Apple's LLM has four variants, and even the smallest one has similar performance to OpenAI's ChatGPT. This information comes from Apple's research paper, shared via Windows Central.

ReALM is an LLM developed by Apple specifically to enhance the Siri assistant, providing better contextual understanding. Notably, the model is capable of processing “on-screen” information, meaning the content on the screen. This gives it a significant advantage over ChatGPT. OpenAI’s AI can process image files and PDF files, but it cannot read the entire screen.

realm 1712071268553222101487.jpg
Apple is the only tech giant that has yet to reveal its internal AI efforts to the public.

Apple is said to be releasing four variants, including: ReALM-80M, ReALM-250M, ReALM-1B and finally ReALM-3B. The letters M/B behind represent the reference number contained in the model (million and billion).

The researchers tested ReALM’s performance against OpenAI’s GPT-3.5 and GPT-4. They found that even the smallest ReALM-80M variant was comparable to GPT-4, while higher-end models outperformed it.

Research shows that ReALM has large improvements over similar systems, and the smallest model (80M) also processes on-screen information 5% better.

Apple is the only tech giant that has so far remained silent about the AI ​​technology trend that has exploded since ChatGPT appeared.

Last week, Apple set the date for its Worldwide Developers Conference (WWDC) and it is likely that ReALM will appear on the agenda this coming June.

(According to Dtrends)

Apple makes extremely difficult demands on suppliers for iPhone 16 iPhone 16 will have the thinnest screen bezel ever, but this difficult demand from Apple could affect the new iPhone launch plan and increase product costs.