Even 21 was only possible by cheating (optimizing away the difficult part using prior knowledge of the results) [1]. Craig Gidney has a blog post that shows the actual quantum circuit for factoring 21 which is far beyond the capabilities of current quantum computers [2].
The processes involved in solar fusion have been well understood since the 1930s [1,2]. Hans Bethe won a Nobel Prize for this in 1967. The problem is that one cannot produce stellar densities and pressures in any kind of apparatus.
All experiments at the AD are strongly limited by the low rates. If there was a straightforward way to improve this by many orders of magnitude, they would have done it a long time ago.
The antiproton decelerator at CERN has been operational for 25 years, and they have plenty of smart engineers there. Unlike in the 1940s, the underlying physics has been well understood for many decades. I would argue that nuclear fission is the counter example that happens to be surprisingly easy to do.
Anything but airplanes. Current airliners have a fuel fraction between 24 and 47%, so any decrease in energy density massively compromises performance. The energy density of jet fuel is two orders of magnitude higher than modern batteries, and jet engines have efficiencies of a few ten percent.
Built with asp.net and jQuery. Nowadays, it would probably be some React monstrosity that takes 30s to load and only shows one item per page (when did information density become evil?).
It's easy to overlook the sample memory spec. These old Teks have decent bandwidth and 1 GS/s, but only a few thousand points of memory. This makes it impossible to zoom into a captured trace and leads to severe aliasing (trap for young players). In comparison, modern designs store millions to billions of points and can operate at high sample rates even on a slow timebase.
I hate the TDS2024 (early-2000s digital scope) with a passion. The LCD is awful, the power supply breaks all the time, and the sample memory is laughable compared to modern scopes, making it very difficult to use. Tektronix kept selling this thing (and similarly awful other ones) until not so long ago for an absurd price (~2000$). For some reason universities kept buying them instead of vastly superior ones from other companies. So now this piece of garbage is a absolutely everywhere.
The fuel is (almost) harmless, it's the fission products that make reactors dangerous. Many of those are water-soluble. Of course the fuel elements should be encased, but drinking pool water is probably not a great idea anyway.
Yes, there is a limit called "quantum projection noise" that determines how much frequency stability one can achieve with a single-atom clock [1]. With N independent atoms, this limit gets smaller by 1/sqrt(N), but with N entangled atoms one can achieve a 1/N scaling. This is the ultimate limit (Heisenberg limit).
This is true, but atomic clocks are about a million times more accurate than any other measurement device. For all pratical purposes, they are never the limiting factor.
The samples from the authors' GitHub are just some text vomited onto slides, and the AI voice reading them point by point. Exactly the opposite of a good presentation.
I always used a Virtual Box VM for Office. After giving this a quick try, I'm impressed. The dockered VM is much less bloated then a normal Windows install, and somehow running the apps via a local RDP connection is significantly smoother than the Virtual Box graphics stack.
The authors carefully avoid giving any actual efficiency figures (or figures from which the efficiency could be easily calculated). Since they barely managed to power an LED under concentrated sunlight, it is probably safe to assume that the efficiency is extremely low. The reference point for the "15-fold" improvement is literally putting a Peltier element out in the sun. Looks like they really don't want their device to be compared to a photovoltaic cell.
Watches tend to be exposed to light a fair bit, so putting a solar panel in the watchface easily outperforms a betavoltaic cell. This has been available for decades, and even some of the high-end Garmins have it.
The US Navy runs 50 MWe reactors inside billion dollar submarines. The requirements are almost entirely orthogonal to economically viable power reactors.
I believe that it may be misguided to focus on compute that much, and it would be more instructive to consider the effort that went into curating the training set. The easiest way of solving math problems with an LLM is to make sure that very similar problems are included in the training set. Many of the AI achievements would probably look a lot less miraculous if one could check the training data. The most crass example is OpenAI paying off the FrontierMath creators last year to get exclusive secret access to the problems before the evaluation [1]. Even without resorting to cheating, competition formats are vulnerable to this. It is extremely difficult to come up with truly original questions, so by spending significant resources on re-hashing all kinds of permutations of previous question, one will probably end up very close to the actual competition set. The first rule I learned about training neural networks is to make damn sure there is no overlap between the training and validation sets. It it interesting that this rule has gone completely out of the window in the age of LLMs.
It is indeed an impressive feat of engineering to make the start menu take several seconds to launch in the age of 5 GHz many-core CPUs, unlimited RAM, and multi-GByte/s SSDs. As an added bonus, I now have to re-boot every couple of days or the search function stops working completely.
No, we do not have an absolute reference point for time. Our best models of physics are time-translation symmetric (this is equivalent to conversation of energy via Noether's theorem), so absolute time cannot have any measurable effect. There is, however, a human made reference time scale (International Atomic Time, TAI) which is generated by the BIPM using (mostly GPS-based) comparisons of the local time scales of many national labs.
The largest uncertainty of the Al+ clock is caused by relativistic time dilation due to the ion moving around a bit in the ion trap. This would affect a 229Th clock in the same way, but maybe the larger atomic mass would help.