According to Counterpoint, with almost 1 billion generative AI (genAI) enabled smartphones set to be launched by 2027, the likelihood of Apple entering the competition with edge-based Apple GPT in its smartphones is increasing.
The company has faced criticism for seeming to enter the genAI arena late. This is evident as even Microsoft Copilot (along with built-in ChatGPT) is now accessible as an iPhone application.
Intentional, Strategic, and Somewhat Slow
Apple has addressed the technology, highlighting the significant amount of system intelligence present in its devices and outlining its “intentional” plans to integrate AI into its products. The implication is that any extensive deployment of such advanced technology may be purposeful to avoid unforeseen consequences.
The company is discreetly investing billions in R&D related to these systems, including AI partnerships with media publishers, with these statements serving as a strategic delay tactic.
Reportedly, Apple aims to introduce a more intelligent, AI-driven Siri and the widespread integration of genAI features across its applications through an internal initiative named “Ajax.”
R&D Progressing Swiftly
The company seems to be making strides. Jeff Pu from The Information suggests that Apple is gearing up to unveil this enhanced Siri by the year-end, strategically aligning with Counterpoint’s projected market expansion. (The current estimate indicates the shipment of 100 million smartphones with on-device genAI capabilities this year.)
The challenge with genAI lies in its server-based nature, often demanding substantial storage and data resources. For instance, when executing a GenAI request on your phone using Microsoft Copilot today, the task is offloaded to the server for processing, with the response relayed back.
This setup is unsuitable for the following reasons:
Apple aims to handle requests directly on the device’s top system without server reliance, emphasizing privacy, security, and sustainability.
Apple’s Initiatives
To achieve this goal, Apple’s R&D teams have unveiled a significant breakthrough enabling iPhones and other Apple devices to execute computationally intensive LLM (large language models) tasks on the device itself.
The researchers claim that their innovation not only overcomes a recent computational bottleneck but also sets benchmarks for future research. Strategies like this are deemed essential for unlocking the full potential of LLMs across various devices and applications as they grow in size and complexity.
It appears that domestic advancements are rapidly unfolding.
Additionally, Apple’s machine learning (ML) teams have recently introduced a new ML framework for Apple Silicon, known as MMX or LM Explore. This development coincided with Apple’s collaboration with Cornell University to release the open-source “Ferret” LLM, facilitating image generation for these devices. To delve deeper into the implications, refer to this article.
A concept devised by Apple’s R&D teams can generate avatars from movies 100 times faster than other systems.
Apple’s Leading Edge in GenAI
In essence, Apple is developing practical task-oriented LLM tools capable of on-device operation.
This progression is not unexpected. As AI becomes more pervasive in society, Apple’s approach is not to establish a vast infrastructure to disseminate information comprehensively. Instead, incorporating on-device AI into its products makes more sense, offering Apple’s teams the opportunity to dominate the market if they can execute this objective more effectively than their current approach to 5G modem development.
Possibly.
Meanwhile, rival products utilizing server-based services are also advancing, with their Siri equivalents generating images, providing intricate responses, and more. Given Apple’s concerted efforts, many industry experts anticipate that the company will showcase initial outcomes of this extensive research at WWDC 2024, highlighting a significant shift for a company that initially led in on-device AI capabilities.
