A startup backed by Khosla Ventures has successfully compressed a 27 billion-parameter AI model to operate on an iPhone, significantly surpassing the typical size of mobile models. This development aligns with Apple’s ongoing strategy to incorporate larger AI models on iPhones, emphasizing on-device AI as crucial for enhancing future software features. Analysts highlight that the model compression achieved by this startup is a key advancement, enabling more powerful AI assistants on smartphones, which is part of a broader industry trend to run sophisticated AI directly on devices to enhance privacy and reduce reliance on cloud services.

Apple: Apple Inc. is a major technology company that designs and sells consumer electronics, software, and services, including the iPhone and its integrated AI features. In the context of this news, Apple has been meeting with PrismML and is exploring using the startup’s compression technology to run much larger AI models directly on iPhones, aligning with Apple’s broader push toward more powerful on-device AI.
PrismML: PrismML is an artificial intelligence startup that develops techniques to dramatically shrink the memory footprint of large language models while preserving their performance, enabling them to run locally on devices such as smartphones. In this news, PrismML reports it has compressed a 27 billion-parameter model to fit within an iPhone’s hardware constraints, which it promotes as the largest AI model yet deployed on an iPhone and a key step toward more advanced on-device AI.
Khosla-backed startup: The Khosla-backed startup in this news is PrismML, an AI company focused on compressing large language models so they can run efficiently on consumer devices like smartphones. In this story, PrismML claims to have compressed a 27 billion-parameter AI model to run entirely on an iPhone, positioning its technology as a breakthrough for on-device AI and attracting interest from Apple.

`json
{
“On-device_AI_trend”: “Recent reporting highlights a broader industry shift toward running advanced AI models directly on consumer devices to improve privacy, reduce latency, and lessen dependence on cloud infrastructure.”,
“Model_compression_significance”: “Analysts note that compressing large models without degrading performance is seen as a potential enabler for more capable AI assistants and tools on smartphones than current lightweight mobile models typically allow.”
}
`