mppm

> You'd get the same result if you asked a random student to fully translate a passage from Hamlet, sentence by sentence, with no prior context. Or asked a random CS student to explain a random snippet of source code from the Linux kernel line by line. Most people don't deeply understand most things unless they get the bug and decide to dig in for fun.

I would rate the amount of specific context necessary to understand a random snippet of kernel code much higher than what you need for that Dickens passage. It's certainly much more dense with metaphor and playful use of language than normal prose, but I don't find it that opaque, even as an non-native speaker.

> The point is that you can't force comprehension on someone who isn't interested or motivated on their own. Most students are just muddling through because they "have to get a degree".

Well, yes, but that doesn't necessarily contradict the article. The bell curve at the bottom basically says that the comprehension they were expecting is in the top 3% or so, not the 60% of the general population who "have to get a degree". Add in all the Netflix and TikTok casualties, and the result ceases to be surprising.

Presumably they mean that the CO2 captured in the wood of the tower can offset the manufacture of blades and other components at some point in the future. Not that reaching net zero in wind power is an important milestone or anything. From their technology page:

> The life-cycle emissions from modern wind power plants made of steel are about 4–7 grammes carbon dioxide per kWh. Building the tower in wood lowers the emissions from the wind power plant by approximately 30 percent per kWh.

That would put wind power some 50-100x below fossil fuels already. Additional improvements are always nice to have, but not really a big selling point.

For all those who are peeved by "stronger than steel" claims: No it's not. Or only in an extremely specific, not very practically relevant way. Their Nature paper [1] has the details, but let me summarize:

Tensile strength along the fiber direction is up to 600 MPa, which is higher than mild steel used for ordinary construction (~400 MPa). Except that mild steel is isotropic and highly ductile, i.e. it will deform by 50% or so before finally breaking, while the densified wood will snap as soon as its strength is exceeded. There are also steel grades that are way, way stronger than mild construction steel.

Flexural and compressive strength is up to half as high (300 MPa, below mild steel).

Strength in the weakest direction (across fibers) is around 50 MPa, or 1/8 of mild steel.

This densified wood is essentially a rather weak fiber composite. With careful planning it can be used instead of steel in construction, but their marketing is simply BS.

1. https://www.nature.com/articles/nature25476

I think at this point it would be preferable for the government to take over the payment infrastructure directly. This is often seen as dystopian, but it's a dystopia that has already come to pass for the most part. If we made it official, at least the rules for what can be blocked or refused or frozen would be out in the open.

It's not illegal. Even though having access to banking is absolutely essential to every imaginable business, bank accounts are still treated as a voluntary business relationship by regulators. Some protections exist in the EU for individuals, but for businesses, it is entirely at the discretion of the bank to open and maintain a relationship.

> Any traditional local bank will usually speak with you first before opening an account. In contrast, neo-banks open accounts quickly and then rely heavily on automated fraud prevention systems, since manual reviews are costly.

Maybe that was still the case in 2015. In 2025, a traditional bank will ask for a ton of documentation, run it through the same (probably outsourced) automated fraud prevention system and tell you that unfortunately an account cannot be opened for you. At least that is a common situation for businesses that are too small to be important to the bank, but carry some compliance risk (international consulting, app developers, digital services, etc.). And then you are back to neo-banks...

This has been a trend everywhere, recently, but the severity of the situation differs country by country, so YMMV.

Unfortunately, this is the reality of modern business banking, and Wise is not unique in this. You can find similar reports about most major EMIs and a fair number of banks. Banks have basically become an inefficient and unfair branch of law enforcement, without presumption of innocence and without due process. Every imaginable abuse is justifiable by Compliance nowadays.

Everything you suggest is in fact part of standard (and mandatory) procedures. If a pilot makes one serious error, you should look closely at the circumstances. If the pilot makes ten in sequence, he probably is, as GP put it, a moron who shouldn't have been allowed near an airplane, ever.

Circumstances do matter, but they do not constitute a general license to dismiss personal responsibility, especially in occupations known to come with very high responsibility.

What you say is true, but I would still disagree with the overall assessment. To put it bluntly, it is the pilot's job to be thrown into unexpected situations. Modern airliners mostly fly themselves, and the pilots are there to a) coordinate with ATC b) step in when the shit hits the fan, which does happen with some regularity. Ultimately this accident was caused by a confluence of many factors, as is often the case, but my read on the situation is that a competent pilot should have been able to handle this emergency without any losses.

