Apple Eyes AI Startup PrismML for On-Device Smarts

Apple is currently in preliminary discussions with PrismML, a startup focused on compressing artificial intelligence models to enable them to run efficiently on smartphones. PrismML's CEO, Babak Hassibi, confirmed the early-stage talks, describing them as progressing positively, though the final outcome remains uncertain.

The startup recently unveiled its Bonsai 27B model, which significantly reduces the storage and memory requirements of large AI models. This compression allows the model to operate on devices as recent as the iPhone 15. This advancement aligns with Apple's ongoing efforts to expand its on-device AI capabilities, potentially enhancing user experience through faster responses and bolstering its privacy reputation.

PrismML's compressed models offer substantial benefits, including reduced memory usage, significantly faster response times, and lower power consumption compared to standard versions. While there is a slight trade-off in accuracy, the gains in efficiency are considerable. The Bonsai 27B series is available in optimized versions for maximum compression or higher quality, both released under a free license.

This development stems from research originating at the California Institute of Technology, with PrismML holding exclusive commercialization rights. The startup secured a $16.25 million seed funding round earlier this year. For Apple, shifting more AI processing to devices like the iPhone could lead to reduced cloud infrastructure costs and strengthen its commitment to user privacy.