xAI recently showed that training a decent-ish model is now a multi-month effort. Granted GPT-4 is still farther along than others but curious how many months/resources does that add up when you have the team that built it in the first place
But also, starting another LLM company might be too obvious a thing to do. Maybe Sam has another trick up his sleeve? Though I suspect he is sticking with AI one way or the other