Start from characters, merge the most frequent pair of neighbours into a new token, and repeat: the tokenizer of GPT and most language models.
This interactive visualization needs JavaScript.