The Timeline Has Consistently Been Too Slow
Hinton's point on timelines is less about the number and more about the error. In the 2010s he thought human-level AI was fifty years out. By 2020, thirty. Now, five to twenty. The error has always been in one direction: capabilities arrive faster than the models predicted. That pattern tells you something about how to calibrate your own assumptions. Every expert in this field has been wrong in the same direction for forty years. The null hypothesis is that they will continue to be.
The play
Recalibrate your working AI assumptions faster, not slower. Decisions made on the premise that 'this will affect my industry in ten years' should probably be made on a three-to-five-year timeline instead. You can always be wrong and still be safer than the version of you that planned for slow.