A new report by Mark Gurman at Bloomberg is shedding light on Apple’s latest plans for AI.
This has to do with carrying out measures that are in line with the company’s privacy-first model, but at the same time, ensure it can collect real-world snippets without sending data back to servers. This might be in the form of emails stored inside models like Mac computers, iPads, and iPhones.
The iPhone maker shared more on this front, including how the goal is to produce synthetic data that’s as close as possible to the real thing. This can assist models in generating summaries, but it’s done without actually collecting the data from a certain device.
The company’s entire AI platform is called Apple Intelligence, and we’ve seen it lag behind other AI giants in the industry, such as OpenAI, Google, and even Microsoft. A major reason is linked to how the company was working on designing AI tools that didn’t really make use of real-time users’ information, as it was against its privacy practices. After all, who wants to engage in another battle about privacy, right? The issue with that was that most information didn’t really feel like actual interactions from users, and that produced poor results.
The company’s tools for writing and generating summaries haven’t been getting a lot of rave reviews. Many of the alerts don’t make sense. Quite a few summaries aren’t on the right mark, and Siri has its own issues where it continues to fail at the one job that it’s supposed to do.
The whole matter could spark a major shake-up, leading to delays in the release timelines and giving off the aura that it’s just not ready to tackle the world of AI yet. At this moment, the latest system wants to lessen the clutter and get the AI job done seamlessly. So what better way than to train models with data that’s as close to the real deal as possible?
The goal here appears to be linked to Apple peeking into emails, without actually storing or reading. This can better calibrate the synthetic data for training, provide better message summaries, and enhance suggestions for writing.
Obviously, the changes would be applicable to all users who opt into this through analytics and settings for product betterment.
Image: DIW-Aigen
Read next: Meta Prepares to Train AI Models on EU Users’ Data Taken from Facebook and Instagram
This has to do with carrying out measures that are in line with the company’s privacy-first model, but at the same time, ensure it can collect real-world snippets without sending data back to servers. This might be in the form of emails stored inside models like Mac computers, iPads, and iPhones.
The iPhone maker shared more on this front, including how the goal is to produce synthetic data that’s as close as possible to the real thing. This can assist models in generating summaries, but it’s done without actually collecting the data from a certain device.
The company’s entire AI platform is called Apple Intelligence, and we’ve seen it lag behind other AI giants in the industry, such as OpenAI, Google, and even Microsoft. A major reason is linked to how the company was working on designing AI tools that didn’t really make use of real-time users’ information, as it was against its privacy practices. After all, who wants to engage in another battle about privacy, right? The issue with that was that most information didn’t really feel like actual interactions from users, and that produced poor results.
The company’s tools for writing and generating summaries haven’t been getting a lot of rave reviews. Many of the alerts don’t make sense. Quite a few summaries aren’t on the right mark, and Siri has its own issues where it continues to fail at the one job that it’s supposed to do.
The whole matter could spark a major shake-up, leading to delays in the release timelines and giving off the aura that it’s just not ready to tackle the world of AI yet. At this moment, the latest system wants to lessen the clutter and get the AI job done seamlessly. So what better way than to train models with data that’s as close to the real deal as possible?
The goal here appears to be linked to Apple peeking into emails, without actually storing or reading. This can better calibrate the synthetic data for training, provide better message summaries, and enhance suggestions for writing.
Obviously, the changes would be applicable to all users who opt into this through analytics and settings for product betterment.
Image: DIW-Aigen
Read next: Meta Prepares to Train AI Models on EU Users’ Data Taken from Facebook and Instagram