iOS 17.4: Apple continues to work on Siri and Messages features with artificial intelligence support with ChatGPT

Apple is widely expected to introduce new artificial intelligence features in iOS 18 in June. Code found by 9to5Mac in the first beta of iOS 17.4 shows that Apple is still working on a new version of Siri based on large language model technology, with a little help from other sources.

In fact, Apple appears to be using OpenAI's ChatGPT API for internal testing to help develop its own AI models.

According to this code, iOS 17.4 includes a new private SiriSummarization framework that calls on OpenAI's ChatGPT API. Apparently, this is what Apple is using for internal testing of its new AI features.

iOS 17.4 also includes several examples of system prompts for the SiriSummarization framework. These include "please summarize", "please answer this question", and "please summarize this text".

The system's prompts also tell it what to do when it receives input in the form of an iMessage or SMS. This is in line with the announcement that Apple is working on integrating artificial intelligence into the Messages app that will be able to "answer questions and automatically complete sentences."

Given the SMS received by the user, containing the fields sender, content, time sent, suggest the appropriate action to the voice assistant, including action type, action value, action value type, and trust score in JSON format. Possible action types: MessageReply, GetDirection, Call, SaveContact, Remind, MessageContact, and None. Possible action value types (ActionValueType) are message, address, phone number, contact, and reminder. The possible evaluation value ranges from 0 to 1 and represents the confidence rating of the proposed action.

Apple is unlikely to use OpenAI models for its artificial intelligence features in iOS 18. Instead, the company is testing its own AI models on ChatGPT.

For example, the SiriSummarization framework can perform generalization using models on the device. Apparently, Apple uses its own AI models to run this framework and then internally compares its results with those of ChatGPT.

In total, the iOS 17.4 code indicates that Apple is testing four different AI models, including an internal Apple model called "Ajax." iOS 17.4 shows that there are two versions of AjaxGPT, including one that is processed on the device and one that is not.

Other models referenced in iOS 17.4 include the aforementioned ChatGPT, as well as the FLAN-T5.

The most important conclusion from these results is that Apple is increasing its efforts to integrate large language models into iOS. It is also noteworthy that Apple is simultaneously developing its own system and comparing its results with models such as ChatGPT and FLAN-T5.

In October, Mark Gurman detailed some of Apple's goals for AI in iOS 18. The report said that Apple and Craig Federighi's development team "have an edict" to fill iOS 18 with "features that work based on the company's big language model."