Just to be clear: this was said by Bryan Catanzaro, the VP of applied deep learning at Nvidia. It is not generalizable to workers using AI tools on their day-to-day work.
A few teams like his are notorious for requiring a ton of computation as they try out different novel approaches, particularly in the context of training rather than inference.
Bryan will sometimes comment on HN, so we may get lucky.
