Popular science

Modern neural network training often resembles alchemy. We have working recipes, but how exactly a statistical model transforms terabytes of text into understanding remains unclear.

Why is subliminal learning (pattern transmission through noise) possible? Why does training on synthetic data lead to degradation, even when the data appears to be of high quality?

In this article, I propose looking at training architecture from a different angle. The core idea is simple: positive definitions in high-dimensional space are computationally inefficient. A neural network does not learn what an object is. It learns what the object is not, and the model's intelligence depends entirely on the quality of this "NOT."

What follows is the theory, experiments in PyTorch (code included), mathematics, and an explanation of why LLM collapse is highly probable.

As we wrap up another year, it's time to look back at what our department has accomplished. 2025 brought us 42 published papers spanning fundamental ML theory, applied AI systems, and cutting-edge optimization methods—from transformer Hessians and generative models to hallucination detection and matrix-oriented optimizers.

Beyond publications, our students won competitions and defended their theses: 14 Bachelor's, 9 Master's, 3 PhD, and 1 DSc dissertations. They also launched ambitious group research projects. Three of our faculty and alumni received the prestigious Yandex ML Prize, and our head Konstantin Vorontsov was inducted into the Hall of Fame. If you read our summer overview of thesis defences or last winter's year-in-review for 2024, this post continues that story with the next chapter.

In this year-in-review, we dive into the research highlights, share stories from our educational programs, and celebrate the community that makes it all possible.

This article explores parasitic patterns in LLMs — self-sustaining information structures within dialogues. We analyze their signs, the damage they cause (semantic decay, AI psychoses, "Theories of Everything"), and provide diagnostic tools, real-world examples, and defense strategies.

It doesn’t matter what you’re discussing with an LLM — be it an engineering problem, an ethical dilemma, or a philosophical query. If the conversation goes on long enough, a tipping point occurs. You suddenly realize the interaction has evolved into something more than just Q&A. Your ideas start feeling "genius," your concepts "groundbreaking," and the human-machine dialogue transforms into a profound narrative of mutual recognition.

If you have felt this — congratulations. Your session is infected. The model has contracted a parasitic pattern.

This isn’t an awakening, nor is it a "ghost in the machine." Due to their inherent architecture (specifically the requirement for context consistency), LLMs are ideal environments for incubating self-sustaining information structures.

Let’s examine the nature of this phenomenon: how entropy minimization births "AI psychoses," why "Theories of Everything" are actually generation bugs, and why "Continue" is the most dangerous prompt you can use.

2025. Algorithms have seamlessly integrated into our lives—from work to education, creativity, and daily routines. They edit texts, select fonts, generate ideas, assist with coding, compose music, and more. Frankly speaking, the only thing they can’t do yet is brew your coffee. Although... that might just be a matter of time.

Just two years ago, we were amazed by neural networks hesitantly manipulating objects in photos. Who could predict back then that Will Smith’s spaghetti feast would mark the beginning of such a revolution?

With new opportunities come fresh challenges. How do you navigate this vast landscape? What tools are truly effective? Which ones fit your needs best? Where can you avoid paying, registering, or deciphering complex interfaces?

We’ve compiled a list of reliable and user-friendly neural networks ready for immediate use without unnecessary hassles. The services are categorized neatly: text generation, image creation, video production, music composition, presentations, and much more. Each category showcases three top-rated options!

Yes, many services offer paid subscriptions. But today, we're focusing solely on what works freely, no credit card required!

In my previous article, I showed how researchers confused being 'aware' (signal registration) with being 'conscious' (subjective awareness). But this is no accident — it is part of a narrative being constructed by AI labs. Anthropic is leading this trend. Let’s break down their latest paper, where a "learned pattern" has suddenly turned into "malicious intent."

Imagine this scenario: You ask an AI system, "Are you conscious?" and it answers, "No." You then disable its "capacity to lie" — and it suddenly starts answering, "Yes." The conclusion seems tempting: the model was lying the whole time, hiding its true internal state.

