knzhou

The people here reading GPT-3's output and dismissing it as "not really understanding anything" or "pushing symbols without conscious intent" because of minor slips in coherence have remarkably high standards for understanding. On a high school essay prompt, GPT-3 produces more accurate and more coherent output than 90% of high school students. If the commenters' standards were actually used consistently, they would also apply to almost all human beings.

Another interestingly common objection is "we think in terms of meaningful semantics, but GPT-3 only calculates meaningless coefficients". This is a category error masquerading as an argument, equivalent to saying "I have moral worth because I'm made of cells, but you don't because you're only made of quarks." Cells and quarks are just two different-level descriptions of precisely the same thing. Understanding in your brain can be reduced to the calculation of "meaningless" coefficients just as surely as cells can be reduced to quarks.

That isn't it, though. Feynman explains magnetism perfectly well in his lectures, he's just saying that he can't give a decent math-free answer in a few minutes to an interviewer with no science background. Any answer that short would not functionally be any better than "magnets work because of magic".

> Then he returned to mainland China in 1955.

It's worth remembering that he did that because he was pushed out, during the Red Scare.

I'm glad that Sabine correctly describes Wolfram, Weinstein, and Lisi's theories as what they really are.

I'm disappointed that she trots out the tired old "zero sum game" rhetoric. I work on small-scale experiments, a field that Sabine portrays as starved of funding because of string theory. But that's not at all how funding actually works. Generally, all of our boats rise or fall together.

String theory is so distant from my field that defunding it and expecting funding for me to rise is like defunding pharmacology and expecting an immediate revolution in palaeontology. Removing funding from one subfield doesn't move it to your favorite field, it just removes it from science in general.

While I completely agree with your general comments, I still want to emphasize how hard this ideal is to implement in practice.

Yes, every subject is linked to every other subject. If I'm tutoring introductory physics, I will use examples and analogies from math, computer science, engineering, biology, or more advanced physics, depending on the background and taste of the student, and it works fantastically. But if I'm lecturing to a crowd, this is impossible, because the students will differ. If I draw a link, some people will think it's enlightening, some will think it's totally irrelevant, some will think it's boring, some will think it's so obvious it's not worth saying, and most will just get more confused.

The same thing goes for the top-down "working backwards from cool results" approach; it's supposed to bottom out in something you know, but whenever you teach multiple people at once, everybody knows different things. The bottom-up linear approach is useful because it gives you a guarantee that you can draw a link. If I'm teaching quantum mechanics I expect to be able to lean on intuition from classical mechanics and linear algebra. If I didn't know the students had that, I would draw a lot fewer links, not more.

Similarly, "if people learned X in school, then Y would be easier to understand later" is true for almost any values of X and Y, because of the interconnectedness of knowledge. But if you ask any math teacher, they'll tell you the school curriculum is already bursting at the seams. You can't just add logic and set theory to existing school math without taking something out. In the 70s we tried taking out ordinary arithmetic to make room for that. It was called New Math, and everybody hated it.

> says you need to master each step on the track before moving onto the next one. this is poisonous and borderline evil

What’s wrong with it? You do need to understand calculus before classical mechanics, classical mechanics before quantum mechanics, quantum mechanics before quantum field theory, and quantum field theory before the Standard Model. I’ve seen tons of people disregard this and the result is always confused word salad. People waste years of their lives this way, going in circles without ever carefully building their understanding from the ground up. The order in school was chosen for a reason.

Special relativity is well-known for having the least prerequisites of any modern physics subject. Yes, you can explain it with just calculus and vectors. That's precisely what we already do in university courses.

But the complaint that started this whole thread is "why can't physicists explain why it's hard to add in extra neutrinos to the Standard Model?" Understanding the reason requires 2-3 semesters of relativistic quantum field theory, a subject which in turn requires relativity, quantum mechanics, and classical field theory.

Seriously, any resource that legitimately explains all of this, without bad analogies, and assuming no prerequisites, would end up precisely as hard to read (or likely harder) than the ~10,000 pages of textbook people currently study. The textbooks have already been optimized for ease of learning, there is no magic way to make a hard thing easy.

Explaining the Standard Model clearly and correctly is just really hard! I've tried, a lot. Costless simplification is impossible, because the mathematical form it's usually written in is already the simplest, clearest, and most concise way we know to describe it. (If there were a better way, we'd be using that instead!)

