Move Aside, Chatbots: AI Humanoids Are Here (www.wired.com)

OpenAI is quietly expanding into humanoid robotics again, hiring researchers focused on embodied systems after earlier robotics work wound down in 2021. WIRED reporting highlights this hiring push as part of a broader industry pivot: advances in motors, actuation and perception—plus progress in machine learning—are making human‑shaped robots viable for real‑world tasks. The move signals that major AI labs now see physical embodiment as a crucial piece of the AGI puzzle: language models excel at abstract reasoning, but they can’t pry open a jar, climb stairs, or learn through physical trial‑and‑error in the environments humans inhabit. Technically, the frontier is now about integrating improved hardware (faster, more compliant actuators and balance control) with algorithms for locomotion, manipulation and long‑horizon decision‑making. LLMs show surprising implicit understanding of physics and affordances, but closing the gap requires sim‑to‑real transfer, sample‑efficient learning, tactile sensing, and robust control. Competitors include Boston Dynamics, Agility Robotics, Apptronik, Figure, Tesla and low‑cost makers like Unitree; each brings different strengths in hardware, scale or data from physical systems. Implications range from factory and logistics automation to fresh debates about safety, job displacement and how much “physical intelligence” is necessary for AGI.