Two users type the same question in Kannada. One uses the Kannada script, the other Roman letters. A large language model (LLM) reads both, interprets both correctly, and even paraphrases them with ...