nthompson

Automatic differentiation only works on functions which are locally complex analytic. It fails on things that fall outside that model, but are still differentiable. Daubechies wavelets are a good example of this.

When I took a tour of the Oak Ridge Leadership Computing Facility a few years ago, Buddy Bland, who is Project Director, told me he could tell when he comes to the Lab in the morning if they are running the LINPACK benchmark by looking at how much steam is coming out of the cooling towers.

https://blogs.mathworks.com/cleve/2013/06/24/the-linpack-ben...

Different programs consume dramatically different amounts of power.

They do not have their own power station. They have the Bull Run coal plant and hydro plants in the area. They do coordinate with TVA before a run.

Pulsed femtosecond lasers have repeat rates in the megahertz. The total energy is very high. When I took laser physics in grad school, they were able to drop a razor blade through the beam and slice it in half. That was in 2013.

So yeah, pulsed lasers can do it, and I'm certain the Air Force would rather be using them. They are finnicky though.

The product lineup is solid; even if they go bankrupt, the debt owners will simply wipe out the shareholders, prune away the less profitable parts of the business, and scale back the expensive R&D ambitions. The Model 3 will still be produced since the fixed costs are already paid.

(Not my view that they actually will go bankrupt. But failure due to overspending on fixed costs tends to work out differently than being unable to cover marginal costs.)

Yeah, but medical bill collections has way less weight on your credit than (say) a car bill past due.

They send so many people to collections that even creditors don't care about it anymore.

Do you work in medical billing? My experience is just the opposite, they try to extract marginal costs from those people.

Buddy of mine breaks his collarbone, gets some pins put in. They drop him a bill for $25k.

He says "I'll give you $1500 now, or you can send me to collections."

They took it; now I use this technique all the time.

They make up a number which is how much they hope to get out of you, and you should return the favor by make up another number. Then scream at each other on the phone for a few hours before coming to terms.

Sad that it's come to this, but medical billing is a shell game. You can play it too.

He's making a joke that the arbitrage could generate infinite amounts of money by generating clean power with coal power.

haha they ought to try that.

The Idaho legislature is totally dominated by farmers (I mean this literally: The Idaho congressmen are primarily farmers.). These farmers are allowed way better deals on hooking up solar and providing it to the grid than city consumers. In addition, a large number of farmers already have solar panels since grid hookups are very expensive.

The "California arbitrage" has been very profitable for Idaho for many years: This involves selling California renewable hydro, for which they pay a premium, and buying coal from Utah to replace it. (Ever driven north of SLC to Idaho on I-15? Those high voltage transmission lines are there for a reason.)

If I had to bet, I would suspect the lines from the Nevada coal plant also supply California, and Idaho Power will sell the solar production there.

Location: Knoxville, Tennessee

Remote: Yes

Willing to relocate: No

Technologies: C++/HPC/Python/JS/Numerical Analysis/Signal Processing

Email: nathompson7@protonmail.com

I have written multiple scientific desktop apps, scientific web apps, written multithreaded and CUDA HPC codes in C++, lots of scientific Python code for low-CPU apps, scientific visualizations in Three.js and D3.js, and I wrote much of boost.math, which you can clone and look at my commits. My boost commits are basically a list of "glue" tools for my projects, quadrature, denoising, interpolators, statistics, so on. Fairly experienced in Eigen for linear algebra.

Can you selectively grow semiconducting tubes, or metallic tubes, or do you only get a mixture of semiconducting and metallic?

That sounds tough. You also have a code size/speed trade off that must be managed. Where did yours land?

Awesome list of questions, btw.

The question about Gracie jiu-jitsu is not so interesting anymore: In the early 90’s, there were simply very talented jiu jitsu practitioners who were close to the UFC. Now we’ve witnessed many strategies used successfully in MMA, like American college wrestling (Ben Askren, Matt Hughes), Thai clinches (TJ Dillashaw), kickboxing (Holly Holmes), and Judo (Ronda Rousey). In fact very few fights end with an interesting jiu jitsu submission anymore, the armbar and the tear naked choke are the most common. These techniques were never unique to jiu jitsu.

Very cool analysis.

I would add: A lot of regex work is still being done out of boost.regex; maybe people still haven't done the switch for most projects? For example, I think boost.regex is still used in Chrome.

Just ran your code on boost, looks like "shared_mutex" is the least used header there.

I use <future> all the time (in boost, no less) so it might be selection bias.

Are you sure it was three miles? Serious wildfires make your eyes burn 50 miles away.

I imagine both steeper and flatter. The flatter curve seems to be easier to handle within my naive mental model of how utilities work.

Isn't this problem caused by fix axis solar? If all solar panels were on trackers, wouldn't the curve change form and partially mitigate the problem?

I should've clarified: I meant portability of code so that is can be compiled by multiple compilers, VS, clang, gcc. The feature set for C++ is so vast that it's almost certain that you'll write code on for one compiler that doesn't compile on another.

At APS March Meeting in LA, there were groups fabricating Josephson junctions. They appeared to be succeeding at ~15mK and individual logic gates, not ICs. I got the impression that their field was advancing steadily, but there was no excitement in their section, unlike anything with buzzwords like "machine learning", or "topological".

I learned x86 floating point assembly from Ray Seyfarth's book:

https://www.amazon.com/Introduction-Bit-Assembly-Programming...

I figured out how to write some basic stuff, then I learned to use perf to performance tune. Being able to read a little assembly was a revolution for me.

I also chuckled about how C is non-portable; omg C++ is so much worse. In some sense, x86 asm is more portable than just about anything now, sure you can't run it on ARM, but your C++ code won't compile on ARM either and you'll be in #ifdef hell for two weeks to get it there, at which point you'll give up.

Look at the git makefile: https://github.com/git/git/blob/master/Makefile

You know what you have to do to build git? Type make. It's amazing.

A-B testing of article titles should be regarded as a breach of journalist ethics.

Really cool tool thanks!

One problem to maybe take care of next iteration:

$ top -o cpu LuluDaemon 29.5%

Yeah, I have dreams of using crystal field splitting to help design semiconductor lasers in bandgap.io, which would be a nice applied group theory problem. Its about 3 years out on my roadmap!

I run bandgap.io off of protonmail. I have had no problems with deliverability, have had great interactions with customer support, and enjoy using the product. After email hack after hack, I just couldn't see how I could ethically store my own users support emails unless they were end-to-end encrypted.

Potential improvements: Searching the inbox isn't great, and I'd like to reduce my attack surface by moving my transactional email from sendgrid to protonmail (not yet supported). Also, it might make sense for protonmail to become an OAuth provider-I'd be willing to support it on bandgap.io

Awesome! Super responsive and fast.

When wolves got reintroduced to Idaho, a few hillbillies went up into the primitive area and stole some pups, and they circulated around as pets. My buddy had one, and it scared the shit out of me. It was 130lbs of killing machine. They could tell it to kill anything, and it would do it. Their neighbor's dog had been biting them, so it was the first to go. Then they'd set it on foxes, which is could run down with ease. But the only person that wolf had respect for was the father, who was 6'4" and would pick it up and slam it whenever it misbehaved. Eventually, it started trying to beat up on the kids to move up the pecking order, and it had to be shot.

So only get a wolf if you can stomach putting a bullet in its head after it brutalizes your kids.