PrismML brings its small LLMs to Qualcomm-powered smart glasses

AI startup PrismML has introduced a version of its small language models (LLM) specifically designed to run on smart glasses equipped with Qualcomm's Snapdragon chips. The company demonstrated the technology at Qualcomm's Snapdragon Summit.

PrismML's 1-bit Bonsai LLM can be run locally on smart glasses built on the Snapdragon AR1 Gen 1 Platform. The startup's broader aim is to create open-weight AI that operates on devices, leveraging their existing computing power. This approach is positioned as an alternative to cloud-based AI services, addressing privacy concerns and high compute demands.

The core of PrismML's technology lies in significantly reducing the size of large language models while maintaining nearly all of their performance on standard benchmarks. The version adapted for smart glasses is a 2-billion-parameter model, optimized for vision and language tasks. This allows users to ask real-time questions about what they are seeing.

While PrismML has now enabled its model to run on Qualcomm's chips, no smart glasses manufacturer has yet announced a product utilizing PrismML's technology. The company's vision is focused on developing more distributed and on-device AI capabilities.