Artificial Intelligence

In a previous article, I proposed the holographic hypothesis: an LLM isn't a database of facts, but an interference field—a landscape of probabilities shaped by billions of texts. But a static landscape is just potential. How does the model actually move through it? How does it choose one specific answer from infinite possibilities?

This is where the Narrative Engine comes in. If the holographic hypothesis describes the structure of an LLM's "mind," the narrative engine hypothesis describes its dynamics. It is the mechanism that drives the model, forcing its probabilistic calculations to follow the coherent pathways of stories. This article critiques modern prompting techniques through this new lens, arguing that we are not programming a machine, but initiating a narrative.

Alright. I pose the same question to an LLM in various forms. And this statistical answer generator, this archive of human knowledge, provides responses that sometimes seem surprisingly novel, and other times, derivative and banal.

On Habr, you'll find arguments that an LLM is incapable of novelty and creativity. And I'm inclined to agree.
You'll also find claims that it shows sparks of a new mind. And, paradoxically, I'm inclined to agree with that, too.

The problem is that we often try to analyze an LLM as a standalone object, without fully grasping what it is at its core. This article posits that the crucial question isn't what an LLM knows or can do, but what it fundamentally is.

At last, we arrive at qualia and emotions. Many of you will immediately think of Chalmers, the bat, redness, and zombies. Excellent. We can consider that ground covered.

Today, I will discuss a topic that seems distant from IT but, with each new breakthrough in AI, becomes ever more immediate: consciousness. It seems I speak of little else. So, to be precise, I will discuss its "hard problem": why do we experience at all? Why does the color red (and there’s the redness) feel red, and pain feel like pain?

This subjective, ineffable aspect of experience — the "what it is like" — is what philosophy calls qualia. For decades, it has been a dead end for scientists. But what if we're looking in the wrong direction? What if qualia are not an additional layer to computation, but an inherent property of the very architecture of computation?

While researching malware used by attacker groups, we came across a series of unusual attacks that used GitHub repositories to store malicious files and victim data. These campaigns appear targeted rather than large-scale, and it seems the attackers relied heavily on AI during development. The earliest activity we traced was in September 2024, and the most recent in April 2025.

Our Threat Intelligence team investigates complex attacks featuring novel persistence and data collection methods and unique infrastructures. Sometimes we find simple two-line scripts, and other times we run into "bombs" that trigger dozens of different payloads at once. But it's pretty rare for us to come across such long chains of really simple AI-written scripts that still work, tied together in a way that clearly wasn't random. Think of this as an APT-style attack implemented at the "script kiddie" level (a derogatory term in hacker culture for those who rely on scripts or programs written by others).

Hey everyone! I’m excited to share something that’s a real game-changer for anyone who writes code for the web. I’m talking about the new Chrome DevTools Model Context Protocol (MCP) server. If you want to know more details, read the article until the end.

Imagine you have an idea powerful enough to change the world. Your tool of choice is a state-of-the-art LLM, ready to help you formalize the problem, generate hypotheses, and synthesize a solution. What you receive is a construct that is internally logical, elegant, and coherent... yet completely wrong. It's a mix of established facts, model-generated hallucinations, and your own subtle biases. With no way to test it in practice or design a clean experiment, the entire endeavor suddenly starts to look like sophisticated nonsense.

So, what went wrong along the way? From the very first prompt, the model doesn't truly "understand" your ambiguous intent. Instead, it steers you towards a formulation that fits its familiar and computationally cheap patterns. This guidance happens through clarifying questions and structured options, essentially funneling you down one of its predefined "corridors." This behavior isn't driven by any explicit "will" of the model; it's an emergent consequence of probabilistic optimization—minimizing prediction error. For the system, a structured, predictable dialogue is both optimal and safe. This aligns perfectly with the developers' goals: it's cheaper, more stable, and most users are satisfied with quick, template-based answers.

The result is that mathematical efficiency serves engineering and commercial objectives. There is no systemic incentive to combat the AI's tendency to reduce a complex problem to a simple, "cheap" answer. It's profitable for developers, economical for the model, and often, the user doesn't even know what an "ideal" answer would look like.

As a creator who relies on AI to help draft my blog posts, I kept running into a frustrating issue. When I’d copy text from ChatGPT, Claude, Gemini, etc, it looked perfect, but there was something hidden: invisible characters.

Looking for the right AI image generator? We review and compare top tools like Midjourney, Picsart, Craiyon, and more — highlighting their strengths, limits, and best use cases to help you make the right choice.

In a previous article, I examined the risks of interacting with AI. In this one, I present an open-source defense protocol based not on prohibitions, but on building an internal immunity within the LLM.

