|
Garry, I said fancy pattern matching. Obviously that's not simple grep or even google's search thingy, that's just silly.
In its barest engineering essence without using any hype, buzzwords, overly technical terms or overstating anything, in as few words as possible what is it technically then?
What are "language models" and "deep learning algorithms" if not a bunch of weighted links on how phrases relate and fancy pattern matchers?
For your GIGO argument to be valid you would need to show the source being wrong in a statistically relevant manner ie: the obvious mistakes it makes with respect to context, relevance and accuracy being predominant in the training material that was used to train it, such that it "learned wrong". To me the "Garbage Out" nature of its responses specifically and in general demonstrate it being merely a fancy pattern matcher, a "next gen google" if you must, rather than anything that would reasonably be called even rudimentary AI. Sure, better training will help mitigate some of the short comings but it will not change what it actually is.
I think it's wildly fascinating, especially the black box nature of the resulting networks.
|