ianjs on Nostr: It feels like an LLM might be the way around this, rather than trying to second guess ...

It feels like an LLM might be the way around this, rather than trying to second guess every possible question, which is an almost impossible quest. Having Speech-to-Phrase hand off to an LLM should give the best of both worlds.

…and yeah, I know, there are a multitude of ways LLMs suck, but they do a pretty decent job in this use case, so once they can be run locally without requiring massive resources we should have a good solution.