Apple was the first big company investing big into AI, when it bought the startup company Siri. Now look at how far behind Apple is in AI field comparing with all of the other big companies. What a new leadership!!!
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 Partners With Alibaba to Bring iPhone AI Features to China
- Thread starter MacRumors
- Start date
- Sort by reaction score
Elon Musk Has Been Promising Self-Driving Cars For 10 Years [Update - We Are Now On Year 11]
Elon Musk Has Been Promising Self-Driving Cars For 10 Years [Update - We Are Now On Year 11] - Jalopnik
Elon Musk has been promising self-driving cars are right around the corner for the last decade. And what a decade it’s been! Full of scandal, intrigue, and poorly manufactured cars and car accessories. We’ve seen the rise of the electric vehicle, the death of the affordable American car and the...
jalopnik.com
I'm sceptical of any of Musk's claims.
I begged to defer. These are features that will make the car a lot safer. I like to drive and I don’t like to cede full control to the cars. But if it makes it safer to address instances when I am tired and inattentive, they can take my money. Anyway, apologies for going off topic 😆
I’ll believe it when it lands. Tesla has too much of a history of being years and years late with their promises to take them at face value.
That's just silly whining.
First, many people are very much against the anti-AI push. What do you say to them?
Secondly, you're presuming that every big company has to do the same thing that every other big company does. Which is absurd.
Well no one forces them to rely on chat GPT in the US
I think they just don’t have the right software development team in place for stuff like this.
They can barely do emojis nowadays.. and even then, at least once a day when I go to select an emoji I get a blank page..! Absolutely pathetic!
Good to hear about this. Should help to bring Apple Intelligence to China and we should hear more about it at WWDC. Rollout might begin with iOS 19
It’s really not. If OpenAI made a censored version of ChatGPT I’m sure they would be able to operate in China. They probably just didn’t want to because of the costs and bad PR involved in building that kind of system.
Well no one forces them to rely on chat GPT in the US
I think they just don’t have the right software development team in place for stuff like this.
They can barely do emojis nowadays.. and even then, at least once a day when I go to select an emoji I get a blank page..! Absolutely pathetic!
You must be joking. Please explain the basics of Apple's AI implementation.
Elon Musk Has Been Promising Self-Driving Cars For 10 Years [Update - We Are Now On Year 11]
Elon Musk Has Been Promising Self-Driving Cars For 10 Years [Update - We Are Now On Year 11] - Jalopnik
Elon Musk has been promising self-driving cars are right around the corner for the last decade. And what a decade it’s been! Full of scandal, intrigue, and poorly manufactured cars and car accessories. We’ve seen the rise of the electric vehicle, the death of the affordable American car and the...
I'm sceptical of any of Musk's claims.
I know.
But this time...
Not much to say, they leverage chat gpt when you ask Siri a question that goes further than “what’s the time” or “how’s the weather” and the user interface is a mess. Without mentioning that if you ask a follow up question sometimes you get an answer that is not related to the previous answer because Siri tries to take over.You must be joking. Please explain the basics of Apple's AI implementation.
The object removal tool in photos is garbage, it’s not even generative, it’s just adding content based on the nearby pixels, photoshop was doing that in the 90’s. It’s a clone/stamp tool basically, with ZERO ways to influence the outcome.
Image playground is a waste of space, I think also kids use it twice and get bored, the capacity to generate content is extremely minimal and the way it’s structured is everything but intuitive.
Writing tools are so fiddly and awkward to use that it’s easier to use Actual Intelligence to write a few sentences…
Oh. I’m adding this… my partner’s name, which I won’t write down here obviously, is not picked up by my Apple watch anymore. If I say call (fake name for the sake of an example) Hannah Jones, the watch keeps telling me I can’t find Anna John in your contacts. There is no way I can get it to call or text my partner on my watch anymore. It works perfectly on iPhone.
Go figure it……
And this is Apple AI implementation my friend….
It's a tiny bit more complicated than that. Apple has built an entire layer that you're missing. And I think some of that additional complexity is what's making the current performance sub-optimal. don't get me wrong, I agree that it needs work, but it's far more complicated than designing a few emojis.
I used AI to generate these comments below as it can explain it better than I can, but there are some good Apple documents on this too, particularly PCC.
There's a couple of good articles here on PCC:
Blog - Private Cloud Compute: A new frontier for AI privacy in the cloud - Apple Security Research
Secure and private AI processing in the cloud poses a formidable new challenge. To support advanced features of Apple Intelligence with larger foundation models, we created Private Cloud Compute (PCC), a groundbreaking cloud intelligence system designed specifically for private AI processing...
Blog - Security research on Private Cloud Compute - Apple Security Research
Private Cloud Compute (PCC) fulfills computationally intensive requests for Apple Intelligence while providing groundbreaking privacy and security protections — by bringing our industry-leading device security model into the cloud. To build public trust in our system, we’re making it possible...
Everything from here is from AI:
The three-tiered architecture of Apple Intelligence, as described in various sources, breaks down into distinct layers that work together to provide a seamless, secure, and efficient AI experience across Apple's ecosystem. Here's a detailed explanation based on the information available:
1. On-Device Tier
- Components:
- AFM-on-device: A language model with approximately 3 billion parameters designed to run directly on devices like iPhones, iPads, and Macs. This model handles tasks that do not require extensive computational resources or privacy-sensitive data processing.
