murbard2

33 ms·

301.
by murbard2 11y ago
Yo, dawg, I heard ya like Von Neumann architecture.
302.
by murbard2 11y ago
Cool project! Was there a particular reason not to use ocamllex and menhir?
303.
by murbard2 11y ago
Having some type of hierarchical structure is probably the right inductive bias to have (it may be one of the reasons why deep learning is so successful), but you also need a search direction. It's hopeless to randomly hop in a high di
304.
by murbard2 11y ago
A virtual campus with thousands of hyperfocused John Von Neumanns collaborating telepathically and accomplish subjective years of works in seconds of real world time.
305.
by murbard2 11y ago
If I had to guess I would say that: - It involves generative models (needed to make inferences from very few examples) - It is still a connectionist approach (typical probabilistic programming is great if you have a lot of insight into the
306.
by murbard2 11y ago
What happens if a malicious forks is introduced in the blockchain? I see two possible outcomes, with the latter much more likely than the former: a) The blockchain protocol is the final arbiter of ownership, and those who purchased the stoc
307.
by murbard2 11y ago
Mostly agreeing, but your numbers on embryo selection assume we have a reliable model for predicting IQ from genome, which AFAIK we don't have. How much of the variance of the IQ can we currently explain out of a genome?
308.
by murbard2 11y ago
I somehow missed that part, thanks for the clarification.
309.
by murbard2 11y ago
Why do you expect to see a difference between the two if the two groups come from the same population?
310.
by murbard2 11y ago
What type of control are you suggesting? If you want to rule out genetic effects, you'd have to look at adaption studies.
311.
by murbard2 11y ago
It's sad because that's an obvious hypotheses (IQ correlates to income and is hereditary) and they don't even put it an the table, at best for fear of being treated like pariahs, at worst because they have internalized the ab
312.
by murbard2 11y ago
It's not the same type of difficult; it's extremely painstaking and it's by no mean an easy task, but it's much less conceptually hard than a 3D engine, let alone an AI.
313.
by murbard2 11y ago
Two comments: 1) it's not surprising: a lot of very simple systems can compute. It seems to naturally happen as you add flexibility to your rules. See for instance Wang Tiles or some cellular automata. 2) it's not a good thing: it
314.
by murbard2 11y ago
They might be able to run Grover's algorithm, which would effectively require the doubling of key lengths to preserve security in symmetric ciphers.
315.
by murbard2 11y ago
Molecular nanotechnology
316.
by murbard2 11y ago
There are two big issues with basic income: 1) Even if you spent the entire $2T annual federal budget on a basic income, it would come to about $500 per person per month. The cost is simply gargantuan. 2) It would not replace welfare, it wo
317.
by murbard2 11y ago
Much more precise as well as much more simple: test whether or not she is able to distinguish some wavelength from any combination of red, green and blue.
318.
by murbard2 11y ago
My money is on Ukrainian based on the handle. (Slovak would be a candidate too but less of a fit for "many programmers, etc")
319.
by murbard2 11y ago
Have you seen Mallat's recent work on the scattering transform? http://www.di.ens.fr/data/scattering/ https://www.youtube.com/watch?v=wHhYvtnY2zI
320.
by murbard2 11y ago
To be honest, I haven't used it seriously enough to give you the good answer this deserves. A friend of mine evaluated it for his own need and went with Meteor citing that reason (plus ease of hiring for js). I've gotten a similar
321.
by murbard2 11y ago
You can also run eliom ( http://ocsigen.org/eliom/ ) which offers reactive DOM elements. Not as seamless as Meteor, but you get something pretty close with the benefits of type safety, down to the HTML you generate.
322.
by murbard2 11y ago
Of course there is: direct enumeration. If you insist on a practical way to do it, gradient descent on the latent vector would probably work well given the type of results DeepDream has obtained.
323.
by murbard2 11y ago
TL;DR the thesis researches how evidence-based programming language design could be conducted, but it does not actually goes into what such a design would be given the available evidence.
324.
by murbard2 11y ago
Not necessarily shortest, but it has many short programs. The algorithmic probability of a sequence x is the sum over the set of all prefix-free programs u that calculate x of 2^-len(u). Think of the shortest description length as a MAP, wh
325.
by murbard2 11y ago
More precisely it would be the algorithmic probability. Not only are there short programs that compute mathematical constants, but there are many such programs, so the total mass is even higher.
326.
by murbard2 11y ago
Many series that calculate pi are very simple, as a result, it can randomly appear in physical problems. See for instance http://ics.org.ru/doc?pdf=440&dir=e
327.
by murbard2 11y ago
"Parasitic worms bore into our organs, steal our nutrients, and sup on our blood—but their effects aren’t all harmful" The point of the article is largely that the same is literally true of babies :)
328.
by murbard2 11y ago
Two disappointments I have with K/Q is that 1) functions that take multiple arguments break the pattern of concatenating functions to build advanced expressions (yes, there is . but it's not nice to use) 2) the right to left evalu
329.
by murbard2 11y ago
No it's not "just math". You need to know what questions you're asking. You can't wave this under "do whatever works best" because evaluating "what works best" requires epistemology, and this is
330.
by murbard2 11y ago
Thanks for the Skype tip, I'll try that. But the fact that I need a tip on HN to know which config files to edit to get Skype working is precisely why Linux on the Desktop isn't quite there yet :) (the fault may very well lie with

More ›