I now assign some non-zero probability to the timeline where coding gets solved, solved swiftly, utterly, superexponentially, to the point that <30B models in Q2 2027 can zero-shot build operating systems – and then there's no "RSI" or singularity.
basically Chollet's take