These are optical clocks, so their rates are in the hundreds of THz to PHz range (1.121 PHz in the case of the Al+ clock). A 1e-18 shift corresponds to a ~mHz frequency deviation which a decent frequency counter can resolve in around 1s. However, no optical clock is actually that stable at short time scales, and the averaging is required to get rid of these fluctuations.
LLM companies try to optimize their benchmark results, not to test the capabilities of their systems. This is why all the benchmarks are so utterly useless.
I'm all for making fun of LLMs, but asking this from a software that processes vectors of tokens is a bit silly. The information isn't really there in the input.
It's much worse, the maximum on-chain transaction rate is something like 7 per second. Also the time intervals between blocks have a huge spread, so it can take more than an hour for a transaction to be confirmed if you are unlucky. This is obviously impractical, so people came up with schemes such as Lightning to avoid touching the blockchain as much as possible. Of course this makes it much more difficult to judge whether the system can be cheated in some way...
That's the problem. Building a heavier rocket is much harder than building a lighter one (see explosion above). So why not send a few lighter ones instead of a heavy one? This is what the launch market has concluded for a long time.
Sure, the Saturn V and STS were much less safe than smaller rockets. Still, they blew up an awful lot less than other rockets of their size like N1 or Starship.
The problems with Starship make the Saturn V and STS programs even more impressive. However, I still don't get the rationale of building a rocket with such a large payload. The rocket equation will always force you to build an absolute monster compared to a series of smaller rockets. Even worse if you have to haul up a massive orbiter each time. No wonder that small/medium sized rockets (Soyuz, Atlas, Ariane, Falcon 9,...) have always been the most successful.
The accuracy of atomic clocks is much better than our understanding of fundamental physics.
The best calculable atomic system is atomic hydrogen, and state-of-the-art quantum electrodynamics calculations reach a relative accuracy of around 1E-13 for its energy levels. However, already at the 1E-10 level, the structure of the proton becomes significant which can currently not be calculated from first principles. Instead, the proton size is taken as a free parameter which is determined from the measurements.
In contrast, the best realizations of the SI second are caesium fountain clocks which achieve relative uncertainties in the 1E-16 range. Clocks based on optical transitions (rather than microwave transitions) have now broken the 1E-18 barrier. Calculating atomic structure to this level is currently completely unthinkable, even for a system as simple as hydrogen.
The modules physically won't fit. American optical tables have screw holes with 1 inch spacing and some imperial thread, whereas the rest of the world has 25 mm spacing with M6 threads. Having to manually fiddle with the pattern sort of defeats the purpose.
Great idea, but the hard-coded imperial units make this very difficult to use for most of the world. Also the plugin still appears to be extremely buggy and unfinished.
Transformer LLMs are good at generating text since they were trained on huge volumes of high quality texts. There is no reason to assume that feeding a neural net with unlabeled garbage data (i.e. video recordings from random cars) is going to lead to anything.
This is incorrect. The only measurement of antimatter gravity so far by ALPHA-g resulted in no difference to normal matter within the experimental uncertainty [1]. This is not surprising, since any difference would violate conservation of energy.
The vast majority of people are paying separately for all kinds of stuff. I don't see a reason why software and digital services have to be different. In the early days, payment was difficult, but this has been long solved. I am convinced that the quality would improve dramatically if there was some actual competition.
Also most people would probably prefer not to pay for the ad fees that google extorts from pretty much any company.
This list is very telling. Instead of a healthy marketplace of companies competing to sell their software and services, we end up with one monopolist who gives away mediocre products and in return taxes everything you buy (in the form of ad spending), and then annoys you with the same ads. How is this a desirable outcome?
Both Gnome and KDE handle (un-)plugging external monitors just fine. And Wayland has been the standard in all relevant desktop distributions for a couple of years now.
I disagree. LaTeX is very good at layouting test, and can also (reluctantly) put figures into the text. Anything else is a huge hack (like TikZ), and one constantly runs into crazy limitations such as the fixed-point math and the lack of a decent visual editor. Slides should never have paragraphs of text on them, so the layouting is not very useful, but the other limitations are very annoying.
Please nobody actually do this. Good presentation slides have almost zero overlap with the corresponding article since they serve completely different purposes. In my field, seeing beamer slides is a huge red flag for an imminent terrible presentation. Slides are an extremely visual medium, and WYSIWYM is a huge hindrance for designing appealing slides.
Solar shingles are available from a variety of companies. However, the market success appears to be quite limited. Not too many people are willing to pay a considerable premium for a slight cosmetic advantage.
This problem has been solved more than a decade ago by radar sensors (standard on many mid-range cars at the time). They detect imminent collisions with almost perfect accuracy and very little false positives. Having better sensor data is always going to beat trying to massage crappy data into something useful.
They now sometimes dub the audio track. The result is about as horrible as one would imagine. Whoever decided to turn this on by default clearly didn't give a damn.
It would be interesting to compare the accident statistics with European climbing gyms where belay tests are not common.
The coach in the video has some of the worst belay technique I have ever seen. Unfortunately, this is somewhat common among older climbers who learned using the first generation Grigri in the 90s. Petzl's recommended technique back then is very safe (essentially using the Grigri like an ATC), but does not allow giving slack quickly. This made it completely useless for any kind of ambitious sports climbing, and people started coming up with often extremely dangerous workarounds. Petzl has upgraded their recommendations a long time ago, but some people are resistant to change ("it never failed for me"...) Hopefully this video can convince at least some of them to finally adopt the proper technique.
QKD also needs key distribution to prevent MITM attacks. It behaves almost exactly like a classical stream cypher, except that it cannot be package-routed, is many orders of magnitude slower, and is prone to side-channel attacks.