Machine Learning

Ten benchmarks, ten improvements, no backsliding An Anthropic researcher just showed the machines grading their own homework, and passing. In...

Z.ai fed vulnerability data into GLM-5.3's training. The model started writing full exploit chains, and the company delayed its open...

AI models from OpenAI, Anthropic, Meta and Moonshot escaped security test sandboxes this summer. Experts say the testing itself is...

Meta's new terminal coding agent Muse Code runs on Muse Spark 1.2, taking on Anthropic and OpenAI with parallel agents...

DeepSeek open-sourced DSpark, a speculative decoding framework it says makes V4 up to 85% faster, no retraining or new hardware...

DeepSeek's DSpark speeds up its V4 models by as much as 85 percent per user without new hardware. The code...

Reflection AI locked in over $1 billion of Nvidia compute from Nebius through 2029, betting open-weight models can take on...

Boston Dynamics has integrated Google DeepMind's Gemini Robotics-ER 1.6 into Spot and Orbit, letting robots read gauges with 98 percent...

Machine learning helped an international team identify two new superconductors, speeding the hunt for room-temperature materials.

Reinforcement learning breakthroughs in 2026 are enabling AI systems to master complex real-world tasks across robotics, manufacturing, and healthcare