AI experts build model that could ‘read War and Peace 5 million times in one sitting’
AI experts Anima Anandkumar and Benedikt Jenik, previously anticipated to lead Jeff Bezos’ Project Prometheus, have unveiled a new AI model designed to understand physics rather than language. Their company, Accelerated Understanding Inc., claims its model can process 5 trillion pieces of data in a single prompt, a capacity 5 million times greater than leading language models like Anthropic’s and Google’s. Unlike ChatGPT, which uses the Transformer architecture for text, this system employs neural operators to predict physical phenomena in space and time. The company aims to apply this technology to enterprise solutions such as chip design, robotics, and weather prediction, focusing on business-to-business deals initially.
Accelerated Understanding Inc. is introducing a physics-centric AI model that diverges significantly from current language-focused systems like ChatGPT. Co-founders Anima Anandkumar and Benedikt Jenik state their model can handle 5 trillion data points in one prompt, a capacity far exceeding typical language models. This system is built on neural operators to predict physical phenomena, moving away from the Transformer architecture. The company is targeting enterprise applications in chip design, robotics, and energy, believing a deeper grasp of physics is key to optimizing materials and predicting complex events. This approach could reduce trial and error in fields like semiconductor manufacturing. The shift from language to physics as the core of AI intelligence presents a new paradigm for industrial applications. For Asia’s manufacturing hubs, particularly in semiconductor production, this technology could offer a competitive edge by optimizing design and reducing development cycles. The focus on enterprise solutions suggests a direct impact on industrial efficiency rather than consumer markets, potentially reshaping how physical processes are modeled and managed across the region’s diverse industries. The challenge will be integrating this novel AI into existing complex industrial workflows.
