I wonder why they didn't start with an open-weights model like Google's Gemma or literally any other offering with a permissive license. Suppose it's easier to throw a cool billion at Sundar and have his guys figure out your problems.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Apple's New Siri Will Be Powered By Google Gemini
- Thread starter MacRumors
- Start date
- Sort by reaction score
Google and Apple may compete in some areas, but clearly there's a fair amount of professional trust and respect in that business relationship. Sam Altman -- and also Zuckerberg -- are looking to play a much different, zero-sum game, and it's clear neither of them are trustworthy. It makes sense for Apple and Google to partner on this. Google gets to trim a billion off the search engine bill to Apple, Apple gets a temporary boost to get its first-party AI working in early 2026, and both companies make plays to box out Meta and OpenAI from gaining too much marketshare.
Last edited:
In the meantime, Siri gets dumber every day. I have been using my HomePods as tools and CarPlay is key, and not the Idiot Siri cannot handle simple directives it had no problem with 6 months ago. Just as one example, there is no reason why Siri has to ask me what phone number to use when I tell it to message my girlfriends who I have an eight year text chain with. Or why when I tell it to go to Trader Joe's it chooses to give me directions to Traverse City, Michigan.
Well yes, but this is rumor reporting. Until it's confirmed, this is still an "if" in my book.
what they did with chat GPT thus far is clunky and awful to use and I just use grok when I need to ask real questions instead. Hoping this is way more integrated.
It’s surprising how many people have no idea how this even works.
What makes you all think that Google will have access to your data?
Well that’s disappointing. The ‘privacy -that’s iPhone’ paying a billion bucks a year to a harvester firm.
There is an order-of-magnitude difference between the capabilities of open-weight models and the SOTA closed-weight models.
Of course, Siri has never natively had AI, the Open AI thing wasn't really part of Siri but now it sounds like it will be integrated and it will make Siri the WWDC 2023 dream*
*you just might have to wait a couple more years due to Apple's track record.
********. gemini nano runs on device and it's fantastic. pretty well optimized , too.You can’t run any useful general-purpose LLM locally on an iPhone. The Gemini model at 1.2 trillion parameters requires several terabytes of RAM.
With Apple’s RAM pricing, that would be a $200.000 or so iPhone.
I think that's a good plan, so they can deliver new, helpful features in the next 6 months, while keeping it on their private Cloud Compute, while on-device stuff I'm sure will still happen through a small (but improved) Apple on-device model. And in the meantime, like they've always done, they'll keep playing catch-up internally until they have something that matches Google's model and in a couple of years they'll replace it with their own.
Read the article, this model runs on apple servers with no access by google
I strongly dislike Google but Gemini is at least in the top 3 of all AI models.
It now seems likely Apple's AI strategy over the next year or two might look like this:
1) Don't spend a fortune on models that are eventually going to become commodities while setting yourself up to be the hardware provider consumers use to access AI services. (Do whatever you can to build a moat around upcoming hardware from Meta and OpenAI.)
2) Have a first-party AI option for your devices that can do personal context and basic search/AI queries. (Sounds like mostly Google tech implemented by Apple with some of Apple's smaller models at the start until Apple catches up on the larger parameter models.)
3) Broaden the current "extension AI" relationships beyond ChatGPT, which might include Gemini, Grok, etc., to that users can call on those using Siri as well. This way you can pick your model, and Apple doesn't mind so long as you're doing it on an Apple device. They may even figure out a way to implement those third-party models into Safari in a way similar to how you can choose your own search engine.
Ultimately, the goal is to keep selling Apple hardware by making sure the first-party AI options are good enough and that you allow users the ability to utilize other AI models until the crazy market sorts itself out.
1) Don't spend a fortune on models that are eventually going to become commodities while setting yourself up to be the hardware provider consumers use to access AI services. (Do whatever you can to build a moat around upcoming hardware from Meta and OpenAI.)
2) Have a first-party AI option for your devices that can do personal context and basic search/AI queries. (Sounds like mostly Google tech implemented by Apple with some of Apple's smaller models at the start until Apple catches up on the larger parameter models.)
3) Broaden the current "extension AI" relationships beyond ChatGPT, which might include Gemini, Grok, etc., to that users can call on those using Siri as well. This way you can pick your model, and Apple doesn't mind so long as you're doing it on an Apple device. They may even figure out a way to implement those third-party models into Safari in a way similar to how you can choose your own search engine.
Ultimately, the goal is to keep selling Apple hardware by making sure the first-party AI options are good enough and that you allow users the ability to utilize other AI models until the crazy market sorts itself out.
From the post....
"The AI model that Google is developing for Apple will run on Apple's Private Cloud Compute servers, so Google will not have access to Apple data."
Person paid to speculate for the purpose of creating volatility in a stock price, that is wrong a lot, says a thing that other company would comment on anyway, and is likely to create volatility in stock price.
I'll believe it when It happens. I'm not saying that I don't think they may not have some potential piping between Apple Intelligence and Gemini. I just don't trust Bloomberg as a source for Apple News.
Also this was part of this week's Gurman newsletter post, so is it really worth reporting on the same story content from the same outlet twice in a week?
I'll believe it when It happens. I'm not saying that I don't think they may not have some potential piping between Apple Intelligence and Gemini. I just don't trust Bloomberg as a source for Apple News.
Also this was part of this week's Gurman newsletter post, so is it really worth reporting on the same story content from the same outlet twice in a week?
Look at it this way. Google is paying Apple 20 billion per year for Google to be the default search engine inside Safari. Apple is just reducing that payola by 1.2 billion per year. Consider it a discount for Google.
I’m bothered by Google winning this contract over Anthropic. They’re the best AI out there and the most ethical, although that is beginning to change, but by Jove, are they full of themselves. For Fricks’ sake, they charged so much that Apple got priced out of a deal with them. Goddamn, I wanted it to be Anthropic over Google.
"The AI model that Google is developing for Apple will run on Apple's Private Cloud Compute servers, so Google will not have access to Apple data."
The continued bugs, lack of substantive new features, and inability to execute on their ideas without outsourcing makes me think Apple engineers are paid too much. It took them 3 years to design Liquid Glass for crying out loud. They haven't improved the skills of the original Apple Intelligence from 18.2. Heck, they haven't fixed App Library since it launched on iOS 14.
Bro, none use this this A company at all. You all flex but most sticked to GPT or Gemini.