olooney

21 ms·

181.
by olooney 7y ago
Oh, it can capture everything. I believe that. It's just that it relies on volunteers to actually do the work of formalizing advanced theories. It has some pretty darn advanced stuff already, like complex numbers. But there's al
182.
by olooney 7y ago
Greg Egan's Diaspora has a rather good illustration of what a fully formalized mathematical system might look like. He called it the Mines[1]: a representation of the full acyclic digraph of all the theorems proved so far by a communit
183.
by olooney 7y ago
Magnets were well known the Greeks in antiquity[1], likely because of the Magnesia region in Greece where natural magnets are commonly found[2]. Words like "magnet" and "magnetite" are named after that region. In 1584, t
184.
by olooney 7y ago
You may enjoy Uncleftish Beholding[1] by Poul Anderson. It's a short essay describing atomic physics explained with pure Anglish words instead of Greek or Latin loan words. > At first it was thought that the uncleft was a hard thing
185.
Principal Component Analysis(oranlooney.com)
138 points by olooney 7y ago|48 comments
186.
Smarter Than Ted – Peter Watts(rifters.com)
2 points by olooney 7y ago|0 comments
187.
by olooney 7y ago
I agree with your criticisms, the Python version presented in the article is rather baroque. In particular, the use __dict__ seems to have little justification and serves only to "uglify" the Python. Here's my version of skew
188.
by olooney 7y ago
On Hacker News? Not if I can help it. :)
189.
by olooney 7y ago
I don't know anything about mission statements, valuations, IPOs, or yoga. I do know a little about regression, though, and that regression line is pure bullshit. (Ironic, considering what they're purporting to analyze.) The downw
190.
by olooney 7y ago
A lot to unpack there, so I'll restrict myself to a single remark: your suggestion that the polarization metric used in the last paper was developed post hoc in response to Trump's election doesn't match the publication tim
191.
by olooney 7y ago
There are things like CONSORT[1] which kind of do this. Statisticians like Fisher[2] have a ton of good general advice of the design of experiments. (A plug for The Lady Tasting Tea[3] and the 7 Pillars of Statistical Wisdom[4] feels approp
192.
by olooney 7y ago
What about astronomy, cosmology, archeology, paleontology, volcanology, evolutionary biology, cladistics, macroeconomics, etc., which do not allow us to "repeat and predict," as you say? Some of these cases can be rescued by consi
193.
by olooney 7y ago
Scientists usually try to distance themselves by saying those are soft science or even pseudo-science. This leads to the embarrassment of the demarcation problem[1] which is that no one can give a bright-line rule[2] to distinguish between
194.
by olooney 7y ago
John Rawls provides a vocabulary for talking about these issues. In Rawls view, society consists of "reasonable citizens" who want to live in a society built on cooperation and are willing to abide by rules if necessary to do so.
195.
by olooney 7y ago
Oliver Cromwell said it best: "I beseech you, in the bowels of Christ, think it possible you may be mistaken."
196.
The Thinking Ladder(waitbutwhy.com)
2 points by olooney 7y ago|0 comments
197.
by olooney 7y ago
According to the Pew Research Center there is evidence that America has become more more polarized since 2016.[1] This is part of a decades long trend.[2] This polarization is happening on both a rational[3] and emotional[4] level. Accordin
198.
by olooney 7y ago
> why not anthropomorphize? It would be easy to be flippant about this suggestion and reply something like "Because we're doing science, not making Disney cartoons?" But the issue is not so black and white. I once saw a le
199.
by olooney 7y ago
Interesting paper but t feels a little like a "paradox of the heap" argument[1]. A human has 9e10 neurons[2], while a rabbit only has 5e8, and there are only 3e8 Americans. I don't see a problem arguing that consciousness is
200.
by olooney 7y ago
The article makes it sound like his ouster had nothing to do with Maria Butina, and everything to do with speculating on shady crypto ventures. > The problem is, with just two tokens—representing Overstock’s and Tzero’s own shares—availa
201.
Why didn’t they say they were sorry when it turned out they’d messed up?(statmodeling.stat.columbia.edu)
1 points by olooney 7y ago|0 comments
202.
by olooney 7y ago
Consider the following lambda calculus expression: ((λ x . ((x x) (x x))) (λ x . ((x x) (x x)))) Under β-reduction, this blows up: doubling with every reduction, never repeating itself, growing infinitely long. Let's name thi
203.
by olooney 7y ago
This feels like it doesn't live up to its own hype because it doesn't really exhibit a production-quality deployment of a Python application but something closer to a dev server. It uses a single-threaded flask server behind a ngi
204.
by olooney 7y ago
Right. The naive way to do complexity analysis of a lambda calculus algorithm is to count β-reductions; but it's immediately clear that the cost of a β-reduction (which is basically a string substitution) is proportional to the length
205.
by olooney 7y ago
I agree. All I see here is: "We know any reasonably smooth function can be approximated by a neural net, likewise it can be approximated by a polynomial. So they're the same!" Imagine how it's going to blow their minds w
206.
by olooney 7y ago
Terminology note: data like images and voice which have strong spatial or temporal patterns are actually referred to as "unstructured" data; while data you get from running "SELECT * FROM some_table" or the carefully des
207.
by olooney 7y ago
I punched "It's actually pretty simple" into http://talktotransformer.com (which generates nonsense text from a seed using an OpenAI language model) and after a couple of tries it gave me this: > It's actu
208.
by olooney 7y ago
I think it's a great start - I'm a big fan of typing games like TyperA, and one of the things I enjoy most about blogging or answering questions on stats.stackoverflow.com is formatting mathematical in beautiful LaTex. It just nee
209.
by olooney 7y ago
Couple of pieces of feedback: 1. There's an order of magnitude difference in the difficulty of problems. Some are hardly more complex than a couple of subscripts and maybe a \frac{}{}, while others need matrix notation. However, they a
210.
Deep Reinforcement Learning(blog.dominodatalab.com)
76 points by olooney 7y ago|1 comments

More ›