I've started pondering what a language would look like that was optimized for LLMs, where token length is the most important budget.
Maybe a language with 10,000 keywords would actually be better in LLM world?
Maybe a language with 10,000 keywords would actually be better in LLM world?