AI News

German researchers compared 10 large language models with 108 psychiatrists in diagnosing depression, mania, and schizophrenia. GPT-5.1 and Gemini-3-Pro-Preview reached the 64th percentile of the physician distribution. The Nature study shows that AI and humans make different types of errors in psychiatry.

Gartner warns that more than 40% of agentic AI projects will be canceled by the end of 2027 — not because of poor models, but due to overloaded infrastructure and unexpectedly high costs. GitHub has already had to limit Copilot, and more companies will follow.

Spanish La Liga is set to revolutionize the game with fully automated offside decisions starting from the 2026/27 season. A chip in the ball, a network of cameras, and AI are transforming football — and the 2026 World Cup is already showcasing just how advanced this incredible technology has become.

A whopping 85% of buyers now use AI in their home search, yet a surprising 89% still rely on a human real estate agent. Why are algorithms transforming, but not replacing, the real estate market? And which powerful tools are already available in Czech?

American AI powerhouse Anthropic is reportedly getting tough on Chinese firms – including financial giant Ant Group – for bypassing access restrictions to its Claude models, according to the Financial Times. What does this mean for the future of AI and for Czech users?

This analysis delves into a pivotal shift in how we understand technology. Artificial intelligence and big data aren't just tools anymore – they're evolving into indispensable infrastructure, as fundamental as electricity or the internet.

Meta's AI chief, Alexandr Wang, just unveiled a massive update to their Muse Spark model, codenamed Watermelon. Insider sources claim its performance now rivals GPT-5.5, promising huge leaps in coding and agent capabilities. The best part? It's set to be a far more affordable alternative for businesses tired of pricey subscriptions.

DeepSeek V4 has arrived in two powerful variants — Flash and Pro, boasting a massive 1.6 trillion parameters — and it's already outperforming Google Gemini in programming tests! Imagine: a million-token context window, completely open source under an MIT license, and a price tag dramatically lower than GPT or Claude. So, what does this game-changer mean for developers in the Czech Republic?

Meta's AI chief, Alexandr Wang, just revealed in an internal meeting that their upcoming AI model, Watermelon, is hitting benchmarks comparable to OpenAI's GPT-5.5! With ten times the computational power of its predecessor and a staggering $125–145 billion investment, it seems Mark Zuckerberg's massive bet on AI is finally starting to pay off.

Beijing's Z.ai has unveiled its GLM-5.2 model, which, thanks to its incredibly low price and top-tier programming benchmarks, is now genuinely challenging market leaders like GPT-5.5 and Claude Opus.