inlineint

I used this approach, but the problem with it is that all the plugins should be synced in it, so the repository becomes larger and larger with time, or kept as submodules what adds additional friction. Also you should manually copy .vimrc to your home directory from the cloned repository.

So I put to the repository only .vimrc and an short shell script install.sh that installs Vundle, .vimrc, and then installs all the plugins by running "PluginInstall" command in vim. Installation becomes just cloning the repo and running install.sh from it.

The additional benefit in comparison to say putting plugins to git submodules is that I don't have to keep my vim repository at GitHub up-to-date with latest plugins versions because Vundle anyway installs the freshest versions.

And Redis, multiple instances of which are supposed to be plugged into gryadka, means "radish" in Russian :)

Some LCD displays use flicker to set backlight LED brightness.

On the other hand SVM doesn't scale as well as neural networks do because it has computational complexity between O(n^2) and O(n^3) [1] where n is the number of samples in the training set. So if you plan to add more data later you may eventually encounter scaling problems with SVM.

[1] http://scikit-learn.org/stable/modules/svm.html#complexity

> But if there are thousands of different such "surprising-to find-particles" to possibly detect, is it actually surprising to observe one of them?

It is hard to correctly identify what exactly particles from those possible thousands are observed in given data. The are two main problems: the properties of those not yet observed particles are not well known (because it is computationally hard to predict them from the standard model) and because the number of useful events is much much smaller than the number of events that correspond to already known events.

> So do elementary particles decay or not?

I don't see a contradiction here: the first generation of elementary particles does not decay (or has not been observed to decay yet), Higgs boson is not from then because the author of the answer are talking about the first generation of fermions and Higgs boson is not one of them.

The standard model provides a set of postulates that could be used for prediction of possible composite particles, their masses, and decay times.

However it is computationally infeasible to calculate them directly, without using various approximations. Physicists try to solve these problems numerically (see for example about the field called lattice QCD), but it is not always possible and leads to introducing various approximations that produce errors and other artifacts in the numerical predictions.

So the particles were allowed by the standard model, but we didn't know for sure their properties. So this provides way to verify already done numerical predictions (I don't really know were they be done for this exact particles or not) and give us data about exact properties of these particles.

One could possibly draw an analogy with (quantum) chemistry here.

Actually it is even a one new particle (Omega_c baryon), but they observed five different excited energy states of it (like observing different excited states of a Hydrogen atom), so called resonances [1]. But the discovery is still exciting because it should have been really hard to find something that we don't really know how looks like in such large amount of noise.

The problem is that it is hard for us to predict masses/energies of new composite particles because although Standard Model provides hypothetical way to do it, it is infeasible computationally.

[1] http://physics.stackexchange.com/questions/64862/resonances-...

There are some speculations about a new ~10.5 inch iPad.

Some high level overviews: phys.org [1] and a detailed answer on physics.stackexchange [2]

[1] https://phys.org/news/2013-07-physicists-publish-solution-qu...

[2] http://physics.stackexchange.com/a/71800

Note that normality of a number in a given base is a sufficient condition for existence of all finite sequences in its expansion in this base, but not necessary.

It might turn out that Pi is not normal say in base 2, but all possible finite sequences of 0s and 1s occur in it, just with some of them being slightly more frequent than others of the same length.

They could add button "Report discrimination" for rejected potential renters and then ask them what type of discrimination they want to report. Then they wouldn't need to guess anyone's race.

Sorry, it was of course 0.02 eV, not just 0.02 :)

k = 1.38e-23 J/K = 8.6e-5 eV/K, so kT = 0.025 eV for T = 300 K.

Just note that in practice nobody wants to keep their data storing atoms frozen near absolute zero, but rather prefer to have them at temperature close to 300 K.

However to store 1 bit of information at given temperature the energy difference between state corresponding to 0 and state corresponding to 1 has to be not less than something of order kT ≈ 0.02, otherwise the information would be quickly erased by thermal motion. But if we take maximum energy gap at atom that might be used for storing information to be upper bounded by atom's ionization energy [1], it turns out that it can't be larger than something of order 10 eV. So it doesn't seem to be possible to store more than hundreds or thousands of bits per atom at room temperature.

[1] https://en.wikipedia.org/wiki/Ionization_energies_of_the_ele...

With condition of being suppled by resources for living I think it's a good idea to go into research.

It's a bit unrelated, but I think that maybe instead of just introducing basic income to suppress social tensions caused by bots taking jobs it would be better to spend the money to create new research jobs at large scale. There are plenty underexplored directions in science and research in lots of them can't be automated anytime soon.

If there were much more PhD positions that paid solid stipend (and colleges were accessible to everyone and paid solid stipends too) and there were a lot of open research positions, people losing their jobs because of automation would go to these programs and then do research which eventually might make the world a better place to live.

