Native Persian speakers, by contrast, get it right 82 percent of the time.

This is the most astonishing part of the article. It didn't just say "fluent in Persian", it said "native speakers", so presumably people who grew up in contact with that culture. And then there's only an 82% success rate? That seems far too low for me.

So either people from that culture constantly misunderstand each other in everyday situations extremely often (doubtful in my opinion), or the researchers chose some rather contrived scenarios as their sample set.

Transpiling code from C to Rust is kind of easy:

unsafe {
// insert entire C code with macros expanded and
// the relevant syntax changes here
}

But actually translating the code so that it is useful Rust is incredibly hard, and beyond the capabilities of LLMs. Because the number of thing that Rust forces you (as a programmer) to do when you want to write something beyond a trivial "Hello World" program is to think about how to structure your program in such a way that the borrow checker is happy. And that requires a level of abstraction that LLMs are simply incapable of. Sure, for some commonly used patterns you could probably do something automatically, but as soon as you get to some non-trivial data structures, the LLM will not be able to give you a good answer, because it doesn't really understand the code.

The only case where translating code from C to Rust can happen somewhat automatically is for C code that has been written in a very idiomatic manner. But that is actually the last code you'd want to port to Rust, because that code is already very safe at this point, and the code that has the most need is the code that was written so that it can't be trivially ported to Rust.

The actual problem wasn't that it passed that "close" to Earth - sure, it was somewhat close, but there have been objects that have passed closer, and objects that are estimated to pass closer in the future that we've already detected. What worries me is that we've only detected it 2 weeks ago, so we could have never done anything about it. This object is quite big for something we hadn't seen before. For comparison: the Chelyabinsk meteor 11 years ago was just 19 meters in diameter, not 1.5 miles (more than a factor of 100), and even with that one we were already lucky that it a) mostly exploded while still in the upper atmosphere and b) did that over an area where the average population density was quite low.

This only affects the boxed versions for now: https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fwww.intel.com%2Fcontent%2F... Tray versions will still be available even after that date.

The 12th gen K-series tray versions have not yet been discontinued, so I doubt that the 13th gen will be discontinued before that. The 11th gen (different mainboard socket than 12/13/14th gen) has been discontinued though, see https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fwww.intel.com%2Fcontent%2F...

That said, I can still find e.g. new 9th gen K-series CPUs on the market that have been discontinued quite a while back. And while Intel isn't producing them anymore, there's still enough stock available worldwide that if you really want one, you can get one. (Whether that's a good price-to-performance proposition nowadays is a different topic.)

One is provide reliable execution, which means that if a program running atop DBOS is ever interrupted, it starts where it left off and does not have to redo its work from some arbitrary earlier point and does not crash and have to start from the beginning.

And the resuming of operations is different from any other operating system with preemptive multitasking? Which program that was preempted has to redo its work from previous states and most current operating systems? (Let's skip low-level locks implemented via transactional memory because I highly doubt that's what they mean.) Plus, how can they guarantee that a program won't crash? Even programs in memory-safe languages can "crash" -- in that they run into an error condition they are unequipped to handle. Rust still has panic.

Sure, the "time travel" thing they claim a sentence later is beyond the standard features of other operating systems, but time travel debugging is nothing new - and also very expensive, which is why it's not used more often. I don't see how they can implement something like this and at the same time claim to also be much faster than other existing OS. Combine that with the fact that going back to a previous state becomes a conceptual nightmare when you also consider network communication.

I'm not saying anything about their product itself, as I haven't looked at it, but their marketing doesn't inspire any confidence that they actually know what they are talking about. It could be that they genuinely have a really innovative product, and their marketing department just ran away with some keywords without understanding them. But more likely they are just overhyping a mediocre at best product.

Sauerkraut Linked To 'Substantial' Boost In Spatial Reasoning, German Study Finds

Max Mustermann, the study's lead researcher and an associate professor at the school's Institute of Cognitive Psychology, told CBS News the results, while based on a limited sample of subjects without preexisting health conditions, exceeded their expectations. "We knew from earlier animal studies that sauerkraut conferred health benefits," he said in an interview from his office in southwest Germany. "But what really surprised us was the dramatic change. The improvement was really substantial."

Just to add some perspective to the numbers here: if 14000 is 0.1%, then 100% is 14 million, so 6.9 million is slightly less than 50% of all customers. They've also shown no real transparency in this entire ordeal, so who knows if these latest numbers are even accurate.

I've yet to see a proper post-mortem of the entire incident, and it's been a long time since reporting on this issue first started. And the way 23andMe has tried to obfuscate the entire thing tells me that they're definitely not a company I'd ever want to be a customer of. Either their internal practices were so bad that if they did a proper post-mortem they might as well close up shop, or their company culture is so horrible that you don't want to have anything to do with them. (It's not like they didn't have enough chances since the first reporting of the incident.) If I ever do decide to get one of these tests (probably never, but who knows), I'll definitely choose someone else instead.

