Apple is widely expected in the industry to launch important new artificial intelligence features in iOS 18 in June. Code discovered by 9to5Mac in the first iOS 17.4 beta shows that Apple is continuing to develop a new version of Siri powered by large language model technology, with some help from other sources.
In fact, Apple appears to be using OpenAI's ChatGPT API for internal testing to help develop its own artificial intelligence models.
According to this code, iOS17.4 includes a new SiriSummarization private framework that calls OpenAI’s ChatGPTAPI. This appears to be a tool Apple uses internally to test new AI features.
There are also multiple system prompt examples of the SiriSummarization framework in iOS17.4. These include things like "Please summarize," "Please answer this question," and "Please summarize the given text."
The system prompt also mentions what to do when it receives input in the form of iMessage or SMS. This is in line with previous reports from Bloomberg, which said Apple was developing artificial intelligence integration in the Messages app that could "answer questions and automatically complete sentences."
Given a text message received by the user that contains sender, content, and sent time fields, it is recommended that the voice assistant take an appropriate action, including action type, action value, action value type, and confidence level in JSON format. Possible operation types are MessageReply, GetDirection, Call, SaveContact, Remind, MessageContact and None. Possible operation value types are message, address, phone number, contact, and reminder. Possible score values range from 0 to 1, indicating the confidence score of the recommended action.
Apple is unlikely to use OpenAI models to support artificial intelligence features in iOS 18. Instead, what it's doing here is testing its own artificial intelligence model against ChatGPT.
For example, the SiriSummarization framework can use on-device models for summarization. Apple appears to be using its own artificial intelligence model to support this framework, and then internally comparing its results to those of ChatGPT.
The code for iOS 17.4 shows that Apple is testing four different artificial intelligence models. iOS17.4 shows that AjaxGPT has two versions, including a version that is processed on the device and a version that is not processed on the device.
Other models referenced by iOS 17.4 include the aforementioned ChatGPT and FLAN-T5. The biggest takeaway from these findings is that Apple is stepping up its efforts to integrate large language models into iOS. It's worth noting that Apple is also developing its own system and comparing its results to systems such as ChatGPT and FLAN-T5.
In October of this year, Bloomberg’s Mark Gurman elaborated on some of Apple’s artificial intelligence goals in iOS 18. The report outlines that there is "an edict" within Apple and Craig Federighi's software team to "include functionality that runs on the company's large language model" in iOS 18.