Topic
Explore English articles about Generative AI.
MIT and others' GIFT feeds a CAD-generating AI's alternative solutions and failures back into training, improving average IoU by about 12%. While it reduces candidate generation at inference time, pre-computing roughly 1 million candidates and verifying manufacturability remain unresolved.
A consortium called Noetra, backed by 44 companies including SoftBank, Sony, NEC, and Honda, has unveiled a national AI factory plan centered on a 140MW-class data center housing 27,500 Rubin GPUs. Japan's Ministry of Economy, Trade and Industry (METI) is committing up to ¥1 trillion, yet the power procurement method and site location remain undisclosed.
Moonshot AI's Kimi K3 offers 2.8 trillion parameters and a 1-million-token context through a low-priced API. Reading the primary sources on its sparse MoE design, the burden of self-hosting, and the constraints ahead of the weight release.
Germany's ZAK issued its first administrative order targeting AI search, covering Google and Perplexity. Generated responses are treated as the operator's own content, while citation links are treated as separate media intermediary functions with distinct responsibilities.
Fujitsu has begun exploring business collaboration with three domestic robot makers—FANUC, Yaskawa Electric, and Kawasaki Heavy Industries—to connect business systems with physical robots through Physical OS. We examine the design of the common platform using NVIDIA technology and the safety and liability issues it raises.
Intel has planned a company-wide rollout of Gemini Enterprise and the parallelization of undisclosed types of HPC simulations using C4/N4. This is a collaboration that separates the roles of AI agents and semiconductor development.
Walden Robotics, a Toyota Research Institute spinout, raised $300 million and announced it has deployed general-purpose robots at Toyota's North American plants. We examine the LBM research findings and the undisclosed performance metrics behind the commercialization.
1Password now handles Claude Cowork logins on the Mac, keeping passwords isolated from the model. Human approval and per-session encryption narrow the exposure surface, but passkeys and post-login action auditing remain future challenges.
Linus Torvalds has declared that Linux will not become an anti-AI project, signaling a policy of accepting AI as a useful tool for development. While maintaining the principle that the human who signs off bears responsibility for generated code, he is pushing forward the introduction and streamlining of AI-based automated review.
OpenAI has unveiled "GPT-Red," which pits attack and defense models against each other in self-play, automating self-improvement in safety. The system rapidly iterates from vulnerability discovery to defensive training, aiming to strengthen the safety of agentic capabilities, though threat model design and other elements remain in human hands.
