I bet if we went back in time 2 years and asked Cory if “next-word-predictors” could produce anything remotely close to gpt-4, he would have the same arrogant /

I bet if we went back in time 2 years and asked Cory if “next-word-predictors” could produce anything remotely close to gpt-4, he would have the same arrogant / confident tone and say, of course not. I think for the simple fact that our ability to predict emergent properties of complex systems is nascent at best, warrants an extreme humility when discussing these topics. We simply are terrible at predicting what will emerge. Also this debate of his is single-dimensional. Even in the unlikely case that significant advances in LLMs were not possible, combining certain technologies / tech stacks here with LLM/LLM(s) is likely (I would say guaranteed) to result in additional surprises in emergent behaviors that will continue to look eerily similar to our intelligence and sentience. We’re unlikely to be near the local maxima.