Isn't it amazing Tim Apple is pushing this AR/VR Headset and they can't even get Siri to work 30% of the time.
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 Engineers Reportedly Working on ChatGPT-Like AI Despite Siri Design Flaws
- Thread starter MacRumors
- Start date
- Sort by reaction score
They really need to rebuild Siri. In fact, they should have had a team working on doing that for years.
The guy who ran Siri basically took the wrong approach; he was going for what was state-of-the-art AI at the time.
It's not really about privacy, it's about using a design that basically didn't pan out.
By contrast, Alexa didn't do AI, it did keyword recognition and used AI for voice transcription. That's why Alexa actually mostly works; it's not really AI, but for the AWS use cases it basically behaved as an AI.
Even ChatGPT isn't AI, as AI was thought of a few years ago. ChatGPT technically is a big statistical model of language, which is not AI as it's been understood. That's not to say it doesn't work: it's actually "smarter" than a lot of human beings (using a broad concept of "smarter").
That said, nobody except the pedants care what AI really means. What people want is results, and they don't care whether it's a bunch of low-cost workers doing "AI" or a computer.
The guy who ran Siri basically took the wrong approach; he was going for what was state-of-the-art AI at the time.
It's not really about privacy, it's about using a design that basically didn't pan out.
By contrast, Alexa didn't do AI, it did keyword recognition and used AI for voice transcription. That's why Alexa actually mostly works; it's not really AI, but for the AWS use cases it basically behaved as an AI.
Even ChatGPT isn't AI, as AI was thought of a few years ago. ChatGPT technically is a big statistical model of language, which is not AI as it's been understood. That's not to say it doesn't work: it's actually "smarter" than a lot of human beings (using a broad concept of "smarter").
That said, nobody except the pedants care what AI really means. What people want is results, and they don't care whether it's a bunch of low-cost workers doing "AI" or a computer.
Siri is a fragmented mess —time to do a rebuild from the ground up, not a retool with fancy features.
I don't think you get what they are saying, in Siri's current state making even the smallest changes requires a Herculean effort. What they are saying is that Siri will require what amounts to a ground up rewrite to enable ChatGPT like features. Any work to add additional languages onto Siri's current code set would just be a lot of work that would go into a system with no future.
So are you saying that they are now willing to compromise privacy by working on a ChatGPT-like feature?
I also hope you understand that ChatGPT is in its very early stages and is not meant to be considered reliable at the moment, but it has showcased insane potential. People who know AI technology better than me are expecting it to improve exponentially. The biggest proof of what the future holds is Google's and Apple's panic.
Yes, "speak carefully" (and slowly!) being sure to over-enunciate is how I use my iPad. "DID. I. EVER. TELL. YOU. HOW. I. GOT. THESE. SCARS?"
I'm used to speaking normally, and quickly, to google with greater success.
OTOH, "Hey Siri" on my M1 iPad instantly works. That colored bubble pops right up. No delay.
"OK Google" works perhaps 25% of the time on my phone. I end up shouting at it until I sigh and go pick it up to trigger it with a button press.
They don’t really talk about their plan to fix the clunky code. I think it’s apparent it needs to be modular so it’s a lot more nimble and they can add features quickly. It blows my mind that macOS was built to be modular from the get go but not Siri. How was that not common sense.?
The good news is, we now have AIs, like ChatGPT, which will replace them.
Remember that Apple bought the company, the creators and the code of Siri itself. They didn’t creat it. I’ve always thought that Apple should have rewritten it shortly after they released the first version, and I’ve akways windered why they didn’t. It’s really very unApple-like.
Which put out information in a way that makes readers trust it, even when demonstrably wrong.
Wait, you’re 100% right, it’s a direct replacement.
They have, they changed it numerous times. The NYT just can’t be bothered to do any actual journalism.
Most the complaints about Siri is that it doesn’t understand you. Did I miss something where chatGTP solved this problem? Of course chatGTP does what it is supposed to - you are typing. The song I want to play always plays when I select it in Apple Music even when Siri missed it (although she is pretty good at it). That’s not surprising just like chatGTP understanding typed text is not surprising. The hardest part about Siri and other digital assistants is understanding speech and people think we are on a revolution because someone solved a different problem.
Stable Diffusion is image ChatGTP. It’s amazing but only works with 512 by 512 images because of the need to minimize data pathways. The data needed to create a ChatGTP of audio regression AI is so large in order to have the fidelity to differentiate tonal differences and close sounding words and inflection - very likely years off. That is the chatGTP of Siri and it’s going to take a bit. I hope I am wrong but it’s an incredibly data intense problem - maybe 1-2 orders of magnitude more data and variation to sort.
Stable Diffusion is image ChatGTP. It’s amazing but only works with 512 by 512 images because of the need to minimize data pathways. The data needed to create a ChatGTP of audio regression AI is so large in order to have the fidelity to differentiate tonal differences and close sounding words and inflection - very likely years off. That is the chatGTP of Siri and it’s going to take a bit. I hope I am wrong but it’s an incredibly data intense problem - maybe 1-2 orders of magnitude more data and variation to sort.
Literally the only thing I use Siri for is reminders like “remind me Thursday at 4pm to ____” cuz it’s faster and easier than opening the reminders app, setting the date/time and then typing in what I want. Everything else seems easier just to do manually.
Altho without Siri I might not even set reminders at all and be too lazy, so I guess it’s nice to be able to easily be reminded of stuff I have to do.
