I write quite regularly about all sorts of strange ways to output information. Recently I came across an ancient, but good review post about alternative displays. There are a lot of interesting things — projectors on the fog, transparent LED screens, kinetic screens, LEDs on the blades of the propeller system, and so on. I strongly recommend reading it.

What else can I add about exotic I/O devices? The other day I learned about the existence of a market of Braille typewriters (although if you think about it, there is nothing surprising in it). Grisha Nosyrev sent me a link to a project to create a Telegraph key from a laptop. Vas3k not so long ago published a great post with an analysis of augmented reality technologies. Also, maybe the other time I'll write about how we played 'Chizhik-Pyzhik' tune with a dot-matrix printer at school.

John Horton Conway (26.12.1937–11.04.2020)

The "corner of cybernecrophile" rubric: An ancient American company, Tektronix (formerly Tek), has been producing oscilloscopes, testers, and other equipment since the mid-20th century. Recently, I came across their official online museum, where enthusiasts collect various Easter eggs in the form of drawings and caricatures, abundantly hidden in ancient Tektronix instructions, diagrams, and boards. Among others, a picture of the Wizard periodically appears. At some point, the museum's holders even assembled the VintageTEK Demo Board, which draws the logo of the Tektronix and this Wizard on an oscilloscope, and also allows you to play Pong.

Also, comrade Gevor brought a link to a post about Easter eggs in geographical maps of Switzerland.

I found an interesting blog by Eleanor Lutz, she is a graphic designer in the field of scientific visualization. Draws maps of Mercury, Mars, Jupiter and other celestial bodies, various medical and anatomical infographics, maps of the movement of wings during the flight of various creatures and more.

The picture above is the combined map of the constellations of 30+ different terrestrial civilizations.

Recently was Zaliznyak’s birthday, also this year I finally got my hands on his famous linguistic problems book. So I was finally able to formulate for myself why his works were so interesting and outstanding, and why his "birch bark" lectures always brought together a bunch of completely different people, not related to linguistics. I think, he had a complete hacker approach to the language - in a fascination, unexpectedness of the methods used, and humor, his lectures resemble stories from some DefCon about the reverse engineering of a rare device. Another example is his linguistic problems. They were reprinted a couple of years ago, so they are available in Russian online bookstores. As an example, I'll quote the first task:

Task 1 ("gizona")
Intended for persons unfamiliar with the Basque language.

Initial data:
Given a text of 12 phrases in an unfamiliar language (Basque). It is known that one of the phrases is grammatically incorrect due to an error in one word (in a more strict form: due to the fact that in one case one sequence of letters between spaces is replaced by some other sequence of letters).

Text:
1. Gizona joaten da.
2. Gizonak zaldia ikusten du.
3. Astoa atzo joaten zan.
4. Gizonak atzo joaten ziran.
5. Astoak zaldiak atzo ikusten zuen.
6. Zaldiak gizona ikusten du.
7. Zakurrak joaten dira.
8. Gizonak zakurra atzo ikusten zuen.
9. Zakurrak astoak ikusten ditu.
10. Zaldiak gizonak atzo ikusten zituen.
11. Zakurra atzo joaten zan.
12. Gizonak astoak atzo ikusten zituen.

The task:
Find a grammatically incorrect phrase and make it grammatically correct by changing (or replacing) in it only one word.

There are almost 3,500 streets in Moscow, rather only about half of them are "streets", the rest are "driveways", "embankments", "boulevards" and so on. And if you paint them in different colors, you can see hidden toponymic patterns ;) I spied the idea in a post by Erin Davis, there are many different cities, but there wasn’t Moscow, so I took the native code for R, slightly modified it and built Moscow plan.

Also, if you're not interested, look at how Victoria Rose embroiders aerial photographs on a hoop, or read how to write a sqlite query to calculate the Mandelbrot fractal.

Meanwhile, a new subgenre in abstract painting is emerging — Martin Calvino draws abstract paintings inspired by images generated by neural networks.

Recently found an interesting project, ThisWordDoesNotExist.com by Thomas Dimson. The system invents new English words and generates descriptions for them in the style of dictionary entries. On the website you can see random words or get a description of your own invented; there is also a twitter bot. As usual, I went into reading the code, hoping to spy on interesting crutches. The huggingface-based code, the main network is GPT2, was retrained on the separately parsed UrbanDictionary.com. The network generates dictionary entries from scratch, then regexp heuristics try to disassemble the output of the network into the desired structure -- the word itself, description, usage examples, etc. The word is checked for presence in the blacklist, which simply contains a large set of existing words and derivatives from them. Just before showing on the site, the whole text is checked against another blacklist. Also I spotted several other attempts but they didn't get to the production.

The projection of a wave shown on a huge LED screen in Seoul. Some more details here.

