inlineint

I'm reading an early release of this book in O'Reilly Safari and confirm that it's worth reading.

However, if you don't get used to deep learning previously it might be worth to combine it with something more foundational about deep learning. I like the lectures by Nando de Freitas (https://www.cs.ox.ac.uk/people/nando.defreitas/machinelearni...) and a book by Ian Godfellow (http://www.deeplearningbook.org/).

I didn't have much trouble by just booting from an Ubuntu 16.04 USB stick and installing it to the entire SSD on MBP 2015. Almost all of hardware works, but require installation of some modules and changing configs.

Shameless plug: I prepared an Ansible playbook that does those manipulations automatically: https://gist.github.com/a-rodin/7930a50befbc7b07e20c5c3b5aca...

I'm not sure wouldn't someone say that it is bloating of such simple format as Markdown, but I really want to see embedding of TeX math formulas in Markdown RFC/standard.

I'm talking about $\frac 3 4$ and $$\int x dx$$ notation. It is already implemented in some Markdown parsers, such as Pandoc and Jupyter Notebook, but because it is not standardized there are also a lot of parsers that don't support it. I think that adding it to the RFC/standard would reduce friction in math communications and make them more accessible, especially if Markdown format become widely used for email communications.

But any measurement device always has a conscious observer that reads in one or another way results of measurements (e.g. spin directed up or down) from this device. Without such observer a combined system of the device and the quantum system of interest would infinitely be in a state of superposition.

Could you elaborate what's wrong with that idea?

As far as I could understand he just talks from psi-empistemic point of view about wave function collapse. Although now we don't have any way to determine whether psi-epistemic or psi-ontic point of view is 'correct' (as like as define what precisely does mean 'correct' for this question), I don't think that you can just call one of these views nonsensical, especially without support arguments.

Did you try GCC with -O3 and -ffast-math? It seems like ICC has its analogue to ffast-math enabled by default: http://scicomp.stackexchange.com/questions/20815/performance...

And Gedit with Kate are ported from GNU/Linux to Mac nowadays.

How much computing power/time did it take to train the model? Had any hyperparameters optimization been done?

Sorry if I missed the reference that has this information, but I think that these numbers are important in any deep learning experiment because they allow the readers to evaluate applicability of described methods to their problems.

I'm by no means an expert, and I may be wrong, but I personally don't think that it's possible to achieve superconductivity at room temperature and normal pressure using this approach at the current state of the art.

If you take a look at the phase diagram of the water [1] you could notice that the pressure required to crystallize water at temperature near 105°C is something near p_freeze = 2.5⋅10^9 Pa. However, the water is inside of the nanotubes of radius r = 0.5⋅10^(-9) m (the article says about the tubes being of diameter 1.05 nm). Such small radius creates significant surface tension (the same force that allows soap bubbles to exist) which is expressed by equation [2] as Δp = γ/r. According to [3] the surface tension coefficient for water at 105°C is γ = 58⋅10^(-3) N/m, so Δp ≈ 1.2⋅10^8 Pa. The required for crystallization and the actual pressures differ in just one order of magnitude.

But the reasoning above used the macroscopic physical laws, and of course one could expect deviations from them when thinking about diameters as small as 4 water molecule diameters. Of course there is still no theoretical model that describes this configuration and calculates exact values of p_freeze and Δp precisely for it. But such model don't have to change the orders of magnitude in the problem, just estimate factors of order of 1, and we could use rough intuition gained previously about the process even without such model.

So if you try to apply these intuitions to the superconductivity you immediately meet difficulties because all known superconductors are in solid state, not liquid. It could be overcome by putting it to the nanotubes in liquid state and then letting it crystallize. Let's assume that this is possible. Evidence [4] suggests that to get a superconductor at high temperatures it requires to have pressure of order 10^11-10^12 Pa. This pressure is two orders of magnitude higher than the one required to freeze water at 100°C. That means that surface tension coefficient in such material should be two orders of magnitude higher than in liquid H2O to create pressure of order of magnitude required for phase transition to superconducting state.

I wasn't able to find data about surface tension in superconducting materials like like H2S mentioned in [4], but I doubt that it is 100 larger than surface tension in water. And it is impossible to significantly increase pressure by decreasing radius of tubes because this radius is already of order of molecules size. So I don't think that it is possible to get high temperature superconductors at normal pressure and room temperature using nanotubes and already known superconducting materials.

[1] https://upload.wikimedia.org/wikipedia/commons/thumb/0/08/Ph...

