The startup claims its new 1‑bit neural engine cuts memory requirements by four times, enabling advanced AI on lightweight wearables.
PrismML announced a breakthrough in ultra‑low‑precision AI, running 1‑bit neural networks on next‑generation smart glasses. By compressing model weights to a single bit, the company says it reduces memory consumption by a factor of four while maintaining acceptable accuracy for on‑device tasks such as object recognition and translation.
The technology leverages Qualcomm’s latest chipset, integrating a custom accelerator that handles the bit‑packed computations efficiently. PrismML’s demo showcased real‑time scene analysis and overlay of contextual information without offloading data to the cloud, promising longer battery life and enhanced privacy for users.
Industry analysts note that while 1‑bit AI is still experimental, PrismML’s claim could accelerate the adoption of AI‑enabled wearables, especially in enterprise settings where on‑device inference is critical. The company plans to ship developer kits later this quarter and is seeking partnerships with OEMs to embed the solution into upcoming AR glasses.