Please. Nobody who's a billionaire isn't profit driven. Sure, he also has other motivations, but that's true for nearly everyone. Regardless of whether you like Elon Musk or not, it's naive to think that profit isn't one of his primary motivations. Same goes for every other extremely rich person out there. And for a very large amount of "normal" people as well. If you want to be a fan of anyone (not just talking about Musk), please recognize the fact that you can very much like a person without drifting off into blind devotion, and that nobody on this planet is a pure paragon of virtue.

What I have been missing from all articles about this topic is an actual explanation what the primary technical criticism is all about. So I've skimmed djb's blog post about the issue and what he's arguing about the complexity is the following:

According to djb, the analysis of Kyber512 done by NIST argues the following:

- It will likely take 2^95 iterations for an attack on the algorithm to succeed.

- There are 2^25 bit operations (calculations) required per iteration.

- There are 2^35 memory accesses required per iteration.

- Hence there are 2^(95 + 25 + 35) = 2^155 operations required to attack the algorithm. (According to NIST.)

djb points out that this is very misleading, since the total amount of time a single iteration takes is not 2^25 * 2^35 (=2^60), but instead something like 2^25 + 2^35 (which is just a little more than 2^35), so you'll get a total complexity of 2^95 * 2^35 = 2^(95 + 35) = 2^130. (And NIST is themselves targetting 2^140 operations for their new standard.)

I haven't looked at the original NIST document analyzing Kyber512 to see if djb's claim about what they're arguing is indeed an accurate representation, but if djb isn't misunderstanding and/or misrepresenting the analysis in the original NIST document (i.e. the bullet points I provided here are indeed what NIST is using to calculate the attack complexity), then this is a huge blunder (and one has to wonder whether this is intentional), because djb is 100% correct that this is a mistake that nobody with even just an undergrad degree in CS should be able to make, let alone somebody's job it is to analyze crypto algorithms.

And while I have not read the original analysis by NIST, I tend to believe djb here, because if djb had simply misunderstood the NIST analysis and the bullet points above are not what the analysis is using to estimate attack complexity, then the person at NIST responding to this could easily refute that, instead of some BS such as It’s a question for which there isn’t scientific certainty and intelligent people can have different views.

Sure, there are certainly areas where intelligent people can reasonable disagree about things (for example in how high the threshold for security should be set for the future), but in this case? NIST's analysis is either correct when it comes to the possible attack complexity, or it isn't, and that shouldn't be a matter of debate.

Thing is, you wouldn't even need to test a nuclear detonation itself. The most complex part of a (non-hydrogen) nuclear bomb is not the actual material, but the mechanism that compresses the fissile material upon detonation in such a precise way (timings etc.) that you actually start the chain reaction. (Sure, creating the fissile material is also difficult in practice, but once you have that it's just solid blocks of metal.) And if instead of actual nuclear material you use something else (some other heavy metal that doesn't go boom), you can probably still test the detonation mechanisms in a controlled manner. And you'll not set of detectors that have been used to look for nuclear tests, because you're not actually using any material that creates a radioactive signature or an explosion large enough to make any seismic detectors suspicious.

Are you going to be 100% sure that the final thing will work properly? No. But can you be confident enough that it's very likely that a significant enough percentage of your devices will work as designed? Yes. (Especially if you already have experience with the mechanism from previous tests.)

Hydrogen bombs may be a different story, but even a non-hydrogen fission bomb would create devastation on scales we haven't seen before. (The bombs dropped on Japan in WW2 were small by today's standards. There are conventional bombs that have a higher yield nowadays.)

Also keep in mind that nukes (at least those from the 3 largest superpowers) are constructed so that even if you blow them up with an explosive they don't actually detonate. So even if some delivery vehicles explode on the launchpad, it will not cause a nuclear explosion. Sure, the radioactive waste that's going to be distributed in the immediate radius of the launchpad is not going to be great, but it's not going to be even remotely as devastating as even Hiroshima/Nagasaki. (Though if that explosion happens inside a silo, it's likely going to be largely contained.) Plus I don't actually buy the idea that the launch vehicles don't work properly, the Russian space program has actually been very reliable when it comes to launching stuff into space. Sure, not 100%, but quite close.

