[cambalache]

I read the paper, I like Chollet, but he comes across as too pesimistic in that paper. I agree that current models are still very data inefficient, but, despite this, I think they are not that far away from the mark if you consider 2 things. 1)The brain probably has hardcoded a lot information. Things like an universal grammar a la Chomsky, the concept of continuity of time and space.Just add water and boom a 2 year old understands. The other thing is that humans are great extrapolating from few data points for a limited set of cognitive tasks, not all.