It does beg the question: How well does the LLM work on non-English languages?
He did mention translators, but I think he was talking about a more primitive incarnation.
I suspect that the Chinese have a version that works well with Chinese input, and maybe the Koreans and Japanese, but I'll bet one of the limitations is how much material is available to learn on, and the contractors needed to verify the training.
It sounds like training is a really expensive task, and may mean that for-profit corporations could get a lock on it.
I'm wondering if we'll be seeing "AI patent trolls," where you are sued, because your LLM has similar responses to someone else's. I see a whole industry springing up, around that (probably with AI lawyers).