Post

Has anyone trained an LLM on single-language documents, using only languages (or just documents) that share no linguistic roots/cognates? Disclaimer: I want to know how infeasible the Star Trek universal translator is: can you just feed it enough new tokens and it can translate a new language? #ai