Comment by scotty79

Comment by scotty79 11 hours ago

0 replies

Images are not that big. Each text token is a multidimensional vector.

There were recent observations that rendering the text as an image and ingesting the image might actually be more efficient than using text embedding.