In the long run, if this approach takes off, I don't see what OpenAI's role will be. They have no moat: all the training data is scraped from elsewhere, the models run on commodity hardware with ever-decreasing costs, and the output of an AI-powered confident-bullshitter is probabilistic enough that users wouldn't notice the underlying model being swapped out in favor of one developed in-house. Maybe if OpenAI were actually "open" somehow then they could at least argue the moral point of supporting OSS or whatever, but AFAICT the "open" part of their name appears to be merely baldfaced openwashing. I guess they're just hoping for an acquisition?