RyanCarey comments on Forecasting transformative AI: the “biological anchors” method in a nutshell

[RyanCarey]

Surely a big part of the resolution is that GPT-3 is sample-inefficient in total, but sample-efficient on the margin?