Statements like "X is can be used to represent any program/function, therefore it is all we are every going to need" are pretty silly.
Show this thread
-
I do think that RNNs are like the brainfuck equivalent of NNs to PLs in the Turing tarpit, but I would think that is obvious and that people wouldn't actually believe hat they can be used to solve what that some C++ NN equivalent could solve, but maybe I am wrong...
I regularly do see this claim verbatim, with Turing-completeness being used as "proof". It's not a straw man, it's a surprisingly common position. Before RNNs were cool, the "universal approximator" argument was being used in the same way to say that NNs were sufficient for AGI.
-
-
Mostly, it seems like an excuse to avoid having to think about what (R)NNs really do, and where their limitations lie.
-
I think that is fair to think in that case.
End of conversation
New conversation -
-
-
Well. I live in a bubble then. I do think these universality results are very cool, but utterly useless. Like I said, I think it is very cool that RNNs are TC, but they are no better than "brainfuck" in the corresponding Turing tarpit IMO.
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.