Is this the thing where the baked an asymptote to infinity into their “model”? It was both bad modeling and also made all of their other input “data” completely irrelevant. I remember someone pointing this out at the time AI 2027 dropped, but there was so many other problems with it in the sneer spaces we didn’t talk about it too much among all the other problems. And someone on lesswrong actually pointed it out, but somehow, “we accidentally baked an asymptote into our model that is supposed to be detailed and carefully thought out” wasn’t a breaking point for lesswrongers.