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 Developing 'Apple Neural Engine' Chip to Power AI in iOS Devices
- Thread starter MacRumors
- Start date
- Sort by reaction score
Siri isn't necessarily behind as far as providing relevant information based off providing results . Siri is behind because it struggles with dictation and deciphering. That's really the main issue. In comparison, I could say a sentence to Siri and use the same sentence to google voice and get completely different results from Siri. At times Siri can be very accurate and other times it's completely off-base. At Times I think it's dependent on Apple servers as well. But When Siri does work, it works efficiently.
How dare MacRumors publish rumors.
It will be Magical.
[doublepost=1495836015][/doublepost]
LOL. Right, just put it all out there.
Last edited:
From my POV Siri will be the killer app but it will be built more and more into the System. In iOS 9-10 we have Siri suggestions for apps, also we have suggestions on maps, face recognition on photos, typping prediction on Keyboard. All these al kind of AI features. If we have a chip which Will be able to do the tasks it might also have some memory built in. What it would mean is; 1st Siri can have context 2nd know some offline tricks (anyway it could send information later to the servers) 3 it can comunicate with other Siri's Chip on other devices, 4 Siri will be able to understand multiple language at the same time because have faster computing available (same as the Keyboard does).
This is like the Secure Element it store your Apple Pay cards, fingerprints etc
Apple is more concerned about privacy than improving Siri, which is wrong, If Siri has access to same data as Google Siri would be great, Siri development is hindered by Tim Cooks privacy first policy.
Two issues with Siri: Reliance on servers for processing and Apple's reluctance to keep dossiers on its customers on those servers.
Moving the processing to the device can help address these problems. The dossier can exist on the device, and the device can provide highly customized assistance even when offline.
One new issue would be the need for each device to have the dossier. There should be a way to keep them in sync while keeping them secure. Otherwise you'll have the situation where your home knows things about you that your phone and laptop don't. We're seeing this already with recognizing faces in Photos.
It's crazy that you're right because Apple had quite a lead on this technology. On board Siri (enclave) could potentially help with this. Maybe on this very chip?
Googles processors are an entirely different level of hardware. They sit in massive chassis in data centers. Apple is looking for a solution that fits into the pam of your hand.
Don't get me wrong I'm excited about the possibility but we can't expect Google Tensor processor levels of performance out of a chip in a cell phone.
[doublepost=1495842025][/doublepost]
Apple does voice recognition very well, the speech to text is very good. Unfortunately they fail at processing natural language. That is they process my voice commands into text pretty good then blow it completely when trying to process that into a reasonable request.
This is baloney! While I agree data is what you are really getting with the current so called AI systems, apples failures ins't in access to data it is in Siri not being able to access that data smoothly. Until I completely gave up on Siri sometime ago, I was extremely frustrated with Siri taking my voice requests, turning it into text and then blowing the processing of that text producing useless data. It wasn't like Siri didn't have access to the data it just couldn't understand what it perfectly translated into text.
I'm thinking it might. I also wonder how Apple it might intergrate this into the Siri speaker. Side note, aside from the Chip, Apple hired a whole team of engineers to revamp Siri. I'm thinking this also will all sync together with the smart speaker and this new chip.
Also, I'm a supporter of Siri. I think it has lots of potential and with the right adjustments, Siri will easily expand.
This is because they aren't AI in the sense that most of us understand. What we have with Siri and these other systems are giant databases that get searched using natural language processing. AI techniques help some with the language processing but it is a stretch to really see the technology as an intelligence.
Frankly even if they use some techniques from the world of AI research, it is a bit of a stretch to call these systems AI's.
Wrong. He wasn't talking about how the market works or how comparisons are made. He made a comment, which I agree with 100%, where they pretty much all suck. Especially with respect to any kind of deep contextual AI.
Siri, as well as others, work OK for being able to give, interpret, and process commands (set a timer, send a message/email, turn on a light, ask for simple information about something, what's the temperature in San Francisco, etc). Beyond that, I haven't seen anything available in the consumer market that's very compelling.
Yep, I've been saying this since the news of an Apple GPU broke. I have a hard time believing that they are going to do a discreet chip, it makes far more sense to roll the tech into the GPU. At least for the hand held devices. Now they could be doing a bigger AI chip for use in the Macs, which could be a discreet chip or maybe not.
