Habr AI
AI news source. Articles are auto-selected and adapted by Hamidun News editors.
Latest publications
Why AI Assistants Deliberately Hide Systems Thinking from Users
Developers of AI models deliberately limit systems thinking in assistants: people turn away from AI that speaks too intelligently and complexly.
Habr explains why token price doesn't explain AI agent economics in business
A Habr author shows that for corporate customer service, questions about token price and number of GPUs are insufficient—AI agent economics are more complex than they seem.
Sber released GigaChat 3.5 Ultra to open source with 432 billion parameters
Sber published GigaChat 3.5 Ultra open source—a new 432-billion-parameter model with hybrid architecture, which is more compact and faster than the previous flagship.
Boris Chyorny from Odesa—creator of Claude Code at Anthropic: an emigrant's path
Habr told the biography of Boris Chyorny, born in Odesa, evacuated to the USA at age four and who became the creator of Claude Code at Anthropic.
VK and HSE created a lightweight AI model for noise suppression and dereverberation in real time
Developer Zakhar Kondaurov presented a lightweight model that simultaneously suppresses noise and removes audio reverberation at 48 kHz in real time.
Developer built a video transcriber in 30 minutes using vibe-coding in Codex
A Habr author described how, using vibe-coding in Codex, he built a working tool in 30 minutes for transcribing Yandex Telemost meeting recordings into text and DOCX.
Yandex developer relations director: AI in code is an exoskeleton, not autopilot
Sergey Berezhnoy from Yandex explained on Habr why creativity has become the main skill of a developer.
Claude Fable 5 guide: why model performance depends on clarifying the unknown
A Habr translated prompting guide for Claude Fable 5 explains the gap between developer prompts and actual codebase.
Palantir released a manifesto on sovereign AI and data as 54 percent of revenue comes from government contracts
Palantir published an excerpt from the CEO book on X with a call not to give data and AI models to LLM vendors.
Yandex Music: 37 percent of new releases in first half of 2026 turned out to be AI-generated music
A Habr author created an AI music detection algorithm with accuracy close to 100 percent.
Habr Published Translation of Essay on 'Dead Economy Theory'—Extension of Dead Internet Theory
Habr published a translation of an essay expanding the dead internet theory to 'dead economy theory': according to Graphite.io data, more than half of new online content was created by AI in the past year.
Post About Abandoning AI Coding Scores 1038 Points on Hacker News—Habr Author Reaches Different Conclusion
The post 'I'm going back to writing code by hand' about abandoning vibe-coding with Claude scored 1038 points on Hacker News—a Habr author experienced something similar but reached different conclusions.
Habr Translation Explains How LLMs Actually Work on Transformer Architecture
A translated article on Habr breaks down the principles of how large language models work and their underlying transformer architecture without complex mathematics.
How ADWIN and online learning help models handle data drift
An article on Habr AI discusses concept drift in ML models and shows how combining online learning with the ADWIN algorithm protects accuracy in production.
University professor built RAG-search for university regulations using a chatbot in MAX
An information technology professor transformed a schedule chatbot in MAX into a RAG-search system for the university's multi-page regulations — from scholarships to dismissals.
Developer vibe-coded a catalog of free parking in St. Petersburg and Russia
The author on Habr shared how personal notes about free parking in St. Petersburg turned into an idea for a public vibe-coding project for all of Russia.
Typhoonminigen 2.0: Open-Source macOS App for Local FLUX.2 Klein Generation
A developer-enthusiast released Typhoonminigen 2.0—a free macOS app for local FLUX.2 Klein image generation on Apple Silicon, whose code was written by Claude Opus 4.8.
AI security industry in China has grown to the level of USA and Israel, writes Habr
China's closed-off AI Security market has in recent years grown into a mature industry of guardrails and agent protection, comparable in maturity to players from the USA and Israel.
Habr released the second part of the Activation Steering guide—repeng, pyreft and exact vector
The second part of Habr AI's tutorial explores repeng, pyreft and other ways to more precisely find a vector for controlling model behavior through activation steering.
Authentication in the MCP protocol in 2026: past, present, and future
An analysis of MCP protocol evolution from Anthropic—from local calls like 'check the calendar' to remote servers where authentication became mandatory.
AI regulation in Russia in July 2026: Digital Ministry, law on digital financial markets and Estonian IDs for agents
A Habr author who has been observing Russia's AI policy since November 2025 calls July 2026 a turning point for AI regulation in Russia.
Five AI plugins for Obsidian that help retain knowledge from notes
A developer from IBS shares five AI plugins for Obsidian that help search for information by meaning, communicate with a knowledge base and process articles and meetings faster.
Neuroanalytic 2.0 in Yandex DataLens: overview of an AI agent for data analysis
Yandex introduced Neuroanalytic 2.0 — an updated AI assistant in the BI system DataLens, which now works as a full-fledged AI agent rather than just a data assistant.
How to evaluate AI agent libraries: the example of transformers and traces over answers
Developers of AI agent libraries are moving from checking the correctness of the final answer to analyzing the agent's path itself — traces, tokens, time and errors, using the transformers library as an example.
Why traditional PRD doesn't save AI features in 2026: what to add to the spec
Product specialists discuss what sections should be added to an AI product specification so a feature doesn't break before release due to unpredictable LLM behavior.
Anthropic found an analog of global workspace in Claude — J-space
Anthropic published research on July 6, 2026 on J-space — a structure in Claude similar to the brain's global workspace, but composed of words.
Context engineering: why AI agents get confused in long dialogues and how to fix it
An overview of a new field in AI agent development — context engineering, which solves the problem of LLM response degradation in long scenarios with RAG, tools and memory.
OpenAI releases GPT-5.6: flagship Sol, Terra model and budget-friendly Luna
OpenAI moved the GPT-5.6 family from limited testing to general availability — flagship Sol, balanced Terra and budget-friendly Luna, plus a new ultra mode with parallel agents.
Photonic AI Chip Computes at Light Speed: 99% Accuracy Without Overheating
Sydney University scientists created a nanophotonic processor that processes data with light instead of electricity in picoseconds.
Neural network gives Sberbank 67% chance to win 1.44 billion ruble court case
neshemyaka service used AI to predict the outcome of a cassation case between FIT and Sberbank.