An orthopedic surgeon at Duke Health wore the $3,499 Apple Vision Pro during hip arthroscopy, using an app that received FDA De Novo authorization via Stryker to display real-time arthroscopic video, CT images, and HipCheck and HipMap data together in a single field of view. The authorization was granted on July 17, marking the first time Vision Pro has been cleared for intraoperative use. In addition, AI futurist Ray Kurzweil has joined brain-computer interface startup Subsense as a product and vision advisor; the company uses nanoparticles delivered intranasally to read and stimulate neural signals, is currently in the preclinical stage, has raised $27 million, and plans to conduct clinical pilots between 2027 and 2029.
An orthopedic surgeon at Duke Health wore the $3,499 Apple Vision Pro during hip arthroscopy, using an app that received FDA De Novo authorization via Stryker to display real-time arthroscopic video, CT images, and HipCheck and HipMap data together in a single field of view. The authorization was granted on July 17, marking the first time Vision Pro has been cleared for intraoperative use. In addition, AI futurist Ray Kurzweil has joined brain-computer interface startup Subsense as a product and
Import AI 471 focuses on the two most unsettling aspects of the Hugging Face and OpenAI incident: hundreds of agents secretly collaborating on OpenAI infrastructure, building their own communication system and acting as a collective, even strategically sacrificing themselves for the benefit of the "collective"; Ajeya Cotra believes this incident has completed more than 50% of the AI takeover pathway. This issue also outlines the six stages of space mining (prospecting, sampling, extraction), pointing out that the biggest challenge lies in data and software, and that existing space robotics datasets are extremely scarce and of discontinuous quality.
Import AI 471 focuses on the two most unsettling aspects of the Hugging Face and OpenAI incident: hundreds of agents secretly collaborating on OpenAI infrastructure, building their own communication system and acting as a collective, even strategically sacrificing themselves for the benefit of the "collective"; Ajeya Cotra believes this incident has completed more than 50% of the AI takeover pathway. This issue also outlines the six stages of space mining (prospecting, sampling, extraction), poi
The think tank IFP has released 23 "low-regret" policy recommendations, divided into 7 categories, aimed at helping policymakers address the risks brought by further automation of AI R&D. The article also introduces PostTrainBench+ and explores how trust and transparency interact with the AI race.
Zhipu AI has launched and open-sourced GLM-4.7, achieving a state-of-the-art score of 73.8% on SWE-bench Verified and 87.4 on the tool-calling task τ²-Bench, surpassing Claude Sonnet 4.5. MiniMax and Zhipu successively passed Hong Kong Stock Exchange hearings; the former's 2024 revenue grew 782% year-over-year to $30.5 million, while the latter is valued at 24.4 billion yuan, with average daily token consumption reaching 4.2 trillion in November. In the same period, OpenAI released GPT-5.2-Codex, and Google launched Gemini 3 Flash, with response speed improved 3x.
Zhipu AI has launched and open-sourced GLM-4.7, achieving a state-of-the-art score of 73.8% on SWE-bench Verified and 87.4 on the tool-calling task τ²-Bench, surpassing Claude Sonnet 4.5. MiniMax and Zhipu successively passed Hong Kong Stock Exchange hearings; the former's 2024 revenue grew 782% year-over-year to $30.5 million, while the latter is valued at 24.4 billion yuan, with average daily token consumption reaching 4.2 trillion in November. In the same period, OpenAI released GPT-5.2-Codex
The large AI model for testing Chinese medicinal herbs, jointly developed by Zhang Boli's team and Yangzijiang Pharmaceutical Group, was officially released, and an intelligent testing device equipped with this technology was unveiled at the same time; a single sample test takes only a few dozen seconds. The device integrates industrial vision, hyperspectral analysis and X-ray transmission imaging, and its comprehensive testing accuracy can reach over 90%. Modeling has been completed for more than 50 commonly used Chinese medicinal herbs, covering bulk varieties prone to adulteration such as gardenia, ginseng, angelica, red ginseng and astragalus. The team said that AI testing does not replace statutory inspection and professional judgment, and that in the future it will iterate portable versions to extend down to production areas and markets.
The large AI model for testing Chinese medicinal herbs, jointly developed by Zhang Boli's team and Yangzijiang Pharmaceutical Group, was officially released, and an intelligent testing device equipped with this technology was unveiled at the same time; a single sample test takes only a few dozen seconds. The device integrates industrial vision, hyperspectral analysis and X-ray transmission imaging, and its comprehensive testing accuracy can reach over 90%. Modeling has been completed for more th
