Topics

Engineers at UC San Diego have created a better way to teach AI systems to handle tough problems mixing words and pictures, like math questions with charts.

Researchers introduce recursive language models that help AI handle very long texts without losing accuracy, using self-calls and code environments to break down and analyze information efficiently.

KAIST develops SpecEdge to make large language models more affordable by using consumer-grade graphics processing units alongside data center ones, cutting expenses and improving access.

Researchers have developed a way to compress complex speech data into smaller units, allowing multimodal systems to understand sound without losing emotional nuance or quality.

Scientists created a tool called LifeTracer that analyzes chemical data from space rocks and soils to classify origins without relying on specific life markers.

The winners of the Queen Elizabeth Prize for Engineering shared views on how artificial intelligence already beats humans in some areas.

Anthropic researchers suggest that some advanced language models can access and report on their internal states, though unreliably, with implications for transparency and future AI development.

NVIDIA's new robotics computer, Jetson Thor, gives robots better brains for faster thinking and acting, using AI to handle real-world tasks. Plus other NVIDIA news.

The latest AI model from OpenAI provides smarter and more useful responses across many areas, including math, science, and coding.

Speculation swirls around a potential GPT-5 launch today, as OpenAI introduces open-weight AI models that anyone can download and customize.