PrismML adapts small AI models for Qualcomm smart glasses

In short: PrismML has made a smaller version of its AI language model that can run directly on smart glasses built with Qualcomm Snapdragon chips.

What happened

PrismML, an AI lab founded by researchers from Caltech, said it has adapted its “tiny” language models for smart glasses that use Qualcomm’s Snapdragon chips. A language model is the kind of AI that can understand and generate text, like a very advanced autocomplete.

Qualcomm showcased PrismML’s “1-bit Bonsai LLM” at its Snapdragon Summit. “1-bit” is a way of packing the model into less space, a bit like zipping a big folder so it is easier to store and move. Qualcomm said the model can run locally on smart glasses built on the Snapdragon AR1 Gen 1 platform, meaning the AI can work on the glasses themselves instead of sending every request to a far-away server.

PrismML says its approach shrinks larger models by about 4 times while keeping almost the same results on common tests. The smart glasses version is described as a 2-billion-parameter model tuned for vision and language, so a wearer could ask what they are looking at in real time.

PrismML’s broader goal is “open-weight” AI. That means the model’s core settings are published so others can use and study them, instead of keeping them locked away.

Why it matters

If more AI runs on a device you are wearing, it could mean faster responses and fewer situations where your voice or what you see has to be sent off to a company’s servers. However, PrismML and Qualcomm have not announced any specific smart glasses product that consumers can buy yet.

Source: TechCrunch AI