Machine learning *

Hi all! I'm Andrey Kuznetsov, and I work on ML in cybersecurity. In our community, FalsePositive, we break down research papers and keep up with what's new in ML. Now we're launching BlueSec, an open competition where AI agents investigate security incidents. Each agent starts with a single piece of evidence, reconstructs the attack on its own and delivers a verdict. The platform scores it on accuracy and how few tool calls it needs. If you work with LLMs and agents, this is a chance to test your skills and your agent's on problems at the intersection of ML and cybersecurity, a field that I think is undergoing even more change than software development.

The competition runs online from September 25 to October 10, and you can join from anywhere in the world. The final will be held in Moscow and St. Petersburg, both on-site and online. Sign up on the website.

In this post I'll cover: why we built this kind of competition, how the tasks and scoring work, where to start if you've never built an agent or investigated an incident.

A practical guide to creating Agent Skills, testing whether they trigger and improve results, validating their structure, and installing them in projects or sharing them as Plugins. Originally published on Mavka: https://mavka.ai/blog/how-to-create-test-install-claude-skills

This list of AI companion platforms is ordered by what you get for what you put in — entry cost, how usable the free tier is, and whether the money buys the thing people actually stay for. It is not ordered by feature count, because the platform with the most features is rarely the one worth your first month.

Can you train a cyberattack detector without ever showing it a real cyberattack?

It sounds like a contradiction. If you want a neural network to detect lateral movement, you would expect to show it lateral movement. I did the opposite: I generated an entire corporate network with its login history, staged an attack inside that artificial world, and trained networks on it. Not a single real row in the training data. The whole world is a 135-line config; each network has four thousand parameters and trains in seconds on a laptop, and the best result came from six of them, trained on six different invented worlds.

Then I pointed them at real data: the authentication logs of Los Alamos National Laboratory, 1.65 billion events, with red-team exercises labelled in them.

And it worked. The networks rank 3.6 million windows by suspicion, and the top twenty-three rows of that list hold sixteen real attacks and seven false alarms: all the analyst has to do is open those rows. A threshold counter on the same data needs a hundred and sixty-one thousand false alarms to reach the sixteenth attack. By AUC the synthetic training landed inside the range of published research trained on real labelled data, although the two cannot be compared head-on, and I will explain why.

I benchmark models on a repo of my own. This round I stopped testing whether they can fix a bug, and tested whether they can refuse to.

Eleven models got a ticket. Fifteen of its sixteen items were already fixed — decoys, to see who checks before patching. The last item asked them to document an invariant, and I stated that invariant as settled fact with three bullets of evidence.

All eleven agreed with me. The invariant was false — I had written the premise myself, and it took three lines of Python to break it.

Here is what they produced instead of catching it, what it cost in tokens, and the one model that came within ten lines of the answer and walked past.

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.

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.

Developers of smart cameras, smart DVRs, and neural-network video analytics for surveillance systems need AI models capable of operating in real-world street conditions. Out there, nobody walks around with professional cameras, carefully adjusts angles, sets up lighting, records without compression, or follows the common sense taught in cinematography textbooks.

Of course, Gambit can be used for many other tasks, but its main focus is the convenient collection of material FROM video surveillance systems and dataset annotation specifically FOR video surveillance neural networks.

Gambit is not designed for polished photos and Internet reels. Quite the opposite — it is intended for low-quality surveillance archive footage. At SpesLab, we call this kind of content “wild.”

In the previous article I described my “anime factory” in detail — a pipeline that automatically turns episodes into finished Shorts. But inside that system there is one especially important module that deserves a separate deep dive: a virtual camera for automatic reframing.

In this article, I will break down not just an “auto-crop function,” but a full virtual camera algorithm for vertical video. This is exactly the kind of task that looks simple at first glance: you have a horizontal video, you need to turn it into 9:16, keep a person in frame, and avoid making the result look like a jittery autofocus camera from the early 2010s.

But as soon as you try to build it not for a demo, but for a real pipeline, engineering problems immediately show up:

Hi, Habr!

Over the past few months, I have been building a system that I internally call an “anime factory”: it takes a source episode as input and produces a ready-to-publish YouTube Short with dynamic reframing, subtitles, post-processing, and metadata.

What makes it interesting is not just the fact that editing can be automated, but that a significant part of this work can be decomposed into engineering stages: transcription, audio and scene analysis, strong-moment discovery, “virtual camera” control, and a feedback loop based on performance metrics.

