TL;DR:
- Apple’s closed environment and in-house hardware and software strategy give it an edge in AI/ML development.
- The company focuses on integrating AI features into its devices, emphasizing on-device processing and transformer models.
- Apple’s control over the software and on-device silicon reduces the strain on its data centers.
- Comparatively, Microsoft relies on GPUs in data centers, while Apple prioritizes on-device processing.
- Apple utilizes on-device ML for autocorrect, Suggestions feature, and videoconferencing in the Vision Pro headset.
- The company’s homegrown M2 Ultra chip enables the training of massive ML workloads, surpassing discrete GPUs’ memory limitations.
- Apple sees AI/ML as a means to enhance existing functionality rather than a standalone feature.
- Google also explores on-device functionality for accelerated AI on client devices.
Main AI News:
In the ever-evolving landscape of artificial intelligence and machine learning, Apple is carving out a distinct path with its closed environment and in-house hardware and software strategy. Unlike its counterparts Microsoft and Facebook, Apple doesn’t need to invest in massive AI factories equipped with an army of GPUs. This approach allows Apple to maintain tight control over its technology stack and ensure seamless integration between hardware and software.
At the recent Worldwide Developer Conference, Apple surprisingly refrained from delving into its machine learning strategy while unveiling its new mixed-reality headset. Instead, the company emphasized its commitment to incorporating real-world AI features into its devices. The focus was on a crucial aspect: bolstering on-device processing capabilities and transformer models to power machine learning applications.
Apple’s unique advantage lies in its meticulous optimization of AI software to align perfectly with its hardware capabilities. By doing so, the company alleviates the burden on its data centers, reducing the need to process and store vast amounts of raw input in the cloud. “Considering Apple’s investment in ML acceleration in Apple Silicon, the strategy appears to prioritize on-device processing, to minimize the amount of data Apple needs to process and store in cloud services,” explained James Sanders, principal analyst for cloud, infrastructure, and quantum at CCS Insights.
In contrast, Microsoft has chosen to equip its data centers with GPUs to facilitate ML services for client devices, relying heavily on off-the-shelf hardware. Apple, on the other hand, has adopted a different approach. The company has implemented a new transformer model and on-device machine learning in iOS 17, bringing features like autocorrect to new heights. Apple boasts that autocorrect in iOS 17 is “state of the art for word prediction, making autocorrect more accurate than ever.”
Furthermore, Apple has been actively integrating machine learning directly into its applications. For instance, the newly introduced Suggestions feature on the iPhone provides users with recommendations for memories to include in the Journal application. Similar to autocorrect, this on-device feature leverages real-time text input that is fed into the on-device ML, which then generates personalized suggestions.
Apple’s visionary Vision Pro headset also harnesses on-device machine learning for its videoconferencing capabilities. With the headset, users can create an animated, lifelike “digital persona” to represent them during video calls in the mixed reality space. This innovative approach eliminates the awkward appearance of a person wearing a headset, ensuring a more natural and immersive experience. The technology behind this functionality utilizes advanced machine learning techniques to create and simulate movement for the digital persona.
During the WWDC speech, Mike Rockwell, Apple’s vice president, shed light on the intricacies of the headset, stating, “The system uses an advanced encoder-decoder neural network to create your digital persona. This network was trained on a diverse group of thousands of individuals. It delivers a natural representation, which dynamically matches your facial and hand movement.”
Apple’s dedication to pushing the boundaries of machine learning is evident in its use of the homegrown M2 Ultra chip, prominently featured in the Mac Pro. The M2 Ultra chip excels at handling massive ML workloads, even surpassing the capabilities of the most powerful discrete GPUs due to its extraordinary memory capacity. This emphasis on powerful, specialized hardware showcases Apple’s commitment to enabling cutting-edge machine learning capabilities across its product line.
While AI capabilities are already widespread on mobile devices, often equipped with neural chips for inferencing, Apple stands apart by integrating AI/ML into existing functionalities or adding new features to its apps. It doesn’t treat AI/ML as a standalone feature but rather as a means to enhance the overall user experience. Sanders notes that Apple is exploring generative AI, although it has not yet been transformed into a marketable product.
Google, too, recognizes the potential of on-device functionality for accelerating AI on client devices. The company has incorporated hardware capabilities in its Pixel devices to synergize with cloud-based AI and ML features. Additionally, Google announced in April that it would enable WebGPU, a protocol that leverages hardware resources on client devices, enabling faster AI processing.
Recently, Apple unveiled ByteFormer, a new multimodal transformer model designed to process audio, video, and other inputs in a contextual manner. This model transforms various input modalities, such as audio, visual, or text, into bytes, enabling the development of models that can seamlessly operate across multiple input formats. It’s worth noting that ByteFormer was trained on Nvidia A100 GPUs, although it remains unclear whether these GPUs were housed in Apple’s own data centers.
Apple’s commitment to advancing AI and ML extends beyond hardware and software optimization. The company has been actively working on Stable Diffusion, a port compatible with the GPU and ML cores of its M-series chips, found in Mac and iPad devices. While not yet feature-complete, Stable Diffusion has made significant progress, including support for ControlNet and image inpainting.
Conclusion:
Apple’s strategic focus on homegrown silicon and seamless integration between hardware and software positions the company at the forefront of AI/ML development. By prioritizing on-device processing and optimizing software to hardware capabilities, Apple reduces reliance on cloud services and delivers powerful machine learning experiences directly to its users. This approach not only enhances existing features but also opens doors for innovative applications. As the market evolves, Apple’s unique AI/ML strategy strengthens its competitive advantage and reinforces its commitment to delivering cutting-edge technology experiences.