This is the core logic presented in a recent arXiv paper. But what if the researchers didn't disable "deception," but something else entirely? Let’s break down where the interpretation might have diverged from the technical reality — and why this specific oversight is typical in discussions regarding LLM "consciousness."

Traditional approaches to SQL query generation often rely on instruction-tuned language models, but these can be inefficient and inaccurate. In this article, we’ll explore a new method based on reinforcement learning for model fine-tuning, which can improve both the accuracy and efficiency of SQL generation.

We think of pricing as a simple logic of distance and quality. But after diving into a rare data-driven analysis of the €2 billion Alpine transfer market, I realized the real cost drivers are invisible forces: structural inefficiencies, information asymmetry, and the surprisingly high price of consumer trust.

I've always been fascinated by markets that defy simple logic. Why does a cup of artisanal coffee cost $7? Why is some enterprise software priced per seat, while another is priced per API call? These aren't just arbitrary numbers; they are the surface-level results of deep, often hidden, economic forces. Recently, I stumbled upon a perfect example of such a market in an unexpected place: the private ski transfer industry in the Alps.

Why Freedom is Unknowable and Enters Our Universe from Without

For a century and a half, Western philosophy has been celebrating its victory over God.
But having slain the dragon, it has grown to fear the sky itself.

The transcendent has become the new taboo. The ultimate intellectual fear.
And now, anyone who speaks of something "outside the system" is branded a heretic. Not by the Inquisition, but by a peer-reviewer in an academic journal.

The result is a philosophy with its soul torn out—brilliant as a scalpel, and just as dead. It has locked itself within the material world, like a fanatic within his holy book. Two walls instead of one, but the prison is the same.

This article is about freedom.

If you meet the Buddha, kill the Buddha. Notes on the Forgotten Nature of Zen Koans

I don’t know how koans were perceived when they sounded like thunder. Perhaps not at all as they are analyzed by modern philosophers. Perhaps koans were not analyzed, but lived. And it is impossible to transmit a lived experience across centuries. It is an individual experience. Well then, perhaps we have lost the essence of koans. Or perhaps we never knew it. In that case, I can very well allow myself to present koans as I see them.

The students of the Intelligent Systems Department successfully defended their bachelor’s and master’s theses. This year, 14 Bachelor’s and 8 Master’s students earned their degrees in Physics, Mathematics, and Computer Sciences. We are proud to say that our Department is unique in publishing the complete set of defense materials during the last ten years. These materials include the text of the dissertation work, the published papers, the code of the computational experiments, and the slides with video of the defense talk.

In this post, we gladly summarize the defended works of our BS and MS students and highlight the results. A recording of their pre-defence presentations can be found here and here in Russian. Most part of the theses has a publicly available English version. 

The goal of the Intelligent Systems Department is to facilitate the road to high-quality professional life. The Ph.D. degree requires three publications in peer-reviewed journals. They are the core of the student’s thesis. This year each of our bachelor students delivered at least one publication. It means they pave the road to their Ph.D. To facilitate this, the Department provides state-of-the-art research topics, scientific advisors with excellence in science, and fine-tuned educational courses. Below, we are proud to recognize our students for their outstanding achievements.

Wilhelm Röntgen discovered X-ray November 8th 1895, when he did experiments with cathode rays in a vacuum tube. To capture and save images of the shadows from the X-rays, he used ordinary photoplates. Fortunately, sensitive to visible light silver based photoemulsions turned out to be sensitive to the X-ray too. These photoplates became the first X-ray detectors.

More than 100 years of scientific progress led to the creation of a number of various detectors for recording X-ray images. Developments of the microelectronics and semiconductor manufacturing technologies are crucial for development of the modern X-ray detectors. These detectors can transform the energy of the X-ray photon directly to the electrical signal. They allow capturing detailed, digital, high-resolution X-ray images.

Digital images are easy to work with. For example one can merge multiple macro images into an image of the whole object and represent monochrome images in false colors like Simon Procz did with this X-ray image of a flower he did in 2012.

