1. I expect it to know it's not possible to increase brightness of an alarm and reject incongruent requests. 2. This particular implementation is limited to op

1. I expect it to know it's not possible to increase brightness of an alarm and reject incongruent requests. 2. This particular implementation is limited to operating on an English text in its finalized form, but a different first-class LLM implementation of an AI assistant could work directly on some form of phonetic input. ChatGPT is pretty good at dealing with such ambiguities, e.g. it already understands questions written in IPA notation. It also understands Japanese written in romaji, which is a non-native phonetic spelling of a language with tons of homonyms.