Taipei, Dec. 30 (CNA) Taiwan Semiconductor Manufacturing Co. (TSMC), the world’s leading advanced chipmaker, officially began volume production of its 2-nanometer chips in the fourth quarter of 2025, according to a recent update on the company’s website.

The low-key announcement confirms that TSMC met its original roadmap for the next-generation technology. Production is currently centered at Fab 22 in Kaohsiung, utilizing the company’s first-generation nanosheet transistor technology. The new architecture achieves “full-node strides in performance and power consumption,” the website said.

The company described the 2nm process as the most advanced in the semiconductor industry in terms of transistor density and energy efficiency, adding that it is designed to “address the increasing need for energy-efficient computing,” particularly for AI and mobile applications.

MIT Associate Professor Phillip Isola studies the ways in which intelligent machines “think” and perceive the world, in an effort to ensure AI is safely and effectively integrated into human society.

Goldman Sachs Group Inc. is co-leading financing for a Texas project to build private power campuses for artificial intelligence.

MagicBot, China’s first humanoid robot achieving multi-robot collaboration, enabling flexible production and efficiently handling complex industrial scenarios.

These days, large language models can handle increasingly complex tasks, writing complex code and engaging in sophisticated reasoning. But when it comes to four-digit multiplication, a task taught in elementary school, even state-of-the-art systems fail. Why?

A new paper posted to the arXiv preprint server by University of Chicago computer science Ph.D. student Xiaoyan Bai and faculty co-director of the Data Science Institute’s Novel Intelligence Research Initiative Chenhao Tan finds answers by reverse-engineering failure and success.

They worked with collaborators from MIT, Harvard University, University of Waterloo and Google DeepMind to probe AI’s “jagged frontier”—a term for its capacity to excel at complex reasoning yet stumble on seemingly simple tasks.

A biologically grounded computational model built to mimic real neural circuits, not trained on animal data, learned a visual categorization task just as actual lab animals do, matching their accuracy, variability, and underlying neural rhythms.