Physical AI

Used bookstores across Japan are reporting roughly fivefold sales increases as bulk buyers purchase entire stocks by the ton. Some orders, including a single 50-ton shipment sent to the United States, are believed to end up at foreign facilities where books are scanned for AI training datasets and then destroyed. Store owners describe a wave of suspicious large-scale purchases draining shelves.

Physics World has published a discussion in which experts assess the future of photonics, weighing the field's main challenges against its opportunities. Contributors examine how light-based technologies are advancing in areas such as communications, sensing and computing, and consider what research, funding and workforce developments will be needed for photonics to reach its potential in the coming decade.

A new systematic review published in Frontiers analyses empirical studies on the use of artificial intelligence in university-level physics education. The review synthesises existing research on how AI tools are being applied in physics teaching and learning, and what evidence exists about their effectiveness. It comes as universities worldwide weigh how to integrate AI into science classrooms.

Commentary is circling around Palmer Luckey's robotics argument, ARK's forecast-beating results, and China's push in embodied AI, with the combined narrative reframing humanoid robots as an interface layer — an 'API' for the physical world. The discussion links hardware, software and engineering communities weighing whether humanoids become the standard platform for real-world automation.

Zenithon AI is working on technology it says can simulate extreme physics up to a million times faster than current methods. The company's ambition, highlighted in recent coverage, points to faster computational modelling of high-energy physical systems. Details about funding, team or technical validation remain sparse, so it is unclear how far the project has progressed or who is backing it.

The accumulated edifice of science, physics, mathematics and computation, built over a rich history, far outweighs the current contributions of machine learning and AI. The author frames these as exciting but perilous times, adding that even if civilization were to crash, people would still do science.

Researchers have unveiled an AI framework that incorporates spatial tissue geography into RNA velocity analysis, allowing scientists to model and steer cell fate decisions computationally, or in silico. The tool aims to improve predictions of how cells develop by factoring in their physical location within tissue. Details beyond the announcement remain limited.

Zenithon has raised $10 million to develop world models focused on extreme physics. The startup aims to build AI systems that simulate physical phenomena in extreme conditions, an area relevant to scientific research and industrial applications. Coverage of the funding round has appeared on European tech outlets, though further details about the investors and the company's plans have not yet been widely reported.

Ten Anthropic Claude AI agents working together reportedly generated a 17,895-line formal proof for a physics problem originally posed in 1904. The feat highlights how multi-agent AI systems can now tackle long-standing theoretical challenges in mathematics and physics. Observers are weighing what this means for the future of formal proofs and the role of AI in scientific research, with debate over the result's verification and significance.