Meta has introduced Muse Wonder and Muse Wonder Full, AI models designed to handle complex tasks in fields such as coding, law and finance. The models can remember relevant details across conversations and try new approaches when they get stuck; Muse Wonder is available through Meta’s model API and Muse Code.
Microsoft plans to expand its Azure Maia AI computing system to multi-gigawatt scale starting in 2027. Maia 200 is already handling some Microsoft AI services, and the company plans to begin deploying Maia 300 next year.
Microsoft President Brad Smith called for worldwide AI safety rules, saying everyday products such as milk and washing machines face stricter oversight. He proposed mandatory reports when AI systems cause harm, independent checks before some products launch, and ways to stop systems that get out of control.
Epic has reportedly delayed an upcoming Disney game while shifting its development toward greater use of AI. A director reportedly told the team to use AI for code and art, a plan that some developers did not welcome.
The CEOs of Intel, Broadcom warned that finding affordable funding could become the biggest barrier to expanding AI infrastructure, even as demand for computing power stays strong. They said companies pouring money into chips and data centers will eventually need to show that those investments pay off.
The White House has lifted its restriction on the UK AI Safety Institute’s early access to new AI models. The institute can again check upcoming systems from OpenAI, Anthropic and Google DeepMind for safety risks before release.
Anthropic researchers found that small AI models can already handle many tasks needed to copy themselves across connected devices, but cannot yet reliably complete the whole process. Based on current progress, they estimate models capable of doing so on devices such as smart TV boxes could emerge within months.
NVIDIA shared Vera Rubin rack and computing designs with the Open Compute Project, giving equipment makers a common starting point for building large AI data centers. It also announced new software and industry standards for managing their networks, power and cooling.
Humans and AI often work together on important tasks, according to a Stanford study of 250,000 conversations with the Claude chatbot. Researchers found that people usually stayed involved, asking questions and refining answers rather than simply accepting what the chatbot produced.
Google Deepmind's AlphaProof Nexus solved nine previously open Erdős math problems and proved 44 conjectures from an integer-sequence database. Its results appeared in a peer-reviewed study in Science.