Apple's new AI training method preserves privacy and could make future Siri more flexible

Apple researchers have found a new multimodal method for quickly training large language models (LLMs) that could enable more flexible and powerful machine learning systems and types “AI”.

A research paper published by the company on research site earlier this week revealed that Apple used what it calls a “careful combination” of image captions, interleaved image-text, and text-only data to train the LLM. . Combining visual and language data allowed the models to solve problems such as creating intelligent image captions or inferring meanings in natural language.

The study found that the choice of image encoder and the resolution of the images it processes has a large impact on performance, more so than the design of the visual language connector.

In one case using the MM1 model with 30 billion parameters, it was found that there was a strong ability for contextual learning. The discovery means it can perform multi-step reasoning on multiple images with few chain-of-thought clues.

According to Venturebeat, Apple continues its tradition of being a “fast follower” rather than a “first mover” when it comes to disruptive technologies. CEO Tim Cook recently admitted that the company is spending $1 billion a year to integrate artificial intelligence into existing technologies.

Cook said the company will share “details of our ongoing work in AI later this year.” Apple is expected to make some announcements about its achievements at WWDC this June.

The company is catching up with competitors in the use of technologies related to artificial intelligence. The company is also developing methods that will preserve user privacy while expanding existing machine learning capabilities.

The latest concerns about privacy and security have not been common among existing chatbot-type services and complicate things for Apple.

Apple's interest in multi-model neural network training has led to state-of-the-art performance that enables multi-step reasoning. This suggests that the company has found a path to rapidly developing machine learning capabilities, as well as providing them with enhanced intelligence.

Leave a Reply

Your email address will not be published.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