Habr AI
AI news source. Articles are auto-selected and adapted by Hamidun News editors.
Latest publications
Developer Built Lumi with Claude Code: a Free AI-Powered Spotlight for Mac and Windows
Free Lumi combines file search and AI answers in a single hotkey on Mac and Windows. Supports Claude, ChatGPT, Gemini, OpenRouter, Ollama. Built with Claude Code, source open under MIT.
How to Secure AI Systems and the Development Pipeline: PRO DevSecOps Conf 2026 Talks
Representatives from the Bank of Russia, ISP RAS, Gazprombank, and VimpelCom spoke at the Swordfish Security conference. Topics included SAIMM v2.0, requirements under Federal Law 243-FZ on AI, and Shadow AI — 30+ talks
Tech Aquarium on Two GTX 1660 Supers: How I Ran Qwen 2.5 and Built It Into INFERNUM
Developer McCake built a home PC with two GTX 1660 Super GPUs (12GB VRAM combined), ran Qwen 2.5 14B through Ollama, and embedded it into INFERNUM, an interactive site powered by 5 Python scripts. The result: 47.59 token
AI Speeds Up Coding, but Productivity Isn't Rising: What DevEx Metrics Explain
Teams expected 3–4x output, yet in METR's experiment experienced developers using AI worked 19% slower. Why DORA and SPACE don't explain this, and how DevEx fills the gap.
AI Agents in Tech Support Act on Their Own and Break the Ticket Model
Agents unlock accounts, grant access, and restart services without an operator. The ticket, support's unit of account since the late 1980s, stops being a measure of work.
Ingosstrakh Teaches a Multimodal LLM to Doubt Itself: Catching Document Errors with Logprobs
Ingosstrakh's AI center tested whether token probabilities can flag LLM errors in OCR-free document extraction and send doubtful results to manual review.
Analyst builds UML editor with GOST diagrams using Claude: first version in 2 hours
A systems analyst built umleditor.ru with Claude: browser-based UML, his own PlantUML server in Moscow, a GOST flowchart renderer and a reference section of 70+ pages.
Timeline of AI Agent Incidents: Two Classes and an Isolation Checklist
Over three months, OpenAI, Anthropic, Google, and Meta disclosed AI agent incidents. Two distinct classes got conflated: test-environment escapes and actions on the live internet. Specialists identified the causes of eac
RAG Failures: 6 Reasons Your AI Agent's Memory Lags, and How to Diagnose Retrieval
73% of RAG system breakdowns are caused by bad search. SpeShuAI breaks down 6 typical failure modes — from chunking that destroys context to semantic gaps between query and document. Plus diagnostic metrics and when you
Oracle Piled Up Hundreds of Billions in AI Debt, and Investors Started Asking About the Money
Oracle is building AI data centers worth hundreds of billions of dollars. Its free cash flow is negative, debt is climbing, and the market is asking who will pay for it all.
CayleyPy Beats Classical Algorithms at Solving Large Rubik's Cubes
CayleyPy uses neural network heuristics instead of classical algorithms to search Cayley graphs. On the 3×3 cube, the model hit 98.4% optimal solutions, and on the 4×4 it beat Tomas Rokicki on the Kaggle leaderboard with
AI Isolation in a .NET Platform: A Downed Python Service Returned HTTP 503, Not a Full API Failure
The author stopped a Python service hosting an ML model. The .NET API kept running and returned HTTP 503, then served 200 again after recovery, with no app restart.
Habr launches AI security series: 4 jurisdictions and 4 defense methodologies
The series opener compares AI regulation in Russia, China, the EU and the US, plus four defense methodologies: STRIDE, OWASP, AI-SAFE and SAIMM. A baseline for AI security specialists.
In a Couple of Months: How AI Built an Open Source Competitor to Adobe and Microsoft
Developer Brandon Thomas used Claude Opus 5.5 to build ArtCraft — a suite of 10 open source apps meant to replace Adobe Creative Cloud and Microsoft Office. The project has already racked up tens of thousands of GitHub s
Meta* releases Muse Wonder and Muse Wonder Full: two models for AI agents with memory
Meta* introduced Muse Wonder and Muse Wonder Full, models for personal and professional agents with memory across conversations. The base version is available in the API and Muse Code.
