Apple is actively seeking partnerships to enhance its generative AI capabilities. According to recent reports, the company will use Ernie 4.0, a generative model from Chinese tech giant Baidu, to power its iPhone 16 and iOS 18. The report added that Apple initially contacted Alibaba but ultimately selected Baidu. As a result, Baidu’s stocks climbed 6% in Hong Kong.
One reason why Apple is interested in using Baidu’s model is that generative AI models require government approval before widespread use in China. In the first six months since authorities began the approval process, more than 40 AI models, including Baidu’s Ernie bot, have been approved for public use in China.
This is not the first time that Apple has sought partnerships. Previously, discussions have been held with OpenAI and Google for GPT-4 and Gemini, respectively. On the surface, Apple has given up on building its own LLM and instead plans to outsource generative models for its devices.
So far, so good
During the annual shareholder meeting, CEO Tim Cook said that the company will disclose more about its plans to implement generative AI ‘later this year’ and was optimistic that Apple will “break new ground” on GenAI this year.
Since testing an internal chatbot nicknamed ‘Apple GPT’ by its employees, Apple has come a long way in generative AI. It recently discussed multi-year deals worth at least $50 million with news publishers to train its generative AI models.
Lately, Apple has dedicated significant research efforts to develop an in-house LLM. Its recent release MM1 comprises a family of multimodal models featuring dense variants up to 30B and mixture-of-experts (MoE) variants up to 64B. These models excel in processing and comprehending both text and images, enabling them to identify objects, scenes, and even relationships between elements.
Last year, Apple open-sourced a multimodal generative AI model called Ferret, which can understand and generate responses based on images and text. Furthermore, Apple recently acquired Canadian AI startup DarwinAI and added dozens of Canadian company staffers to its AI division.
Earlier this year, Apple shelved its electric vehicle (EV) dreams to focus on generative AI. As part of this shift away from EV, Cook announced that Apple would reassign ‘many employees working on its cars’ to generative AI projects within its AI division.
Apple’s moat lies in its hardware
Unlike other major players in generative AI, such as Microsoft, Google, and AWS, Apple is not primarily a cloud service provider where LLMs have extensively found use cases. Instead, Apple’s focus lies predominantly in on-device generative AI solutions.
Many speculate that Apple will announce something huge regarding Siri at WWDC 2024 in June. The new AI-powered Siri is expected to enable more human-like, contextual conversations and likely offer greater personalisation based on the user’s preferences and habits.
Apple’s strength lies in its hardware. Its chips, specifically the M3 Max and the A17 Bionic, show strong potential for running generative AI applications on edge. Last year, Apple open-sourced MLX, an array framework for machine learning on Apple silicon. The company shared examples of MLX in action, performing tasks like image generation using Stable Diffusion on Apple Silicon hardware.
Currently, apart from Apple, Samsung too is heavily invested in AI. It is the first to partner with Google Cloud to deploy Gemini Pro and Imagen 2 on Vertex AI via the cloud for their Galaxy S24 series. Unlike OpenAI, Google has specifically developed an LLM called Gemini Nano for Android phones.
It makes sense for Apple to partner with Google as the search giant is already paying Apple billions of dollars yearly to be the default search engine in Safari on Macs, iPads, and iPhones. This could be a great extension of the partnership and could lead to a better deal as well.
Apple is poised to enhance its devices, yet it’s wise of them to steer clear of the AGI race. Pedro Domingos sums it up: AGI for Apple means ‘Apple’s Giving up on Intelligence’.