This is an extremely long and unfocused analysis of what was a fairly straightforward incident. Following the lightning strike, which created a dangerous, but manageable situation, the main contributors to the catastrophic outcome were (roughly):

80% - Pilot error. Poor adherence to procedures and checklists. Poor choices all around. Poor piloting in manual mode and botched touchdown. Part of the blame for this rests with Aeroflot, for putting such a pilot in the air.

15% - People retrieving their luggage slowed down the evacuation and increased death toll.

5% - Aircraft design. Could be improved in some areas, but no really serious bloopers.

~0% - Delayed emergency response. Not good, but partly caused by incorrect communication from pilot. Also, fire spread so fast, it's not likely they could have changed anything.

Key distribution is (as always) an important, but solvable problem. There are some tradeoffs involving centralization vs cost of replacement, but those apply generally, not just in this particular case.

As for replay attacks, that's where the button press comes in (like on a hardware security token) -- the key only responds to challenges within a second or so of a button press and the car sets a similar timeout for validity.

Re price tag: you can buy a smartphone for 100$. Surely it is possible to mass produce cheap key fobs with send/receive capability and a tiny crypto module.

Re power: Key fobs already do some form of crypto and broadcast. Adding reception capabilities ought not to be that power hungry.

I'm confused why this is still an unsolved problem. A simple cryptographic challenge with pre-shared keys + button press ought to make key fobs perfectly secure for all practical purposes. Is there something I'm missing here?

From what I can tell, they didn't. The theoretical basis for it is the Linear No-Threshold Model, which is, bluntly put, garbage. The empirical evidence seems to come mostly from so-called case-control studies, which are, bluntly put, garbage.

You conduct a case-control study as follows: 1) Select some number of lung cancer patients and a control group without lung cancer. 2) Try to establish why the cancer patients got cancer and the non-cancer group didn't, by retroactively estimating their smoking habits, radon exposure (from radon maps or retroactive measurements of places they lived at), as well as other factors. 3) Use statistics to try to disentangle suspected radon cases from smoking (90%) and other stuff. 4) Pretend that all this is not totally biased, sampling or otherwise.

Please take a look at slide 11 of this presentation [1]. Stopping power and Bragg peak (slide 30) are also relevant concepts. What you are describing applies to photons and neutrons, but not to charged particles.

1. https://indico.cern.ch/event/975141/contributions/4137563/at...

This is somewhat inaccurate from a physical perspective. What you say would be applicable to the attenuation of gamma rays, which is governed by low-probability interactions and is therefore exponential. But alpha particles, being charged an heavy, lose energy continuously through electron interactions and have a relatively fixed range in matter, beyond which the incidence is practically zero.

Yes, comptime evaluation is restricted to a configurable number of back-branches. 1000 by default, I think.

Yes, I've seen such research, but I am highly skeptical, to put it very mildly. This actually being the case is simply implausible from the physiological perspective, considering the very high (and constantly changing) concentration of CO2 in the lungs and in the bloodstream. Early research with Navy submariners was consistent with this -- no detectable impairment below 1% (10,000 ppm) and clear impairment at 3-4%. Later long term studies with 0.5% exposure (5,000 ppm) also found nothing. I put the recently popular research showing cognitive impairment as low as 600 ppm in the same mental niche as vitamin megadosing -- underpowered, low quality research with no basis in human physiology.

That said, CO2 can often serve as an easily measurable proxy for stuffy air, which can contain particulates, formaldehyde and other organic exhalations which may actually have some effect (psychological, if nothing else). But this just calls for ventilation and air filtering, not CO2 scrubbing.

Very cool!

It would be cooler still if this technique could be used for future VR technology, creating full immersion by targeting all photoreceptors individually. But unfortunately... the optics of the eye does not actually allow individual cones to be fully isolated, as the spot size would be below the diffraction limit. They discuss this in Fig. 2 and the first section of the results.

Even with a wide-open pupil and perfect adaptive optics, there would be 19% bleedover to nearby cells in high-density areas, while what they achieve in practice is 67% bleedover in a lower-density (off-center) area. This is enough to produce new effects in color perception, but not enough to draw crisp color images on the retina. :(

Honest question: Why would you want to? The average CO2 concentration in your lungs is something like 20,000 ppm, so it doesn't matter much whether ambient CO2 is at 200ppm or 800ppm. Other aspects of air quality do matter, but I'm highly skeptical about the value of capturing CO2 in your house.

