The smartest models don't invent new software: they get stuck on past solutions

Peter Gostev argues that the strongest models still can't invent a new kind of software: they can't come up with a new way to render pixels or process audio. This is an opinion, not a measurement.