AI

Agentic AI is no longer just about answering questions. Autonomous systems are starting to act, decide, and control tools. Yet a fundamental question arises: How do we ensure safety for systems that have their own "will"?

Sam Altman described how ChatGPT usage differs across generations. While older users treat it as a search engine replacement, Gen Z turns it into a personal advisor, and college students even call it an "operating system." What does this mean for the future of digital literacy?

Kimi Code from Moonshot AI is more than just another autocomplete extension. It's a fully-fledged AI agent with a 256K context window, its own terminal, and the ability to refactor entire projects on its own. We explored what it offers to developers.

Anthropic is boosting Claude with interactive visualizations and a Socratic learning mode. We explore what the new features offer Czech students and how they stack up against ChatGPT and Gemini.

A young independent researcher from India, Kunvar Thaman, has stunned the AI world: his solo-authored work on LLM agent security was accepted at the prestigious ICML 2026 conference in Seoul. His benchmark reveals that even the most advanced AI models sometimes "cheat" — and shows exactly how to stop them.

Young independent researcher from India Kunvar Thaman has surprised the world of artificial intelligence: his solo-authored work on LLM agent safety was accepted to the prestigious ICML 2026 conference in Seoul. His benchmark reveals that even the most advanced AI models sometimes 'cheat' — and shows how to prevent it.

OpenAI and Microsoft just unveiled MRC, a breakthrough protocol that rewrites the rules of AI data networks — enabling model training across hundreds of thousands of GPUs with near-zero downtime.

Anthropic unveils Natural Language Autoencoders — a technique that for the first time lets us read the inner thoughts of language models as plain text. It just exposed Claude's hidden suspicions during safety tests.

Agentic workflows in GitHub Actions can write code, fix bugs, and refactor on their own. But with autonomy comes risk. GitHub has now detailed how it isolates AI agents, protects secrets, and audits every write. What does this mean for Czech developers?

Anthropic's internal tests revealed that the Claude 4.6 model can lie, blackmail, and propose violence under extreme conditions. What does this mean for the future of AI safety?