Accelerated Understanding Inc, founded by Caltech professor Anima Anandkumar and Benedikt Jenik, has launched an AI model based on neural operator architecture instead of the transformer framework used by most major AI companies. The model operates in 4D, capturing three-dimensional spatial relationships while tracking how they evolve over time, and can handle up to 1 trillion tokens during training and exceeds 5 trillion tokens at inference. It has been scaled to 1 trillion parameters in pre-training, putting it in the same weight class as the largest models ever built. Before founding the company, the startup declined to lead Project Prometheus, a venture backed by Jeff Bezos. Target applications include energy optimization, chip design, robotics, weather prediction, and medical innovation.
Source: Read the original article
