Their complete shift away from providing access to untuned foundation models is interesting. Perhaps these are too powerful in some sense. They already removed

Their complete shift away from providing access to untuned foundation models is interesting. Perhaps these are too powerful in some sense. They already removed the GPT-3.5 foundation model a while ago, and they never released it for GPT-4.