Five days after initial reports, PrismML has confirmed it is in discussions with Apple regarding its AI model-shrinking technology.

A significant challenge in AI processing is managing large models that exceed the capabilities of typical smartphones. PrismML emerged on July 9 with a solution to compress these models into a more manageable size. By July 14, the startup confirmed ongoing talks with Apple and other companies, according to CEO Babak Hassibi.

Hassibi stated that Apple is currently evaluating their technology, with discussions being in their early stages but "progressing nicely." These talks could lead to licensing the technology or even an acquisition.

PrismML's technology could greatly benefit Apple's goal of on-device AI processing. Their method significantly reduces model sizes, potentially allowing them to run on iPhones without cloud reliance. For instance, a 27-billion-parameter Qwen model, normally 54GB, can be shrunk to under 4GB, fitting within an iPhone's memory. While there's a slight performance trade-off, it enables more private and secure on-device AI.