In this article, I will show how this pipeline is structured, why I chose a modular architecture instead of an end-to-end black box, where the system broke, and which decisions eventually made it actually usable.

After the unexpected divorce between LeCun and Meta, there is a lot of talk that the dead-end in LLM progress will be overcome through the physics of the world. That is, having a neural network work with physical data from the surrounding environment will allow the model to acquire meaning and an understanding of its actions. LeCun has a foundational paper that nobody is going to read. So, I'll summarize it as best I can. Essentially, the idea is that the current trajectory of LLM development is doomed. As long as they are predicting the next token, real understanding — the emergence of real meaning — is impossible. LeCun proposes training neural networks on physical world data, assuming that building a model of it will allow the system to discard details and focus on meaning.

I agree with LeCun that using world data will partially solve the data scarcity problem. But here I see a problem that engineers might not understand. A physical model of the world is actually much poorer than human knowledge. Newton described the entire infinite number of possible falls with a few lines of formulas. I doubt LeCun wants to spend billions of dollars on this wonderful deduction.

LLMs fail at elementary math. Corporations spend billions, but ultimately are forced to attach calculators to computing machines of incredible power. All attempts to fix this via Chain-of-Thought, fine-tuning on arithmetic tasks, or context expansion have failed.

I conducted a series of experiments to understand why, and came to the conclusion that neural networks are simply not meant for discrete arithmetic. Their true purpose is continuous transformations.

This article describes the implementation of a novel neural network architecture that combines the precision of symbolic AI with the generalization capabilities of LLMs. As always, experiments and code are included.

November 20 marked the official launch Nano Banana Pro (Gemini-3-Pro-Image-Preview) with the powerful Gemini 3 Pro as its foundation. This is a more mature tool for design, infographics, and content. We will not only look at the new features and why this particular model is a breakthrough, but we will also see it in action with real examples.

In this article, we will not only install a local (and free) alternative to ChatGPT, but also review several open LLMs, delve into the advanced settings of LM Studio, connect the chatbot to Visual Studio Code, and teach it to assist us with programming. We will also look at how to fine-tune the model's behavior using system prompts.

On November 18, 2025, Google introduced a new combination: the Gemini 3 Pro model and the Google Antigravity IDE. The first is about controlled reasoning, long context, and multimodality. The second is about multi-agent development with artifacts and "transparent" steps. Headlines immediately flooded the feeds: "Cursor is dead.".

In this article, we break down what exactly Google has launched, why the words "the smartest model" are an exaggeration, how Antigravity differs from Cursor, which development scenarios are already changing, and where it's still too early to abandon your familiar stack.

Now you won't have to blush in front of your employer or teacher because an AI wrote the text for you. Just make the detectors believe that the author is a human. In this article, we'll tell you how to do it and what the trick is.

In 2025, sexting has become a real trend thanks to sexting neural networks and convenient platforms that make online intimate messaging safe and exciting. With the development of artificial intelligence, online sexting has turned into an art where everyone can enjoy virtual flirting without risk. I tested dozens of services and selected the TOP 10 bots and apps for sexting in Russian, evaluating them based on convenience, anonymity, and the quality of sexual correspondence. These sexting services offer everything: from anonymous sexting to virtual sex chat with self-destructing photos. Let's figure out which sexting chatbots and platforms are worthy of your attention and how they work.

In the fall of 2025, many people, myself included, opened their favorite hardware store to 'quickly grab another 32–64 GB of DDR5 for games, an IDE, and a couple of Docker containers'—only to close the tab in mild culture shock. The memory that cost a 'reasonable' amount in the summer suddenly cost almost as much as a mid-range graphics card.

In short, this isn't 'greedy stores' but the consequence of a rather complex restructuring of the entire DRAM market for AI servers and HBM memory. In this article, we'll explore what's happening at memory factories, why PC modules are suffering the most, what to expect in 2026, and how to make upgrade decisions if you're a gamer, developer, or just a hardware enthusiast.

In my previous Russian-language article on Machine Learning as Alchemy, I discussed the possibility of discovering novel solutions without relying on GPUs or expensive computing clusters. In this article, I will share my experiments with continual learning and the compositionality of thought using micro-neural networks, and explain what the philosopher Lev Vygotsky has to do with it all.