As everyone knows, the Jules sets corresponding to points inside the cardioid of the Mandelbrot set are simple and boring, and those that correspond to points far outside the Mandelbrot set are disconnected and dusty. The most interesting things happen at the border of the cardioid, and in the video above, Matt Henderson calculated a beautiful "torus", swept by Julia sets corresponding to the point moving along the border of the cardioid.

Also, readers who dislike fractal images can study the history of x86 register naming or read about the new shining GPT-3 model released yesterday.

Today I’ll write about April Fools Day's Request For Comments. RFCs are documents describing Internet protocols and various technical issues related to their use, in fact, the main base of Internet standards, if the word "standard" is generally applicable here; at the moment there are almost 9 thousand of RFCs. Engineers sometimes like to joke, and April Fool’s RFC began to appear from time to time (Wikipedia already knows more than 50 of them).

The most famous (albeit not the first) joke was published in 1990 in RFC 1149, "IP over Avian Carriers". The standard indicates that "bandwidth is limited by leg length", "broadcasting is not supported", "MTU increases on average with carrier age" and other useful information. Later, the standard was expanded in RFC 2549 (QoS support) and RFC 6214 (IPv6 support). In 2001, the protocol was practically implemented in the Bergen Linux user group, 9 ping packets were sent and 4 replies were received:

--- 10.0.3.1 ping statistics ---
9 packets transmitted, 4 packets received, 55% packet loss
round-trip min/avg/max = 3211900.8/5222806.6/6388671.9 ms

Among other April Fools' Day RFCs, I’ll mention RFC 2324 + RFC 7168 on the "Hyper Text Coffee Pot Control Protocol" (very similar to my old coursework), RFC 3251 "Electricity over IP", RFC 5514 "IPv6 over Social Networks" (later also implemented) and RFC 7511 "Scenic Routing for IPv6."

This year RFC 8774 is published, "The Quantum Bug," but I have not read it yet. Also, once upon a time, I drew interactive "RFC constellations" — a graph of mutual citations of these documents. You can check it here.

Topi Tjukanov trained StyleGAN2 on old maps, wrote a detailed post about it, and also published a trained model, which he called Mapdreamer.

A while ago Grisha Sapunov (CTO in Intento) and I made a Russian-language telegram channel @gonzo_ML, where we occasionally analyze interesting scientific articles on machine learning and neural networks. Now, 1.5 years after, we decided to make its English-language mirror on Twitter https://twitter.com/gonzo_ml. Welcome!

Found beautiful visualizations of migratory bird flows in the Americas; there are more such charts on the National Geographic website.

Also check: a poster with Britain’s most complex motorway junctions and a thread about anime floppy disks.

In collaboration with the @NeuralShit channel, we made a StyleGAN2 network for coats of arms generation. Check the telegram bot @this_coatofarms_doesnotexist_bot.

So far, the generation does not use texts in any way, but it may still be possible to experiment.

I've run a demo with a Mobius lent on a cheap camera, played tunes on disk drives and printers, made a text scroller though favicons of browser tabs, drew pictures on the clock, and made a neuromusic-box. I've seen Doom on an e-paper reader, Quake on an oscilloscope, and cartoons on a screen made from 640 bottles.

But the first time I see a demo on loading indicators of 64 processor cores. Source

A fresh sketch from Starkey Comics about the genetics of alphabets. At least some places cause questions, but, in general, the map looks interesting.

Long time no nerdy jokes, so now here is a bunch for you:

Switching from qwerty to dvorak generates a cyclic group of order 210 (while the order of the group generated by colemak is 42). At the top left picture is dvorak^2, and at the bottom left is dvorak^-1. The source is an excellent twitter thread. On the right is the truth table for the == operator in js. Source. And here you can see what happens if you iterate it according to the rules of Conway's Game of Life.
Also, check the Scunthorpe Sans font, which automatically censors abusive English words using ligatures.

As usual, I came up with an idea, but it already exists: Bas Uterwijk from Amsterdam recreates portraits of various mythical and historical characters using Artbreeder, for which he started a special instagram, GANBrood.

No time to explain, but there is a great twitter thread where different people describe step by step how they calculate 27+48 in their head.
For sure, there should be some special scientific words for this stuff, anyway it looks just fascinating.

Several years ago, my colleague Anna Shishlyakova drew an excellent ligature with an internal slang quote (top left). Then I came across the works of Viktor Pushkarev with different types of Slavic calligraphy, called Vyaz. Of course, I wondered how one could automate the formation of ligatures, at least using a relatively formalized example of monospace vyaz.

Last weekend, I finally managed to find time and assembled a kind of prototype. It works stochastic and doesn't always do well, but you can usually play around with the parameters and get a nice result.

The code is available on github, and a short post with the description is on the medium.