Abstract
Ajeya Cotra proposes a framework to estimate when the computation required to train a transformative AI (TAI) model becomes affordable. She defines key terms, anchors his estimates to biological quantities, and models algorithmic progress, willingness to spend, and falling hardware prices. Cotra finds a median estimate of 2052 for TAI affordability, with a range from 2036 to 2100. – AI-generated abstract.