Apple has been facing some challenges in integrating AI with its voice assistant, Siri. Recently, Siri has been experimenting with describing images received in Messages when using CarPlay or the announce notifications feature, as reported by 9to5Mac.
Nevertheless, Apple remains committed to advancing AI. In a newly published research paper, Apple’s AI experts discuss a system that goes beyond image recognition. Interestingly, they claim that one of their models outperforms ChatGPT 4.0, a well-known large language model.
The paper introduces ReALM (Reference Resolution As Language Modeling), a system designed to enhance the capabilities of a voice assistant. ReALM considers both the content on the user’s screen and the ongoing tasks.
It categorizes entities into three types: on-screen entities, conversational entities, and background entities.
Apple believes that ReALM could significantly improve Siri’s intelligence and usefulness. Furthermore, Apple is confident in the system’s speed and usability. In benchmarking against OpenAI’s ChatGPT 3.5 and ChatGPT 4.0, Apple’s on-device AI shows promising results.
The paper demonstrates substantial improvements over existing systems, with Apple’s smallest model achieving performance that is comparable to ChatGPT 4.0. Additionally, larger models outperform ChatGPT 4.0 by a significant margin.
Apple aims for on-device performance without sacrificing quality. This approach aligns with their commitment to user privacy and security. The upcoming iOS 18 and WWDC 2024 on June 10 may unveil further developments in Apple’s AI technology.
