Is it likely that hardware will allow an exponential takeoff?
-
AI Impact has tried to answer this already.
-
https://www.lesswrong.com/posts/75dnjiD8kv2khe9eQ/measuring-hardware-overhang
-
https://www.lesswrong.com/posts/N6vZEnCn6A95Xn39p/are-we-in-an-ai-overhang
-
Training requires way more compute than deployment. Thus when we have enough compute to train an AI which is human level at deployment time, we'll have enough compute to run that AI many many times very quickly, so we'll have a faster-than-human AI. Worse if the algorithm is parallelizable and the AI takes control of several machine farms.