With the advent of LLMs and AI, Apple faces an unprecedented set of challenges. The race to integrate AI into consumer and professional products is intensifying, placing Apple before the need not only to adopt these technologies but also to adapt them to its unique and secure ecosystem, while preserving standards high levels that its users expect.
Apple is currently overwhelmed by what is happening in the field of artificial intelligence, with OpenAI, Meta, Microsoft, Google, Mistral… But we already know what Apple is up to. The American firm wants to integrate artificial intelligence (AI) tools like ChatGPT directly into its devices.
The objective? Catch up in the generative AI sector while offering significant innovations to its users. A recent document, “ LLM in a Flash”, highlights this ambition by describing how these large language models could work on devices with limited memory capacity.
Running LLMs directly on smartphones
To understand Apple’s ambition, you need to understand what it means to run LLMs directly on an iPhone. An LLM, for Large Language Model , is a type of AI specialized in understanding and generating human language. Until now, these models required significant computing resources, generally available only in the cloud. Apple wants to change that by optimizing these models to work efficiently on its devices, despite memory, power, and compute constraints.
Inference is the technical term for the process by which an AI model responds to user input. Optimizing inference on devices like the iPhone means users will be able to get fast, intelligent answers straight from their phone, without relying on a constant internet connection to the cloud.
The future of AI on mobile devices
Apple is not the only company that wants this autonomy of intelligence. Other tech giants, like Google with its Pixel 8 Pro and Gemini Nano, are also working to embed AI capabilities directly on devices. Microsoft introduced Phi-2, which is also a small language model. This trend shows a shift in manufacturers’ strategy: to deliver faster, private, and integrated AI experiences.
For users, this could mean a revolution in the way we interact with our devices. Imagine asking your iPhone complex questions and receiving answers instantly, or having a personal assistant who understands and anticipates your needs more intuitively. This should make what Siri, Google Assistant or Amazon Alexa offer today as an experience outdated. Siri, Apple’s assistant, would become a much more powerful and useful tool.
Exactly when Apple will launch these new AI features remains unknown. Although on the surface Apple maintains a certain reserve, behind the scenes, the excitement caused by ChatGPT seems to have taken them by surprise.
They are now actively exploring ways to incorporate generative AI into a wide range of applications. At the same time, Apple’s intensifying research efforts are evident. Ars Technicareveals that the company has published a second significant document in a short period of time. Furthermore, according to theNew York Times, Apple is in talks to access the archives of several news publishers, aiming to use them to train its AI models, with potential deals worth up to $50 million. Apple not only has a lot of money, it has many other assets.
Apple’s capabilities
First, Apple’s expertise in ARM chip design represents an undeniable strategic advantage, particularly in developing specialized chips suitable for running large-scale language models (LLM).
ARM chips, known for their energy efficiency and performance, are at the heart of Apple devices and enable deeper and optimized integration of AI technologies. By developing custom chips, Apple will be able to optimize every aspect of LLM operation, from query processing to memory management and power consumption, to provide a smooth and responsive user experience, even for data processing tasks. complex language.
Furthermore, Apple enjoys another significant advantage: its ability to deploy software updates on a massive scale. Unlike OpenAI, which relies on partnerships and third-party platforms to distribute its models like ChatGPT, Apple can potentially integrate and update its own LLM directly through iOS to hundreds of millions of active users. This rapid and universal deployment capability, coupled with the existing infrastructure of the Apple ecosystem, uniquely positions the company to drive the adoption and integration of generative AI on an unprecedented scale.
The LLM challenge for Apple
However, despite these advantages, Apple faces considerable challenges. Although it is a key player, Apple is lagging behind in the race for generative AI, an area already heavily invested by other technology giants.
Additionally, Apple’s philosophy around data privacy and security presents unique challenges for integrating LLMs, which typically require processing large amounts of data, often in the cloud. Apple therefore faces the challenge of developing a powerful and efficient LLM that can run locally on resource-constrained devices, while adhering to its strict privacy policies. Faced with these challenges, Apple is giving up.