Safety of Self-Assembled Neuromorphic Hardware — LessWrong