They did slip in a bit of tech to support AI at last years WWDC which seemed to have breezed right past most of the Apple user community. I don't see a killer app per say, I see killer features that get leveraged all over the place. Obviously the libraries are available to developers but also to Apple. We could quickly see AI techniques added to Mac OS and IOS to improve many of Apples already shipping products.
[doublepost=1495843809][/doublepost]
Yep this is the key problem with Siri.
Voice translation to text is almost perfect. Processing that text into something that actually delivers the data the user wants fails miserably. This has nothing to do with DATA that some here are implying, rather Siri can't process English into rational queries. Garbage queries deliver garbage data.
[doublepost=1495844075][/doublepost]
This statement is complete garbage. It doesn't take a rocket scientist to figure out that Siri fails at the point of creating the query to pull the data forth cloud. Siri otherwise has all the data it needs to do the job. This can be easily proven too as you can eventually get the at a you want by refining the original question.
In the end Siri is a massive failure at natural language processing.
I was with you until the end. This isn't something that separates Apple. Again this is something Apple has been learning from others. Coprocessors were really common in video and audio cards.
In phones, Motorola integrated a chip for voice recognition even with the phone in idle w/ screen off, and sensing context like if your hand was near the phone (the system was dubbed Motorola X8 Mobile Computing System) and later Qualcomm had created a dedicated place in their CPUs to allow any phone to have the same feature), and so the M7/8/9 Coprocessors was Apple doing what others had already started. I wouldn't call it "copying" either, it was a natural progression to where everyone was going to go as you want to wake the power hungry main CPU as little as possible.
I think Siri struggles with translation. But when Siri does work, it works great. I'm still convinced Apple's servers have a role in Siri's performance.
I've seen many people in Macrumors repeat this claim, but I don't know what sort of experience you may have had with Google Now (or whatever it's called now).
Every Friday I visit my parents and very quickly Google figured this out and on Fridays started giving me travel times to my parents' home. Google Now, using Gmail in the backend, also surfaces travel times, e-mails and reminders automatically. This isn't the sort of AI we expect (as we've seen in Star Trek / Star Wars) but it's a step in the right direction, and from my experience further along than what competitors currently have.
Machine learning is what will likely power AI learning, and as Google demonstrated at their recent I/O, the Photos service will begin using this to automatically start recommending who you probably ought to share your pictures with (i.e. you went out with Sam and Jess, you should probably share these pictures with them, and they should share pictures of that same event with you, and all you have to do is press this button).
I hate to think this is true, but might be... that if Siri was doing all this and Google was behind, that some people here would be singing its praises and saying "baby steps, at least Apple is this far ahead," but because it's competitors that are ahead people are prone to downplaying the advances made thus far.
[doublepost=1495844882][/doublepost]
Agreed. Echo's Alexa is, in my experience, better than even Google Home, though it's pretty close.
Out of curiosity I just (verbally) asked my Apple Watch the travel time to San Francisco. Within a second it gave me the correct answer. While that's nice, and might wow some people at a party or bar, it's not particularly impressive from an AI perspective. As I said, Siri handles simple commands like that just fine - as others do as well.
"I hate to think this is true, but might be... that if Siri was doing all this and Google was behind, that some people here would be singing its praises and saying "baby steps, at least Apple is this far ahead," but because it's competitors that are ahead people are prone to downplaying the advances made thus far."
I've never found it necessary/useful to thump my chest saying one product is doing better in a particular area than another. I'm guessing that's more the domain of people still in high school, needing to validate their favorite technology, and, being competitive.
I don't have knowledge of Motorola's industry with coprocessors, But nor are they really relevant in the smart phone market anymore either. I think my point is Apple is tuning their devices more specific to various products. Hence S1/S2 Chip, W1 Chip, M9/M10, A Series, to each device to Create a sense of exclusivity if you will that simulates different functions. Apple may not be the first to implement the coprocessor, but they can better the process and user experience. Also, Apple's coprocessors are not taxing on the battery as well, which helps with managing the efficiency from the processor itself.
Last edited:
So you feel Echo is advanced AI? Can I tell it to "Book me an evening flight to SF June second on JetBlue and make sure you get me an isle seat but if the flight is more than $500 then let me know before you book it". Obviously you can't because we do not yet have AI. We instead have well written algorithms that are extremely limited in scope.
Apple is not far behind on something that is in its very infancy.