Hype article. Don't get me wrong, I'd be happy if it was true, because I was one of the (many) people disappointed by OpenAI abandoning its scale obsession - and frankly cutting training costs that much would possibly be the most significant part of such a model (it'd be an absolutely huge change to the field). Nonetheless, this is... dubiously sourced lets say, despite how interesting the whole Gwern rumor thing was.