While rumors swirled about Apple potentially using Google’s AI technology, Apple has been quietly building its own. In a recent research paper, the company offered a glimpse into the development of its innovative MM1 AI model.
Imagine an AI that can not only understand your words but also the pictures you show it. That’s the goal behind MM1. Apple plans to train the model on a massive dataset of text documents with images, image captions, and even standalone text. This, they believe, will allow MM1 to excel in tasks like generating image captions, answering questions about pictures, and even making logical inferences based on combined visual and textual information. The aim? Unmatched accuracy in these areas.
By using a diverse training approach, Apple is essentially teaching MM1 multiple languages: the language of words and the language of images. This will allow the AI to not only understand but also potentially generate text based on what it “sees” in a picture.
Apple acknowledges the fierce competition in the AI landscape, with companies like Google and OpenAI already making significant strides. Their strategy? Leverage existing AI training methods while adding their unique twist. Just like with their hardware and software design, Apple is taking an independent path to achieve reliable and competitive AI. This research paper offers a fascinating window into their approach, and it will be exciting to see how MM1 develops in the future.
