Welcome Here  Shenzhen Mingjiada Electronics Co., Ltd.

On October 13, 2024, Apple released the latest version of its multimodal AI Grand model MM1.5, an upgrade that significantly improves on the previous generation MM1.

Multimodal AI refers to AI technologies that can process and understand multiple types of data (such as text, images, audio, etc.). The MM1.5 model has a scale of 30 billion parameters and is capable of processing image recognition and natural language reasoning tasks. The model continues to follow the data-driven training principle, focusing on the effect of mixing different types of data in different training cycles on model performance.

The MM1.5 model not only improves OCR (Optical character recognition) capabilities, supports arbitrary image aspect ratios and resolutions up to 4M pixels, but is also good at understanding text-rich images. In addition, supervised fine-tuning (SFT) with additional high-quality multi-image data further improves the model's contextual learning and out-of-the-box capabilities. This version also enhances the ability to understand text-intensive images, visual reference and localization, and multi-image reasoning.

Apple has improved the data mixing strategy in MM1.5 model, which significantly improves the model's performance in multi-text image processing. The model provides multiple parameter scales from 1 billion to 30 billion, capable of handling a variety of complex tasks. Documentation of the model has been published on Hugging Face for researchers and developers.

Apple's AI strategy includes the development of a multimodal large model, MM1. In the field of artificial Intelligence, Apple's development history can be divided into several important stages, from the early hardware layout to the later software and ecosystem integration, and then to the recent launch of Apple Intelligence system, from closed to open, from hardware to software transformation.

Apple is increasing its investment in the field of AI, especially in the application of generative AI. For example, Apple plans to invest $1 billion a year to apply generative AI to a range of products and services such as iOS, Siri, Apple Music, and Xcode. At the Worldwide Developers Conference (WWDC) in 2024, Apple introduced a new personal Intelligence system called Apple Intelligence. The system integrates generative large models and is deeply integrated into operating systems such as iOS, iPadOS, macOS, and will be expanded to watchOS, tvOS, and visionOS in the future. Apple Intelligence has enhanced the AI capabilities on the device side and optimized the data processing and user experience through the end-to-end cloud integration architecture.

In terms of hardware, Apple focuses on AI chips and AI hardware innovation. In December 2023, Apple released its new machine learning framework MLX, an open source framework designed specifically for Apple Silicon chips to optimize the training and deployment of machine learning models on Apple chips. At the same time, Apple is also developing its own AI servers, and plans to produce 70,000 to 80,000 units in the next three years, most of which will use Nvidia chips.

On the software level, Apple is redesigning its Siri digital assistant to deliver queries in a smarter way and integrating AI chatbots into its software ecosystem. In addition, Apple is working with OpenAI to focus on the progress of its own Ajax grand model.

AI layout, in addition to self-research and acquisition, Apple acquired DarwinAI at the beginning of this year to build its artificial intelligence team. According to market research, from 2017 to 2023, Apple acquired a total of 32 AI startups, which is more than Google, Meta and Microsoft. The acquisitions include companies that specialize in speech recognition, image recognition and facial expression recognition.

Apple's strategic adjustment in the field of AI is reflected in several aspects, the company abandoned the 10-year electric vehicle project "Titan project" and shifted resources to generative AI projects.

Company website: www.hkmjd.com