LLMs also have other inputs, like audio and images. They get encoded (just like a human eye encodes an image) and passed to the weights.