It's strange that the goalpost seems to have moved from "AI is net negative to productivity" to "only 2x improvement isn't worth it"