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 Could Use ChatGPT or Claude to Power Siri
- Thread starter MacRumors
- Start date
- Sort by reaction score
That would be a good thing. Latest version of Android is far more useful and helpful than iOS.
Apple really dropped the ball on AI, they got caught sleeping doing their incremental tech as the tech world passed them by. AI is the biggest and most distributive revolution in tech since the Internet age of the mid 90's so they must do something dramatic to close the gap and partnering or purchasing an AI solution seems the most judicious path forward. Keep in mind its been over a year since Apple previewed the infamous Siri vaporware and we still have seen nothing from them even now and the whole team has been revamped so that implies a hard internal reset occurred, so one can assume they are still falling further behind every week. But the good news is Apple has the money and resources to catch up quickly but they need to attack this in short order to get back into the game and these rumors imply they are looking at all options. Lets all hope they make the right choice.
When outdoors, Siri can’t hear me with the phone right to my face. It’s unusable unless I use type to Siri.
I feel like every time I ask Gemini to do something, Claude could do it a tad better for my taste. I hope Apple chooses to partner with Anthropic.
What's funny is that Apple fired Forstall over the Maps fiasco, and that was an easy fix. Apple got completely caught flatfooted on AI, and all heads are still on shoulders. This gaffe was so big that it put Apple in a position where it will be playing catch-up for several years now.
LOL there's beef between the editors at Macrumors and Gurman. Also, Gurman is off-putting or unlikable.
The situation with Apple and AI is similar to Apple and search. During the early days of the iPhone, there were rumors of Apple doing its own search engine but it fell so far behind, it had no choice but to partner with Google and make a deal to make Google Search the default search engine. We're seeing the same thing with Apple and AI.
Meta might make a comeback, they recently (like within the past week) hired away top people from Google and OpenAI with $100M dollar signing bonuses like they do in sports, on top of their salary and stock
Apple has been sleeping for YEARS on their mobile and iPad OS, not just AI. It’s pitiful what they’ve accomplished, or lack there of, given their budget.
Anthropic is the right partner for current human and programming language based GenAI and honestly Apple should probably buy them and also fund them to work on world models.
LLMs as we’ve known them are going to be ancient history in 5-10 years though, but Apple will probably hedge and partner which might wind up being a mistake. Meta and Google are going to dominate this space unless someone else figures it out. We know some of what Meta is working on and I believe Google is actually a year or two ahead internally vs. what they show the public, unlike everyone else. They had an enormous head start with transformers and the enabling technologies and threw an incredible amount of money at the problem. I suspect they are doing the same thing Meta is, but not telling us.
GenAI as it stands should never have left labs, but since it is public it has to be iterated on because it just isn’t good enough, particularly OpenAI when context gets long, I run into issues that would be solvable with an algorithm that doesn’t even need “AI”.
~100 lines of code could help a lot, old artifacts are referenced often once you exceed a certain context length, it should just always prioritize the most recent document if heuristically it >90% matches an older one. I digress.
Anthropic is the right partner for now despite this technology being fundamentally flawed and incorrect outputs being something that cannot be purely trained out with scale. We’ll see what the future brings, but of what I know exists my money is on Meta’s team despite absolutely loathing them as a company. Once they crack and implement the world model paradigm where it can self-learn ground truths via real-world multi-modal inputs, we may possibly see ‘actual’ AI, not these flawed but sometimes useful facsimiles.
If they work, world models are going to be absolutely nuts and probably the closest thing to ‘AGI’ until at least 2040+. They will try to tell us they are AGI, which will be a incredibly annoying debate.
From my POV Yann LeCun is possibly the only rational person heading up any of these efforts, at least publicly. It’s a shame he’s doing it with such a terrible company at Meta.
LLMs as we’ve known them are going to be ancient history in 5-10 years though, but Apple will probably hedge and partner which might wind up being a mistake. Meta and Google are going to dominate this space unless someone else figures it out. We know some of what Meta is working on and I believe Google is actually a year or two ahead internally vs. what they show the public, unlike everyone else. They had an enormous head start with transformers and the enabling technologies and threw an incredible amount of money at the problem. I suspect they are doing the same thing Meta is, but not telling us.
