You got it-- a perfectly rational voter ("rational" meaning "optimizing according to game theory") will use only 0 or 10 scores (assuming simplifying assumptions, of course).

Canada is a parliamentary system. Turns out to behave differently.

Arrow's theorem doesn't actually cover approval voting, since it doesn't quite fit the requirements (short answer, Arrow's theorem has a built-in assumption that the choice function is deterministic based on individual voters' ordered preferences, while approval voting adds an additional voter preference that is not deterministic, the cut-off between "like" and "dislike". You can model this as an additional parameter, but all the ways I know of to modify Arrow's theorem in this way have problems.) Nevertheless, the basic point of Arrow's theorem is important: don't try to find a system that's perfect in every possible case, just look for one that works better in real-world cases.

Run-off voting is, indeed, much better than plurality-takes-all. I will argue that it may be better, but still has problems (e.g., a centrist candidate who would beat either of two opposite wing candidates on a one-on-one election could be eliminated in the first round). But, possibly more important, approval voting completely utterly simple.

Another system that is straightforward is simple numerical scoring: everybody scores all the candidates from zero to 10, and you add up all the scores, highest number wins. That is also trivial to explain. Fast to count votes (only one pass, consisting of addition) but can't be done without modifying existing equipment. (Oddly, this is mathematically identical to approval voting if voters are perfectly rational. But voters aren't, of course.)

It is the net result of Duverger's law, which is a consequence of the voting system where the plurality takes all; which tends to suppress third parties.

If you want to see more than a binary choice, advocate for a system that does not squeeze out third parties. My choice would approval voting, a system which has the advantage of not needing any change whatsoever in the existing voting process, only requiring removing the current constraint that if a person votes for more than one candidate, their vote is discarded.

OMG, I'm so stealing that ;)

"Woke" simply means "I'm conservative, and the thing I'm calling 'Woke' is something that I hate". It has no well-defined meaning beyond that. I've heard things as diverse as "the concept of the Metaverse" and "removing copyrighted content so you don't get sued" described as "woke".

No call by push value or P vs NP or complexity theory etc.

So a different set of things that are also useless to the vast majority of people?

Even when Chrome adds support, we'll have to wait ages before we can actually reliably use the format without having to implement fallback logic and fallback formats for legacy browsers.

AVIF is also painfully slow. And if I recall correctly, is outperformed by JPEG XL at moderate and low compression (but - again, if I recall correctly - AVIF wins at highly compressed images). Also, AVIF faces some patent threats. And misses a lot of JPEG XL's interesting features.

A practical issue with a circle is that it is not a circle until it is finished,

That's not the reason at all, AFAIK. The reasoning is, okay, we want people to be able to move from one place to some distance place in the city at the maximum comfortable speed, which is limited by G-forces. You have some guaranteed G-forces from first accelerating and then decelerating. But if it's linear, that's your only G forces. If it's curved, however, you also have radial G-forces.

The Line's train going from one end to the other (170km) nonstop is supposed to do it in 20 minutes, aka with a mean speed of ~510 kph. Let's say a peak of 800 kph. Now if we shape that 170km into a circle, that's 54km diameter, 27km radius. From the centripetal force formula a=v^2/r, that's 222,22...^2 / 27000 ~= 1,83 m/s^2, or a constant ~0,2g to the side. This is on top of the G-forces from your acceleration and deceleration. You can probably deal with ~0,2g in a train if everyone is seated without much discomfort, though it's double what's acceptable for standing passengers. But you can eliminate that if the city is linear (at the cost of increasing the mean distance that the average person has to travel to go from one arbitrary point in the city to another)

That's not to defend this concept. Because the city doesn't need to be 170km long; you can just made it more 2d and have the distances be vastly shorter (at the cost of just needing some extra lateral travel within the city). Honestly, if I were building a "designer" city from the ground up, I'd use a PRT (Personal Rapid Transit) system rather than trying to make it super-elongated.

What got me is that I don't see why this isn't readily resolved by active damping, the same systems that many tall towers now use to resist earthquakes or resonant wind forces. Big heavy weight at the top (or in this case the bottom) hooked up to actuators that make it move in an inverse direction to the sway.

Again, this is not to defend this colossal waste of money. I just don't see why there aren't ready solutions for this specific problem.

Agreed - but that said, there are space elevator alternatives, like the Lofstrom Loop / Launch Loop, which at least theoretically can be built with modern materials (and have far better properties anyway - not latitude-constrained, provides dV, vastly higher throughput, far more efficient, stores energy / can add cheap energy at off-peak times, etc). One could always "waste" money on them trying something new :)

History didn't ignore her; her name is all over the histories of the discovery of the structure of DNA. She wasn't awarded the Nobel prize because Nobel prizes aren't awarded posthumously, but I agree if she had been alive, she should (and very likely would) have been added.

By the way, her work after leaving King's College, in elucidating the structure of viruses, was also groundbreaking.

The person whose work was vital in the x-ray crystallography of DNA and also deserved the prize is Raymond Gosling

Nope. The technique was invented by Gosling back when he was working for Wilkins, before Franklin arrived at King's College. And, yes, in doing that work he learned that humidity was the key.

Not to downplay Franklin's role-- she was doing the hard work of interpretation of the x-ray diffraction patterns-- but Photo 51 was taken by Gosling.

Correct: he should have received some measure of credit. And, to be fair, he did: the Nature paper (in the same issue as Watson and Crick's) was authored by Franklin and Gosling.

But Watson, Crick, and Wilkins got a Nobel prize. Gosling, who did the actual work, finished his degree, couldn't find a job in Britain, and left the field. If there's a person who was unfairly erased from the histories, it's Raymond Goslling.

Maybe not. That was the excuse for why Jocelyn Bell didn't share the Nobel prize for discovering pulsars. But in this case Gosling was the one who devised the method and protocol, and did so before Franklin arrived at King's college.

There's several good books on the details (although I advise you to skip Watson's book the Double Helix, which is sensational but glosses over the contributions of everybody else.)

That's what happens when you cut the cybersecurity office, fire half the people doing the work, and then put the rest on furlough because cybersecurity isn't essential.
  https://www.nytimes.com/2025/1...
  https://www.axios.com/2025/10/...

No. Like any software, AI requires maintenance, and that maintenance costs money, lots of money.

It does not. Models need nothing more than the storage of some gigs of weights, and a GPU capable of running them.

If you mean "the information goes stale", one, that doesn't happen at all with RAG. And two, updating information with a finetune or even LORA is not a resource-intense task. It's making new foundations that is immensely resource intensive.

Can you integrate it into your products and work flow?

Yes, with precisely the difficulty level of any other API.

Can you train it on your own data?

With much less difficulty than trying to do that with a closed model.