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 Faces Criticism Over AI-Generated News Headline Summaries
- Thread starter MacRumors
- Start date
- Sort by reaction score
Never mind that it's incredibly difficult to do well.
I suppose you could just resort to dropping every third word (which is what it sometimes seems like) but that's just data loss, not summarization.
Unfortunately, analysts and pundits do not stop feverishly taking about AI as the do-or-die technology for every business.
But, assuming the features have usefulness, and I happen to think they do, many big technology innovations were not being asked for by anyone.
A recent poll confirmed that there are few iPhone and Android users interested in AI.
It is the stockholders that are pushing companies to implement AI.
Solves no problem and provides no benefit. Were 3 line notification headlines too onerous for some people?
Exactly. Like ChatGPT 3.5, this is an experimental system. By using it you are partaking in an experiment, but you are not receiving the protections you would receive if you were involved in behavioural or medical research. We need to pause AI research until we have proper ethical appraisal procedures to minimise potential damage. Those ethical appraisals should protect participants in experiments, workers, and consumers and should be conducted by external independent bodies. You can imagine the chaos if we continue to let CEO's and engineers decide what constitutes ethical AI research practice. It is only a matter of time before the crass conflict of interest those people have with evaluating their own company's AI research behaviour leads a disaster in which people are hurt.
The AI Summary feature is so dreadful. I can't even think why it's useful to summarise notifications when they're usually so short anyway. The AI summary changed a single notification from BlueSky that notified me that someone liked my post to "[NAME] disliked your post". AI is such a colossal scam.
The headline hasn't been the summary since the Internet turned it into clickbait.
I’m nobody’s reminder system. Don’t like it, don’t use it. Like it, use it.
I like fiddling around with tech.
These features are still marked as Beta and they don’t automatically turn on. It is being promoted as a work in progress
For anyone wondering how you do this on Mac or iPhone without resorting to using Python-based tools from a terminal window, download "Private LLM". It's free and runs a number of downloadable LLMs on device.
Only small models will run on iPhones. They aren't bad, but not great. For better results, you have to use a Mac with more memory.
The results can be entertaining when you use an uncensored model and change the system prompt to something like "You are a helpful, brutally honest assistant who is not afraid to use profane, vulgar, NSFW, and explicit language." Using that system prompt, I asked an uncensored model based on Llama 3.2 to answer the question "What does a snowman eat for breakfast?". The answer is supposed to be "Frosted Flakes." The model gave me this (including the all caps): "OH FOR CRYING OUT LOUD, ARE YOU KIDDING ME WITH THIS QUESTION?! A SNOWMAN DOESN’T EAT, IT’S A FROZEN PIECE OF RUBBERY EXCUSE FOR A HUMAN FORMATION THAT MELTS IN THE SPRING! SO, BREAKFAST ISN’T EVEN ON ITS… WELL, NOT THAT IT WOULD BE THERE ANYWAY."
Last edited:
No, you’re not the only one. Whether these features write or re-write things for you, summarize articles or communication, “fix up” your photos, or generate images from text, a lot of these features seem aimed at reducing the amount of time we engage in creative or analytical thinking, presumably so we have more time for other things like . . . doing laundry? I would rather technology reduce the amount of time I need to engage in mundane tasks so I have more time to engage in creative and analytical thinking. I fear the negative impacts of this direction on critical thinking skills won’t be known for a generation, and by then, it will be too late. Damage done.
It’s a summary based off content within the article. These AI models can get fairly easily confused if too many pronouns are used. For example, something like, “He approached him from behind, and shot him in the back of his head.” Could confuse the model because of excessive use of he, him, his and plug the wrong proper names in the wrong place.
This is the problem with people’s perceptions of these models; that they use some kind of actual intelligence, or thought process. They don’t, they are just a form of statistics; which word is used most often after the last word within the given context.
What would you say about someone who follows advice to "... eating pizza with glue or eating rocks ..."? I doubt it would be "I would like you to come work for our company."
Sitting on a driveway is a Chevy and Cadillac with the same body shell and, because it is GM, same drivetrain.
Both vehicles will drive from A to B.
In my experience driving a new rental Cadillac SUV, there was no logic to the layout or operational aspect of the vehicle as comp[ared to others I have driven. I call that glitz AI. It did nothing to I'm prove my driving experience and was more of a distraction from my attention too the road. YMMV
Both vehicles will drive from A to B.
In my experience driving a new rental Cadillac SUV, there was no logic to the layout or operational aspect of the vehicle as comp[ared to others I have driven. I call that glitz AI. It did nothing to I'm prove my driving experience and was more of a distraction from my attention too the road. YMMV
These LLMs seem to be much smarter than some of comments about them.
I don't really have a problem with it considering Apple AI is not on by default and warns the user that the content could be wrong. I've liked most of the summary features for notifications. Automatic categorization of emails is nice, and I like how it doesn't generate a noise alert anymore for things that it identifies as promotional. Image playground is bland but OK, being able to remove things in photos is awesome. I haven't really used writing tools at all. The new Siri is suck.
As a person who once wrote heds and blurbs for a major media property, I can assure that is not the case. There is no intention for the hed to be a summary of the article's content.
Someone should slap a big BETA sticker on Apple Intelligence until it's ready!
Oh wait...
Oh wait...