Learned something new today, thank you!

If I understand correctly, the Hermite functions are the eigenfunctions of the Fourier Transform and thus all have this property -- with the Gaussian being a special case. But sech(x) is doubly interesting because it is not a Hermite function, though it can be represented as an infinite series thereof. Are there other well-behaved examples of this, or is sech(x) unique in that regard?

This submission is not technically inappropriate and flagging it would be an abuse of that feature. It's just very low quality and very surprising that it made #1 on HN.

1) Why is this even on HN? Some guy complains about third wave coffee on his trashy, possibly AI generated, blog. Very interesting.

2) Since people take this seriously for some reason: Fine coffee is neither a hallucination, nor a theater performance, nor a sign of ultimate decadence. Or at least no more so than fine tea and wine. Different producers, roasts and preparation methods give markedly different coffee with a lot of nuance that you can learn to discern and enjoy. Or not -- to each his own.

I get that. But if you want to work in relative isolation, would it be too much to ask to not advertise the project publicly and wax poetic about how productive this (unavailable) language makes you? Having had a considerable interest in Jai in the past, I do feel a little bit cheated :) even though I realize no binding promises have been made.

Jai's perpetual closed beta is such a weird thing... On the one hand, I sort of get that the developers don't want to waste their time and attention on too many random people trying to butt in with their ideas and suggestions. On the other hand, they are thereby wasting the time and attention of all the people who watched the development videos and read the blog posts, and now can do basically nothing with that knowledge other than slowly forget it. (Except for the few who take the ideas and incorporate them into their own languages).

Very unlikely. Here's a back of the envelope calculation: The human energy requirement per day is about 9MJ. This corresponds to about 500g sugar (or starch), which releases around 750g CO2. Metabolic activity is reduced at night, so 250g CO2 is the upper limit for a full night's sleep. At typical temperature and pressure, this is < 0.14 m^2 of CO2. Assuming a very small (20 m^3) and hermetically sealed bedroom, you'll end up with a concentration of 0.7%, or less. Serious physiological studies (with divers and submariners) show that CO2 has a measurable effect starting at about 1% concentration and only becomes pronounced at 3% or so. This is consistent with the fact that exhaled air contains about 4% CO2 during normal breathing and can go much higher (>10%) during breath holds. In summary, sleeping in a stuffy room might give you respiratory problems, but no improved CO2 tolerance.

Maybe they've changed it in the past hour, but as I write this comment the 25$ plan is called "Ultimate" and promises unlimited search, but not unlimited assistant.

I agree about the need for appropriate wording and advertising, but other than that, the new limits seem entirely reasonable and in line with what other aggregators like Abacus and Poe are doing. The paid plans of the major AI labs themselves always have usage limits too. It simply can't work any other way if you include costly models in the mix.

I'm not much of an expert, but maybe this can be useful: AlN is a somewhat widely used insulating substrate that is chosen where sapphire is insufficient (~40 W/mK), but BeO (~300 W/mK) is too expensive or toxic. The intrinsic conductivity of single-crystal AlN is very high (~320 W/mK), but the material is extremely difficult to grow into large single crystals, so sintered substrates are used instead. This reduces thermal conductivity to 170-230 W/mK depending on grade. Can't comment on pricing though.

> Whose gonna pull the trigger on beryllium oxide mounting packages first?

Nobody, presumably :)

Why mess with BeO when there is AlN, with higher thermal conductivity, no supply limitations and no toxicity?

Edit: I've just checked, practically available AlN substrates still seem to lag behind BeO in terms of thermal conductivity.

> The "real" point of an argument is not to persuade the other side (though that is what you aspire to nonetheless) but to exchange views.

Maybe this is just a matter of definitions, but for me the point of an argument is to convince or be convinced. When two incompatible views exist on a subject, at least one of them must be wrong. Some topics of conversation allow for diverging views or values, but then we are just talking or sharing experiences, not arguing.

That said, it is my experience as well that actually changing someone's (or my own) mind on an important issue is unlikely. Especially on complex topics with partial and uncertain information, like political issues, our life experience and cumulative knowledge significantly influences our selection of sources and interpretation of the facts, so converging on a common point of view may require the exchange of a prohibitive amount of information, even among rational arguers.

Productive argument usually occurs in a sort of semi-echo chamber, with people who mostly agree with us on the context, and are only arguing about the top layer, so to say. But when trying to argue about the deep stuff, we are mostly just "exchanging views", in the end.