Apple Partners with OpenAI for On-Device AI in iOS 18, Enhancing Privacy

Apple Inc. has reignited discussions with OpenAI to integrate advanced AI features into iOS 18, utilizing large language models entirely on-device. According to a Bloomberg report, this move promises to significantly enhance user privacy and system responsiveness.

Apple Explores OpenAI Collaboration for iOS 18 AI Features, Balancing Online and Offline Capabilities

According to a 9to5Mac report, Apple has "renewed discussions with OpenAI" about adopting its technologies to power at least some of the new AI features available on the iPhone and iPad later this year. According to sources familiar with the topic, Apple intends to integrate OpenAI's features into iOS 18.

According to Bloomberg's Mark Gurman, negotiations are focused on adopting OpenAI technologies to power an AI-based chatbot in iOS 18. It's unclear whether this chatbot is meant to replace Siri or if Apple will launch it as a new virtual assistant. 9to5Mac previously discovered references to a new iOS "SiriSummarization" framework that works with ChatGPT.

At the same time, it appears Apple is still conversing with Google about adopting Gemini on iOS. The organization is most likely seeking the most cost-effective option. Last month, The New York Times confirmed these conversations. Apple is also said to be talking with Baidu for users in China, where Google has no operations.

The business plans to deliver several generative AI-based features with iOS 18 and other software updates, which will be announced in June at WWDC 2024. Apple has also been working on its language models to run offline. However, these models have limitations, therefore having online alternatives would be excellent for providing more accurate findings.

A source recently theorized how Apple would mix the online and offline models, possibly even putting some AI-based functions behind a paywall.

Apple's AI Strategy Leverages On-Device Models for Enhanced Privacy and Responsiveness in iOS 18

As seen by most of the research data Apple has published recently, the corporation is heavily investing in several types of artificial intelligence technologies. In last month’s report, Apple will unveil its AI plan in June at WWDC, alongside iOS 18 and other new OS updates.

According to Mark Gurman's weekly Power On newsletter, Apple's offline, on-device, huge language model will power the new iPhone AI features. You should expect Apple to highlight the advantages of this strategy, such as privacy and speed.

9to5Mac previously discovered code references in iOS 17.4 to an on-device model known as "Ajax." Apple is also working on server-hosted versions of Ajax. On-device LLMs have the disadvantage of being less powerful than models operating on massive server farms with tens of billions of parameters and constantly changing data.

To make the most of an on-device approach, Apple engineers can most likely leverage its platforms' complete stack vertical integration with software tuned to the Apple silicon chips within its devices. On-device models typically react far faster than routing a request through a cloud service, and they have the added benefit of working offline in areas with no or limited access.

While on-device LLMs may not have the same embedded rich reservoir of information as ChatGPT to answer questions about random trivia facts, they can be tailored to perform well in a variety of jobs. Consider how an on-device LLM could provide smart auto-replies to Messages or improve the interpretation of many typical Siri inquiries.

On-device models may also be capable of performing generative AI activities such as document or image production using prompts with reasonable results. Apple can still partner with a firm like Google to fall back on something like Gemini on the server for specific tasks.

It also aligns well with Apple's strict privacy policies. Running all your downloaded emails and text messages through an on-device model is safe, as the data remains local.

Photo: Bangyu Wang/Unsplash