Apple asserts superiority over GPT-4 with new AI incorporating on-screen content and background context
Apple’s team of AI researchers introduces Reference Resolution As Language Modeling (Realms), claiming its ability to surpass GPT-4 in specific tasks. Published on the arXiv preprint server, their paper unveils Realm enhanced information-gathering capabilities.
Despite the dominance of Large Language Models (LLMs) like GPT-4 in recent years, Apple’s Siri digital assistant has notably lagged in artificial intelligence advancements. With Realm, Apple aims not only to catch up but to lead, asserting its outperformance over other publicly available LLMs for certain queries.
In their research, Apple’s team elucidates how Realm delivers more precise responses by leveraging ambiguous on-screen references and accessing conversational and background data. By scrutinizing the user’s screen and active processes on the device, Realm discerns contextual clues preceding the query, enhancing its search process and likelihood of providing relevant information.
Tests against multiple LLMs, including GPT-4, validate Realm’s superiority in specific tasks. Apple plans to integrate Realm into its devices, augmenting Siri’s capabilities for users who upgrade to iOS 18, slated for release this summer.
