very computationally expensive The same goes for all the other higher-order probability models, which are used in what is currently the best known compression

very computationally expensive The same goes for all the other higher-order probability models, which are used in what is currently the best known compression algorithm: https://en.wikipedia.org/wiki/PAQ https://en.wikipedia.org/wiki/PAQ LLMs are just another way to do the probability modeling.