Byte by Byte: Optimizing Tokenization for Large Language Models
Do image classifiers deepdream of electric sheep?