We’ve discovered neurons in CLIP that respond to the same concept whether presented literally, symbolically, or conceptually.
We’ve trained a neural network called DALL·E that creates images from text captions for a wide range of concepts expressible in natural language.
