Wave of Apple Generative AI Tools on Track for WWDC Debut

[J.A.K.]

Does anyone honestly benefit from these types of things?

Like, they all essentially tend to amount to "Company thinks it knows what's best for the user so does it for you."

None of it is true A.I. It's all just glorified spellchecking. And Apple's dictionary is godawful.

 

[ric22]

I'll be happy if Siri just stops asking me "Who is speaking" several times a week or actually increases/decreases the volume on my playing HomePod instead of increasing/decreasing the volume on my not playing iPad that is in the room...🙄

I wrote about this last week. It's mind numbing... I have no idea how Siri is seemingly worse than 6 years ago.

 

[picpicmac]

"Better AppleCare" is probably not a feature that gets people hot and bothered, but I'm surprisingly excited for this...

Apple has a great deal of information on its support website... but I often find myself getting lost in too much of irrelevant information. If Apple can implement an interactive support tool, integrated into MacOS for example, replacing the lame "Help" menu, then I'm all for it.

 

[picpicmac]

but with the tools I can expedite everything and save hours to months of time.

But how do you validate the code? In other words, how do you know the code is doing what really is desired?

 

[xxFoxtail]

Seriously. All I care is when will AI clean my toilet. The rest I don’t care 🤣

Enjoy talking to your new voice-controlled smart bidet

Alexa, clean my butt.

www.theverge.com

This might not be exactly what you’re looking for though.

 

[kiranmk2]

I agree, but so far there is no algorithmic or ML understanding of music yet. The recommendation algorithms of music streaming services are fairly serviceable (my main experience being with YouTube Music), but they clearly rely on “what others also liked who liked stuff you already liked”, not on any understanding of musical contents. And I suspect that Apple’s “AI” play lists will likewise only be based on metadata and user behavior.

The music genome project which formed the basis of Pandora worked really well when I used it 20 years or so ago. Not every song it recommended was right, but there were a lot of artists I discovered through that before it got blocked for non-US locations.

In trying Spotify & Apple Music, Spotify is half way decent in recommending new music, but I find Apple Music pretty poor (as you say, just content that is pushed is either promoted or like by people who listened to things I like).

 

[picpicmac]

and wokeness dominate the algorithms.

 

[picpicmac]

Better and smarter software will ALWAYS benefit you as the consumer.

HAL will like a word with you.

 

[AndiG]

But we all need an iPhone 16 for the full AI feature set. This is quite evident since the latest rumors say that iP16 pro and iP16 models will get the same CPU.

The same is also true for the iP17. Apple will soon discover that the new iP16 processor is too limited to run full scale LLM locally and the iP17 will offer even better and more sophisticated AI support.

 

[Chuckeee]

But how do you validate the code? In other words, how do you know the code is doing what really is desired?

I’ve actually played with that a little bit and from experience it was very useful, the code wasn’t very complex by itself but AI excelled at finding existing libraries and specialized functions that were ideally suited for the problem. Much better than any search engine, more like having a grad student available for grunt research.

 

[garranhado]

AI AI AI but it will work for non-US countries?

 

[originalmagneto]

I just hope we’ll get Siri and AI functions in more languages. I’m from Slovakia and it’s ridiculous that Siri still doesn’t work here 🤦‍♂️🤦‍♂️🤦‍♂️

 

[ebika]

But how do you validate the code? In other words, how do you know the code is doing what really is desired?

You write tests to validate it as you should do for most code that isn’t trivial.

You do touch on an interesting emerging concern. Generated code tends to be a force multiplier for experienced devs that can read and double-check the code. Less experienced devs have a risk of missing defects in the generated code, making them worse at their tasks. This is something we’re watching closely where I work. Best practices are still emerging as the tools evolve.

 

[picpicmac]

Much better than any search engine, for like having a grad student available for grunt research.

Less experienced devs have a risk of missing defects in the generated code, making them worse at their tasks.

Not long ago a presentation (now on Youtube) by a panel of profs (at UC I think) had one of the Comp Sci profs simply accepting that tools like ChatGPT was going to be used and encouraged it.

But from a software engineering perspective this opens up a great concern. As you noted, someone who is a SME (subject matter expert) can check if an LLM is outputting something that is correct or not.

Yet someone without expertise is unlikely to do so, or to do so easily.

Much like when people use a word because they heard it somewhere else, but really don't understand the various denotations and of course not the connotations.

"V&V", as it used to be called, is probably the most important aspect of doing software and system engineering. Lots of people get into coding as some sort of activity, but simply writing code to be processed by a machine is not the hard part.

The hard part is getting what is specified, and only getting what is specified.

 

[jayducharme]

Have you tried ChatGPT?

I have, and found it to be partly amusing and often wildly inaccurate. I'm sure the tech will gradually improve, but for now it's still a novelty (though it's certainly more functional than Siri currently is).

 

[StyxMaker]

I agree, but so far there is no algorithmic or ML understanding of music yet. The recommendation algorithms of music streaming services are fairly serviceable (my main experience being with YouTube Music), but they clearly rely on “what others also liked who liked stuff you already liked”, not on any understanding of musical contents. And I suspect that Apple’s “AI” play lists will likewise only be based on metadata and user behavior.

I don’t want a lot, I just want to be able to tell it to play all of the Beatles music ordered by release date and have it happen.

 

[CarAnalogy]

I have, and found it to be partly amusing and often wildly inaccurate. I'm sure the tech will gradually improve, but for now it's still a novelty (though it's certainly more functional than Siri currently is).

