Full paper: https://www.nber.org/system/files/working_papers/w31161/w31161.pdf https://www.nber.org/system/files/working_papers/w31161/w311... The most interes

Full paper: https://www.nber.org/system/files/working_papers/w31161/w31161.pdf https://www.nber.org/system/files/working_papers/w31161/w311... The most interesting thing to me is how little of a productivity boost you are seeing among higher skilled agents. Top quintile is getting an expected 2% boost with no change in the confidence interval. This jives with a lot of the heterogeneous commentary you see on HN among engineers. GPT-3/4 are a godsend when approaching new systems; the productivity gain is quite small if you are already skilled.