Scientists are making progress on multiple fronts: new methods help language models make complex reasoning processes more transparent, while other techniques make training control systems more efficient. At the same time, researchers are working to accelerate the inference speed of diffusion models.
A study day full of alarm signals: simulations show that AI models almost always resort to nuclear weapons in crises, the UN demands fair distribution of AI profits, and a philosopher warns against therapeutic chatbots. At the same time, teachers report genuine added value in the classroom.
Researchers have grown human brain cells on a microchip and trained them to play the video game Doom. The experiment demonstrates how artificial neural networks could leverage biological intelligence — and opens new questions for brain research.
US company Anthropic is providing its language model Mythos to the European cybersecurity agency ENISA. The AI system can detect network vulnerabilities — but also poses significant risks to critical infrastructure.
Uber is bringing autonomous driving to Europe: the company will test self-driving vehicles in Munich in 2026 together with Israeli AI startup Autobrains and Nvidia. The service is expected to enable significant savings in driver costs.