GenAI as it stands should never have left labs, but since it is public it has to be iterated on because it just isn’t good enough, particularly OpenAI when context gets long, I run into issues that would be solvable with an algorithm that doesn’t even need “AI”.
~100 lines of code could help a lot, old artifacts are referenced often once you exceed a certain context length, it should just always prioritize the most recent document if heuristically it >90% matches an older one. I digress.
Anthropic is the right partner for now despite this technology being fundamentally flawed and incorrect outputs being something that cannot be purely trained out with scale. We’ll see what the future brings, but of what I know exists my money is on Meta’s team despite absolutely loathing them as a company. Once they crack and implement the world model paradigm where it can self-learn ground truths via real-world multi-modal inputs, we may possibly see ‘actual’ AI, not these flawed but sometimes useful facsimiles.
If they work, world models are going to be absolutely nuts and probably the closest thing to ‘AGI’ until at least 2040+. They will try to tell us they are AGI, which will be a incredibly annoying debate.
From my POV Yann LeCun is possibly the only rational person heading up any of these efforts, at least publicly. It’s a shame he’s doing it with such a terrible company at Meta.
Last edited:
What happened to “we have a working version in par with ChatGPT”?
You would think they would have done that last year. We are two years in the AI revolution!
Why they just don't admit they've crapped their pants, give store credit to all iPhone 16 buyers and leave AI to people who actually know, what they're doing?
Tim just keeps digging deeper...
Tim just keeps digging deeper...
I really hope that’s meant as humour
Why are they trying to build their own LLM in the first place? There are companies that are much much better at this and Apple has the money. No one expected them to have their own search engine. Using Google for this was perfectly fine.
Honestly I do not understand what this "Apple is loosing it because Siri is so bad" is all about. If you need AI you have all options available now on iPhone. Just use the ChatGTP app or Perplexity or whatever. Problem solved.
Honestly, I believe Apple’s best move is to stay the course—keep focusing on developing their own LLM and conserving resources. Desperate measures could end up costing them billions, and we already know they can’t change that reality. In the meantime, it might even be cheaper for them to offer free access to Claude Pro or ChatGPT Plus—that could actually work.
Apple may fix its Siri failings by partnering with ChatGPT creator OpenAI or Claude creator Anthropic, reports Bloomberg. Apple has been meeting with OpenAI and Anthropic to discuss a potential deal that would see a third-party large language model used as the backbone for Siri.
Both companies are training versions of their models that would work with Apple's Private Cloud Compute servers, and Apple is running tests, but hasn't made a final decision on whether to abandon Siri. So far, Apple has found that Anthropic's AI models are the best fit for Siri, and Apple and Anthropic have already discussed some preliminary financial terms. Bloomberg says that Anthropic is asking for a multibillion-dollar annual fee that increases over time, which is why Apple is also considering partnering with other companies like OpenAI.
Apple has been training a large language model for a future version of Siri, but it is no secret that Apple has been struggling with AI. Apple previewed several new Siri capabilities at the June 2024 WWDC keynote and advertised the iPhone 16 models with the functionality, but ultimately wasn't able to develop the technology in time.
In March, Apple said that the Apple Intelligence Siri features would be delayed until 2026, sparking customer outrage and spurring several lawsuits. Apple now plans to introduce those features in an update to iOS 26.
Apple's work on an LLM version of Siri is separate from the personalized Siri features that are coming in iOS 26. LLM Siri is planned as a follow up to the new features coming in iOS 26, and it will presumably debut in iOS 27 in the fall of 2026.
Partnering with Anthropic or OpenAI would give Apple time to work on its own AI models while still delivering modern AI features that customers have come to expect. Apple already partners with OpenAI to integrate ChatGPT into Siri, and iOS 18 users are able to have Siri pass requests on to ChatGPT for more detailed answers.
Article Link: Apple Could Use ChatGPT or Claude to Power Siri