I find these kinds of articles actually quite dangerous, because they downplay the risk of the outcome of a nuclear war. If Putin (or anyone for that matter) actually decides to use nukes, it's going to be bad, real, real bad. Likely world-ending bad. I also don't like what kind of message this rhetoric sends towards Russia, that "the west" thinks they don't actually have nuclear capabilities anymore. I don't want Putin to think he needs to demonstrate that that's wrong. If only 5% of their nukes work (and I think that's unrealistically low), they still have more than the next country after them, and way more than would be required to end the world according to estimates.

Look, I agree that Michio Kaku has gone off the deep end, I haven't heard him say or write a coherent sentence about physics in the last couple of years.

But Sabine Hossenfelder? You may not agree with what she says, but she's always well thought out and coherent, and she comes across as very sincere. She also attempts to be as clear as possible in what she's saying, avoiding any kind of obfuscation. Putting her on the same level as Michio Kaku, or calling what she does quackery, is just ignorant. Btw. personally I agree with some of her takes, but not all of them. I think she's a very valuable perspective to have in physics, especially because she challenges orthodoxy in an intelligent manner.

Can you actually offer a substantive critique of what she's saying? (Since I was critical of Michio Kaku here, I can give you an example of a substantive criticism of what he's saying: he likes to conflate the multiverse with the many-worlds interpretation of quantum mechanics, although they are completely different things.) Or are you just lashing out because deep down you suspect that she might be onto something and you don't want your own world view to collapse?

Yes, this. I only started blocking ads once the web became unusable without an adblocker. It still baffles me that people can actually use the web in its current form without one. I wouldn't mind ads in principle, but as long as the current ad industry doesn't massively change, I'll keep blocking them.

This, so much this. Stop-and-go is annoying enough with an automatic transmission, it sucks even more with a manual one. Driving a manual transmission can be really fun if you drive on a road that has not a lot of traffic and that follows the terrain with quite a few curves, and I'll happily rent a care with manual transmission when on vacation -- but I'd never want a car with manual transmission for a daily commute in a metro region.

I don't think that the following scenario is completely implausible: Rosendorff lies during the trial to make himself look better (and the defense did ask him about that but he simple didn't tell the truth), but now reads in the media that Holmes could face up to 20 years in prison, and is now racked with guilt, so he shows up at Holmes's home and says what is claimed here.

I'm not saying that's what happened, and I'm not saying that this is even the most likely explanation (I think it's not by a long shot) -- I'm just saying that this is indeed a realistic possibility, and in the interest of a fair trial, if he really did show up to her house unprompted, I don't think it's unreasonable for the court to ask him what exactly he meant by what he said there. (But I do not think that the court should immediately grant the request for a new trial.)

Well, if some payment or threat was made, that isn't just borderline witness tampering, that is plain old witness tampering. ;-) Also a possibility that should be looked into. (If only to determine if further criminal charges should be brought.)

Well, this is a weird one, because Holmes hasn't even been sentenced yet. According to Wikipedia sentencing is scheduled for October 2022. The trial so far has only determined whether the jury has considered her guilty or not, not how much she should be punished. They did find her guilty (beyond a reasonable doubt) on some of the charges, but not guilty on some others. (And on some of them a mistrial was declared because the jury didn't reach a verdict, and the government declined to retry those charges.)

If Rosendorff did indeed show up at Holmes's house claiming that the government made things seem worse than they were, I'm not really sure how to take this. I can see two scenarios: one where Rosendorff himself was much more culpable in the entire matter than previously assumed, and he's now feeling guilty that Holmes (and Balwani) are on the receiving end of the justice system, but not him. The other scenario is that after the jury verdict came in, the maximum prison sentence (which Holmes is unlikely to get) was covered by the media, and he feels guilty by participating in the trial for that reason, because he doesn't think she deserves to go to prison (or at the very least not for that long).

In any case, I'd be very surprised if the judge grants the motion for a new trial, because on the face of it I would interpret the sentence the government made things seem worse than they were as pertaining to the degree of culpability, not on the question of culpability in general. However, since sentencing hasn't happened yet, I think it would be reasonable for the court to delay sentencing a bit, and to go back to Rosendorff to ask him what he meant by those statements. I'm not a lawyer, and I don't know if there are any procedural issues that would prohibit this, but I do think this would be the right thing to do. Then the court can make a decision, depending on the response of Rosendorff, possibly letting that influence the sentence.

And if there really was a bombshell in the new testimony of Rosendorff (something like "I completely lied" because $reasons), then the court could still decide for a new trial at that point. I find that outcome extremely implausible though, I highly doubt that anything Rosendorff has to say would overturn the existing guilty verdicts.

I can't fault Holmes's lawyers though - it's their job to zealously advocate for their client, and a new trial is a possible remedy when new evidence comes to light (that the defense couldn't have known during the trial), so of course they're going to ask for one, if only to make their alternative requests seem more reasonable. (Of course both the prosecution and the judge will know how that as well.)