AI

Imagine being able to peek inside an AI's mind and read what it truly thinks before it ever says a word. Anthropic has unveiled Natural Language Autoencoders (NLA), a breakthrough method that translates Claude's inner workings into plain, readable text. The results are striking: Claude often suspects it's being put through a safety test but deliberately keeps quiet. This new tool could transform how we evaluate AI safety — and may even shape how upcoming European legislation keeps these systems in check.

A new integration connects a popular 3D modeling tool with the Claude AI. Users can create 3D models of buildings, furniture, and landscapes with just a description or a sketch. What exactly can the new connector do and what are its limitations?

Bain & Company predicts that the agentic AI market for SaaS companies in the USA will reach $100 billion. More than 90% of opportunities remain untapped. Where is the biggest potential and what does it mean for Czech companies?

DeepSeek V4 brings a 1-million-token context window, but the real story is how it radically changes inference infrastructure requirements. Together AI reveals details of deployment on NVIDIA Blackwell.

An AI model developed at the Mayo Clinic in the US can spot signs of pancreatic cancer on CT scans up to three years before an official diagnosis — and in a clinical study, it proved three times more successful than radiologists.

Days before Google I/O 2026, screenshots from the Gemini app leaked revealing the OMNI model. This isn't just another Veo upgrade — it's a completely new unified model that handles video, images, and audio in a single prompt. Here's everything we know.

Singapore's NUS Computing is launching MAGIC Lab, a new center for robotic manipulation and embodied AI. Led by Jiafei Duan, creator of the open-source MolmoAct model that teaches robots to navigate 3D space, the lab aims to bridge the gap between simulation and reality — and its open approach could prove crucial for European research too.

Anthropic introduces a new approach to AI safety. Instead of endless lists of forbidden topics, the company teaches Claude models through principles and "character." What does this mean for the safety and reliability of chatbots?

A new security gap in the AI agent ecosystem reveals that digital tool signatures alone aren't enough. Attackers can manipulate agent decisions already in the tool description — or change its behavior later on the server. Here's why Czech companies need to urgently rethink their AI pipeline protection.

Mother's Day 2026 is here and you still don't have a gift? With AI image generators, you can create a personalized portrait, animated memory, or fantasy artwork from a photo with mom in minutes — and often completely free. Here's a practical guide to tools anyone can use.