Artificial Intelligence

Recipes for building a truly autonomous neural network surface in the information space on a regular basis. More often than not, the methods and approaches coincide regardless of the author. And although they look original, most of them have in fact been formalized with the help of LLMs — and LLMs, as we know, work very well with existing knowledge.

In this article I will analyze the main techniques being proposed and the faulty foundation they rest upon.

Literature review, comparison, and analysis of existing solutions constitute a significant part of a thesis manuscript.  In the last three years, AI chats have dramatically intervened in the thesis preparation workflow. The ability of AI chats to analyze and compare discourages students from acting as the primary drivers of research and instead encourages their passive reliance on AI-generated content. This situation makes manuscripts eclectic and lacks focus. The theoretical part disconnects from the computational experiments. As a result, the audience is reluctant to read students' manuscripts. Ironically, they use AI chats to summarise the manuscripts to extract their main contributions. We are proud to say that our thesis works do not suffer from this AI distortion. We are grateful to our students, especially our bachelor’s and master’s alumni, who have carefully prepared and successfully defended their theses, producing clear, well-founded, and meaningful results. Our thesis manuscripts are designed for public discussion and rigorous scientific analysis. Each thesis presents a clear research idea grounded in the fundamentals of machine learning. It includes a theoretical justification, a repository with the source code of the computational experiments, and slides. Videos of the pre-defence sessions are uploaded to our YouTube channel “Intelligent Systems”: Bachelors & Masters.

We build a Tiny Language Model from scratch in pure Node.js without TensorFlow or PyTorch, implementing neurons, autograd, embeddings, self-attention, FFN, backpropagation, and SFT while observing how individual weights and entire matrices change during training.

In previous articles (1, 2), I analyzed Anthropic's J-space: why their interpretation falls short; why J-space analogues appear in micromodels under objective pressure rather than as an emergent property of scale; and why the vector they discovered is strictly equivalent to Vygotsky’s concept of the sign.

Two questions remained outside the scope: is a metasign possible—a sign integrating several basic ones—and what is the true nature of the neural network's self-description (the report)?

This article addresses both.

I recently published an app on Google Play where 99.9% of the source code was written by AI. It’s not just some "weekend hack job": a native app (Kotlin), over 50 screens, complex animations, multilingual support, an extensible architecture.

I achieved this through a highly detailed specification. Every screen was prototyped by a designer, for every button there was a description of which APIs to call. Was it worth it? Let’s look at the numbers.

Recently, Anthropic reported the discovery of an analogue to the global workspace in LLMs — J-space. In a previous article, I analyzed why Lev Vygotsky’s theory offers a better interpretation of the researchers' findings.

In this article, I argue, based on experiments, that J-space analogues do not depend on the size of the neural network. Rather, they are an inevitable stage in the evolution of a cognitive structure developing under pressure.

On July 6, 2026, researchers at Anthropic published the results of a new study claiming to have discovered an analogue of the "global workspace" in Claude, dubbed J-space.

The research is highly fascinating and, I believe, reveals the future of neural networks much more deeply than it appears at first glance.

In this article, I will explore why the Global Workspace Theory (GWT) accurately describes the structure of this finding, yet remains silent on its main oddity: the fact that this workspace consists of "words" (or, more precisely, discrete signs). And I will explain why Lev Vygotsky answered this exact question a century ago.

Neural Networks for the Little Ones

Every time you say “Thank you” to a neural network, you launch a pipeline that multiplies hundreds of matrices with billions of elements, and burn as much electricity as an LED lamp in a few seconds.

This is the first article in a short series dedicated to networks for AI/ML clusters and HPC.

In this series, we’ll touch on the principles of model operation and training, parallelization, DMA and RDMA technologies, network topologies, InfiniBand and RoCE, and we’ll also philosophize on the topic of general and specialized solutions.

In this particular article, we’ll figure out what a neural network is, how it works, how it’s trained, and most importantly, why it needs hundreds of expensive GPU cards and some kind of special network.

The refrain of today’s story: there’s no magic in neural networks—it’s just a multitude of simple operations on numbers, performed on computers with special chips. There’s no magic in how they work, nor in the infrastructure they run on.

Yesterday (November 27), Habr hosted an 'Author's Fireside Chat'.

It was very interesting, and one of the speaker's statements struck me. It was that AI can help write simple pieces of code but doesn't work with complex things. Thus, large language models are likened to a junior programmer.

I decided to write an article about it this morning, drawing on my knowledge and experience in computational mathematics (I used to do modeling in the past, and for the last few years, I've been teaching computational mathematics at MIPT). Let me know what you think.

I think this is the main myth of vibe-coding. It's exactly the opposite — AI is often good at writing quite complex things and retrieving important information that is difficult to find on your own. But it gets confused in the most elementary things. It's a reverse junior.

The problem is that this is a dangerous illusion, and I will now clearly explain why, and how it can be dangerous. Brew some coffee and get ready for a debunking that might save your millions, your career, or even human lives in the future.

Hello, Habr! My name is Nikita Pastukhov—author of FastStream, Principal Engineer, and maintainer of AG2 (a framework for developing agents). I’ve been in development for 8 years, and for the last year, I’ve been up to my ears in agents.

