Centered Study on LLM-Driven Conversations