curuinor

31 ms·

361.
by curuinor 9y ago
I've long thought that there was a creation of trust when the firm is created and it remains small, and another breakdown of trust when the firm becomes huge. So I think the fundamental transactional cost analysis applies when, say, th
362.
Backpropagation for literal credit assignment, in the firm(howonlee.github.io)
37 points by curuinor 9y ago|4 comments
363.
by curuinor 10y ago
Something about MeFi that makes it an oasis of basically what the lurid 90's folks imagined the Web would become. Probably because it's from the 90's, but it's a dingy dirty utopia of sorts
364.
by curuinor 10y ago
VC's and tech financing really have awful technology in and of themselves. You can think of it as the cobbler's kids having no shoes or you can think of it as really technically competent folks only really being attracted to VC to
365.
by curuinor 10y ago
moved that needle I guess
366.
by curuinor 10y ago
Carl Hewitt (of Actor model fame) has this thing where he claims that contradiction alone allows you to defeat Godel Incompleteness, so he shops it around places and everyone's like, "man, wtf, this isn't how that works"
367.
by curuinor 10y ago
Similary to St. Petersburg paradox ( https://en.wikipedia.org/wiki/St._Petersburg_paradox ). The expectation is a stranger and more dangerous beast than would be easily expected.
368.
by curuinor 10y ago
RNN is natively and obviously dynamical but FFNN can also be seen as transient dynamical system. Little Ramsey theory ever used in neural network land, but theory of dynamical systems used from quite a long time ago, from Werbos-BPTT which
369.
by curuinor 10y ago
Well, they already had a big instance of that, in the Minsky Papert book "Perceptrons", talking about linear separability. Talking about backpropagation (as opposed to delta rule) in opposition to that is a sort of mushing of hist
370.
by curuinor 10y ago
It's not like we don't know anything about NP complete problems, neither. Critical phase transition in transition of alpha on random kSAT and other stuff, the realization that this is neither necessary or sufficient for NP complet
371.
by curuinor 10y ago
Why can't we pull out the statistical mechanics? You can't understand the 10^27 variables in a lump of coal, either, but you can still do thermodynamics. People already do it, Ganguli, some stuff from Bengio
372.
by curuinor 10y ago
Bak of the Bak-Tang-Wiesenfeld model called his model the sandpile, basically explicitly to remind one of the Sorites paradox. Interesting thing to talk about. You can definitely take it as a sort of different descendent of the Ising model:
373.
by curuinor 10y ago
Usually if you properly use GPU the real limit is on the order of 20k. I think the graph in his post is old as crap, actually (if you sparsify connections and do some tricks with sparsification sky's the limit, O(n * small multiplier o
374.
by curuinor 10y ago
Many of the most loyal Soviet apparatchniks recognized that they got their positions due to them being freed up by the various purges and such. Piketty claims that the lowering inequality of the early 20th century that Kuznets claimed was t
375.
by curuinor 10y ago
Well, HogWild works well enough. It's another one for the giant "weird shit you can get away with in neural network land" bucket, I think
376.
by curuinor 10y ago
As I said in Rob's thingy, I hope you get the tenure committees and job committees, because they don't have to respect it but they're the ones you have to get to respect
377.
by curuinor 10y ago
Kratom is indeed not an opoid but still is a painkiller. However, unfortunately, it still can be addictive.
378.
by curuinor 10y ago
There were mass depopulations of cities at the end of the Western Roman (due to Germanic invaders), Eastern Roman (due to Muslim invaders), Abbasid (due to Mongols) and certain of the Chinese (due to Mongols, Xiongnu, etc etc) empires. This
379.
by curuinor 10y ago
After listening to him for a bit, reading some of his books, annoying him a fair amount, I think that my opinion of J Searle is that he doesn't know jack shit about AI. Gedankenexperiment as a methodology has had considerable success i
380.
by curuinor 10y ago
Strang's complaint is that there's too little linear algebra. This is true. This doesn't overshadow the fact that you're not going to get out of using some partial derivatives in neural net land (and many other AI subfie
381.
by curuinor 10y ago
If you're shipping anything after 2010, you're not going to get within an order of magnitude within state of the art with that book (PAI), unfortunately. There's basically no numerics in that book about anything that'll
382.
by curuinor 10y ago
Saxe Ganguli McClelland, 2013, about linear nets and orthogonal initialization. But then, read Li Jiao Han Weissman 2017 (maybe preprint), "Demystifying ResNet", which makes a nice claim about the niceness being conditioning of He
383.
by curuinor 10y ago
Strang's class is very pretty and excellent and just a little bit off from the center of the sorts of linear algebra used in machine learning. Not a lot off, but a little off. A field that does inspire a lot of deep learning folks and
384.
by curuinor 10y ago
Unfortunately, there is not a drop of numerics in that book. It's a good book for learning about symbolic AI. There is, to a solid first order approximation, zero symbolic AI in a system like, say, Google speech recognition.
385.
by curuinor 10y ago
There is lots of content in this book but there is no practical technical content in this book. Interesting philosophy. Much of AI philosophy is done by extremely non-practitioners. John Searle can't code. Nick Bostrom came to coding e
386.
by curuinor 10y ago
Less symbolic failures. There was a huge and abiding torrent of neural net stuff that dealt with evolving topologies in late 90's. I see very little of it in any way shape or form in industry or academia today, because it's a lot
387.
by curuinor 10y ago
More failures. Minsky and Papert gave a criticism on single-layer perceptrons in '86 where they proved that they could only make linear discriminators and therefore were useless for any real practical purposes harder than the XOR probl
388.
by curuinor 10y ago
Well, it's not very nice to talk about people spending decades on failure, but I guess I can give some examples. Cycorp still exists, from 1984. However, D. Lenat's approach to AI by ontology engineering has basically been complet
389.
by curuinor 10y ago
AI is not a field where the practitioners can safely ignore the academic. A huge number of people got into rabbit-holes in its history and basically failed completely and wasted decades of their lives, in some cases.
390.
by curuinor 10y ago
ag is faster ack is simpler and more OG

More ›