PrismML Claims 93% AI Model Reduction, Apple in Talks

Startup PrismML has developed a method that it claims shrinks an artificial intelligence model by 93%. The Caltech spin-off, backed by Khosla Ventures, asserts it has reduced a 54GB AI model to under 4GB, potentially allowing it to run on devices like the iPhone 15 or newer.

PrismML CEO Babak Hassibi confirmed that Apple is currently evaluating its technology, describing the early-stage discussions as promising. The company raised $16.25 million in seed funding in March. PrismML has also released two free compressed AI models designed for Apple devices and Nvidia-powered PCs.

The breakthrough could enable Apple to run more sophisticated AI locally, reducing reliance on cloud processing. This could lead to improvements in speed, privacy, and battery efficiency for features like Siri. However, industry analysts caution that the claims require real-world validation.

While the potential for on-device AI is significant, further testing is needed to confirm PrismML's performance claims. The broader impact on the semiconductor industry also remains to be seen, as reduced model size does not automatically equate to a reduced demand for processing hardware.