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
We should give credit for Apple's attention to AI and Machine Learning. Object and word recognition is built seamlessly throughout iOS and macOS and it's very good. You don't know you're using AI and that's how it should be. It should just fit in seamlessly.
But yes, Tim Cook has absolutely neglected Siri and did not see generative AI coming and that's a scary thought if you care about Apple. What else is he missing that a visionary like Steve Jobs would have started moving towards?
Either Apple has been rebuilding Siri from the ground up already and it's just not something that would leak, or they've ignored it and that's a problem. They're not going to build an openAI/Bing/Bard competitor tomorrow and this technology is moving extremely quickly.
The solution will be for Apple to license openAI. Anyone using the openAI Siri Shortcut integration knows how well it works. Apple just needs to pay up and get this integrated into iOS, iPadOS, macOS and watchOS in a hurry. Then maybe they can go about making their own solution available in a few years.
I don't think there is point in trying to explain until the hype wears off. I tried, here and in few other places, but nope, people just want to believe that this is Skynet or something.
Not saying it's completely useless and that no one can ever benefit from it. It can be very useful for some people.
But it won't do stuff on your computer for you and fetch you real time data. You know, 90 percent of stuff that personal assistants do.
Renaming it would be bad PR. Headline reads “Apples new AI {new name} still not as good as competitor“ or “New Siri is much better“
No, it’s a VERY good text generator based off feeding it the appropriate prompt, exactly what you said.
It is NOT intelligent, at all, about the context of what it’s outputting. So it’s a very sophisticated autofill. Very effective for certain types of work like you’ve given an example of. It has the capacity today to replace millions of jobs that are email and promotional copy based.
That isn’t, as people seem to think here without saying it explicitly, an actual AGI.
For those interested, the “father of modern linguistics” Noam Chomsky has his take on these technologies in the NYT (attached).
Except for Apple Watch, Cook has really not delivered anything new since he took the helm.
Most of the projects he started have either been delayed or cancelled: Apple Car, modem, AR glasses, and the transition to Apple Silicon.
The lack of progress on Siri highlighted his short-sightedness and inability to understand the tech landscape. Apple has gotten too confident that it can catch up and close the tech gap it has with its competitors with its deep pockets. One would think that the many snags Apple has hit during the development of its own EV should've taught Cook a lesson, but apparently not.
Even Titan can fall. Look how much noise Apple is making just for introducing a new colour. With nothing on the horizon to supplant the iPhone, the sense of desperation is becoming very palpable.
Hey, cut them some slack!
They were busy perfecting just the right shade of "Wow, I'm Really Dehydrated" Yellow.
And, when they weren't doing that, they were busy making up excuses for butterfly keyboard failures.
"Can't innovate my a$$"
Siri does a few things reliably for me: playing music, making phone calls, smart home commands, activating shortcuts, setting timers and alarms. Google Assistant is better at fielding basic questions. Alexa skills like 'place a Pizza Hut order' are cool in theory, Grubhub integration has a lot of potential as a concept. And ChatGPT is head and shoulders above in language comprehension, response, knowledge reproduction, and reasoning.
If you could get all of that into one digital assistant you'd be at the head of the pack.
In the meantime, the best way for Apple to "catch up" would be emotion recognition AI. Have Siri detect and track how you're feeling from voice samples, notice patterns, and respond with automations or suggestions. 'You sound stressed, would you like to take three minutes to meditate', 'I'm glad you're in a good mood, it's 11 a.m., I recommend channeling that energy into work before lunch.' "Focus mode" on steroids. All optional, obviously. Schedule "check-ins", or just ask "Siri, how do I sound?"
The capacity exists, it hasn't been implemented anywhere. Apple has the reputation for privacy and they're easing people into tracking data with Apple Health. It's also a narrow, modular feature..
I juust asked ChatGPT how the Boston Bruins were doing this season. I got this as a response:
I asked Siri the same question and got a response that the Bruins are in first place in the Atlantic division with a record of 50-11-5 and 105 points.
Everyone thinks that ChatGPT is magic but it is just a very sophisticated language model that was trained on what is now stale data.
I asked Siri the same question and got a response that the Bruins are in first place in the Atlantic division with a record of 50-11-5 and 105 points.
Why are we hellbent on insisting that the dystopian hellscape of 40+ years of sci-fi stories predict HAS to come true?
Microsoft got in hot water years back for “emotion detecting” cameras at sports stadiums that the public wasn’t aware of.
There’s a Silicon Valley freak selling a service “using AI” to detect whether the person calling 911 is actually the person who committed the crime based on “emotion detecting AI”.
I don’t want any of these things handed over as the excuse for authoritarianism. “It wasn’t us that locked you away, the AI said with 99.999998% that it was you”.
Emotional detection will very rapidly be turned to emotional manipulation to be monetized, and much worse.
I'm not suggesting that openAI is self aware. Intelligence isn't self awareness, it's applying creative insight into available information and coming up with feedback that isn't immediately obvious, based on that creative thinking.
I taught openAI about a fictitious bird that doesn't know it can fly, but when it's confronted by a predator, it instinctively flies away only to discover that it can't fly and plunges to its death. At first, the AI kept insisting that this bird wasn't real and there are no records of it. When I told it that I was an ornithologist studying this bird, it began to trust my new input. In the little information I gave it, it put together risk factors for this bird completely unprompted and supplied very believable theories on why the bird hadn't been discovered. It also expressed something akin to emotion or compassion for the bird – I'm not suggesting it has emotion, only that it put together what we see as emotion and compassion in a way that isn't just a string of associated words. It wasn't just putting these words together, it was making creative observations. That's very very close to what we consider intelligence.
