Do you think that, much like how OpenAI scaled up GPT-2 to GPT-3, someone else can train a larger model like Dessa on, say, multiple GB200s?