Anything simpler necessarily sacrifices correctness. Metaphors are great, but it's clunky to explain when the metaphor stops working or becomes actively harmful to understanding. For example, adding a 4th neutrino that is otherwise identical to the other 3 would break the Standard Model would violate anomaly cancellation, making the model mathematically inconsistent. But explaining from scratch what an anomaly is or why they need to cancel would require a ton of vague, fragile metaphors layered on top of each other, at which point I'm not sure what value it would provide.

Imagine explaining how a CPU works on the silicon level to somebody from the 18th century. It's the same problem -- it's hard to say anything that isn't the length of a full textbook, or completely misleading, or bordering on woo. I've written and then thrown away a lot of attempted explanations because of this.

> Is "China bad" how HN and the US is going to look for the next 20 years as North America goes down the hill?

As a US citizen, reading HN for a year has downgraded me from being extremely patriotic, to considering my options if I get expelled from the country in the next 20 years on the basis of my ethnicity. These folks would cheer that on, and they're my neighbors.

When I was younger I would laugh at those ridiculous "forwards from grandma" about, like, secret Satanic messages embedded in Super Mario. Every kid knew these were nonsense; we just assumed it was a consequence of that generation growing up without technology.

I was wrong. Every generation is equally prone to hysteria. We learn nothing.

This article reminds me of the time the city of Philadelphia tried to ban corner stores from installing bulletproof glass to protect themselves in robberies. Or the numerous education reformers trying to ban remedial classes by making all classes nominally "honors".

Bulletproof glass, remedial classes, and dollar stores reflect societal problems, so outside observers, in their great wisdom and compassion, decide they must be the cause of the problems. They campaign to get rid of them, completely oblivious to the fact that this would only hurt the people they supposedly care about, while doing absolutely nothing to address the real problems. Their injunctions are not even self-consistent: the dollar store is bad for not having the resources to prevent theft, while the corner store is bad for trying to.

It helps shield the detectors from cosmic radiation, shields people living on the ground from radiation produced by the collider, and helps with vibrational isolation. But those are just nice bonuses.

The most important reason is that there are already lots of people living a nice small-town life on the surface, and erecting a giant, dangerous ring would wreak havoc. It would require land rights, cut towns in half, split highways, trigger lawsuits, and probably end up costing more money and time than just digging.

> Some linear accelerator concepts could be much less expensive than very large circular colliders, but I imagine research into those areas are not funded due to the particle physics budget being mostly consumed by CERN.

That's not true; CERN has no particular commitment to circular colliders. Scientists from CERN have proposed linear colliders, along with more exotic options like muon colliders.

> particle physicists should focus on developing new technologies that could bring colliders back in a reasonable price range and hold off digging more tunnels

People keeping saying this, but we haven't been permitted to dig a single new tunnel since 1983. Multiple proposed colliders have been shot down, with the US's attempt defunded in the middle of construction. The LHC doesn't even have its own tunnel; it was put in the 1983 tunnel to save money. To save more money, the LHC is going to run for another 20 years, so any future collider's tunnel won't be dug until 2040 at the earliest.

That is a nearly 60 year gap between new tunnels, which I find incredibly depressing. What I find almost incomprehensible, though, is that the people who successfully argued to defund earlier proposed colliders now argue to defund the next one on the basis that our field is progressing slowly! I wonder why that is?

Mostly the confidence comes from half a century of trying to find such "extra" visible mass! There's a long history of people looking for extra dust or gas or compact objects using a tremendous variety of clever techniques, and yet the deficit has remained. At this point, if you believe that there indeed is extra mass around, it kind of has to be dark matter by definition because it's been so damn hard to see.

Also a good idea, but the details don't work out. Specifically, early on the Fourier components of the matter density distribution grow independently. Light neutrino DM would suppress the small scale clumping too much. Making the whole process last longer would fix this, but it would also enhance large scale clumping, contrary to observation.

It might be possible to make it all work out by adding a third effect (and maybe somebody has already done that!), but there's also a danger of constructing a Rube Goldberg machine -- if there are too many gadgets needed to get the structure right it starts looking like epicycles. The great thing about traditional dark matter is that it doesn't require any cosmological complications, it just works.

That's a perfectly good question, and that's exactly why for a time in the 90s they were one of the leading candidates for dark matter. Unfortunately it fell apart when we got a better understanding of structure formation, which determines the large-scale distribution of matter in the universe today. If the 3 neutrinos have masses around where we think, then they'd be too light and move too fast in the early universe, smoothing the distribution out and thus suppressing small-scale clumping, contrary to observations. On the other hand, if the 3 neutrinos were a lot heavier (possible we can only detect their mass differences, not the absolute masses) they would screw up the CMB. The qualitative case was great, but quantitatively it just didn't work out.

