Apple reveals new information about its AI

We know thanks to Mark Gurmande BloombergthatApple reportedly invested more than a billion dollarsin order to develop artificial intelligence.

This is also consistent with the allegations of the informant Revegnus who gavemore information about AI for Siri on the social network [X]. In his publication published in November, he announced that Apple was creating a large language model (LLM) in order to integrate it with Siri, specifying that it could be unveiled during theApple's WWDC 2024 conference. According to him, Cupertino would like “completely transform Siri into the ultimate virtual assistant».

A month and a half later, it was the informant eyes 1122 who spoke on this subject. He came to support the rumors spread by Revegnus, “AI-generated version of Siri on Apple's Ajax-based model to be announced at WWDC”, then specifies that “AI should add more personalization and natural conversation to Siri».

Then, new, much more serious reports, including one fromFinancial Times,came to confirm previous rumors, and provided new information. Some of these are now officially supported by Apple, thanks to the first beta version of iOS 17.4, which it has just deployed.

References to OpenAI in the code

After delving into the program, our colleagues were able to observe some astonishing lines of code. Among them in particular, references toOpenAI, who is, as a reminder, the creator ofChatGPT. This does not seem to mean that Apple would work with this type of tool, but rather thatit uses it as a comparison model for its own AI.The code further shows that Cupertino would test four separate models. We find there in particularan LLM called FLAN-T5, or eventhe AjaxGPT firm’s own model.

As stated above, we expect Appleannounces this major new feature during WWDC 2024,which should take place next June.