[Question] Work on Bayesian fitting of AI trends of performance?

I remember seeing a report on trends of performance in AI. It would be quite useful for me to find it again. As I remember it, the author looked at things like ImageNet top5 accuracy and extrapolated how it increased over time.

I can’t remember the author, though I think it was sponsored by OpenPhil. Anyone knows what I am talking about?

I’d also be interested in related work about predicting trends of performance in technology more in general :)

EDIT: I found the report I was thinking about (which had nothing to do with OpenPhil oops). Still interested in other work in forecasting trends of performance if anyone has any recommendations!