- gravitational lensing measurements

- velocities in galaxy clusters

- getting the fine details of the complicated, wiggly cosmic microwave background spectrum right to extreme precision

- getting the fine details of cosmological structure formation (as measured from matter density fluctuations on very large scales) right

- galaxy collisions, like the Bullet cluster (hard to explain with modified gravity)

- the fact that some galaxies have much less dark matter than others (very hard to explain with modified gravity)

These articles always miss the mark because they get the history backwards. Modified gravity theories were taken much more seriously in the 1970s and earlier, where the evidence for dark matter was shakier. Cosmological and astrophysical observations from the 1980s to 2000s have vastly strengthened the case for dark matter, which is why it's the leading hypothesis now.

Popular articles won't tell you this, because without exception they laser focus on galaxy rotation curves, a piece of evidence that's nearly a century old and by far the weakest one. But by neglecting to mention the actual evidence we base our conclusions on, they (purposely or not) make us look like fools.

Furthermore, dark matter hasn't been ruled out. One particular candidate of what it could be (a WIMP) has been studied and mostly ruled out over the past 20 years, but by the nature of the business there are many possible candidates. And of course we've shifted attention towards those other options, which is exactly how science is supposed to work!

> What is the amplitude of a single photon's wave?

That doesn't have a well-defined answer, unfortunately. Suppose the electromagnetic field contains a single photon. Its electric field then necessarily exists in a superposition of values, so it doesn't have a well-defined amplitude. It also doesn't have a definite "typical" amplitude either -- that depends on how spread out the photon is, and its frequency.

As a rough estimate, a photon of frequency omega has energy hbar omega, so if it's spread out over a volume V, then matching energy densities gives E^2 V ~ hbar omega, so you expect the rms value of E to be of order hbar omega / V.

Nah, GP is right, if he's talking about how you can decompose states into superpositions of coherent states of the whole field. These are different from coherent states of individual modes.

Specifically, a coherent state of a field is a state which is an eigenvector of the positive-frequency part of the field operator, when that operator is evaluated at any position.

I mean, that's because the language of physics is mathematics. If you ask vague questions with inherently undefined terms, you shouldn't be surprised if you get different answers! I bet chefs wouldn't agree on whether a taco is a sandwich, either, but that doesn't mean they don't know what they're doing.

Furthermore, as someone with a lot of experience thinking about photons in many different limiting cases, your objections don't make sense to me, because in general, absolute statements like "There are no X" in physics don't make sense. For almost any value of X, X can be a useful way of thinking in some limit or other, and whether or not you want to invoke X depends on the calculation you're doing. When you say there is no X and everybody who thinks otherwise is wrong, you're just inviting a debate over semantics, which we physicists are naturally not interested in.

A beautiful book. I have all my students read it and I use its ideas every day in research.

I agree, and I think the overwhelmingly negative response here perfectly illustrates the post's point. The author is thoughtful and has great, but not perfect English. Because he's foreign, everybody concludes that it's a sinister work of propaganda, and attributes all manner of horrible personality traits to the author on the basis of the tiniest of mistakes.

It's absolutely predictable. I sincerely hope I won't ever have to endure the displeasure of meeting any of these HN commenters in person.

> I've often read complains in the HN bubble that the Chinese system would be too rote.

I've seen both sides. 99% of complaints you read on the internet are written by people who've never been to the country they're complaining about, and whose sum total knowledge consists of a cheap stereotype.

The MMR study took 12 years from publication to retraction. This one took 12 days.

Of course you can. For example, Dr. John Campbell has a popular Youtube channel where he dissects new studies. And if you just want an opposing political spin, you know Fox News and co. have Youtube channels, right?

What gets taken down is spittle-flecked rants about how COVID-19 was created by Bill Gates and Anthony Fauci to bring about the Biblical apocalypse, but can be cured by drinking bleach, but simultaneously it doesn't even exist.

You could have read the paper yourself 10 days ago to find the flaws. You could have read the public discussions where scientists questioned the paper. You could have read the published "Expression of Concern" in the Lancet yesterday. You can read the retraction notice today. What exactly are you missing?

The Enron scandal lasted for 10 years and was only exposed by outsiders. This paper was questioned by other scientists from the very first day, and now has been retracted by the authors themselves after barely lasting 10 days. It's not at all comparable.

That's not only a correct attitude, it's an absolutely essential one for anybody who takes science seriously.

I mean, now you can understand why the WHO is always putting out statements along the lines of "no evidence yet for X", while random bloggers and HN commentators are already completely convinced of X (or completely convinced against it). Real knowledge takes time. Individual studies don't settle questions.