std_throwaway

Gains in instructions-per-clock start to flatten out. And that's where the gains were coming from in the last years. Some time ago a paper was posted here that showed how even if you have an infinite amount of transistors, you will still be limited in the range of 3-10 instructions-per-clock for typical programs.

Clock speeds seem to have leveled and IPC will only see another gain of 50-100%. Single threaded performance is close to the limit. What after that? Is this the end?

Our body already has this ability but does not use it often.

Does it lock the GIL so numpy can release it again immediately afterwards?

That's a better analogy, thanks.

Successful companies follow the way of the money. Everything else is charade.

I can't seem to find the kickstarter link.

Analysis paralysis goes well with bikeshedding when people who are not into the technical details are involved in the decision making process and start arguing about not-so-relevant details.

I'd like a situation where the affected people give input and management decides who shall be involved in the decision making and delegates the decision and parts of the execution to those people.

> Different situations call for different approaches.

This leads to a situation of growing chaos where not everybody can understand what everybody else does. Your engineers are no longer easily replaceable and creating interacting systems gets growingly complicated.

Mix this in with limited time and resources of management, it's easier to give out guidelines than to deal with the mess that'll be created when loosening the leashes.

(It all totally depends on the situation, though.)

Good engineers solve problems well.

Excellent engineers solve better problems.

Usually it's a long path from something a customer wants (or actually needs) to finding an appropriate implementation (if there even exists one). Many things can go wrong along the way and miscommunication happens in both directions.

In many ways Tesla is the China of America.

Why is Airbnb responsible for the action of criminals? Shall they ban each and everyone who gets accused of something without proof? Don't we have to police and justice system to sort such things out?

It's not overvalued if people are still paying the price. Only the last fool in a long line will get hurt.

It's not much better than searching for the cure for all illness.

Drink enough, sleep enough, eat well, do some sports, meditate. If that doesn't help, good luck with going to the doctors.

Those 70 are not controls but each may have a different disease.

Things you can take on a plane would probably be alright.

You won't get fires if there is nothing that burns.

The best programmer is the one who can solve the problem effectively and efficiently without even touching the keyboard.

You don't have to be "good". You have to solve the right problem efficiently. Find the right problem and be the solution. If you need to write a program, make an easy one. Easy to use and easy to maintain. You'll get what that means after a few years of maintaining. Nobody ever cares if you're a "good programmer" or how you feel about yourself.

This statement is both correct and useless. It deflects but does not add depth or insight to the discussion.

> I am unsure what you mean-- do you mean with different training sets but the same testing set?

Yes, assuming we have 10000 different training images. Divide these into 5 sets of 2000 each and train 5 networks with them. Assuming that 2000 images are plenty for this application, we will have 5 well trained networks that have similar performance for a test set.

BUT

They will work slightly differently internally and those "inverse gradient search" methods (or what they are called) might only be able to manipulate an image for one network at the time with "specifically chosen additive noise" while the other 4 are unimpressed.

That's assuming that the manipulation can't be targeted at all 5 classifiers at the same time.

Does this effect carry over to classifiers which were trained with different training data?

So, we're going to get compile time reflection?

I see they use another approach to file systems. They sound very confident. A bit too confident to my taste. Are they experienced enough in building file systems that they won't run into huge problems arising when it actually goes into the dirty details?

In python there is the term "pythonic" describing if something is in the spirit of the languages founding fathers. As in: "The most pythonic way to write this down is ..." or "Map/filter don't feel very pythonic."

What is the equivalent term or expression for being in the spirit of C++?

The article, while a joke (just like C++ originally, look it up), feels very C++y.

It's a joke.

You have to go with the times or you have to go in time.

Also, change your code often and hire/fire programmers at a high rate.

In order to fight the virus you have to become like a virus in many ways. Like them, you have to hide your own code in order to avoid detection. You have to get down to their level and use dirty tricks.

Also keep in mind Heisenberg while traveling along state lines.

If one meeting ends at t1 and the other meeting starts at t1, do they conflict?

To which meeting does t1 belong? One, both, neither? Can we set up an event at t1 which is not in conflict of either? How about two events at t1?

You don't buy a CPU alone. You buy a whole server. If the CPU increases 1000$ in price to gain 10% performance but the complete server already is 20000$, it's quite worth it to go from 20000$ to 21000$ for 10% of performance.

Also if you factor in a performance critical software for 100000$ that runs on that single node, you will buy the fastest hardware that you can get for a few thousand dollars more.