As we test and improve new technology like AI, our understanding of its safety concerns and how to address them can rapidly evolve over time, leading to a developing philosophy for AI safety. However, many prior predictions about AI have turned out to be incorrect.