[2] https://en.wikipedia.org/wiki/Surface_tension#Surface_curvat...

[3] https://srd.nist.gov/JPCRD/jpcrd231.pdf

[4] https://arxiv.org/pdf/1506.08190v1.pdf

Android patches had been merged into upstream a few years ago [1], so its kernel is actually Linux now, not a fork.

[1] https://kernelnewbies.org/Linux_3.3#head-b733d694037e0b34ad4...

> because from a physical perspective the two are simply geometrical transformations of one another

Not really, at least if you want to keep simplicity of Newton's laws which are formulated for inertial reference frames. The reference frame in the heliocentric model is inertial, while in the geocentric one it is not.

Have you considered Thunderbird?

I think machine learning can be employed to filter out accidental touches from the meaningful ones. It is already done in iPad Pro that allows to use Apple Pencil while holding your hand on the screen, and still detects meaningful touches.

It is worth noting that this kind of chat is fundamentally vulnerable to MitM attack: anyone can just open two chat windows and copy all messages from one to the other and vice versa, even by hands or using browser automation.

"...to this day he owns neither a TV nor mobile phone, and only acquired his first computer on his 80th birthday."

https://www.theguardian.com/science/2013/dec/06/peter-higgs-...

Perhaps he used his computer to watch it.

There is a service http://www.cryptograffiti.info/ which can be used for posting sensitive information that should not be removed or hidden. It allows to write a message and store it in Bitcoin blockchain as a transaction. It costs near 0.0015 BTC or $1 per kB. Large files can be posted as magnet links to torrents with them.

Even if the service's site had been shut down, everyone would always be able to obtain the transaction from it's hash using any bitcoin client/blockchain explorer, convert it to ASCII and read the text.

I'd like to note that it is worth to sign with GPG all messages posted that way in order to have ability to post updates and verify authorship.

"Democratize" word meanings from the Oxford dictionary:

1) introduce a democratic system or democratic principles to: public institutions need to be democratized.

2) make (something) accessible to everyone: mass production has not democratized fashion.

The second one seems to be exactly what they say about their system.

The company probabaly can't entirely be replaced by machines because there would be still a need for someone who is familiar with decent tech and is responsible for selection of technology stack, because there would be a number of competing AI workers manufacturers.

> The operator can use foot pedals to control the robot’s movement and a virtual reality headset like the Oculus Rift to experience what the robot is seeing.

Maybe it's a bit off topic, but I'd like to put here fiction sources (which I know) that explore this kind of setup:

1) A Cognitive Discontinuity story by Andrej Karpathy (2015): http://karpathy.github.io/2015/11/14/ai/

2) Sleep Dealer movie (2008): https://www.rottentomatoes.com/m/sleep_dealer/

It is open source. If Mozilla wants to discontinue it you always can host it yourself.

Yes, it's a valid point. That's said I personally prefer Docker for installing databases of all kinds on my development machine.

For Node.js you can use NVM for the same purpose. It is specially suited for handling multiple node verions and installs everything to user's directory.

Thank you for detailed answer and especially a note about two hands. I didn't think about it, but actually it could be a problem.

Right now a MIDI controller with faders/knobs seems to be the most simple and reliable solution and it can be used with one hand as like as with two.

I'm not sure that it good for fine tuning of multiple parameters because it's easy to move hand in a wrong direction. But I can't say it for sure because I haven't use it. The fun thing is that the process of controlling would probably remind playing on Theremin.

Yes, some MIDI controllers with motorized faders look good.

> Not so much if you're a web developer that regularly installs new software to make your dev environment easier to use

If you are a web developer you can run all untrustworthy tools in containers/virtual machines that doesn't have access to your system, it would be an implementation of these principle.

Not all developer tools can be run that way, but pretty much of them can.

The article assumes that you don't find your work interesting enough to keep your attention focused for a prolonged period of time. Maybe it is better to try to understand why it is not interesting and find ways to make it actually catching rather than figure out workarounds. Because why should we spend time on things we are not interested in?

In case of an article (which is cheap) there is no need to wait for confirmation, it's enough to wait for transaction to appear in mempool and have enough fee.

But it is not applicable to humans because in order to use it with human brain one has to modify genes and breed humans from embrions with modified genes.

I believe that AI. But while there are a lot of AI start-ups now, it's hard to pick right ones that can make you rich.

Of course you can invest in Google which claims to be an AI company, but it is already not cheap to buy.