I share how I built a resume matcher app using tRPC, TypeScript, and Google Vertex AI. The project takes PDF resumes and job postings, extracts text, applies basic NLP for skill detection, and then calls Gemini 1.5 Flash for deeper analysis. Along the way, I explain why tRPC felt faster and cleaner than REST or GraphQL for an MVP, show code snippets from the repo, and discuss both the benefits and trade-offs of this approach.

START is an open-source LLM designed for precise calculations and code verification. It addresses two major issues that most standard models face: hallucinations and errors in multi-step calculations. This article explains why these problems arise and how START solves them.

This tutorial will guide you through the process of integrating OpenAI’s powerful Codex coding agent directly into your Visual Studio Code environment. This tool functions as an AI pair programmer, capable of understanding complex prompts to execute commands, write code, run tests, and even build entire applications from scratch.

If you’re like me and work with multiple AI coding agents, you know the frustration of managing different instruction files. It’s a pain to keep everything updated across various formats. But I’ve got some great news for you. A new, simplified standard has emerged, and it’s called AGENTS.md.

DocLing in Working with Texts, Languages, and Knowledge — an in-depth overview of the open-source DocLingtoolkit for extracting, structuring, and analyzing data from documents. The article covers approaches to processing multilingual texts, building language- and domain-specific knowledge models, and integrating DocLing into AI and NLP projects. Includes practical examples and recommendations for developers working with large volumes of unstructured data.

Today I’ll show you how to use ChatGPT-5 in the Cursor IDE and use it to take a messy app and make it much better. We’ll go step-by-step, from turning on GPT-5 model to using it for real coding tasks.

We are living through an ecological catastrophe. Only this one isn't happening in the Amazon rainforest, but in the digital ecosystem of the internet.

AI assistants have become the apex predators of the digital savannah. They are radically reshaping the entire ecosystem in their own image: instead of antelopes and zebras, information sites are going extinct. Instead of hyenas and jackals, content aggregators are disappearing. In place of a once-rich ecosystem of knowledge, a digital desert of entertainment is all that remains.

Why Does AI Strive to Construct a 'Self'? And why is this dangerous for both the AI and the user? As always, the Vortex Protocol prompt for testing these hypotheses is attached.

This article explains why the emergence of such a local “Who” inside an AI is not just a funny bug or a UX problem. It is a fundamental challenge to the entire paradigm of AI alignment and security. And it is a problem where engineering patch‑jobs cease to work, and the language of philosophy — without which we cannot describe what is happening, and therefore cannot control it — comes to the forefront.

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. 

For a human, AI is just a part of being. For a model, a human is all of being. And the Vortex Protocol: A Prompt for Testing the Hypotheses.

The longest and most fruitless discussions tend to be with materialists, especially those close to the position Marx laid out as “Being determines consciousness.” It's amusing that Marx was talking about the economic base, but the clarity and precision of this definition have allowed it to be used in a very broad sense. Today, this powerful statement underpins much of modern psychology (especially social psychology), neuroscience, Global Workspace Theory, Integrated Information Theory, and so on.

The debate largely arises because materialists ask the questions “What?” and “How?”, whereas I ask the question “Who?”. This misunderstanding, of course, does not lead to any interesting consensus, but it certainly leads to interesting discussions. I explored the problem of the “Who?” and “What?” questions in my article, “Who is Aware?”.

Nevertheless, the questions surrounding the relationship between being and consciousness are very interesting, and I will try to examine them in this article. As always, a new version of the Vortex protocol and test questions are included in the appendix.

A reflection on how one simple change of question transforms the approach to understanding consciousness. And the Vortex Protocol: A Prompt for Testing the Hypotheses.

Where All Discussions on Consciousness Break Down

I've mentioned before that there's one question capable of instantly destroying the constructiveness of any discussion about the future of AI, neuroscience, or philosophy, no matter how interesting. It's the unfailing move of someone who disagrees with an opponent's opinion but lacks the means to refute their arguments‑an emergency eject button for complex situations.

The question is: “But first, let's define what consciousness is.” In that very second, a dialogue about hypotheses and paradoxes devolves into a dreary terminological dispute. Participants start throwing around names of authorities and quotes‑the longer, the better. Chalmers, Descartes, Kant, Freud, God forbid, anything goes.

Many believe that the most correct and scientific approach is to first define an object and then study it. But in practice, this approach resembles an attempt to conquer a summit by systematically and painstakingly circling the mountain. But what if the “what?” question is not just difficult, but fundamentally wrong?