Altho without Siri I might not even set reminders at all and be too lazy, so I guess it’s nice to be able to easily be reminded of stuff I have to do.
Here's a simple example of something that Siri cannot understand, but I bet a ChatGPT could - Hey Siri, play some quiet jazz dinner music without vocals.
I've never been successful asking for it, or for any other genre when I request music with no vocals. Yes, I know I should request specific genres that are usually sans vocals, and that's what I do, but really, I'd like to hear music from many genres, just the examples that are vocals free. Siri is too dumb to comprehend. ChatGPT, and Bing, definitely seem like they could handle it.
(also, if anyone has a magic incantation they've used for this with Siri, please reply! other than to simply request 'deep house' or etc)
I've never been successful asking for it, or for any other genre when I request music with no vocals. Yes, I know I should request specific genres that are usually sans vocals, and that's what I do, but really, I'd like to hear music from many genres, just the examples that are vocals free. Siri is too dumb to comprehend. ChatGPT, and Bing, definitely seem like they could handle it.
(also, if anyone has a magic incantation they've used for this with Siri, please reply! other than to simply request 'deep house' or etc)
From a technical perspective, I appreciate seeing why Siri was so behind/lacking when compared to other competitors like Google and Amazon's Alexa. I always questioned what Apple was doing wrong, but seeing that it was how the phrases were being stored, I can see why it would take Siri quite some time to complete an action or respond. The time it takes to search through the whole database of phrases and then formulate the response is just slow.
This is one time when Microsoft got it 100% right. For AI you need large training models released to the public and Apple don’t really do that type of offering. They can’t just rebuild Siri and it becomes GPT-6. It also takes a large cloud infrastructure which MS has and Apple is still leasing cloud from AWS.
I understand why Siri sucks is mostly because of “privacy”, but if the A and M chips are so capable, they should process everything on device, and personalize based on each person.
Please rebuild Siri to be more efficient and flexible. Please do NOT make Siri more chatty. One of the things I love about Siri vs Assistant/Alexa is that Siri doesn't keep talking after it does what I ask. I've probably told Google Assistant to shut up a thousand times now
Siri's always been a gimmick, it's good for three or four things. Shortcuts help a little.
Apple's still sitting pretty with Jobs' legacy designs, but the flagship products basically sail themselves. iPhone, iPad, Macbook.. iMac and Mac Studio are nice but they seem to be fumbling their production differentiation on the desktop.
They don't need new products. But for the existing ones, there's no innovation. No creativity. In some cases, not even effort. The touch bar, Stage Manager were misfires. 3D Touch (gone) and LiDar didn't add any real value to the iPhone. Siri, Safari are dead in the water.
Dynamic Island is genuinely interesting but under-utilized, and they're headed for under-display sensors anyway.
They're stagnating.
Apple's still sitting pretty with Jobs' legacy designs, but the flagship products basically sail themselves. iPhone, iPad, Macbook.. iMac and Mac Studio are nice but they seem to be fumbling their production differentiation on the desktop.
They don't need new products. But for the existing ones, there's no innovation. No creativity. In some cases, not even effort. The touch bar, Stage Manager were misfires. 3D Touch (gone) and LiDar didn't add any real value to the iPhone. Siri, Safari are dead in the water.
Dynamic Island is genuinely interesting but under-utilized, and they're headed for under-display sensors anyway.
They're stagnating.
This is joke. "Microsoft's LEAD in AI...?" You mean the lead b/c they were the first to hit the BUY button on GPTChat.
Apple will be forced to either adopt and integrate an existing AI chatbot into all of their OSes or overhaul the underlying technology of Siri. Whichever way Cook decides to get Apple out of the hole they dug for themselves, reputational damage is inevitable.
How will Apple integrate all the advances made in AI in the past few years into its closed ecosystem?
It doesn't really look like Cook has a roadmap.
How will Apple integrate all the advances made in AI in the past few years into its closed ecosystem?
It doesn't really look like Cook has a roadmap.
It's not just randomly connecting words to sound intelligent. It actually delivers useful content if you know what to prompt it with. You seem to truly be missing the potential of this technology. However method it's using to generate text, it's already saved the company I work for hundreds of hours parsing information. It's been able to scour the web for information about our products and compile them into something useful. It's been able to summarize long articles from our blog into newsletter tidbits that lead our customers to those articles. We don't just post what it outputs; we read it and pretty much every time, it's been bang on with the information and accuracy.
One example where it's saved us hours of work: I noticed that our founder has made a dozen appearances on podcasts which could be useful for written content. I had one AI listen to all the podcasts and transcribe them. This step alone saved us days of work. Then I fed those transcriptions to AI and asked it to summarize the key points into an article size. It did so perfectly. Again, days worth of work. Finally, I had it look at all those transcriptions and asked that it give me a dozen notable quotes. Not just random quotes, it found some pretty incredible quotes from our founder, each one something we'll use in social posts and ads.
Mind you, we just started using openAI 2 weeks ago and it's already given our team members back valuable time to work on other projects, time that we were desperate for because we have a small team. This thing is going to absolutely change the world. Anyone not seeing that is going to have to catch up.
It is always lurking there in the background even if it is disabled.
ChatGPT isn’t a voice assistant, how do propose an extremely effect *predictive text* engine do that?
I’m starting to get the bigger picture here, everyone knows chatGPT is the future but no one seems to understand what it actually *is*.