Crabs in Love: Monogamous Parasites of Sea Turtles | April 15 2026, 21:56

WOW, it turns out that under the shell near the anus of sea turtles, the parasitic crabs Planes minutus make themselves at home, and there’s only enough space for a cozy duo, so they form a monogamous pair and live happily ever after inside the turtle’s butt (had no idea what to do with this information, so I brought it here). In relation to the turtle, this is commensalism. It’s when it’s good for one (or in our case, two), and the third doesn’t give a damn. I see a scientific paper claiming that they sometimes mistake the turtle for ocean debris, where there’s room for more than one wife, and then, goodbye monogamy. But, at least, no butts involved.

Navigating the Depths of High-Dimensional Spaces | April 13 2026, 23:17

I am now working a lot with high-dimensional vectors, and some things that I hadn’t fully realized before are really starting to tickle my brain. Our 3D intuition doesn’t just not work there—it lies.

It turns out that any two random vectors in high-dimensional space are almost certainly nearly perpendicular to each other. Almost all the space is one continuous “equator”.

Much of machine learning is built on exactly this. If your embeddings suddenly show high cosine similarity (for example, 0.8 — this is not a statistical error, but a powerful signal. It’s almost impossible to randomly converge like this in a 1000-dimensional world.

In such spaces, almost all the mass of data is concentrated in an extremely thin surface layer. The “insides” of objects are mathematically empty.

This can be easily verified with such an imaginary example. Take the “skin” of a multidimensional sphere with a thickness of just 1% of the radius. The volume of the sphere is proportional to the radius raised to the power of its dimensionality.

• In three-dimensional space, the pulp (0.99 of the radius) occupies 97% of the volume, you raise 0.99 to the third power.

• In 1000D, the pulp occupies just 0.000043%.

You can understand it differently. For a point to be closer to the origin, it requires that along all axes the coordinates need to be close to the origin. If one axis has a high value, that’s it, the point has gone. If you take points randomly, the mere probability that they all at once will be below any value decreases with the growth of dimensionality, and decreases quickly.

All the “meat” of the data always ends up in the skin. Any sample in High-D is essentially a set of boundary values.

For white noise in high dimensions, the distance between the closest and the farthest neighbor becomes almost the same. The concept of “closeness” simply degrades.

Sky-High Prices at the CIA-Adjacent Gas Station | April 11 2026, 21:16

We have one gas station near the CIA that simply sets gas prices 40 percent higher than anywhere else. Just an ordinary shabby station that follows the principle of “if it works, don’t fix it.”

CPU vs GPU: A Speed Challenge in Embedding Creation | April 11 2026, 18:08

When working with certain tasks, the difference between a CPU and a GPU is simply astounding. For example, I need to create many (millions) of embeddings, model BGE M3. Running this on my quite powerful 24-core Intel Core Ultra 9 285K processor takes 45.85 seconds to create 500 embeddings, while using an NVIDIA 5090 GPU, the same task is completed in just 0.36 seconds. It is so fast that I specifically wrote this benchmark to figure out whether my GPU is being utilized at all. The program that sends requests to TEI does it in test mode not actively enough (roughly a couple of times per second), and the GPU load graphs are practically zero.

— Testing http://localhost:8080/embed — <– CPU version

Requests completed: 500

Total time: 45.85 sec

Throughput: 10.90 req/sec

Average latency (Avg Latency): 4386.11 ms

P95 latency: 5021.88 ms

— Testing http://localhost:8090/embed — <– GPU version (NVIDIA 5090)

Requests completed: 500

Total time: 0.36 sec

Throughput: 1398.69 req/sec

Average latency (Avg Latency): 31.38 ms

P95 latency: 53.18 ms

========================================

RESULT: http://localhost:8090/embed is 99.22% faster

Yuki’s Mysterious Bi-Annual Behavior Shifts | April 09 2026, 14:31

Yuki’s “ooooh” mode is activated again (April 7, 2026). It usually lasts a few days in April and October.

Previous occurrences were –

– October 15-20, 2025

– April 11, 2025

– April 1-4, 2024

– February 2, 2023,

– October 27, 2022,

– March 15, 2022

Behavior changes during this period include:

1) He might sing songs for hours on end. For instance, at six in the morning.

2) Suddenly, he likes to go for walks. Usually, he does not. Even though he always has access to the yard, he specifically needs to go on a walk. He might go to the door and knock on it with his paw. Usually, at the word “walk,” he rushes to the third floor.

Now, he looks into your mouth when you’re talking to him. Always seems to be waiting for something, possibly expecting the question of whether he wants to go for a walk. He knocks on the window and the front door with his paw.

And yes, he starts wanting to walk at around six in the morning, and then again soon after returning from a walk.

3) On the walk, he sticks his nose in the grass every five minutes, and it’s hard to pull him away. Usually, this is rare, but now it’s constant.

4) He might sit and watch the sunset for half an hour.

