AI Lab PrismML, founded by Caltech researchers and advised by Caltech’s Ion Stoica, has created a version of a small language model for smart glasses that runs on Qualcomm’s Snapdragon chips.
On Wednesday at Qualcomm’s Snapdragon Summit, the chipmaker showed off PrismML’s 1-bit Bonsai LLM. This can be run locally on AI smart glasses built on the Snapdragon AR1 Gen 1 platform.
As TechCrunch previously reported, PrismML’s claim to fame is its ability to significantly shrink large models (in this case, by a factor of 4) while preserving nearly all of their performance on standard benchmarks. The smart glasses version is a 2 billion parameter model tuned to vision and language, so the wearer can ask what they are looking at in real time.
Prism’s larger goal is openweight AI that runs on the device and better leverages the computing power the device already has. The startup is touting this as an alternative to relying on the privacy promise of its own AI lab or the insatiable need for more computing.
Releasing a model of Qualcomm’s chip is a step toward that vision. However, smart glasses that run PrismML have not yet been announced.
