Seems like the API is less reliable than GPT-4 so far, but I guess it makes sense for the endpoint to be popular at launch!