3 ms·
Since the predictive algorithms in use mostly come down to glorified dot products, I question their general validity. Sure, it works sometimes, but just look a
by varelse 14y ago
Since the predictive algorithms in use mostly come down to glorified dot products, I question their general validity.
Sure, it works sometimes, but just look at some of the wacky stuff Amazon, Google, and Netflix derive from such things. Now replace search results and product recommendations with waterboarding and extraordinary rendition based on the results of those dot products and hilarity ensues...
- yid 14y agoAnd those dot products come down to glorified FMUL and FADD machine instructions! Specious reductionism goes all the way down.
- varelse 14y agoAnd those FMUL and FADD instructions could either be fused multiply/add or separate multiplies and adds with twice the truncation issues, Holy FPRE! But I'm curious as to why you place such faith in the sooper top sekrit(tm) basis vectors for these dot products? It seems to me that it's politically expedient for such a system to generate false positives rather than risk letting the terrorists win(tm) and that's where I think things go wrong. Because it would appear from the outside that the powers-that-be have decreed that it's better to torture the occasional innocent than risk letting the guilty slip through their fingers.
- yid 14y agoI think you might be a bit too wrapped up in conspiracy theories, friend.