People keep saying it’s wrong, but I haven’t had issues with that. Side by side comparison to Siri as something you can just ask a simple question and get a simple answer, Siri just doesn’t work and ChatGPT just gives me the right answer like a person would.

 

[Miles Teg]

I am looking forward to the realization of the Knowledge Navigator

 

[neuropsychguy]

But how do you validate the code? In other words, how do you know the code is doing what really is desired?

That's a good question. I know it works because it does what I expect. A semi-short answer is I have very specific tasks I need scripts or other code to do, mostly using other software or tools. Many times these are done only once for a project and then never done again. Essentially I mainly need to create various pipelines to handle and analyze data. Sometimes that pipeline is used once, sometimes, I'll or others will use it regularly. I can code these pipelines/scripts/analyses by hand or I can use various "AI" tools to do all the boring stuff or help me figure out the boring stuff (I say boring, not that coding is boring, but because I'm a scientist and so my main focus is on the analyses and publishing data). In most cases it's mostly automating things I could do with various manual steps. Rather than having to spend X hours or days or weeks coding for one analysis, I can now get most of the coding done in minutes to hours. Or, I can do much more complex coding than I have skills to do. Again, I know it works when it does what I expect it to do. I already know what I want and what it should look like, I just use the LLMs to help me get there faster.

If it's a statistical analysis (R or Python), I validate using other statistical tools and see if I get the same answer. I always have to date, so now I mostly just go with what I can generate using ChatGPT and/or other tools. It's either that or spending 2-10X more time Googling and searching through various sites to figure out how to do what I need to do. If I can have a LLM generate code for me that not only works but is also commented, that's a huge win for me and for other people who might use the code when I post it online.

Another thing I use "AI" tools to do is help me make things simpler for other people to use. If I can take some time to make tools other scientists and researchers created easier to use, that benefits a lot of people -- okay, maybe only a few people but it helps out the scientific field. It saves me time in the future and saves other people time.

 

[deepspacecowboy]

Music playlists have never been something I got too excited about, but in this case it seems like there is massive potential. Imagine being able to describe exactly the kind of music you want to hear, which era, genre, and band/s you want the music to sound like, and even which you don't like, all at once. Or, describe the exact context of the event you're playing music within. Or a kind of project you're working on and seeking creative inspiration for.

How boring. Let the AI spoon feed us… instead of making a playlist because one loves music. Back in the day we made mixtapes for friends and lovers. It took a long time to craft the perfect mix. The whole process was full of thought, intention and affection. As someone who loves music, I think it’s quite sad the way everyone wants some algorithm to think for them these days. No creativity. No appreciation for the art. Music has become another disposable commodity.

 

[deepspacecowboy]

"Better AppleCare" is probably not a feature that gets people hot and bothered, but I'm surprisingly excited for this... just because I've found that for tech support issues, I can just ask ChatGPT or Bard and get a genuinely helpful reply in seconds. As opposed to, say, searching on Google and having to spend an hour or more wading through years-old, semi-related posts on Reddit and random message boards to maybe find something that can help.

Now if I can just get my older relatives to rely on AI for their tech support instead of coming to me, all my problems would really be solved! 😩

And what’s the inevitable conclusion of all this “progress”? One day we’ll be so dependent on the machine and, at the same time, completely ignorant of how anything works. This Outer Limits episode from the 90s was unbelievably prescient:

Stream of Consciousness

 

[filmantopia]

How boring. Let the AI spoon feed us… instead of making a playlist because one loves music. Back in the day we made mixtapes for friends and lovers. It took a long time to craft the perfect mix. The whole process was full of thought, intention and affection. As someone who loves music, I think it’s quite sad the way everyone wants some algorithm to think for them these days. No creativity. No appreciation for the art. Music has become another disposable commodity.

I used to make mix tapes. Then I burned CDs. So I get that. It's not like you can no longer craft your own playlists.

But AI working as I described it would enable thought and creativity to shape what you're listening to, and also work as an amazing tool to discover new music you would never have otherwise heard, by explicitly asking it to find deep cuts, music by obscure independent artists, and tracks that are tailored to your very specific set of tastes and interests.

 

[deepspacecowboy]

I used to make mix tapes. Then I burned CDs. So I get that. It's not like you can no longer craft your own playlists.

But AI working as I described it would enable thought and creativity to shape what you're listening to, and also work as an amazing tool to discover new music you would never have otherwise heard, by explicitly asking it to find deep cuts, music by obscure independent artists, and tracks that are tailored to your very specific set of tastes and interests.

Personally I think most people these days are more interested in finding new music than actually enjoying and appreciating the music they have found. That’s why no one even remembers the artist or album they downloaded - and loved - only a few weeks prior. I refuse to let any algorithm find music for me. Everything I find, I find myself. I’ve tried letting Apple Music suggest things and it’s always something derivative and uninspired. I just think if you really love music, you don’t let some AI spoon feed you.

 

[svish]

Really happy to hear that Apple is focusing on AI. Expecting Siri to become better in iOS 18

 

[Biro]

I have no idea how Siri is seemingly worse than 6 years ago.

The first thing I do after buying any new Apple device is turn off Siri… and prevent Siri from learning or suggesting anything from any app or program. The only time Siri’s voice is ever heard on my devices is when I’m using directions via Maps. I don’t want or need a “personal assistant” and don’t need or want what’s now being marketed as AI.