Is sample-efficient learning a singularly important step on the path to AGI?
Almost definitionally, learning as efficiently as top humans would suffice for AGI. (You could just train the AI on way more data/compute and it would be superhuman.)
AIs will probably reach milestones like full automation of AI R&D before matching top human sample efficiency in broad generality (though they might be better in some/many cases).
Almost definitionally, learning as efficiently as top humans would suffice for AGI. (You could just train the AI on way more data/compute and it would be superhuman.)
AIs will probably reach milestones like full automation of AI R&D before matching top human sample efficiency in broad generality (though they might be better in some/many cases).