Bill Zito messaged me the following:
I think the numbers you reference in the talk might be from "Preliminary prices for human-level intelligence," AI impacts, 2015, https://aiimpacts.org/preliminary-prices-for-human-level-hardware/ which is also referenced by Luke's 2015 openphil post "what do we know about AI timelines" https://www.openphilanthropy.org/focus/global-catastrophic-risks/potential-risks-advanced-artificial-intelligence/ai-timelines#Trend_extrapolation
I think that Drexler's estimates based on task comparisons in CAIS are more compelling (and I've spent some time fact-checking the numbers / making revised estimates if it's of interest to you in the future) https://www.fhi.ox.ac.uk/wp-content/uploads/Reframing_Superintelligence_FHI-TR-2019-1.1-1.pdf?asd=sa section 40.2 estimates brain compute
Also, section 40.3.2 references some other calculations, which are also referenced in the AI impacts post: "Sandberg and Bostrom (Sandberg and Bostrom 2008), for example, consider brain emulation at several levels: analog neural network populations, spiking neural networks, and neural electrophysiology; the respective implied RPFLOP values are 1, 10−3, and 10−7. Again based on a proposed neural-computational correspondence, Kurzweil suggests the equivalent of RPFLOP = 0.1 (Kurzweil 2005)."
I was actually referencing AFAIK unpublished numbers which I heard from someone else, but these resources might be interesting.