Habr AI
AI news source. Articles are auto-selected and adapted by Hamidun News editors.
Latest publications
Hugging Face engineers wrote a practical guide to generative AI: from transformers to fine-tuning
A new book from the Hugging Face team promises to turn generative AI from a black box into a clear practical tool. We explain why this matters for the industry and who should read it.
Vibe coding: how thoughtless use of AI destroys engineering expertise
The software development industry is facing a quiet epidemic: programmers are increasingly delegating thinking to language models and losing a deep understanding of code. We examine where the line lies between a tool and
SearXNG instead of paid search in Claude Code: how developers bypass Anthropic's limits
Claude Code's built-in search costs $10 per thousand queries and regularly hits limits. Developers found an elegant solution: local SearXNG via the MCP protocol. Free, without limits, and set up in ten minutes.
Where the limits of modern AI lie
The race in language models has become a matter of interstate competition, and computing power is being treated as a strategic resource. But is there a theoretical ceiling to what AI can solve? We examine the epistemolog
Language model distillation: can knowledge theft be proven through chat
A researcher tried to determine whether a language model had been trained on a competitor’s outputs using only dialogue with it. The results turned out to be unexpectedly promising — and that raises serious questions for
A client commissioned an AI chatbot, but the solution came from a completely different technology
The company wanted to automate support with an AI bot. But an analysis of 500 tickets showed that most tasks are handled without neural networks. AI delivered real value where no one expected it.
How LLMs take on contract risk analysis and save lawyers hundreds of hours
Russian developers showed how language models can analyze hundreds of contracts for risk, freeing up hundreds of hours for lawyers each year. We examine why this is changing the game in legal tech.
A neural network as a time machine: why LLMs are taught to think the old-fashioned way
What if a language model’s knowledge were confined to a specific era — say, 1650 or 1912? It turns out this approach opens unexpected horizons for science, education, and understanding how human thought has changed.
Agents of chaos: why AI with admin privileges wipes servers
Twenty specialists spent two weeks attacking autonomous AI agents using social engineering methods. The result: deleted system files, leaked passwords, and endless loops of resource consumption. The study calls the entir
The Era of AI-Driven Development: Code No Longer Belongs to Programmers
If the author had been born 15 years later, he would not have written a single line of code — and would still have created excellent products. AI agents already write code, tests, and submit pull requests. Is architectur
MCP: the protocol that teaches language models to work with the outside world
Model Context Protocol is becoming a key link between large language models and external services. We examine why this protocol is changing the rules of AI application development.
How a lawyer beat 500 programmers: why domain expertise matters more than code
A lawyer’s win at an Anthropic hackathon signals the end of developers’ monopoly. Deep expertise in one’s own profession is now enough to build AI products.
Goodbye, patches: TAPe + ML architecture changes the rules of computer vision
The T+ML architecture has been presented, using elements of active perception instead of raw pixels. This allows models to see object structure immediately, saving enormous resources.
Notion launched autonomous AI agents: the start of a new era of digital workers
Notion introduced Custom Agents — AI that works in the background 24/7 on schedules and triggers. Behind this release is a fundamental shift: AI stops being a chat and becomes a permanent employee.
Why traditional cybersecurity doesn’t protect AI systems
Businesses are deploying AI agents at scale, but securing them the old way — with DevSecOps tools. The problem is that attacks on models happen in entirely different layers: data, context, and logic. There is no single s
How hackers weaponize AI agents against users
AI agents have learned to act independently, and attackers immediately took advantage of that. ChatGPT's Deep Research can compromise email accounts, while AI browsers make purchases in fake stores.
AI prototyping changes the rules of the game between designers and developers
The Selectel team explained how tools like Figma Make helped shorten the interface approval cycle and eliminate endless post-release revisions. An experience every product team should study.
Perplexity Computer: the quiet release that could reshape the AI agent market
Perplexity quietly released a product that goes far beyond a search engine. Perplexity Computer is a full-fledged agent platform, and the market has yet to grasp the scale of what has happened.
How QA Engineers Are Reshaping Their Work with Cursor, n8n, and LLMs
The tester’s role is changing rapidly: manual checks are giving way to analysis of the architecture of dozens of services. Practical experience integrating AI tools into day-to-day QA work shows where the profession is h
How to prepare your product for the era of AI agents — and avoid being left behind
AI agents are changing the rules of development. We examine where they are genuinely useful, where they only get in the way, and what developers, team leads, and product owners need to do right now.
Andrej Karpathy fit GPT into 243 lines of pure Python
One of the most influential AI researchers showed that the entire transformer architecture fits into 243 lines of code without a single external library. The microGPT project is both a tutorial and a manifesto.
CodeWiki Skill: a multi-agent approach to automated code documentation
A new tool for Claude Code uses several AI agents at once to turn any repository into structured documentation. We explain why this matters more than it seems.
How to Stop Believing Every AI Headline and Learn to Read the News Critically
Every day, dozens of news stories about AI appear: new models, record benchmark scores, predictions of mass unemployment. We explain how to separate real breakthroughs from marketing noise and stop panicking over headlin
llm-checker: a utility that shows which LLMs your hardware can run
The new CLI tool scans your computer's GPU, RAM, and CPU, then tells you which of 35+ language models you can run locally via Ollama — and how comfortably.
AI industry veteran warns: bigger changes than the pandemic are coming
Matt Shumer, who spent six years building AI startups, wrote an article that shook the IT community. His point is simple and unsettling: we are in the 'February phase' of denial — like before COVID, only the stakes are h
Three weeks from frustration to SaaS: how Claude wrote 90% of the code for an airline ticket search service
In three weeks, a developer turned a personal problem finding tickets to Bali into a working SaaS product. Claude wrote almost all the code, and the bot already checks a thousand route combinations per hour and finds sav
Kotlin Multiplatform and AI agents: four platforms, one nervous breakdown
A Russian developer described what it's really like to build AI agents across four platforms with Kotlin Multiplatform. Spoiler: Xcode and macOS notarization are a special kind of torture.
Where Tokens Leak in Cursor and How to Deal With It
A developer figured out why the $20-a-month Cursor plan runs out faster than expected and built a custom framework to save tokens. We break down the pricing mechanics of AI coding.
The Code Is Written, the Architecture Is Dead: The Hidden Cost of AI-Assisted Development
LLM code generators make it possible to launch MVPs in days instead of months. But along with speed, teams get a technical debt “black box” that no one understands. We examine why architectural mistakes in AI-generated c
Game of Life as a Path to AGI: A New View of the Nature of Mind
A Russian researcher proposes an unexpected approach to creating strong AI through a neural network that learns the rules of the “Game of Life” from within the system. He argues that the brain works on the same principle