Summary
As AI companies approach "recursive self-improvement," where AI models can autonomously enhance themselves, a physics professor warns that full AI autonomy could be "the worst idea in the history of humanity" due to potential risks to human control.