robryk

There's always the first time you hear a particular story. There's a continuously increasing mapping from how likely such an event is to when you should expect to hear about it for the first time.

Your reference for [1] just states that bypassing the turbine is a thing, not that it's normally used.

First, reactors are in a stable equilibrium when operating, so one will actually increase their power by increasing the rate at which heat is removed (and v.v.). Alas, that's workable only within some small range.

A reason[1] load-following with PWRs was originally difficult is that traditionally PWRs use boron concentration in primary loop to regulate power and that can be decreased only slowly. The reason it's done that way is that it's the easiest way to ensure that power is adjusted uniformly throughout the core; if instead some control rods were partially inserted, the top part of the core would operate at lower power (and thus lower fuel burn-up) than the bottom part, which would cause compounding control issues later on.

France is using their PWRs in load-following mode by (a) having additional less absorptive control rods ("gray rods") that can be inserted fully to adjust power by smaller increments (b) more complicated schemes to decide which combination of available actuations to use to change power. See https://hal.science/hal-01496376/document for a paper that tries to optimize control designs so that power changes are more possible (and describes how the control schemes work).

Note that the total heat capacity of even just the primary loop in usual reactors is quite large: in PWRs it usually requires ~0.5s of full power output of the reactor to warm it by 1degC, so this can easily cover, say, ~5% variations for something like a minute.

[1] Another is that reactors are not stateless due to xenon poisoning.

The header compression in http2 has atypical requirements placed upon it: contents of different headers in a single request must not impact each other's compressed size[^], lest an attacker able to manipulate one of them can use that to guess the other. Thus, LZ77-style compression is out of the window, as well as using Huffman codes selected based on _all_ the header values. In essence, each header value (in a single request) has to be compressed independently. IIUC the dictionary you reference is used for header keys (as opposed to values).

[^] this statement is a bit of a lie for reasons of simplification

> Isn't the earth's eccentricity incredibly low? (To the point that human-noticeability or non-precision-machinability are out of the question?)

It's one-percentish, and causes the length of time between equinoxes to be shorter on the northern-hemisphere-winter-side by a day or so. The date scale on the model includes single-day divisions, which made me think that such accuracy was intended; perhaps I was wrong in that.

I don't get how one can get realistic modeling of eclipses without taking into account the eccentricity of Earth's orbit around the Sun (it not only minorly affects the distance, but more imporantly it causes the angular speed of Earth around the Sun to vary).

Am I overestimating the impact of that error, missing a description of how it was accounted for, or something else?

Why is the delivery service concerned about the food quality at all? Post office is not concerned about quality of products it ships and I don't see anyone expecting it to be.

In that case it's hard to say which deaths are "caused by government". If a government could have imposed a policy that prevented them, are they caused by government? Should we balance that against side effects of such a policy? What about side effects of a feasible way of arriving at such a policy?

If we want to discount such deaths as caused by the government, we will end up with lots of cases of "would not happen but for policy X, but policy Y would have prevented them". From some POV _every single death_ is somehow caused by some policy that's government-imposed. To have a reasonable way to define "caused by government" we need to have something to compare with.

I can't imagine how a society in a world where $DEITY would strike down anyone who tried to create a government would look like; I don't know what they'd use for coordination and whether it would be more or less vulnerable to e.g. creating correlations that defeat the disaster-prevention of the law of large numbers.

If we want to compare two situations, we should not only point out the differences in one direction. Otherwise we'd end up claiming that e.g. a society on a desert will have many fewer deaths on account of drowning being vastly less likely.

What are you comparing with (both for government and for criminals)? In case of criminals it's easier to imagine a similar world where just criminals don't do crime; however, that comparison fails to take into account indirect effects of crimes (which are both negative -- e.g. costs spent on protection against crime -- and positive -- more incentives to influence the society so that crime is less beneficial).

In the case of government, I find it hard to see what do you mean by "same world, but without government". If everyone behaved the same in absence of government, nothing would really change. If we take into account changes in people's behaviour due to lack of government, then how do we stipulate that no government forms in that world? Are we imagining a world where everyone knows that forming any government will end the world and thus no one does that? In that case I'm hard pressed to even guess the sign of the difference in magnitude of harm, let alone its order of magnitude.

In this setup we want to be able to decode a cipherstream with holes into a plainstream with holes. This is very much possible with CTR-MAC, as long as MAC is done on some finitely-sized segments.

When talking about expected magnitudes of loss due to fraud, it's much more useful to talk in terms of relative (as opposed to absolute) magnitudes.

As long as you simply transmit the count in the packet this is not an issue. You mostly have to do that anyway, because packets can be reordered and you want to stitch them together in order (with holes).

CTR with no authentication is strictly better than EBC with no authentication in that case. Similarly, CTR-then-MAC (with MACs being e.g. per packet) is strictly better than EBC-then-MAC.