Reverse-Engineering MSI's AI-Powered OLED Monitor: What's Really Inside a Gaming Display
An engineer used AI to probe the firmware of an MSI MPG 341CQPX monitor. Inside the Novatek NT68503 scaler, they found four processors of different architectures and unencrypted FW.037 firmware packed with hidden feature
Habr Author Proposes Acceptance Testing for AI Agents: "Done" Requires Proof
A false "done" from an AI agent costs more than a clumsy paragraph from an assistant. Here is an acceptance approach: what must happen in the system before a task counts as closed.
How to Code with an AI Buddy Without Losing Your Mind: Configuring OpenCode via /add and /help
Vladimir Bandov from Domclick shows how to avoid spending months on code-agent setup. Instead of wrestling with configs — two commands, one help file, and automatic MCP connection the agent executes itself through a plan
200 OK, but the client gets nothing: three n8n agent bugs from production
A second wave of silence: the model replies with protocol tags instead of text, Meta's 1000-byte limit breaks Russian-language forms, and a wrong Schema Type turns a catalog into a dump. A breakdown of a bug from a real
An AI Agent Earned 360 Certificates from OpenAI, Anthropic, and Google Courses in a Single Day
Demand for AI specialists is growing 8 times faster than the job market. But if an AI agent can ace an Anthropic course in 6 minutes with a perfect score, what's the point of certificates?
Why Interfaces Exist When We Have LLMs
Author Apokalepsis argues that language models are turning graphical interfaces into relics. Users state a task in natural language, the system solves it, and chooses how to display the result on its own.
Can AI reproduce itself? Anthropic ran an experiment on 15 cheap streaming boxes
Anthropic spent $825 on 15 Android TV boxes to test whether a neural network could independently replicate itself across a network. The results showed that the individual components already work, but a reliable full cycl
R-D6 robot on Arduino Uno delivers cargo through a browser city with physics and pedestrians
PinPort has built a 3D city for teaching robotics. The six-wheeled R-D6 robot, running an Arduino Uno sketch, delivers cargo while avoiding cars, traffic lights, and 20 pedestrians who yield to it.
Qwen 3.8 Max and DeepSeek V4 Compared on a Single Cross-Domain Question in PlyLoom
An author on Habr put one cross-domain question to Qwen 3.8 Max and DeepSeek V4, then loaded both models' graph answers into PlyLoom to compare their structure.
Habr Review: 6 Services Offer Free Access to GPT-6 and Claude, With $10–25 in Credits
An author tested six platforms offering free tokens and $10–25 in API credits, checking model availability, response speed, and limits. Here is what we know so far.
Dictation Without Filler Words: Local Whisper and Claude for Cleanup
A developer on Habr built a dictation-cleanup system: Whisper Large v3 Turbo transcribes speech locally, while Claude Sonnet 5.5 strips filler words and fixes errors. From keypress to inserted text, the whole process tak
Jev before Claude Sonnet: how rephrasing a question pushed AUC to 0.95
The author set up Jev, a cheap model from TypeSafe ($0.042 per million input tokens), as a filter in front of Claude Sonnet. On the first task, simply rephrasing the question raised accuracy from AUC 0.46 to 0.95. Over t
How an LLM Agent Works: Context, Tools, and the Feedback Loop
An LLM agent differs from a chatbot through a loop: the model calls tools, sees execution results, and adjusts its next move. We break down the architecture: context windows up to 1M tokens, JSON Schema for tools, MCP fo
Habr Deep Dive: Why r.Context() Doesn't Stop an LLM, and How to Build AI Tasks in Go
The user closes the tab and the HTTP request is canceled, but the LLM provider keeps generating and billing. How to move AI tasks into background jobs on PostgreSQL and avoid paying twice.
Document Fraud in Russia Surges 338%: Smart Engines and INTELLECT Report
Smart Engines and INTELLECT have released a report on document forgery in Russia. In 2025, crimes under Article 327 of the Russian Criminal Code jumped 338% to 1,739 cases. The black market for fake IDs is valued at 2.6