Apple Using ChatGPT to Test AI-Powered Siri, Shows Code
The code indicates that Apple is developing a new version of Siri, enhanced by large language model technology, and is testing its features—by using OpenAI’s ChatGPT API for internal purposes.
The beta version includes a ‘SiriSummarization’ private framework, which actively utilizes OpenAI’s ChatGPT API. This is part of Apple’s internal testing strategy to refine its AI capabilities. The framework involves various system prompts, such as “please summarize,” “please answer this question,” and others related to processing iMessage or SMS inputs. Bloomberg previously said Apple’s AI plans within the Messages app would be for fielding questions and auto-completing sentences.
The SiriSummarization framework is designed to suggest appropriate actions based on the user’s received SMS, incorporating details like sender, content, and send time. It is programmed to provide suggestions in JSON format, including action types such as MessageReply, GetDirection, Call, SaveContact, Remind, and others, with a confidence score ranging from 0 to 1.
Despite these tests, Apple is unlikely to employ OpenAI models directly in iOS 18. Instead, the company is focusing on developing and refining its own AI models. The SiriSummarization framework, for instance, performs tasks using Apple’s on-device models, which are then internally compared against the results from ChatGPT.
iOS 17.4 code suggests that Apple is examining four different AI models. This includes Apple’s internal model, ‘Ajax,’ as previously reported by Bloomberg. There are two versions of AjaxGPT in the code: one processed on-device and another that is not. Other models referenced in the code include ChatGPT and FLAN-T5.
Let’s see if Apple will actually be able to make Siri great again (fingers crossed).