I'd say it is good for anything that does not need to be robust. Image search on google does not need to be robust. Google translate does not need to be robust. Autonomous car has to be. Any mission critical application has to be...
Sure @ylecun the sweet spot of wheels is not mountain climbing, but if deep learning isn’t good for solving MNIST robustly, what is deep learning good for? MNIST was supposed to be DL’s sweet spot. Your analogy only adds to @bethgelab’s #robustness concerns, once thought through.https://twitter.com/ylecun/status/1001718737995812866 …
-
-
-
That being said I stand by my opinion that the current hype could only be justified if this method worked flawlessly in mission critical domain. It does not, even in the most beaten to death datasets. Hence I think we will have an AI winter.
End of conversation
New conversation -
-
-
Well it’s even more amusing if you consider how wheels are actually used to “climb elevation”... we modify the inclines to fit wheels or legs best use. This is what is going to happen with all ai things.
-
Yes exactly, particularly in the early days while we're largely using single algorithms in standalone fashion.
End of conversation
New conversation -
-
-
There will be no progress toward AGI unless mainstream AI experts come to realize that deep learning is just GOFAI on steroids. It's symbolic AI's last hurrah. Deep neural nets are just glorified expert systems with the additional problems of overfitting and local optima.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.