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.
Report: Apple's Next-Gen Version of Siri Is 'On Par' With ChatGPT
- Thread starter MacRumors
- Start date
- Sort by reaction score
Making an LLM is easy at this point. Apple just has to take Siri off device and remove privacy. Thats the bitter pill to swallow.
Everyone arguing this is too hard or some magic impossible task are way behind the tech. If apple's version was one year behind ChatGPT it would still be an amazing upgrade. Every few weeks the major LLMs leapfrog each other. This week the leader is Grok, then Gemini, then Claud, then ChatGPT again. These technologies are now well understood. Making a decent one isn't some herculean task.
Everyone arguing this is too hard or some magic impossible task are way behind the tech. If apple's version was one year behind ChatGPT it would still be an amazing upgrade. Every few weeks the major LLMs leapfrog each other. This week the leader is Grok, then Gemini, then Claud, then ChatGPT again. These technologies are now well understood. Making a decent one isn't some herculean task.
Last edited:
If Siri is on-par with chatGPT right now as of mid-2025 I will eat my hat. No way no how. If Apple caught up with them this quickly that would put them near the top of the market for something that they were just saying a couple months ago at that internal meeting only even works 2/3 of the time. No shot Apple has improved this quickly. And if it was this good you can bet your ass Apple would be monetizing the hell out of it already.
I hope they give it a new name, because Siri is cursed. And while they’re are it, let us name it what we want! Siri is a joke.
I hope they give it a new name, because Siri is cursed. And while they’re are it, let us name it what we want! Siri is a joke.
If Siri is on par with ChatGPT right now it ought to be delivered in September‘s iOS 19 at the latest. Waiting even longer gives it even more of a bad look.
I am in the Camp that wants to keep the Siri name. But then, I also preferred the orb to The light ring around the exterior of the device because a game I play, crossword jam, can do the same thing already. It’s no big deal to me, however, I could see the orb from across the room if the iPad was not within reach. Also the Siri name is very much linked to Apple whether they like it or not. Give it a big stage event and talk about how great it has become and that should suffice.
Hey if deepseke could do it, musk could do it, and pretty much everyone could do it then Apple sure as hell can. It’s not about people or talent, it’s about leadership. ChatGPT is the worst AI out there though, they got surpassed pretty quickly.
Exactly. Everyone is coming up with their own AI and each single one is better then Siri
Look, I get the frustration in this thread. Everyone wants to chase the next shiny. BUT ignore the shininess for a minute and read something like this:
www.ben-evans.com
We all agree (including people who are spending truly massive amounts of money) that there is *something* there in LLMs. But how should that something be packaged as a commercial product?
Again it's simply a FACT that the current packaging is of limited commercial appeal (that's the whole point of the above article). So what's wrong?
Some of this is technical. For all their vaunted brilliance, LLMs are still mostly not much better than fancy search engines. People are trying to move them beyond that, but it ain't happening so far.
I encountered this recently trying to resolve a specific question about memory barriers.
On the one hand, the "answers" generated by (paid version) of Grok were impressive insofar as that they were technically correct, and even included [correct] code generated to supposedly illuminate the issue.
On the other hand, the answers were, ultimately, nothing more than fancy search engine. They were a more convenient gathering together in one place of everything that's available on the web about how memory barriers work. BUT they were also no more than that. A probing question of the form "why was this choice made, not that" could not be answered, because the answer is not present on the web. It's probably present in the minds of the people on the committee, but they haven't written it down. I can hypothesize some answers, and I hoped Grok could too, but no, not yet.
But this is minor. Most people don't use LLMs at this level of technical detail, and "intelligence" in the form of being able to answer technical questions better is being added all the time.
The real issue is probably this tweet:
"Your butler doesn't need a PhD... but he does need to live in your house"
What people probably want is something akin to a butler. Meaning access to your ENTIRE digital world and agentic capabilities. That package doesn't yet exist, and everyone's slightly nervous about both elements of it, the access side and the agentic side.
And this is why worrying about Apple "being behind" is foolish. The problem of creating a smart LLM is very interesting, but it's not the only problem that needs to be solved; the other problem is the twin elements of integrating this into your digital world, and giving it agentic capabilties. And on this side no-one has much of a clue what they are doing. MS has tried to put something in place with Recall. Personally I think the sentiment was absolutely correct, but the usual hysterical crowd reacted as they usually do. So how do we give the AI in the machine awareness of how I'm using my digital environment (so that it can act as a better butler)? Presumably not by the easiest method of taking occasional screen shots...
