nemonemo

Older doctors can probably diagnose TB better, but I would wonder if that can be used as a core competency measure of a doctor. Also, I believe there are cases younger doctors diagnose a lesser-known disease when older ones could not.

There can be all different sorts of anecdotes, but the link was about a study with more than 100k doctors, which would be a stronger evidence that the probability of survival is better for patients with younger doctors.

Thank you for sharing your story. Last year, I and my wife deliberately decided to minimize applications to schools with my kid who seems to have similar social/academic tendency as your first one. We applied one school where such kids could also be thriving, and ended up not getting accepted. She would need to go to a school this year, and within a year, my kid became much more mature than a year before, and we & people around her think the delay was probably good for her. Just wanted to add my anecdote.

Ah, I forgot talking about reading others' code. Reading good code is important. Understanding good design is important. I've read tons of Linux kernel code, source code of Linux packages, golang code in github, etc.

Also, there are many books that discuss software design. The one that helped me most was "Principles of Computer System Design". Not all the parts are great, but reading the first few chapters & recommended books changed my software writing.

Write your own code, get feedback, read others code. Repeat them all again, and again.

At an earlier stage, building my own sizable project helped me the most. I still remember writing bunch of Java code in the middle of Iowa when I was an exchange student, which ended up helping a lot of my friends course registration in my university and got great feedback about the UI of the program. I learned a lot, especially about all the basic system building, and the best thing was that it gave me the confidence that I am capable of building something useful.

Since then, one of my weekend hobbies was doing some fun project that does not need a whole lot of effort -- like making a simple dashboard for the family in Ruby, building a home automation bot in Go, writing some my own investment tool in Python, etc.

It's just the same as picking up another language or any other skill: there is no one path people learn it. The way I learned English was mostly by reading many books, and many other people learn a second language mostly by watching TV shows or talking to people. There is no set path -- although there may be necessary things people would need to do regardless to achieve high-level of mastery, like talking and reading in a language -- but one thing necessary is an incentive, especially a positive one, and building confidence over time.

My undergrad university had the same issue, and the last glassblower retired 15 years ago founded a company and kept supplying the necessary equipments to the research cluster.[1]

The glassblowers there seemed to think that the repair of broken glass equipments is potentially hazardous to the glassblower, and one of the founders who handled more repair requests had two cancer surgeries. It's only an anecdote, but it could be a reason why it is difficult to find a successor of this 71 year-old person.

[1] https://translate.google.com/translate?u=http%3A%2F%2Fwww.ha...

Then you need to rely on the checks and balances between branches of the government. As you said, the checks and balances are ripe for disruption or at least improvement.

Lee asked to be black, because there's 7.5 points advantage for the white who follows the black, and Lee won as a white this time.

I am from South Korea, and I know this fan death thing could seem superstitious, but as another comment pointed out, it is a cautious approach toward unknowns, in addition to mass version of confirmation bias. (News reports a death with a fan on -- oh, that scary scythe of the fan..)

But who knows? There could have been a death that could have been prevented by turning off a fan, and nobody can verify its nonexistence. My attitude toward this is like my agnostic attitude toward God: I just can't know its existence or nonexistence. And I think what the people in the article are expressing can be understood in a similar line (though 100% is an overstatement.)

Still I sometimes argue with Koreans with relatively less background in scientific thinking about scientific things like if fan deaths are possible, or using metal spoons for yogurts kill more microbes than plastic spoons, etc, but many times, my conclusion is that I just can't have any conviction about either ways.

This article contains good tips for building a GPU cluster with RDMA. One thing I would like to add is that there are two types of GPUDirect depending on CUDA versions. Previous CUDA supported GPUDirect through CPU memory, and now CUDA supports "true" GPUDirect between the RDMA device and the GPU memory. However, some chipsets may not support the "true" GPUDirect very well, and two of our old machines had up to 20x times of throughput asymmetry with GPUDirect (which is, send was much slower than recv.) There are several papers that discuss this limitation. Our work, GPUnet[1], overcame this performance issue with GPUDirect by using fairly recent chipsets, but you can probably imagine our pain when we saw around 150MB/s throughput with GPUDirect, when 3GB/s is the expected one.

[1] GPUnet: Networking Abstractions for GPU Programs, OSDI 2014 https://sites.google.com/site/silbersteinmark/GPUnet

I think the fundamental reason of not being finisher may come from several reasons: lack of detailed planning (oh, there's an unanticipated obstacle), lack of energy to finish things, and little barrier of attention shift.

I would say I am not a finisher, but what I am trying to do is reducing these reasons: trying to understand/plan better to the deepest level, sleeping/eating well, getting more satisfaction from finishing things than starting one, and doing smaller things, selectively.

This is a very bold statement. IO-bound means that the bottleneck is at IO. GPU has little benefit in IO.

Also, mapreduce is not embarrassingly parallel. Out of mapreduce, map is the only embarrassingly parallel phase. It has input parsing, shuffle, reduction phases that are not embarrassingly parallel.

GPU acceleration sounds all great.. until anyone actually generalizes the framework, implements applications, and measures even the most trivial workload. There should be some niche applications you may get 2-10x benefit, but 1000x is either measurement error, apple-to-banana comparison, or a new marketing gimmick. Unless you have a very fitting application, it is very easy for GPU to have less performance than CPU. I would be very curious what application actually could have 1000x performance enhancement. Even memory bandwidth, the main benefit of GPU, does not differ by that amount.

Your comment conveys another implication that UN is supposed to be impartial to issues between member countries. That cannot explain veto power of five permanent members of UN security council, which no other country has.

On consistency: I don't think UN, as a group of nations with conflicting goals, tries to behave in a consistent way, although people are free to blame it for its inconsistency.

NK is aiming the weapon at US and its allies, unlike India or Pakistan who aim at each other.

B.R. Myer's talk explains the motive behind NK's nuclear program pretty well.

http://www.booktv.org/Watch/11315/The+Cleanest+Race+How+Nort...