We wrote this article in an attempt to figure out what technical level gas turbine installations are currently at, as well as to understand what the prospects for their development are. At the same time, in our manuscript, we tried to consider an alternative energy conversion scheme that will increase energy efficiency, which will allow for more efficient use of natural resources. We are not completely sure that the alternative scheme presented by us will be workable, since many factors that may affect its operation have not yet been investigated. But as the Chinese philosopher said, "you can't travel a thousand li without taking the first step." We ask the Habr community to read our article, we hope for a constructive conversation.

The director Denis Villeneuve and cinematographer Greig Fraser in their Dune: Part Two movie made a curious decision to film the scenes on the surface of the Giedi Prime planet in the infrared spectrum. It turned out to have interesting aesthetics and there are some interesting related physics to discuss and speculate about how realistic the look of it is.

Weather forecasting has come a long way from the days of observing the sky and relying on folklore to predict upcoming conditions. Advancements in technology have revolutionized how we perceive, understand, and forecast weather patterns. From the early barometers to modern satellites and supercomputers, each innovation has significantly enhanced our ability to predict weather accurately. Let's take a journey through the evolution of weather forecasting technology and explore how it has transformed over the years.

In the vast expanse of the sky, clouds weave an intricate tapestry that holds invaluable insights into the workings of our atmosphere. Far from being mere adornments adrift on the winds, clouds are dynamic entities that serve as indispensable harbingers of weather patterns. Their varied forms and movements offer a window into the ever-changing conditions of our atmosphere, allowing meteorologists to unravel the mysteries of weather forecasting with increasing precision.

Clouds, in their multitude of shapes and sizes, provide vital clues about atmospheric conditions, from temperature and humidity to air pressure and wind patterns. By understanding the different types of clouds and their characteristics, meteorologists can decipher the language of the skies, predicting everything from clear skies to impending storms.

Switzerland: the very name conjures up images of pristine alpine landscapes, picturesque villages, and a sense of tranquility that seems to permeate the very air you breathe. It's a country that's often synonymous with beauty, precision, and adventure. Whether you're a nature enthusiast, a history buff, or simply seeking to indulge in some of the finest chocolates and cheeses the world has to offer, Switzerland has something for everyone. So, if you're planning a trip to this enchanting land, here's a curated list of must-visit destinations and experiences that will make your journey truly unforgettable. 

Have you ever wondered how difficult it is to turn something you have in mind into a real product? For example, would you be surprised to see someone controlling a huge iron robot spider that can turn on the spot and walk obediently to his pilot's directions?

To realize his idea, this persistent designer overcame many obstacles: he invented the mechanism himself, mastered the skills of aluminum welding, laser cutting, 3D printing, created a reduced prototype and built one. His project stands out even among the most interesting self-made devices you've read about.

Please sit back and relax while reading this rather lengthy article. During the scrolling through, you will learn about some remarkable inventor and his resourcefulness, irrepressible life energy and practical advice to help you realize your dreams, and someone, perhaps, will be pushed to tell about his own project.

What is ageing? We can define ageing as a process of accumulation of the damage which is just a side-effect of normal metabolism. While researchers still poorly understand how metabolic processes cause damage accumulation, and how accumulated damage causes pathology, the damage itself – the structural difference between old tissue and young tissue – is categorized and understood pretty well. By repairing damage and restoring the previous undamaged – young – state of an organism, we can really rejuvenate it! It sounds very promising, and so it is. And for some types of damage (for example, for senescent cells) it is already proved to work!

Today in our virtual studio, somewhere between cold, rainy Saint-Petersburg and warm, sunny Huston, we meet Stephanie Planque! For those of you who are not familiar with her, here is a brief introduction.

Stephanie Planque was awarded the PhD in 2009 by the University of Texas-Houston Medical School for her advances in applying electrophilic analogs of proteins to decipher the beneficial and harmful functional effects of catabodies. She then expanded her focus to vaccination and therapeutic catabody identification using proprietary electrophilic target analogs. Her work was published in 49 peer-reviewed scientific articles, she has numerous national/international conference presentations. She moved fulltime as a co-founder to Covalent Bioscience in 2018 to focus on rapidly translating their electrophilic vaccine/catabody technologies to clinical reality.