The same will probably turn out be true if anyone tries to provide agentic capabilities via simply puppet'ing of the UI (ala things like QuicKeys and similar keynboard/mouyse replay engines of the nineties).
In other words, the quick and dirty solutions appear to be out, what's required is a huge suite of bespoke APIs that give the LLM access to your data, to your on-going activity, and to some degree of agency.
And, what a surprise, that is EXACTLY what Apple has been working on!
It's slow because it's a massive project. (The equivalent of putting together the first GUI, in world of nothing but CLIs, and like that same project, hampered by the massive gap between the HW that's required and the HW that's actually available at the price [and power] point you want to hit.)
And it's invisible because no-one else is doing it, and because it's much more difficult to understand than the obvious interaction of a user with a chatbot.
It's a lot easier to write pong, and a lot easier for the public to understand pong, than it was to write the first MacOS and fit it into a 128kB 68000-powered little box!
But the second of these was the project that changed the world.
GenAI’s adoption puzzle — Benedict Evans
Generative AI chatbots might be a life-changing transformation in the nature of computing, that can replace all software, but so far, most of its users only pick it up every week or two, and far fewer have made it part of their lives. Is that a time problem or a product problem?
www.ben-evans.com
We all agree (including people who are spending truly massive amounts of money) that there is *something* there in LLMs. But how should that something be packaged as a commercial product?
Again it's simply a FACT that the current packaging is of limited commercial appeal (that's the whole point of the above article). So what's wrong?
Some of this is technical. For all their vaunted brilliance, LLMs are still mostly not much better than fancy search engines. People are trying to move them beyond that, but it ain't happening so far.
I encountered this recently trying to resolve a specific question about memory barriers.
On the one hand, the "answers" generated by (paid version) of Grok were impressive insofar as that they were technically correct, and even included [correct] code generated to supposedly illuminate the issue.
On the other hand, the answers were, ultimately, nothing more than fancy search engine. They were a more convenient gathering together in one place of everything that's available on the web about how memory barriers work. BUT they were also no more than that. A probing question of the form "why was this choice made, not that" could not be answered, because the answer is not present on the web. It's probably present in the minds of the people on the committee, but they haven't written it down. I can hypothesize some answers, and I hoped Grok could too, but no, not yet.
But this is minor. Most people don't use LLMs at this level of technical detail, and "intelligence" in the form of being able to answer technical questions better is being added all the time.
The real issue is probably this tweet:
"Your butler doesn't need a PhD... but he does need to live in your house"
What people probably want is something akin to a butler. Meaning access to your ENTIRE digital world and agentic capabilities. That package doesn't yet exist, and everyone's slightly nervous about both elements of it, the access side and the agentic side.
And this is why worrying about Apple "being behind" is foolish. The problem of creating a smart LLM is very interesting, but it's not the only problem that needs to be solved; the other problem is the twin elements of integrating this into your digital world, and giving it agentic capabilties. And on this side no-one has much of a clue what they are doing. MS has tried to put something in place with Recall. Personally I think the sentiment was absolutely correct, but the usual hysterical crowd reacted as they usually do. So how do we give the AI in the machine awareness of how I'm using my digital environment (so that it can act as a better butler)? Presumably not by the easiest method of taking occasional screen shots...
The same will probably turn out be true if anyone tries to provide agentic capabilities via simply puppet'ing of the UI (ala things like QuicKeys and similar keynboard/mouyse replay engines of the nineties).
In other words, the quick and dirty solutions appear to be out, what's required is a huge suite of bespoke APIs that give the LLM access to your data, to your on-going activity, and to some degree of agency.
And, what a surprise, that is EXACTLY what Apple has been working on!
It's slow because it's a massive project. (The equivalent of putting together the first GUI, in world of nothing but CLIs, and like that same project, hampered by the massive gap between the HW that's required and the HW that's actually available at the price [and power] point you want to hit.)
And it's invisible because no-one else is doing it, and because it's much more difficult to understand than the obvious interaction of a user with a chatbot.
It's a lot easier to write pong, and a lot easier for the public to understand pong, than it was to write the first MacOS and fit it into a 128kB 68000-powered little box!
But the second of these was the project that changed the world.