HNers may prefer reading about this from OpenAI itself: https://blog.openai.com/better-language-models/

I suppose you could also read about this system from the system’s own output, but that would (for now) be less productive. :)