Curse of knowledge and Naive realism: Bias in Evaluating AGI X-Risks — LessWrong