Reactions from ranked influencers
3 posts
For your weekend reading pleasure a somewhat geeky post arguing that we must ground AI in human knowledge (and should avoid being too bitter-lesson-pilled)
This is very much open to debate. To explain what works vs to only work on what is explainable. There is solace in the idea that we must build only what is explainable. Yet, multiple trends in data-driven AI (even before deep learning) had moved to post-hoc analysis of scalable systems. Humanity has frequently developed "know-how" before the "know-why”. https://praxiscurrents.substack.com/p/the-age-of-empiricism-in-physical Post-hoc explaination to understand a working artifact (even though it is machine generated) is a discipline of science itself.