Surprisingly in some jurisdictions it does. At least in cases related to criminal defense in Poland the existence of a attorney-client relationship is protected (obviously if it comes to trial, it's eventually revealed, but it's not subpoenable up to that point).

> And even if it were true, so what? Last I checked Tesla didn’t have software which can perfectly emulated a human agent.

GP was arguing against:

> The feature is impossible to implement with the sensor suite that comes with the car.

_For that statement_ it's immaterial whether Tesla does something right now, but rather whether it's in principle possible. Modulo your previous objection humans provide a counterexample to that (very general) statement.

This note makes it very obvious that customer's evaluation of the expected timeline might impact their purchase decision. In my non-expert opinion Tesla puts itself on notice that any false statements that might impact that evaluation would be fraudulent. If my opinion matches the legal consensus, then the doctored video would be such a false statement.

> Passengers would in fact have a lower chance of injury in a car crash if they were wearing helmets.

I once thought so. Someone pointed out that helmets might make whiplash worse (because they extend the head backward, so will change the angle at which the head impacts the headrest) and that we're essentially using airbags for (a.o.) the same functionality (protection from direct strikes to the head). So, I'm ambivalent on the topic now.

I have very little doubt the effect's real.

Note though that getting base rates from one place and getting rates of accidents from another is usually a bad idea: they will have various ways of measuring the rates, so will have different biases (based on location, time-of-day, time-of-year, ... where the measurement happens).

> The author was trying to stress the importance of wearing a life jacket, I don't think they need to get into statistical robustness in their summary paragraph for something that is so _obviously_ correct.

Then don't provide statistics and just state that it's obviously correct. If doing so is less convincing, then it means that the statement _was_ misleading.

> The only reason you would want to argue this point is if you think they are being intentionally misleading, which I highly doubt you do.

I think they believe the conclusion is true and are intentionally saying whatever they can to make people think it's true, regardless of whether the reasoning they imply is correct.

E: So, in my estimation they are intentionally misleading in the area of valid arguments and not in the area of water safety.

This is exactly my point. Without knowing how common {wearing helmets in cars, wearing PFDs in kayaks} is, the statements imply nothing about safety benefits of either, and yet that one was presented in a way that suggests that it does.

What would be a non-misleading value to quote would be absolute change of risk of drowning (e.g. that a typical trip without a PFD costs X micromorts, while a typical trip with one Y micromorts). What would be less problematic (but still potentially misleading[1]) would be a relative risk reduction (wearing PFD makes you X times less likely to die).

[1] Because it doesn't tell you anything about the original risk that you're lowering. E.g. helmets are a good protection against falling bricks: equally good at construction sites as when you're walking down the street. The only difference is that falling bricks are more likely on a construction site.

> Of those people that die, over 80% of them are not wearing a life jacket or Personal Flotation Device (PFD).

Making this statement in such a context borders on statistical misconduct, because it ignores base rates: Consider whether you'd find a statement that ">90% of injured car passengers were not wearing helmets" as a valid argument for helmet-wearing. This is presented as an argument for PFD-wearing.

I'd guess that the context model (based on last 1-2 bytes) can be used to separate opcodes and their parameters.

UNDO makes sense if the server is keeping a bag of responses. For example, ActivityPub has an Undo action that points at a previous action punished by the same user. You use it to unfollow, unvote, unfavorite, etc. Obviously, handling of Undo has to be at least partially specific to the thing being undone, but they are all expressed using the same primitive.

If the human is doing the traction control job, they know how far from the edge of the envelope they are (because the difficulty changes). If it's a computer doing that, it's hard to communicate that to the human (iiuc we don't even try). Unless this is improved, this means that sometimes it's better to leave a human in control of both strategy and tactics even if the computer would outperform on tactics.

Switzerland had that, but abandoned the distinction in 2019: now there are no transmission restrictions on Swiss licenses.

Source: https://www.admin.ch/gov/de/start/dokumentation/medienmittei..., grep for Automateneintrag

The UK authorities agree with you: if you pass the exam with a manual car you can use both, but if you pass it with an automatic one, you can only drive automatic ones.

In snowy conditions you don't actually want all the torque available: then you'd want to control the torque applied instead of the power delivered when starting to move (similar to placing a restriction on the top torque by shifting into second gear on manual shift cars).

Would

   int foo(int x, int y) { return x+y; }

compile? After all, this function can be called in a way that causes UB.

TBF putting a reverse proxy that does authn and rejects requests that do not authn correctly in front of everything (except for the login pages) is nearly as good: the exposed area is then login page, reverse proxy, and the authn login in the reverse proxy. That might be vastly simpler to implement.

> It's a great argument against the scale that cloudformation has, and regulation would help.

Agreed. It's very annoying that such services like ddos protection have an ever-growing scaling advantage (because the sizes of ddos attacks grow).

> If you can't figure out that one of your clients is doing this bad things, you shouldn't have so many clients

What kinds of entities would you extend this to? I would guess you wouldn't day the same thing about hardware stores (which sell dangerous tools).