LeegleechN

This idea is common in Shogi piece designs for those who can't read the traditional Kanji designs.

The molecules in a crystal still vibrate kinetically. A solid with no molecular motion would be at absolute zero temperature.

Ableton Push 3 (standalone version). It's only an 8x8 grid though.

The Airthings View Plus is a good one. It has high quality (for consumer) particulate, CO2, VOC and Radon sensors and a good interface. Also tracks temperature, pressure, and humidity for completeness.

Practice like you are learning a musical instrument or chess. Break it down into very specific subskills (like a specific chess endgame or guitar picking technique) and practice those in isolation in addition to your usual practice. Also find a coach if possible.

Justinguitar.com for guitar.

Feynman diagrams are a representation of an integral so moving beyond them is a question of reformulating the mathematical basis of physics.

There is some work in that direction (look up 'Amplituhedron') but overall it's not that surprising that physicists are using computers to do the same math with much more precision than they could before rather than shaking up the fundamental mathematical tools of their field which have stood since the 1600s.

He's not wrong about the narrow algorithmic use cases (so far), but he's completely missing the utility for simulation of quantum phenomena (chemistry, microbiology, materials science). That use case alone completely justifies investing into them even if you don't care about advancing science.

I think one of the biggest applications for quantum computers which isn't well appreciated is simulation of quantum systems: chemistry and materials science. Similar to the old analog computers which were used to simulate differential equations long before they could be fully "virtualized" (simulated deterministically without noise) using digital algorithms.

There are always alternative ways to describe the same phenomena. For example QM could be done with only real numbers (you can emulate imaginary numbers with matrix operations). There is no clear cut notion of what is "required".

No, they can solve integer factoring. This result shows that polynomial factoring is fundamentally different from integer factoring. The latter is believed to not be NP Hard.

Every apparent exponential in the real universe must actually be a logistic or some other bounded curve.

If you mean important in the sense of potential economic impact, this seems like a case where you should form a company to develop the idea.

This reminds me of this chart showing underestimation of solar growth by the International Energy Agency for decades:

https://twitter.com/aukehoekstra/status/931869221386104832

One of the best charts I've ever seen at making a point.

Since I'm sharing solar charts here is one of solar panel efficiencies which is another of my all-time favorite charts: https://www.nrel.gov/pv/cell-efficiency.html

To avoid misleading, I want to clarify that the reduction in cost of solar energy has been driven far more by manufacturing scale than by efficiency improvements (although they are also relevant).

He had grants for 0.25% of the company but left when only 1/3 of them were vested . Gitlab will probably IPO for at least a $10 billion valuation. Assuming no further dilution since the grant, the shares would be worth $8.3 million for a $4050 option exercise.

The headline is misleading. The work tightened the range of possible strengths of a fifth force by a factor of 10. In other words it ruled out the existence of a fifth force within a wide range of parameters that were previously open.

It's just an absolutely giant construction project. The mass and volume of the construction goes as the cube of the major radius of the reactor. Back of the envelope, SPARC is (6/1.5)^3 = 64 times smaller than ITER. The construction budget for SPARC is ~$500M, so ITER being tens of billions is in line.

Thanks for the links. This is a really well written case which relies on an easily swallowed fallacy. The argument is that the main limiter to improvement of batteries is research which needs time, so we shouldn't quickly scale up battery production and end up with suboptimal batteries.

Of course the main way to improve batteries at this point is to make them cheaper, which comes directly from production scale (look at what happened with solar panels). And even if that wasn't the case, while we wait for improved tech before we scale up battery production, we're continuing to dig up and burn fossil fuels.

This reminds me of some of the propaganda spread by fossil fuel companies decades ago. They position themselves as being on the side of environment, and yet the conclusion is always to delay, do more research, and continue with the status quo.

Wind acts like a battery (or capacitor) that smooths out fluctuations in solar energy. Radiant solar being more "direct" really has nothing to do with which is the better option.

Globally distributed, consistent databases. See this Google paper: https://static.googleusercontent.com/media/research.google.c...

My understanding is that octopus limbs are semi-autonomous, whereas giraffe legs aren't doing any more decision-making than they are for people or cats.

Which is not to say that I disagree that most people underestimate animal intelligence.

They're working on a Steam release with a more accessible interface, including mouse support. I suggest trying it again when that comes out.

Their competition are the ones who played marketing games already, skipping forward a node relative to pre FinFET scaling. Intel's rename aligns it with the rest of the industry in terms of transistor density on each process number.

None of these numbers have aligned with physical features on the chip for a while, so they are just an arbitrary convention now. It's better for all of the manufacturers to use the same convention.

Neural network models have millions of parameters, which means you are trying to optimize in spaces with millions of dimensions. If you can't use differentiation you get hit really hard with the curse of dimensionality.

A fixed rate plan for something with unknown and highly variable costs is just about the definition of insurance.

Actually doing the material (as opposed to sitting in a lecture) is the way to learn, but that could be done in class.

The fact that Black-Scholes has only one unknowable parameter makes it quite usable, more so than more complicated option pricing models. You can work backwards from the market price to solve for the implied volatility, treating it as a generalized 'price' for the option after factoring out things that are easily adjusted for. You can also abuse the implied volatility (adjusting it up or down) to account for factors outside of the idealized model.

Most prescribed drugs are generics that will cost you a couple bucks. The drugs advertised on TV are precisely the ones with high margins that support expensive customer acquisition efforts.

The gradient we want is the gradient with respect to the process which generated the dataset. The gradient we get is an estimate based on only a handful of samples from that process at a time. The analogy holds up fine.

The encoder is a neural network which is a nonlinear and highly expressive function family. The encoder function is optimized to produce a set of "factors" which are designed to be linearly comparable. If the "factors" aren't expressive enough, you can add more (increase the embedding dimension) or make them more sophisticated (increase the encoder network complexity). There's nothing stopping you from allowing them to interact in a more complex way as well, except that you give up the ability to easily do operations like indexing.