To think about it, SIAI name worked in favour of my evaluation of SI. I sort of mixed up EY with Kurzweil, thought that the EY has created some character recognition software and whatnot. Kurzweil is pretty low status but it’s not zero. What I see instead is a person who by the looks of it likely wouldn’t even be able to implement belief propagation with loops in the graph, or at least never considered what’s involved (as evident from the rationality/bayesianism stuff here, Bayes vs science stuff, and so on). You know, if I were preaching rationality, I’d make a bayes belief propagation applet with nodes and lines connecting them, for demonstration of possible failure modes also (and investigation of how badly incompleteness of the graph breaks it, as well as demonstration of NP-complete in certain cases). I can do that in a week or two. edit: actually, perhaps I’ll do that sometime. Or actually, I think there’s such applications for medical purposes.
Well it won’t be useful for making glass eyed ‘we found truth’ cult because it’d actually kill the confidence, in the Dunning-Kruger way where more competent are less confident.
The guys here haven’t even wondered how exactly do you ‘propagate’ when A is evidence for B and B is evidence for C and C is evidence for A (or when you only see a piece of cycle, or several cycles intersecting). Or when there’s unknown nodes. Or what happens out of the nodes that were added based on reachability or importance or selected to be good for the wallet of dear leader. Or how badly it breaks if some updates are onto wrong nodes. Or how badly it breaks when you ought to update on something outside the (known)graph but pick closest-looking something inside. Or how low the likelihood of correctness gets when there’s some likelihood of such errors. Or how difficult it is to ensure sane behaviour on partial graphs. Or how all kinds of sloppiness break the system entirely making it arrive at superfluous very high and very low probabilities.
People go into such stuff for immediate rewards—now i feel smarter than others kind of stuff.
“Singularity Institue? Oh, Kurzweil!” It’s as if he has a virtual trademark on the word. Yeah.
To think about it, SIAI name worked in favour of my evaluation of SI. I sort of mixed up EY with Kurzweil, thought that the EY has created some character recognition software and whatnot. Kurzweil is pretty low status but it’s not zero. What I see instead is a person who by the looks of it likely wouldn’t even be able to implement belief propagation with loops in the graph, or at least never considered what’s involved (as evident from the rationality/bayesianism stuff here, Bayes vs science stuff, and so on). You know, if I were preaching rationality, I’d make a bayes belief propagation applet with nodes and lines connecting them, for demonstration of possible failure modes also (and investigation of how badly incompleteness of the graph breaks it, as well as demonstration of NP-complete in certain cases). I can do that in a week or two. edit: actually, perhaps I’ll do that sometime. Or actually, I think there’s such applications for medical purposes.
A simple open-source one would be an actually useful thing to show people failure modes and how not to be stupid.
Well it won’t be useful for making glass eyed ‘we found truth’ cult because it’d actually kill the confidence, in the Dunning-Kruger way where more competent are less confident.
The guys here haven’t even wondered how exactly do you ‘propagate’ when A is evidence for B and B is evidence for C and C is evidence for A (or when you only see a piece of cycle, or several cycles intersecting). Or when there’s unknown nodes. Or what happens out of the nodes that were added based on reachability or importance or selected to be good for the wallet of dear leader. Or how badly it breaks if some updates are onto wrong nodes. Or how badly it breaks when you ought to update on something outside the (known)graph but pick closest-looking something inside. Or how low the likelihood of correctness gets when there’s some likelihood of such errors. Or how difficult it is to ensure sane behaviour on partial graphs. Or how all kinds of sloppiness break the system entirely making it arrive at superfluous very high and very low probabilities.
People go into such stuff for immediate rewards—now i feel smarter than others kind of stuff.