Apple’s Local AI Model ‘Substantially Outperforms’ GPT4, Researchers Boast

Apple’s Local AI Model ‘Substantially Outperforms’ GPT4, Researchers Boast

Games

Products You May Like

Apple is continuing to hammer away at its ideal on-device AI assistant, and according to its own researchers, they’re already on pace to beat the biggest player in the industry. In a paper published to the arXiv Thursday, Apple AI specialists boasted that their local model “substantially outperforms” GPT4, the technology behind ChatGPT, Google’s Gemini, and Microsoft’s Copilot. This is largely because the model considers a handful of conversational variables as it tackles user requests, allowing for a more natural and productive experience.

These variables include what the paper refers to as on-screen entities, conversational entities, and background entities. The first of these considers what’s on the user’s screen—a capability likely made possible by Apple’s model being local, not web-based. The second incorporates details provided by the user throughout the conversation, even if it was several “turns” (or requests) ago. Meanwhile, background entities wrap in processes occurring out of sight: an alarm, for example, or music playing in the background.

A phone with the Gemini and ChatGPT apps displayed.


Credit: Solen Feyissa/Unsplash

Why are these important? To use Apple’s lingo, most AI assistants only have two entities to work with: information provided by the user during a request and information gleaned during training. Even then, this first variable is a bit iffy; AI is notoriously bad at remembering what users have told it, even in the short term. This problem, known as “reference resolution,” can make for a rigid and unproductive user experience when a model doesn’t properly understand a request’s context.

If they’re integrated correctly, on-screen, conversational, and background entities should enrich an AI model’s knowledge and enable it to provide more insightful answers. And, according to Apple, that’s exactly what they’ve done. Incorporating these entities into a small AI model reportedly resulted in performance “comparable to that of GPT4,” while wrapping them into a large model allowed Apple to “substantially outperform” the same competitor.

That’s fantastic news for the Cupertino-based tech company, which has set aside more than a billion dollars to catch up with its rivals in the AI race. Analysts have speculated that Apple will introduce an AI-equipped Siri upgrade along with iOS 18 later this year—a welcome advancement even for users who aren’t thrilled about generative AI, thanks to Siri’s historically abysmal performance. Apple is also said to be working on an AI-powered health coach that would help users optimize their exercise, diets, and sleep habits, though it’s unknown whether that particular feature would use a model like the one described in this paper.

View original source here.

Products You May Like

Articles You May Like

Everyone in Dragon Age: The Veilguard is too nice, and I can’t stand it
Sloths and Sequels with Writer/Producer of Slotherhouse, Bradley Fowler! [Interview]
Nvidia Lights Up Microsoft Ignite With Tools for RTX AI PCs
Werewolf By Night: Red Band – Volume 01 Issue 04
‘Bomb Cyclone’ Pounding The US Will Be Strong And Unpredictable : ScienceAlert