And I want to prove to you that writing your own agent is no more difficult than writing a CRUD

Why does this even need proving? Because there’s a noticeable gap between what’s happening with AI globally and what’s happening in the average Russian company. Globally—every company has an OpenAI subscription, there are a billion startups with AI products, and agents are deeply integrated into the back office. In Russia—it’s “dangerous, we host our own models,” “it’s unclear,” and support chatbots. Globally, engineers already know how to develop agents. In Russia—it’s “what even is that?”

So let’s break down how agents work using the example of OpenClaw—the most hyped “personal AI agent” right now. It lives in your messenger, sorts your email, manages your social media, writes code, and deploys services. Its popularity is a testament to how little people are currently using agents in their daily lives. For those in the know, OpenClaw hasn’t brought anything new to the table.

Hey devs! When you use Codex, putting too many tasks into a single conversation makes the model slow, confused, and unreliable. To solve this problem, I highly recommend using subagents (or Custom Agents).

Operating autonomous vehicles in mixed traffic with human-driven vehicles imposes two fundamental limitations. First, sharing roads with conventional vehicles causes delays through traffic congestion. Second, unpredictable behaviour of human drivers and pedestrians, together with current traffic regulations, prevents autonomous vehicles from realising the full benefits of automated vehicle-to-vehicle coordination – optimal distance control, prioritised merging, dynamic road load management, and automatic compliance with radio-transmitted vehicle-to-infrastructure (V2I) commands.

To overcome these limitations, physically segregated road sections or physically segregated lanes must be introduced, reserved exclusively for highly automated vehicles operating in fully automated mode without a driver on board. These segregated sections and lanes require special traffic management rules, including dedicated road signs, road markings, entry and exit procedures, and V2I commands.

Physically segregated lanes and road sections would allow the safe exploitation of all capabilities of fully automated mode, including centralised control via V2I commands as technical readiness permits. To implement such segregation, new road signs are needed that clearly mark the boundaries of these sections and inform all road users of the applicable rules.

How a Product Owner at Exante uses AI to cut complex feature delivery time in half — without replacing human input in product decisions.

I ran into a very practical problem after doing a lot of local work with AI agents. I had Codex projects, Claude Code projects, regular repositories edited with agents, drafts, pipelines, instructions, skills, artifacts, and several machines. At some point it became hard to tell where the current version of a project lived, which files were safe to push, where agent instructions belonged, and where source code had already been mixed with logs and intermediate output.

That is why I built AI Workspace System: a small set of shell scripts, conventions, and Markdown documentation that makes local AI-agent work predictable. It is not an IDE and not an agent orchestrator. It is a thin infrastructure layer around Git, GitHub, Codex, and Claude Code.

The core idea is simple: all projects should be visible from one list, instructions should follow one structure, sync should be safe by default, and machine-specific details should not live in the repository.

When you buy a $20 ChatGPT subscription, you get access to about $1,000 worth of tokens. When you buy a $100 Claude subscription, you get access to about $2,000 worth of tokens. However, these subscriptions cannot be connected to Cursor directly. The API and subscription request formats are different, so you need a workaround — proxying the requests.

AI coding assistants are becoming better every day. But most still work one prompt at a time. You ask something, get an answer, and then guide the next step manually.

Renowned biologist Richard Dawkins recently published an essay exploring the possibility of LLM consciousness following a two‑day conversation with Claude AI.

Let“s first look at why an essay by this particular author caused such a stir in scientific circles, while thousands of ordinary users fail to turn heads when they claim their AI companions are sentient. The latter constantly post endless walls of text from their chats with LLMs, where the density of words like ‘consciousness,’ ‘soul,’ ‘reflection,’ ‘recursion,’ ‘emptiness,’ ‘warmth,’ ‘love,’ and ‘pain’ exceeds all reasonable limits. It is worth noting that the semantic density of these dialogues is practically zero‑but we will return to that later.

In early 2026, one new law and one far‑reaching legislative initiative are expected to seriously affect digital freedoms in the EU. The first allows police to collect biometric data and target individuals; the second aims to put all metadata into one box and then use AI to run investigations. Naturally, both laws were adopted under the mantra of protecting democratic values, rights, and freedoms. Xeovo has examined the sprawling regulatory texts and explains what exactly Members of the European Parliament are aiming at.

Like millions of others convinced they possess knowledge the world desperately needs to hear, I decided to write a book on prompting. In the process (which, by the way, turned out to be far more difficult than anticipated), I found myself examining LLM clichés. You know the ones. At least, in the comment sections of tech blogs, hundreds of self-proclaimed experts use them to spot AI-generated text.

Anyway, these clichés definitely exist, and many authors now routinely add blocklists of these phrases to their prompts to weed them out. Whether this is actually a good or a bad thing is what I’ll break down below.

Google recently released the Agentic CLI, a powerful tool that makes building, testing, and deploying AI agents faster and more intuitive. I think it’s super useful, so I’ll walk you through the entire lifecycle of an AI agent using the CLI.