Despite reportedly discussing a deal to license Google’s Gemini AI, Apple is pushing ahead with significant AI research. A new paper from the company details a new method for training large language models (LLMs) with seamless integration of text and visual information.
The ‘MM1: Methods, Analysis & Insights from Multimodal LLM Pre-training‘ paper focuses on a new approach to creating more flexible, intelligent AI systems. Specifically, Apple’s research looked at combining different types of training data and model architectures.
Doing so gives the resulting AI the ability to understand and generate language based on a mix of visual and language cues. That capability is critical for complex tasks like questions involving visual elements.
Apple’s MM1 model also has excellent in-context learning capabilities, according to the paper. This is especially true for the largest configuration of the model, with 30 billion parameters. It can apparently handle multi-step reasoning over multiple images with techniques that allow AI to do complex, open-ended problem-solving with minimal examples.
Those interested can read the full paper here.
MobileSyrup may earn a commission from purchases made via our links, which helps fund the journalism we provide free on our website. These links do not influence our editorial content. Support us here.