5) Unstable appetite, occasionally. You put meat on top of his food, and he doesn’t even look at it.

Navigating Nabokov: A Companion Glossary for “Lolita” | April 08 2026, 11:24

I have finally finished the book The Reader’s Glossary – essentially a 5200-word dictionary for “Lolita” by Nabokov, but organized not alphabetically, like regular dictionaries, but in order of the occurrence of complex words, divided by chapters and indicating the context of the word or phrase. The website – readersglossary dot com (see the first comment). It is expected to be used, among other things, as a companion book while reading the original. Yes, it’s twice as thick 🙂

The dictionary turned out quite thick – 600-700 pages. It is available in four languages – Russian, English, French, and German. Moreover, the translations (RU, FR, DE) or clarifications (in ENG) are not abstract but contextual, taking into account how Nabokov himself translated the fragment from English (“Lolita” was first written in English, then translated into Russian).

On my website, there are huge fragments of these dictionaries RU, FR, DE, EN available for review (each about 1/3 of the total volume).

There is also a full-fledged interactive dictionary on the site, where you can enter a word and see its translation or explanation. The dictionary mainly contains complex words, but we know that complexity has its own definition for everyone, so all words are divided into three categories and highlighted with different frames. Probably for a well-read Anglophone, the first category (dotted) is completely useless (about 50% of the dictionary), for the less-read, maybe 20% are useless. But I decided not to cut it further, because the book is not only for Anglophones but also for those for whom English is a second language, and there those dotted frames are very handy.

Overall, I did this “for myself and friends,” just for fun, not as a commercial project. Therefore, I am quite sober in understanding that it has a super niche audience, and if even once a week someone finds it useful, it’s already nice.

Although it was something like a hobby, the book took a lot of time. To achieve what I did, I developed a dozen applications/scripts, a couple of which have their own interactive UI, in which I spent many hours over two months of work. And of course, I learned a lot in the process, which is actually the main fun of it.

So, come to the website – readersglossary dot com. Link in the comments

P.S. In Russian – only as a PDF for now. Amazon doesn’t allow selling books in Russian, only in a small number of European languages in addition to English. The French and German versions of the dictionary will be released on Amazon about a week from now.

Nikolai Nosov’s Anticipated Instagram: Art Templates and Avatars | April 07 2026, 13:04

As early as 1954, Nosov predicted the emergence of Instagram

P.S. It’s just not very clear why there are different templates for different eye and hair colors.

Space Exploration: From Pioneering Days to the ISS as a Playground | April 06 2026, 19:59

essentially, a person in space is about these days and about the days 50 years ago. The International Space Station is already kindergarten: in terms of the Earth, it is only slightly higher than airplanes fly.

Precision in the Sky: Aerial Refueling of HH-60 Pave Hawks by a KC-130 Hercules | April 05 2026, 12:59

An interesting photo from Iran. An American Lockheed Martin KC-130 Hercules tanker and Sikorsky HH-60 Pave Hawk helicopters connected to it. If you think about it, it’s incredibly complex. Look, the plane has to fly at a super low speed for it – close to the stalling speed – while the helicopters, in contrast, must push to their limits to keep up. To avoid entering into a spin, the plane is forced to rapidly lose altitude, consequently, the helicopters must also purposely drop altitude. The helicopters are positioned lower than the plane, so if the pilot slows down even more (though how much more can he slow down?), and the helicopters don’t slow down, the hose could hit the rotor blades and that’s it. The helicopters also gain several tons during refueling, which adds to the complexity. Why refuel two at once? It’s more complicated. Actually, it’s both more complicated and simpler, because the load on the wings is distributed symmetrically, making it easier for the plane to maintain a stable course. It’s also interesting how the issue of static electricity is handled in the dry air.

A good addition from the comments (Sergey Snegirev):

1. It is noted that the Hercules stalls at 100-110 knots (depending on the air temperature and altitude, which is important in Iran), and the photo shows the flaps deployed, allowing it to stay up to even 90 knots. Meanwhile, the Pave Hawk can accelerate up to 190 knots (but obviously, nobody performs AAR at max speed), with a cruising speed of 150 knots, so there’s quite a sizable overlap. It’s assumed that AAR takes place around 120 knots on the video.

3. It is noted that AAR happens at exactly the same altitude, so there’s no need to lose altitude

4. It is noted that the tanker – like any other aircraft – can dissipate static electricity using an electrostatic discharger. The refueling hose has a contact that equalizes the potentials of the two aircraft before fuel delivery.

5. It is noted that the fuel used to refuel external aircraft is stored in tanks in the aircraft’s cargo section, separate from the tanker’s own fuel system. Even if the pumping were from the tanker’s own fuel tanks, it would come from the central tank, which is set up to transfer fuel from the side tanks.

It is noted that on the underside of the plane’s wings, several metal “antennas” are installed on the flaperons to discharge static electricity into the air. It is noted that when the hose is connected, a wire passes through and the potentials on the bodies are equalized.