4 ms·
I just feel bad about this "my intuition is that you'll need more layers than 10 in the NN, and much much more data." I still find current AI as basically a fi
by nojvek 9y ago
I just feel bad about this "my intuition is that you'll need more layers than 10 in the NN, and much much more data."
I still find current AI as basically a filter. Image to text, speech to text e.tc
A smart AI would be able to figure out sudoku rules from a very small sample of games, it would then figure out a rudimentary backtracking algorithm. It would then sleep over it and figure out more patterns like naked singlets e.t.c and build a really fast sudoku solver with 100% accuracy. All this happening in a matter of seconds.
- jacquesm 9y agoWhen a pig suddenly starts to dance the remarkable thing is not that it can not do a tango, but that it can dance at all.
- dkarapetyan 9y agoPigs dance all the time for suitable definitions of dance and time. I don't think optimizing a rats nest of functions with SGD is that impressive. If a model lacks explanatory power what use is it to people? In this case when the solution from the NN is incorrect what recourse do you have? The actual solution might be an arbitrary permutation of what the NN gave you and there is no way to tell which rows and columns will have to be reshuffled to get the actual solution or even if there is a solution. The constraints might be inconsistent and you will never know.