This is dumb. It's kind of like saying: "once more processing power is no longer useful, we will no longer need our large human brains." SOTA will remain at th

This is dumb. It's kind of like saying: "once more processing power is no longer useful, we will no longer need our large human brains." SOTA will remain at the edge of what compute can produce for a long time to come. SOTA is a moving frontier, and there will be demand for incrementally smarter models, because they save you time by making fewer mistakes. No matter how much more efficient algorithmic innovations make ML model training, compute will make those algorithms smarter. It's a coefficient.