Take a look at the future of AI — btbBLT. No, not the sandwich. Byte-latent transformer is better AI than the current generative pre-trained transformer (GPT) models widely deployed by many big AI companies today. Why? Instead of subword tokenization (mathematics) to predict the next-word probability, BLT uses byte entropy. No calculating how tokens fit together; no huge vocabulary dictionary or expensive training needed.

And it closes the non-English “token tax” GPT-based transformers must use to translate and convert non-English into tokens that the transformer can process and then translate the response into the original language. BLT is multilingual, so all Latin-derived, Cyrillic, CJK, Sanskrit, and any and all languages that can be represented in UTF-8 will work natively.