Just replying to someone who is going to say that not everyone is able to meaningfully participate in research, I think it is not true. Yes, different people might have different productivity and abilities, but science needs not only researchers generating new ideas, but also laboratory assistants and experimental scientists mostly just doing meticulous measurements.

However it's simpler to just throw money at people and tell them that they don't need to work rather than scale the educational and research systems to the extent that would allow them to engage a large part of the population into them.

It is unclear what kind of technology are they planning to use for this shield. If it is usual electromagnets, it's unclear where would they take energy from, if superconducting ones than it's unclear that it would be possible to keep them in superconducting state for long.

Wish someone more knowledgable about this kind of tech comment on this.

Did you try Qt? Creating GUI with qtdesigner is similar to Delphi, and Qt Studio is well integrated nowadays.

I think that Continuum Analytics does incredible work in this direction, so it is really possible to just do `conda install matplotlib` and install matplotlib on any distribution.

It seems that something like entropy-based helpers that predict probabilities of next letters and make those letters large could increase input speed. One of the first examples of such helper was done in David McKay lab: https://en.wikipedia.org/wiki/Dasher_(software) .

I don't like this use of dual numbers notation for differentiation because division by a dual number is not defined. For example, how would one calculate ε^2/ε? If ε is a matrix [0,1;0,0] then it doesn't have inverse and thus the expression could not be evaluated.

On the other hand, little-o notation [1] was invented for exactly this purpose. It is easy to evaluate derivatives using it, for example (x+ε)^3 = x^3+3x^2ε+o(ε), and so ((x+ε)^3 - x^3)/ε = 3*x^2 + o(1), and o(1) tends to 0 for ε tends to 0.

[1] https://en.wikipedia.org/wiki/Big_O_notation#Little-o_notati...

It would be really fascinating if you added GPU instances with GTX 1080 cards. These cards could allow you to make the prices much lower than those of AWS GPU instances that use K-80 and make it a perfect fit for Deep Learning applications that don't require double precision.

And it is processed not by Vim itself, but by the terminal emulator that sends Esc character to Vim. So the shortcut Ctrl+[ was not invented by vi author, but is a consequence of using Esc to exit from the edit mode.

There is also another fundamental limit, the speed of light.

Because of the special relativity theory nothing including electrical current can transmit information faster than the speed of light (c = 3*10^10 cm/s). So if maximum size dimension of your CPU is say 2cm (I'm ignoring that they actually can have multiple smaller cores), then the maximum physically achievable frequency is upper bounded by size/c = 15 GHz.

Although it has not been reached yet, it has the same order of magnitude as the record frequencies achieved by overclockers using liquid nitrogen cooling (something a little below 9 GHz). Also it shows that reducing transistor size and as a consequence overall CPU size increases maximum physically allowed frequency.

> Once we've accepted the hypothetical that technology allows for a single car to fly from A to B safely (which is almost certainly automated), it seems clear that having cars move at different altitude will be trivially possible.

In real world we can't accept this because there are always unpredictable factors, we can only say that probability of a safe flight in a single car is say 99.999%. Then if there are 20 cars one right above another, then probability that none of them crash is only 99.98%, and it could damage all cars below it, so the resulting probability of safe flight in a car reduces to something of order 99.99% instead of 99.999% for a single car. And 99.99% is really not safe enough for everyday flying during many years.

The article recalls teletype terminals. I'm just putting a video showing how it worked in close: https://www.youtube.com/watch?v=MikoF6KZjm0 . It hypnotizes IMO :)

I'd like to see how looks editing and running BCPL/C programs using ed on such a terminal.

Take a look at Google's "Best Practices for ML Engineering" [1] and "Machine Learning: The High Interest Credit Card of Technical Debt" [2]. They are not tutorials, but cover a number of interesting topics related to deployment of machine learning models to production.

[1] https://news.ycombinator.com/item?id=13414776

[2] https://news.ycombinator.com/item?id=10338575

This MBP model is a little bit dated and the photo is blurry: http://www.nerdalize.com/wp-content/uploads/2016/06/mockup-e...

An alternative to Cython is Numba [1], which speeds up some cycles in pure Python by just adding a single decorator.

[1] http://numba.pydata.org/

I think it's worth to mention here that Wolfram site also has a so called "Wolfram Programming Lab", which is basically a free access to Mathematica hosted in a cloud and not requiring registration.

Link to create and open a notebook: https://lab.open.wolframcloud.com/app/view/newNotebook?ext=n...

There is an open source webcam driver now [1], although it has some problems.

[1] https://github.com/patjak/bcwc_pcie

This submission seems to be a kind of follow-up to the one about Braess's paradox (https://news.ycombinator.com/item?id=13352513).