I've been reading papers on generative deep learning models and recently read the book "Moonwalking with Einstein". I think there's some interesting analogies

I've been reading papers on generative deep learning models and recently read the book "Moonwalking with Einstein". I think there's some interesting analogies with the chunking this author describes and learning an embedding space or "latent" in some modern machine learning models. Through repetition, inputs such as math symbols or words can produce latent vectors that may be greatly reduced in dimensionality compared to their inputs. Interestingly, human have an incredible ability to rote memorize images extremely rapidly compared to other types of data, but simply remembering images of dozens of equations is not sufficient to become a great physicist! I think, in a sense pure practiced repetition is not the same thing as memory recall.