Hi, I'm Rauf Aliev.

A fabulous interactive self-study on the basics of building Large Language Models (GPT). Highly recommend everyone interested to at least take a look. It’s better accessed from a computer, as you won’t grasp anything on a phone due to the need for screen space.

my 31st painting. Little fires everywhere

Today we shopped at Lotte Plaza. Look, they sell frozen sea cucumber here, or holothurian, or look, it’s frozen; it’s an edible trepang. It’s quite a funny creature, really. Essentially, it is the marine equivalent of a worm. Only it’s much more “vegetable” than a worm. On one end it has a mouth, on the other an anus, and a tube in between. They breathe through these anuses and as it was recently discovered, they also eat a little bit through them (though the mouth is still the priority). It’s also interesting that sea cucumbers can quickly become stiff or soften into a semi-liquid jelly, depending on the circumstances. Moreover, they can regenerate all internal organs and recover even after being cut into two or three parts. Cases of asexual reproduction through splitting in half have even been described. All in all, they’re interesting creatures, but I won’t be eating one.

When I started delving into school-level biology from the perspective of a forty-year-old, I was hooked by something that perhaps wasn’t covered during my school years, and I am not sure about now. But it’s very intriguing.

It’s about how the body knows which cells to create and where. This is quite a non-trivial question, and there are so many layers of meanings and explanations that I don’t even know where to begin.

First, the basics, from the school curriculum. Look, our DNA contains only 19969 coding genes in DNA (plus another 63,494 non-coding DNA genes, which are also called “junk”). By the way, the Y-chromosome was sequenced just this year, but overall the human genome has been decoded. It is mostly not entirely clear what it does (at least not clear), but at least we have the source code. Essentially, it’s a 3.3 GB file.

So, each of these 20000 genes encodes proteins for various purposes, for example, like building material for 230-400 types of cells. Protein is a three-dimensional structure made up of amino acids. There are 26 different amino acids. It’s easier to imagine it as a tangled string with beads, where some beads of 26 types repel each other, others attract each other, in short, if something doesn’t work out, the whole protein can end up as a tangled string with beads, or it may not. As mentioned, there are 26 amino acids, and the sequence determines which protein will result. Actually, the “program” consists of “triplets” (a three-digit code), which are translated into amino acids, and these in turn create the protein ultimately churned out into the body and then put in its proper place. Interestingly, the genetic code is the same for all living beings on the planet (Well, almost, very very rare exceptions exist, but there’s a very slight difference). Also, it’s worth reminding that this program forks billions of times from the first day of life, and each of us has 30 trillion instances. With copying, errors also accumulate, so a significant part of these 30 trillion (probably all) are slightly modified versions relative to each other. Yep, it’s also worth noting that people differ by one percent of DNA. That is, 99% makes us such that we generally survive somehow, and 1% adds uniqueness—from facial features to the way of thinking.

So, why this introduction. The question is how such a complex organism can be programmed by such a simple program (remember, just 3GB of code, of which maybe 1% is actually functional, the rest is who knows what).

And here’s what I found out. Evolution made this program such that it changes itself in a specific, well-adjusted direction, allowing very little randomness in this process. Just imagine, you have a program that calculates and outputs the number pi. Then three bytes are removed from this program, and it starts to output the number e, then some more three bytes are removed, and it starts to produce the Planck constant, then another five bytes are removed, and there you have the constant g. A good analogy is a record. You play it—it plays one music. Then cleverly made scratches cause the needle to jump back and forth, and the music changes, but it’s still music, not noise. If you imagine the structure of the organism as a set of interactions within the cell, each producing protein and attaching it somewhere, then on every iteration N+1, this program undergoes small, nature-intended corrections through trial and error, and thus conventionally through about 100 generations, the cell begins to multiply proteins for conventionally the liver, and then after about 300 generations for conventionally the skin. These corrections are called gene expression factors. Effectively, these are various things floating in the cell, which either came from outside, or were born in previous generations. If it came from outside, it could even be from the environment, or they could be chemical signals from other parts of the organism. For example, a child is born to a mother, and obviously, the cells “know” about it. This mechanism really surprised me back in the day. How can you write a program that works differently over the organism’s lifetime depending on the results of previous iterations! But as we see from our experience, it somehow works, and even quite reliably.

If you draw an analogy with programming, it’s the same program, but instead of parameters, it takes a “mask,” filtering some of the instructions. The mask is selected such that the resulting program continues to operate and do something useful. It’s like with that example, having a program for generating the number pi, applied a mask, it turned into a program for generating the number e. Evolution tried different masks and those that resulted in useful progress survived to the next iteration, N+1. For multicellular beings, this still happened asynchronously and synchronously at the same time. Asynchronously, because cell A doesn’t know about the state of cell B and in which iteration it is. Both cells’ programs output something and progress to the next iteration. Synchronously because enzymes (consider part of mask A) may depend on the results of B’s work, so B must precede, thus there’s synchronicity. And this mechanism obviously often fails more than it works, but since it’s tested at the cellular level (we have 3 trillion of them) and at the level of a living unit (in the case of bacteria it’s a number not even countable, then unicellular, then multicellular, those are some wild numbers) and all this over millions of years, it is quite believable that random “masks” leading to advancement further to N+1 are not so rare. So gradually, that’s what happened.

