Siri will open up to ChatGPT competitors come early summer. (Picture: generated)Previously, Siri would hand off more complex questions to ChatGPT when it couldn’t handle it itself — but that’s about to change, according to Bloomberg (paywalled).
Starting in June, if users have Gemini or Claude installed on their phones, Siri will be able to use those bots instead, by recording their preferred «Extension» in Settings.
That would end the ChatGPT monopoly that OpenAI has enjoyed since 2024, and opens up the chatbot ecosystem to other players, likely staving off regulators.
Opening up the platform is for the system level Siri queries native to iOS itself, and must not be confused with the standalone Siri app, which will use Gemini in a billion dollar deal.
With Gemini running on Apple’s own servers, they have wide access and permission to customize it. (Picture: generated)With Google’s bespoke Gemini model running on their internal servers, Apple will have full access to the AI, The Information (paywalled) writes.
That entails that they can run «distillation» on the model, meaning they can use it to provide answers and reasoning over a wide array of tasks and use that to train smaller, more capable Apple models, MacRumors says.
Distillation is a controversial technique, and many of the big AI labs have been accusing Chinese startups of doing it to make their own models more capable.
Expect more advanced AI features in June, not spring, a new report says. (Picture: generated)More detail is coming out on the Apple-Google AI deal, and it is clear it will come with absolutely no Google or Gemini type branding or references on the Apple side.
The LLM will supposedly function like any other, and be better at things such as scanning your Calendar app for upcoming events and your Contacts for sending messages. It should also be better at providing «emotional support» through «conversational responses» according to a report from paywalled The Information, seen by 9to5Mac.
The Google model will not be in a glass house over at Googleplex, but rather sit on Apple’s Private Cloud Compute servers, and, according to The Information, Apple can choose to ask Google to implement improvements to it, or tinker with the model themselves — thereby evolving it to their particular needs.
Timewise, «some features» will launch already this spring, but others — like remembering past conversations, or things like warning of the weather ahead of an Apple Calendar event, won’t be launching until the WWDC in June, apparently.
You’ll find Google’s AI under the hood of your Apple devices in the near future. (Picture: generated)In a joint statement, Apple says that «Google’s Al technology provides the most capable foundation for Apple Foundation Models,» and will now be driving AI for Apple.
All of Google/Apple’s models will continue to run on Apple’s Private Cloud Compute, ensuring industry leading safety and privacy for queries and prompts.
Unpacking the statement reveals a lot. Firstly that Apple is not limiting Google to make a long awaited smarter Siri assistant, but will be using it as a foundation for «innovative new experiences» for all Apple users.
The deal was tentatively agreed in November, and the reason for such a late statement could be that Apple has decided to expand the scope.
This means Google will be delivering AI services to Android and iOS, cornering the smartphone market, and to Chrome, which basically has a browser monopoly — spurring antitrust ideas for many, including from Elon Musk.
Apple has been rumored to announce a revamped Siri assistant in March or April this year, writes MacRumors
Siri will take on Perplexity and OpenAI with its coming revamp. (Picture: Apple)World Knowledge Answers, as it is known internally, will be a massive upgrade for their voice search assistant Siri, according to Bloomberg.
Rumor is that Apple will use an underlying, custom model from Google for the brains of the assistant, and Apple has been looking at it for quite a while.
Siri might be getting smarter, with a little outside help. (Picture: Apple) Bloomberg reports that the iPhone company is considering opting out of using its homegrown LLM for future versions of the chatbot Siri.
They have instead asked Anthropic and OpenAI to train some of their models on their Private Cloud Compute servers.
The Samsung model
This mirrors Samsung’s approach to integrating Large Language Models in its Galaxy phones, where they have some in-house, lower level AI doing the legwork and passing the rest off to Google’s Gemini, writes Engadget.
According to Bloomberg, Apple has recently focused on Anthropic as the most promising LLM, being more compatible with Apple servers and offering the best experience.
Apple is still playing catch-up in the LLM market, but that may change soon, according to recent reports. (Picture: Apple)In an article mainly about how Apple got AI and LLMs so wrong, Bloomberg’s Mark Gurman, a noted Apple-watcher, also notes a few things it’s getting right.
One of those is the significant progress it is having during testing of an AI chatbot version of Apple’s personal Siri assistant, developed by a team in Zurich.
MacRumors reports that Apple is testing a completely new architecture — not the current patchwork Siri that punts to ChatGPT — but a standalone LLM designed to eventually replace Siri outright.
