A new startup recently caught the attention of Apple as it looks for fresh ways to bring better technology directly to your pocket. Just days after early reports surfaced online, PrismML confirmed that it is actively talking with the smartphone maker about its new software compression tools.
The startup built a clever method to squeeze massive artificial intelligence models down to a fraction of their normal size. This breakthrough could let high-end programs run locally on standard phones without destroying battery life.
Don’t miss the best of The Mac Observer
Set us as a preferred source and our Apple reporting ranks higher in your Google Search results and Discover feed — one tap, no account changes.
The startup squeezes massive software models to fit inside phones
PrismML CEO Babak Hassibi shared in a recent interview that Apple is currently evaluating its new technology. He noted that talks are in the early stages but progressing nicely. The main problem with modern AI is that the best programs require huge amounts of memory and computing power. Normally, these models live on massive cloud servers because normal computers and phones simply cannot handle them alone.
PrismML claims it can reduce a 54-gigabyte model down to just under four gigabytes. It does this by aggressively reducing the memory values used by the underlying code. This massive drop in size means a device like the iPhone 15 Pro could run the software entirely on its own built-in memory.
The startup did admit that the compressed version loses a tiny bit of performance compared to the full-size model, especially when handling complex math or coding problems.
Running tasks locally on devices keeps user data highly private
Keeping things local instead of relying on cloud servers is a major priority right now. Recent news that Apple met with a startup to squeeze massive AI models onto iPhone lines up perfectly with its long-term hardware plans. By running tasks directly on the device, the company can protect user privacy since data never leaves the phone. It also means people would not need a constant internet connection just to use basic smart features.
If this technology works at scale, it could change the current hardware rush. Right now, companies are spending billions to build giant data centers just to process user requests. If phones can handle the heavy lifting themselves, it lowers the demand for expensive server chips and cloud infrastructure.
Apple and PrismML might agree on a licensing deal, or the tech giant could simply decide to buy the startup outright. The ability to run full-sized models locally could give Siri a massive intelligence boost without compromising user privacy or requiring expensive cloud processing fees.
Discussion