Researchers have found that Apple’s new artificial intelligence system, ReALM, surpasses the capabilities of OpenAI’s GPT-4.

The paper titled “ReALM: Reference Parsing as Language Modeling” studies the problem of reference parsing. Reference is a linguistic process in which one word in a sentence or utterance refers to another word or entity. The task of resolving these references is called reference resolution.

Researchers say that while large language models (LLMs) are extremely powerful for a variety of tasks, their use in reference parsing, especially for non-conversational entities, remains underutilized.

Research shows that the smallest version of ReALM, benchmarked against GPT-3.5 and GPT-4, performs on par with GPT-4, while larger models significantly outperform GPT-4.

Ahead of WWDC 2024 and the expected June release of iOS 18, expectations are high for the debut of the advanced Siri 2.0. It is still uncertain whether ReALM will be integrated into Siri by then.

Apple’s recent forays into artificial intelligence haven’t gone unnoticed, and have been marked by the launch of new models and tools aimed at improving the efficiency of AI on small devices, as well as strategic partnerships. These developments underscore the company’s strategy to put artificial intelligence at the forefront of its business operations.

The launch of ReALM represents the latest and most targeted effort by Apple’s artificial intelligence research team to refine and accelerate existing models, driving them to greater speed, intelligence, and efficiency.

Key features of Apple ReALM AI

ReALM reportedly uses a new method of converting screen information into text, allowing it to bypass the need for image recognition parameters and enable more efficient processing on AI devices.

It also takes into account what’s on the user’s screen or running in the background.

Therefore, LLM should allow users to scroll through a website and instruct Siri to call a business. Siri can then “see” the phone number on the website and call it directly.

Therefore, ReALM can significantly improve the context awareness of voice assistants. Updates to Siri are able to interpret on-screen information and use additional context to help provide a smoother, hands-free user experience.

ReALM can also handle a variety of references, including those that rely on conversational context, screen content, and even background information. This is critical to developing AI systems that are more intuitive and responsive, able to adapt to the complexities of human language and context.

The paper reports huge improvements over existing systems with similar capabilities, as its smallest model apparently achieves an absolute gain of more than 5% in screen reference.

Featured Image: Canva


#Apples #model #ReALM #GPT4

Leave a Reply

Your email address will not be published. Required fields are marked *