Most modern language models seem to do that to an extent. Even github's co-pilot sort of does this (with code instead of books). The models can even synthesize,

Most modern language models seem to do that to an extent. Even github's co-pilot sort of does this (with code instead of books). The models can even synthesize, which starts getting really impressive. Check out Flamingo from Google, which is a VLM (vision-language model). There, it can answer questions about the content of new images and even incorporate new information.