Apple’s new ReALM AI model could make Siri faster and smarter than GPT-4

For the longest time, Apple has shied away from artificial intelligence, giving several other industry bigwigs (and competitors) such as OpenAI, Google, and Microsoft a crucial first movers’ advantage. That’s set to change, however, for Apple is planning to finally enter the generative AI space, as recently confirmed by CEO Tim Cook.

What is it’s offering, you ask? Well, the iPhone maker has just unveiled a new small language model called ReALM (Reference Resolution As Language Modeling), designed to make its voice assistant, Siri, both smarter and faster. Here’s all we know.

How does ReALM work?

Now, ReALM being a Small Language Model (SLM), is naturally tiny compared to LLMs such as Gemini Pro and Gemini Ultra, GPT-4, and others. This is because unlike powering a generative AI platform, ReALM provides other AI models with context.

ALSO READ:  Apple’s new AI-based image editing tool lets you edit pictures using textual prompts

How it does this, is by reconstructing entire screens, and labeling each icon, image, or anything else on that screen. This ten creates a text-based representation of the visual layout, which can be passed on to voice assistants such as Siri to provide it with context and clues for queries and commands from users.

Despite being exponentially smaller then, Apple claims that ReALM can surpass GPT-4 in terms of performance since the latter can process image files and PDFs, but can’t read your screen.

With this, at least for now, it looks like Siri will be smarter and faster than ever before since it will now be able to figure out a context within conversations with users.

When can we expect ReALM to be rolled out?

Details regarding the model were first published in an open research paper by the AI research and development team from Apple.

ALSO READ: Apple reportedly working on new AI features, including a ‘smarter’ Siri

Rumour has it that the company could officially unveil ReALM, and with it, effectively, “Siri 2.0” at the Worldwide Developers’ Conference (WWDC) 2024, which kicks off on June 10. However, there has been no confirmation from the company regarding the same as of yet.