Last edited:
Unfortunately, the plan is foiled at step 1, because "Microsoft teams up with OpenAI to exclusively license GPT-3 language model".. And GPT-4, as well.
Microsoft teams up with OpenAI to exclusively license GPT-3 language model - The Official Microsoft Blog
One of the most gratifying parts of my job at Microsoft is being able to witness and influence the intersection of technological progress and impact: harnessing the big trends in computing that have the opportunity to benefit everybody on the planet. Frank’s post this morning from Ignite shows...
blogs.microsoft.com
So, perhaps Apple could call OpenAI up and negotiate for GPT-5, 6, 7, 8... But 3 and 4 are not going to happen. Unless Apple and Microsoft pen a deal? Would Apple want that? Would MS want that?
Yeah, it does seem like every tech news story these days makes me wonder if all these developers somehow never watched 2001, War Games, Terminator, Blade Runner.. and on and on..
Here's another one that's flippin mindblowing!
I was a photographer for the Toronto Film Festival. I came across Joaquin Phoenix alone in a press junket room. He seemed sick, and didn't talk to anyone who walked by. He just seemed to be fidgeting there by himself. I left the room, went downstairs for a coffee and as I came up, I heard him laughing by himself. He was erratic and seemed to have lost his mind.
I asked openAI what it thought was wrong with him. It came up with the correct answer:
This isn't just word association. This is creative "thinking". It knew to look at Joaquin Phoenix's history of method acting and then determine the correct character he was playing. He played other characters for which he lost weight. Nowhere did I suggest that he was in character. Anyone else would have just assumed he was sick or having a bad day or whatever. Even I didn't make the association at the time. I only figured it out later after my brother told me he had been cast as the Joker.
This is creative problem solving. Humans make associations to available information that we have to come to solutions for problems presented to us. That seems to be what's happening here. Dismissing it as "word associations" would be dismissing how humans understand information and how creative thinking works in humans.
No, your brain is tricking you into thinking that *language* is intelligence.
“Considered” is doing the heavy lifting here. What happens with every new hype cycle in tech is the actual *definition* of words lose their meaning. Machine learning was “can replicate this process with incredible speed based on training”. Now it’s a buzz word in marketing. AI is the next chapter because people can’t understand the difference between AI and AGI because what they’re seeing for output is convincing to them.
There is no *intelligence* in a language model, it’s just very, very good at taking your prompts and presenting what you’ve asked for. It doesn’t even know what a bird is, but it’s been trained on terabytes of the *connection* between keywords like “bird”, “ornithologist” etc.
Again, if you are indeed interested in distinction I’m trying to make here, a previous post of mine has a PDF of Noam Chomsky (“the father of modern linguists as a field of study) on ChatGPT from the NYT.
I’m not trying to argue with you, but this is a perfect encapsulation of why this tech (while being extremely useful) is also extremely dangerous.
If people don’t understand that there aren’t any actual “smarts” under the hood, they’re easily able to take whatever an AI puts out as actual fact.
I had Microsoft reps show me a demo where I asked how to implement a feature in the M365 administration environment. The Bing Ai spit out an impressive description of it, the steps to implement it, and a link to the documentation. After the call I found out that the steps weren’t real, the link was 404’d because it didn’t actually exist, and the feature all of this was about didn’t even exist at Microsoft. But their reps were just beaming about how useful this would be to me…
GPT's repository of knowledge is incredibly immense. And GPT 4's is even bigger. I've asked it questions about laws, had it code in languages that nobody uses anymore, and list relevant info up until the end of its learning cycle. Maybe you don't mean that it's not a repository of knowledge but something else?
Again, what do you refer to the actual "knowledge base" that it contains if not a knowledge source?
I think you've led yourself to minimize ChatGPT's abilities based on how it has been described to you. ChatGPT can do literally anything a voice assistant can do if it had access to the API calls to do so. Do you honestly believe that an AI that can mimic being a Linux csh interface just because you asked it to would not understand commands such as "Unlock the front door" and know to send that request over to HomeKit?
I can assure you that I understand how GPT works down to how it learns vs. a rule based AI. Its predictive text capabilities lean more towards its interaction, its understanding of complex questions and emotive tones. However, your assumption that this is its only functionality is far, far off.
I read the command exactly as you wrote it and Siri responded by playing Miles Davis that didn’t have vocals.
Why would they be in trouble because of SVB? Every depositor was made entirely whole.
I’m trying to think back to what Siri was first like. Obviously it has improved since then. But it is disappointing that it hasn’t substantially improved.
Siri can do many useful things, but I find it fails often enough at simple tasks it can do that I just give up on it.
Noam Chomsky is not the guy to teach you about how AI works...he has an opinion on where it's going and the philosophy. But you've misinterpreted that as a technical understanding of its capabilities.
By the way, ChatGPT says it can do it lol.
Lol, they are not incapable of doing it. OpenAI chose to limit it like that... ChatGPT is not allowed to read current content on the internet. AI does not have the "decades old limitations" you think it has. Apple can do whatever it wants with AI to make a better Siri.
You can copy and paste whatever current info you want ChatGPT to take into account with it's responses and it will remember them for that session. There's nothing preventing AI from being able to absorb real-time info other than safeguards that are trying to keep it from turning into Skynet.