PERSON
Ilya Sutskever
The researcher who was in the room for AlexNet, sequence-to-sequence, and GPT—who bet his career on the scaling hypothesis when it was unfashionable, watched every installment of the prediction come true, and then warned, in nearly the same breath, that the thing he had spent his life building might destroy us.
Ilya Sutskever is the closest thing the history of artificial intelligence has to a person who has been right about the most consequential question in the field, continuously, for two decades. Born in the Soviet Union in 1986 and trained under
Geoffrey Hinton at Toronto, he co-built AlexNet in 2012, the experiment that demolished the image-recognition competition by a margin so large the entire field reorganized within a year around its lesson: that scale and learning beat human cleverness, and that a machine left to learn its own representations from raw data would find better ones than experts could devise. He took that lesson to Google, where he co-invented sequence-to-sequence learning, and then to OpenAI, which he co-founded in 2015 with the conviction that artificial general intelligence was not a distant fantasy but a foreseeable engineering target, and that the most important question was therefore