l3robot

Not to mention that even the utility function of the "millenials dilemma" makes no sense here, it's not even consistent like it has to be. What it represents here?

- More wealth is better? Then 0/10 for chill/work maybe is ok, but it make no sense to have 8/8 for chill/chill. - More fun is better? Then 8/8 is ok, but it makes no sense to have 0/10 for chill/work. - More fun in the near future is better? But it is clearly stated that fun will be unachievable in the game since work will prevail as the dominant position.

It is true. It is crucial to understand any kind of growth situation which are predominent in life. For instance, one of my friends who is a teacher wanted to explain sustainable development to his highschool students by showing the effect of cutting too much trees if this doesn't match the trees growing rate. He asked me to solve the not so trivial equations with a series and found myself returning to the same maths used for the interest computation.

Not also to mention something often oversight. An autonomous car can park far away from it's user, meaning we could drastically reduce de number of space taken by parking in the city. First, giving back this spaces to the people as green parks and services places. Second, lower one of the biggest problem in our current city: urban heat island.

I think the reasoning of the Fermi paradox is that "at some point in time" a distant civilization must have been using EM wave to communicate. Now, considering the number of planets, it must exists a civilization in the universe that sits at a distance that let us sense this EM dominated past. And this even if in their own local time, they use fancy new tech. Even if they were not trying to communicate with us per se. Their own radiation should be measured.

Interesting points

Yes, that was my point: they choose one and it works fine. It is easier in that case to ensure and keep a certain mass of users and so, for a long time. That leads to the possibility of blitzscaling and monopoly I think.

To me, the problem comes when you provide a service. The service market seems much more difficult to diversify than the product market. When the need is well filled by one company, it is difficult to keep competition alive. It is obviously not binary. It depends on the service.

For instance, for the product side; Nike, Adidas, etc. are constantly trying to improve and change their products to keep up with the complex shoe market even if they are pretty colossal companies. Because the user need for a new shoe pair is coming on a really short scale and the competition is fierce.

I'm not seeing the same kind of competition with services. When you use a service that fills the need, you don't change in general. Or at least, people change on a very long scale. Some times, it even takes a cultural shift to change habits.

On the google vs bing example: Well, ok some uses bing, but in majority because microsoft makes it easy to do so on their systems ;-). There exists some very good alternatives, like duckduckgo, but it represents a minority.

On Monopoly: It has been a though I have for some time. I wonder how could they not form monopolies. Who has 4-5 taxi apps on their phone? Who has 2-3 social profiles? Who has the habits of jumping from a searching engine to another? A minority. Their product are almost natural monopolies.

They are very difficult to fight against monopolies because when they have user commitment, the need for the product is filled entirely by one company. In addition, it is very difficult to make the user change, it's part of a habbit.

I might be wrong there. Just a though I had, but interesting to discuss. Is it really possible to avoid monopoly with these products?

What a beautiful text. It just made my day. I think, for most of scientists, these previously thought impossible outcomes is as, if not more, rewarding as aligned with theory outcomes. Certainly this is one of the reasons that encourage me to continue in science. Curiosity of course, but also the feeling that impossibility is only temporary.

TL;DR

- New JIT feature that lets you run your model without python. It now seems trivial to load a pytorch model in C++

- New distributed computation package. Major redesign.

- C++ frontend

- New torch hub feature to load models from github easily

Disclaimer: I am not an historian nor a sociologist nor a political scientist. It is pure intuition, so please correct me if you find me wrong.

This kind of article recalling the war declared between Apple and Google by strong characters (reaching almost royal admiration) like Steve Jobs and showing their alliance now forced by economical benefits, brings to my head an observation I make for some years now.

It feels like the world is more and more returning to a, though different, late Middle Age state where kingdoms, merchants/bankers and religion were in control. It feels like we are close to the end of the powerful democratic state.

Yes, these forces has always existed and played a major role in political affairs, but during thr XXth century, we managed to keep it marginal, at least in the West.

Now there is an inversion, Kingdoms are coming back.

There is a lot of example: formation of alliances over sole benefits of oligarchs, growth of organisations never seen from a long time, customer attachment to a company like it was a dogma, tax and regulation ducking by companies, companies wars often more important than wars between states, states bending to organisations wishes, etc.

Maybe to return to this state is a human reflex. Like were unable to create a stable democratic state.

I don't know what to think about it atm, but I'm pretty sure we are on the edge of a great schism with the modern era.

I’ve always wondered if it would be possible to train a crow to pick up loose coins for food.

Yep, this guy did it and it is quite fascinating; TED talk: https://www.ted.com/talks/joshua_klein_on_the_intelligence_o... Website: http://www.josh.is/crow-machine/

And also, this dutch startup is aiming at training crows to pick up cigarettes: https://www.crowdedcities.com/

RIP to one of the great. Not only he was one of the genius minds of our time, but he also was an important symbol of perseverance for all of us.

Still on Sierra. The High Sierra installer has been downloaded without me knowing.. I receive an annoying weekly notification that try to make me install High Sierra. And now all these really concerning bugs too.. it makes it worse Apple..

"It was a time when companies weren’t afraid to invest in basic science." No they were probably afraid, but they were forced to invest in science by states. AT&T did not decide to invest massively in science and risky projects like Unix, they were forced to. Please stop thinking companies are behind innovation. A great piece of article that demestify this myth: https://www.theguardian.com/technology/2017/may/11/tech-inno...

Where does the article state that zero training error is a good thing? The authors only show that almost every modern neural network can reach 0 training error, even if the labels are randomized (generalization impossible). Hence, they can learn the dataset by hearth. The authors can, from that, use the testing error as a generalization indicator.

Indeed a careful hyperparameter choice is the only key now to have good generalization. As I understood it, the goal here is more to show that the correlation between the regularization of the network and its generalization power is far from being clear as it is for other ML algorithms like SVM.

In short, NN hyperparameters help to reach generalization, but cannot "explain" it. It's the key difference here between practice and theory.

It's almost the same in Quebec and I believe everywhere in Canada. Thanks for making this point clear!

You are absolutely right, I wanted to say the ribosome was. I confused both. My bad.

There's computations based on information theory that can give insights about the amount of information in a signal. A mathematician could probably give a more complete answer, but I think it's the idea.

What a great blog post! Totally agree with him. And, personnaly, it is why i'm having so much fun doing maths. Everytime it's a new exploration, a new challenge. My best math teacher I had was seeing math with this philosophy in mind and his class was like discovering new lands every time. I'm sure that if we explained in a way that failing a math problem is as normal and challenging that failing a Mario Bros Level, more people would be in peace with it.