Getting more in bed with Google is incompatible with "iPhone means Privacy"
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.
LLM Siri With 'World Knowledge' Search Feature Coming in Early 2026
- Thread starter MacRumors
- Start date
- Sort by reaction score
It’s already kinda built into Apple Intelligence, you can get answers from their LLM with Shortcuts.
Humans (the ones that can't get past skin colour) already took that one.
Here's to AI being smarter than tariffs, religion and nuclear bombs 📣
Same reason people pay ChatGPT, Claude, or Gemini. There’s a $200 a month tier for people who need it (researchers wannabes).
Looks like they don’t, thats why they are going with Google Gemini for there Apple Intelligence LLM.
Maybe the writing is on the wall, and Apple need to forego the AI battle for now, and focus on being the best of the best of what they do do. Yep, I just said do do. This path is not my preference, but clearly they're losing the AI battle, so something needs to give? How would you turn this around?
I was excited about Visual Intelligence until I learned it was powered by google.
I’ve been really excited about Siri improvements.
But if Apple is going to leave the hard problems for other tech companies to solve then count me out.
Hope any new use of google comes with a toggle. 🫤
I’ve been really excited about Siri improvements.
But if Apple is going to leave the hard problems for other tech companies to solve then count me out.
Hope any new use of google comes with a toggle. 🫤
This sounds eerily familiar. My guess is that by 2030 they will launch a system that is state of the art. For 2024.
According to an insider who only goes by Hugh Janis, the new and improved Siri is due on or about the 5th of never.
It was reportedly confirmed by Genny Taliyah. Who is known for their intimate knowledge of the ins and outs of deed.
It was reportedly confirmed by Genny Taliyah. Who is known for their intimate knowledge of the ins and outs of deed.
A decade and a half ago, we had Lumia Vision that identified objects thru its camera. Way to play catch-up, Apple.
Apple hasn't shown itself to be worth anywhere near $20, or even $2. Almost no one trusts Siri and isn't willing to pay anything until Apple has a solid track record.
Even tech illiterate know Siri sucks.
I’ll always stick with Apple products. But I asked Siri to define a word that was a little edgy. Siri said she couldn’t. So, Siri is deciding for us which words are worth defining?
So far the only use for Apple Intelligence I've found is cleaning up pictures. With the Apple Pencil on the iPad, it takes seconds what I used to dick around in Photoshop for hours with.
The summarized notifications are mostly amusing, and the image playground app is as useful as the beer drinking app on the OG iPhone.
I’d totally forgotten about that. Those were the days. Thanks for the trip down memory lane.
Actually what will actually define his Apple legacy is taking Apple to a nearly $4 trillion dollar company
But sadly, not by providing a better product, but by finding more ways to get money out of users. They call it "value creation", sane people call it rent-seeking.
First of all they have to get rid of the name Siri. It has the stench of stupidity, worthlessness and clunkiness.
Using the term World Knowledge and there being a term of art, that is very important, called World Models, is going to confuse so many people.
I really hope they don't use that phrase in marketing until they actually ship a World Model (if they do), those are going to be revolutionary, far more than GenAI to date.
Some reading if you're interested, although this is a bit short and outdated: https://en.wikipedia.org/wiki/Foundation_model#World_models
Yann in particular has spoken since then about multimodality being used to train World Models, and I think that specifically is going to push these over the edge into something we can hardly comprehend.
Synthetic data is also getting a huge boost, see Genie 3: https://deepmind.google/discover/blog/genie-3-a-new-frontier-for-world-models/
It's going to get nuts faster than people are prepared for:
The combination of true real-world multimodality training, combined with the 'classic' definition of World Models which include things like large memory along with a comprehensive system of cognitive architecture, and synthetic training like a future version of Genie that is informed by multimodal training including real-world physics being fed back into the system will be used to generate effectively unlimited training data.
The best use case for GenAI right now is going to look like child's play when/if they pull all of that off, and the signs are starting to show up now that it will be possible with enough time, money, and brainpower. Brace yourselves.
So yeah, anyway, Apple using this term for "better Siri" is silly. Society will be largely unrecognizable in the coming decade(s) due to technology advances yet right now I can't get my HomePod to tell me when it's going to rain tomorrow. There's some dark humor in that.
I really hope they don't use that phrase in marketing until they actually ship a World Model (if they do), those are going to be revolutionary, far more than GenAI to date.
Some reading if you're interested, although this is a bit short and outdated: https://en.wikipedia.org/wiki/Foundation_model#World_models
Yann in particular has spoken since then about multimodality being used to train World Models, and I think that specifically is going to push these over the edge into something we can hardly comprehend.
Synthetic data is also getting a huge boost, see Genie 3: https://deepmind.google/discover/blog/genie-3-a-new-frontier-for-world-models/
It's going to get nuts faster than people are prepared for:
The combination of true real-world multimodality training, combined with the 'classic' definition of World Models which include things like large memory along with a comprehensive system of cognitive architecture, and synthetic training like a future version of Genie that is informed by multimodal training including real-world physics being fed back into the system will be used to generate effectively unlimited training data.
The best use case for GenAI right now is going to look like child's play when/if they pull all of that off, and the signs are starting to show up now that it will be possible with enough time, money, and brainpower. Brace yourselves.
So yeah, anyway, Apple using this term for "better Siri" is silly. Society will be largely unrecognizable in the coming decade(s) due to technology advances yet right now I can't get my HomePod to tell me when it's going to rain tomorrow. There's some dark humor in that.
Early 2026. So, a demo at WWDC 26. Then a beta version with limited features by iOS 27.1. Got it.
Apple has always been in bed with Google, have you seen the iPhone 2G with YouTube preinstalled? Not sure how you can make that statement.
I don't know if it's Siri that can't do that, or Apple's horrible HomeKit networking. This is a process that should take less than 500ms after the command is finished, but on average takes 10 seconds or more, if it completes at all.