PrismML adapts its compact AI language model for Qualcomm smart glasses
AI startup PrismML has created a version of its small language model designed to run on smart glasses powered by Qualcomm chips. The model lets glasses locally recognize and describe what a wearer is looking at.
PrismML, a startup founded by Caltech researchers and advised by UC Berkeley professor Ion Stoica, has built a version of its compact language model to run on Qualcomm's Snapdragon chips. On Wednesday, at Qualcomm's Snapdragon Summit, the chipmaker showed off PrismML's 1-bit "Bonsai" model, which can operate directly on AI smart glasses built with the Snapdragon AR1 Gen 1 platform.
Smaller size, similar performance
PrismML is known for dramatically shrinking large language models — in this case by a factor of four — while keeping nearly all of their performance on standard benchmarks. The smart glasses version is a 2-billion-parameter model tuned to handle both vision and language, allowing wearers to ask, in real time, questions about what they are looking at.
A push toward on-device AI
The company's broader ambition is to promote open-weight AI models that run directly on devices, making fuller use of computing power those devices already have. PrismML frames this approach as an alternative to relying on the privacy assurances of proprietary AI labs and their ever-growing demand for computing resources. Releasing a model for Qualcomm's chips marks a step toward that goal, though no smart glasses product using PrismML's technology has been announced so far.
