Groundbreaking Technology Emerges from UCLA

A team of visionary researchers at the University of California, Los Angeles (UCLA), led by Aydogan Ozcan, has unveiled a game-changing technology that could revolutionize how generative artificial intelligence (AI) works. This innovative approach replaces traditional reliance on electricity with incredible optical power—utilizing light instead.

The Heavy Energy Cost of Current AI Models

Today's generative models, particularly diffusion-based systems like DALL•E 2 and Stable Diffusion, are notorious for their staggering energy consumption. Training these sophisticated models can devour tens of thousands of kilowatt-hours—equivalent to the yearly energy usage of several households. Even when generating a single image, conventional graphics processing units (GPUs) can consume a substantial amount of energy, comparable to running an electric kettle for hours!

A Bold Shift: Light Over Electricity

In stark contrast, this new optical technique leverages a beam of light, marking a profound departure from energy-hungry processors. Initially, a digital encoder crafts a 'noise circuit'—a low-power pattern. This pattern is then projected onto a laser beam via a spatial light modulator (SLM). As the light carries this noise pattern through a second SLM, it morphs into a fully rendered image, effectively shifting the workload from the computer to the laser.

A Promising Future for AI