- Semantic Index: A feature that organizes and surfaces personal context from the user's apps and data. It helps in understanding context relevant to the user's query without transferring personal data off-device.
- Functionality:
- This tier is responsible for processing AI operations on the device itself. It ensures privacy by keeping personal data on the device, leveraging Apple's powerful M-series silicon chips for on-device inference.
- Tasks include basic language understanding, instruction following, reasoning, writing, and tool use. Specialized adapters enhance performance for specific tasks like summarizing emails or notifications.
- Advantages:
- Enhanced privacy and security as data does not leave the device for most operations.
- Faster response times due to on-device processing.
2. Private Cloud Compute (PCC) Tier
- Components:
- AFM-server: A larger language model hosted on Apple's servers, which is used for operations that require more computational power than what's available on-device.
- Functionality:
- This tier is invoked when the task at hand exceeds the capabilities of the on-device model or when privacy considerations allow for cloud interaction.
- It deals with data minimization, where only the minimum data necessary for the request is sent to the cloud.
- The system uses cryptographic measures to ensure that devices only communicate with verified servers, enhancing security.
- Advantages:
- Access to more powerful computing resources for complex AI tasks while still maintaining a high level of privacy and security.
3. Integration and Orchestration Layer
- Components:
- AI Orchestration: This is not a separate physical tier but rather a conceptual layer that manages how tasks are routed between on-device and server-based models based on the nature of the request, privacy implications, and performance needs.
- App Intents Toolbox: Tools for developers to integrate AI capabilities into their apps, leveraging the Apple Intelligence system.
- Functionality:
- This layer ensures that AI tasks are efficiently distributed between the on-device and cloud tiers. It decides when to use which model or combination of models for optimal performance, privacy, and user experience.
- It also facilitates the interaction between different applications and services, allowing for a more integrated and context-aware experience across Apple's operating systems.
- Advantages:
- Provides a seamless user experience by dynamically choosing the best execution path for AI tasks.
- Enhances the functionality of apps through AI, making them smarter and more responsive to user needs.
Conclusion
Apple Intelligence's three-tiered architecture combines on-device processing for privacy and speed with cloud capabilities for more complex tasks, all managed by an orchestration layer that ensures the right balance between performance and privacy. This model not only leverages Apple's hardware for efficiency but also reflects their commitment to user privacy and data security.
------
Private Cloud Compute (PCC) is a key component of Apple's AI strategy, specifically designed to enhance privacy, security, and performance in cloud-based AI operations. Here's a deeper dive into what PCC entails:
Core Concepts:
- Privacy by Design:
- Data Minimization: PCC only sends the necessary data to the cloud for the task at hand, reducing the exposure of personal information.
- Stateless Processing: Each request to the cloud is treated independently; no data from previous operations is retained, ensuring that personal data does not linger on remote servers.
- Security Measures:
- End-to-End Encryption: Data is encrypted in transit and at rest, ensuring that only the user's device can decrypt and use the processed results.
- Verified Boot: Devices only communicate with servers that have been cryptographically verified, preventing man-in-the-middle attacks or rogue server interactions.
- Performance Optimization:
- Dynamic Task Allocation: PCC allows for tasks that are too complex for on-device processing to be handled by the cloud, providing access to more powerful computational resources when needed.
Architecture and Operations:
- AFM-server: This is the server-side model of Apple's foundational language models (AFM). It's much larger than the on-device model, allowing for handling of more complex AI tasks, such as detailed image processing, advanced text generation, or complex data analysis.
- Orchestration: When a task is initiated, Apple's system decides whether it should be processed on-device or if it requires the capabilities of PCC. This decision is based on factors like the complexity of the task, the sensitivity of the data involved, and the current capabilities of the device.
Technical Implementation:
- Hardware Security Modules (HSMs): These are used to manage cryptographic keys securely, ensuring that data encryption and decryption are handled with high security.
- Secure Enclave: On the device side, the Secure Enclave, a part of Apple's chips, plays a role in securing data before it's sent to the cloud and in verifying the integrity of the cloud compute environment.
- Custom Silicon: Apple's use of custom silicon like the M-series chips in their devices helps in running secure, hardware-accelerated cryptographic operations both on-device and for securing cloud interactions.
Transparency and User Control:
- Transparency: Apple commits to transparency in how data is handled by PCC. Developers can view the software images running on PCC nodes, and there's a promise of no backdoors or hidden access points.
- User Control: Users have control over which applications can use PCC, and there's an option to disable cloud computing for all operations if preferred, though this might limit some functionalities.
Future Implications:
- Scalability: As AI models and tasks become more complex, PCC provides a scalable solution that can grow with the demands without compromising on privacy.
- Innovation in AI: By allowing more intensive AI tasks to be offloaded to the cloud securely, developers can create more sophisticated applications without worrying about the limitations of on-device processing for all scenarios.
In summary, Private Cloud Compute is Apple's approach to balancing the need for advanced cloud-based AI computation with their foundational commitment to privacy and security. It's a critical part of their broader AI strategy, ensuring that users can benefit from powerful AI features while maintaining control over their personal data.