A great example
Apple have honestly still never gotten me as an Apple Maps user as a result
Last edited:
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.
BBC Calls Out Apple's AI Feature for Creating More Fake News Headlines
- Thread starter MacRumors
- Start date
- Sort by reaction score
A great example
Apple have honestly still never gotten me as an Apple Maps user as a result
It's basically creating (or trying to create) headlines on the fly. I don't know nor do I care about Mr. Nadal's sexual orientation. But if that's incorrect, I would sue the crap out of Apple.
AI is a gold rush at the moment, and isn't a good thing. Every company has to have AI in their software otherwise they think they will miss out.
AI / LLMs are a good thing, but people should treat it like a tool to use. Although they are far from perfect, still very useful and give good value. Like any tool, use it wisely.
AI / LLMs are a good thing, but people should treat it like a tool to use. Although they are far from perfect, still very useful and give good value. Like any tool, use it wisely.
The notification summarization tool is pretty good for texts in my opinion but it kinda sucks for some other apps (like news apps) all you need to do to stop the summarization for apps like BBC is just turn off summarization for the BBC app
people should keep in mind that the summaries were designed more for communication and not news apps
people should keep in mind that the summaries were designed more for communication and not news apps
Apple is so stubborn that will not pull it. AI for news and sooooio many other things is stoopid idea. Really. Just read your news people. Oh, but all news are click-bait?. Garbage in, garbage out. AI can’t fix that. Is not magic.
Of course it's useless. What's the point of a summary that you can't trust to be accurate?
...and we're just seeing the blatant mistakes that are newsworthy/funny/shocking. Not the ones that are just too garbled, vague and inaccurate to be of any practical use.
See Apple Pay as well. But both turned out okay as I expect apple intelligence to.
Last edited by a moderator:
Yep I've turned all the summary stuff off now, after it said a parcel was delivered which I'd just ordered.
Apple panicked and moved to to quickly. This stuff is not ready for prime time. It gets things wrong - important things wrong.
Apple panicked and moved to to quickly. This stuff is not ready for prime time. It gets things wrong - important things wrong.
I get a lot of hate every time I bring this up to people who don't know any better. LLM's are great for putting me on the right path, helping me get through things here and there, but most of the time the information given is far from "cut and paste".
Apple Intelligence? You mean the feature that is not on by default? Labled heavily as a beta release? And warns you again and again you should double check its results because it is a beta? Ok.
It cracks me up how bad the summarizing is. I’m betting they were so behind and caught off guard by OpenAI and the rest that they rushed training a model capable of on device inference. But instead of holding back, they decided to use it so they could also collect real world user data for those that opt in. Hopefully this will make better models and summaries going forward.
BBCs hate for everything Apple is globally well known
Siri is TRASH. Apple Maps and Apple pay didn't take 10+ years and still be a dumpster fire of garbage.
I agree that this feature would be "killer" if it was reliable. Though I've rarely had Apple AI accurately summarize text messages, top priority emails, etc. It needs to get A LOT better before users trust this feature.
It's the worst part.
The problem is that it's generative in nature. It's taking a bunch of different sources and then creating something out of it. That means it's not accurate to it's source anymore, but it's still being attributed back to the source. It's literally putting words in your mouth.
If it's a simple curation service that doesn't create it's own headlines, then sure.
Is this AI in general or were you talking about Apple Intelligence. I ask because I regularly use ChatGPT and although I ever read before I post, it has never made things up and saves me lots of time.
How so?
Clearly says he's through to the final, doesn't it? Admittedly it uses a bit of that weird language that's only used in headlines (though only a bit - just the word "cruises"), but still, I don't think it was ambiguous.
I think this is just what you get when you try to summarise summaries. It's a stupid thing to do.
More importantly - somebody not winning a sports competition is not news!
I doubt it'll ever work reliably. Language is subtle and irregular lexically and semantically speaking. You can't conclusively define a language well enough for there to be any determinism in interpretation of it. If you look at how humans react to language which is carefully crafted to have meaning other than what is obviously written or uses idiomatic or metaphorical speech rather than simple conceptual speech then it gets even dirtier.
What pops out is a statistical approximation of the information that went in. If you include a metaphor which is obvious but has been contextually adapted then it'll throw the weights off and garbage comes out every single time.
Statistical approximations also have a significant error. The current error margin is 30-40% on the best models on even basic things. That is large enough with low enough confidence that validation is required for every output, which is likely more costly than doing the work yourself in the first place, at least from experience.
Add to that, the cost of generation and execution of the models grows exponentially whereas the error decline is almost linear, suggests it'll never converge on a usable product. We're already hitting a hard wall now and investors are losing interest in propping it up.
Really though, if you assume that 1 in 3 things the summary says is bullcrap and you have to Google everything and find a source and check it yourself in case it's talking bullcrap, is that a viable technology?
Nope!
And trust it you must not from day one. Unless what you are creating or using it for is of zero value.
You're not sure? I am 100% confident it was FOMO in Apple's case!
Eh, it's no substitute for human intuition and people that use it in my industry without deep knowledge of the subject matter come off looking stupid. It requires constant supervision. By extension this all makes Apple look clueless, just like the little brother in A Christmas Story:
"Hey guys, wait up!"
I dunno where you live, but that has never been and continues not to be true in the UK. It's absolute wank at knowing the correct locations for things. I've submitted numerous corrections too, and I get a notification saying it's been corrected, yet it has in fact not changed. I can verify at a glance that Maps shows a bunch of businesses in my area that don't exist (and have never existed, they're just houses), and doesn't mention a bunch of businesses that do exist, while at a glance Google Maps doesn't have either of those problems.
I want to prefix by saying I have not seeing the exact headline in this case.
But I would wager that this might be the result of these news headline being misleading in the first place. And the AI summary just changed it from a question mark to a statement. I think news companies or any publication should stop doing clickbait title. But that would be a long shot.
But I would wager that this might be the result of these news headline being misleading in the first place. And the AI summary just changed it from a question mark to a statement. I think news companies or any publication should stop doing clickbait title. But that would be a long shot.
Right?! It's only a handful of large US cities that are anywhere close to accurate. Comparing Apple Maps and Google recently, Apple Maps had me going 45 minutes in the wrong direction to an office park instead of a trailhead in Sedona. Apple Maps continues to be awful.