On average, cells divide 52 times, then die (apoptosis) (exception – cancer cells) This is called the Hayflick limit and is related to telomeres—maybe you’ve heard about experiments to extend life through telomere restoration. We have three trillion cells, with different lifespans for different cells. DNA in each cell is its own, plus due to inevitable mutations slightly different than its neighbor. In the end, the program for fingers, toes, hands, ears there takes the form of a single event, which triggers symmetrically at the same time in some iteration. Subsequent iterations just finish what was started. If something malfunctions at the gene level, then problems arise immediately on both sides. And if with something critical, then in the following iterations it starts driving unexpected enzymes leading everything awry, mostly ending in miscarriage. In general, life is bustling 🙂

The second interesting thing is epigenetics. It’s exactly about the factors of gene expression. It is believed that when a child is born, what he gets are genes from dad and mom. In reality, he also receives some unique set of these external gene expression factors from the mother’s body. Moreover, the result of the DNA work of this child in the mother’s womb ends up in the mother’s body (though there is a placental barrier, but it’s overcome), and then participates in the processes of the next pregnancy, or even affects the mother’s body. How it participates is unclear, maybe it’s all on the edge of “nothing,” but the fact that in the second child’s body there are greetings from the first child’s body is a scientifically proven fact. Google fetal microchimerism.

In general, all this is wildly interesting. And it seems like a bottomless barrel of knowledge. If I were a school kid now, I would recommend not going into IT, but into life sciences. The most significant discoveries in the next 10-20 years, I believe, will be there. And yes, pretty soon they’ll connect the brain to AI. Everything is already there, just needs a little “fine-tuning” (about 20 years should be enough).

Minute breakfast. This is potato

Checkmate just half a second before the end of the game!

Interestingly enough, I don’t remember what I was coding, but during the process, I realized that I’ve never been bitten by mosquitoes around here. I’ve never heard a mosquito buzzing at home, even though we leave the balcony wide open all summer. There’s a pond with frogs just a 5-minute walk away, or ten minutes if you’re a mosquito. Today I read about a homemade laser anti-mosquito system with radar and realized that if I built it, I’d have nowhere to test it. So I wrote on nextdoor, neighbors, where have the mosquitoes gone?

I was expecting links to successful stories about the introduction of mosquito saboteurs. I wanted to write a post about the results. Indeed, such an experiment is being conducted in our state – they catch mosquitoes, inject them with GMO potion in the butt, and release them into the wild. They happily start families at the nearest pond, and soon the mosquito wife starts noticing that the foreign dad is only making sons, while the daughters vanish into mosquito heaven before even being born. With dad’s genes, sons fly off to start their own families, resulting in the same outcome, a shortage of brides begins, and eventually, there’s a significant mosquito die-off.

Anyway, I asked the neighbors where the mosquitoes had gone. And they started replying. That there are tons of mosquitoes around here, and they’re devouring the neighbors insatiably. Every second person responds that I don’t have any because they all flew to his or her place. Overall, I don’t know what they’re being injected with, but now the imbalance is elsewhere. I don’t know what’s wrong with us, maybe because we’re not yet true Americans.

I think it’s all Yuki’s doing. He clicks his teeth and eats anything that flies, and apparently, in the mosquito community, our house is marked with a red cross. And when Liz’s cat comes over, it’s marked with two.

Published an article on hybrismart about my utility for comparing file system snapshots. It’s amusing that this is actually the second iteration—the first was made under DOS, to share a library of warez according to friends’ choices. Of course, the utility now has broader applications possible 🙂

Briefly, it works like this. Vasya does a dir /S on his Windows machine or ls -lR on his Mac and sends the file to Petya. Petya opens it in my program, selects a hundred files from different directories, and creates a request file, which is essentially a script or a flat list of files, and sends it back to Vasya. Vasya receives this script, executes it on his machine, retrieves an archive with only the selected files, and passes this archive to Petya. Petya unpacks the archive, and all the files are placed into the corresponding spots in his file system, preserving all directories.

The second case is when both Vasya’s and Petya’s machines contain the same huge archive of something, which is supposed to be identical, but things happen, and something gets corrupted or deleted that wasn’t supposed to be, or, conversely, some extraneous files are created that alter the system’s behavior on Petya’s end differently from Vasya’s, even though you would think the source code, which is under version control, is the same. How to figure this out? That’s where the utility comes in. Each takes snapshots of the directory with all subdirectories, and then the utility compares them and shows the difference. Comparing them in other ways is very tricky, especially if Vasya and Petya are on different operating systems, or on the same operating system but in different languages, or on the same OS but different versions or modifications.

The source code is included. Requires Python/QT.

My first little Swift program for macOS! I am now 0.01% a junior developer for Macs! Look, I can do this in any editor, on video it’s Outlook, select any text, and by pressing a key combination, rephrase the selected text (Cmd-Option-R). If it’s in Russian, it will be translated and rephrased. Translation only — Cmd-Option-T. There’s also a feature to correct the grammar without changing the sentence – Cmd-Option-G. ChatGPT is used for this, of course, so a paid account is needed.

From start to finish, developing this little program took two hours. Including installing XCode and trying to figure out how everything works there. Of course, ChatGPT helped. It’s my first time seeing Swift.

I’ve uploaded it on GitHub. Just insert your own OpenAI key 🙂

When buying paper on Amazon, the main thing is not to accidentally make a mistake with the volume. Otherwise, you might come home to find 20 tons of printer paper in front of your house, and your credit card $35700 lighter. And it looks like rain.

By the way, in English, 500 sheets of paper are called a ream. I just ordered 3 reams.