Apple claims that he has found a way to improve his AI models without training user data, and even copy them from their iPhones and Mac. IN He reported the blog for the first time ON By BloombergThe company has presented its plans so that the devices compare a synthetic set of data with samples of the latest E -Mail messages or messages from users who decided on the device analysis program.
Apple devices will be able to determine which synthetic inputs are closest to the real samples that they convey to the company, sending “only the signal indicating which of the variants is closest to the sampled data”. In this way, according to Apple, it does not gain access to user data, and the data never leave the device. Apple will then operate the most frequently selected imitation samples to improve its AI text results, such as E -Mail summaries.
Currently, Apple trains its AI models only on synthetic data, potentially resulting in less helpful answers, in accordance with BloombergMark Gurman. Apple struggled with the launch of the flagship functions of Apple Intelligence, because it pushed the premiere of some possibilities and replaced the head of the Siri team.
But now Apple is trying to reverse, introducing its modern AI training system in the beta iOS and iPados 18.5 and MacOS 15.5, according to Gurman.
Apple talked about the operate of a differential privacy method to ensure the privacy of user data from at least 2016 with the launch of iOS 10 and has already used it to improve the function of Genmoji powered by artificial intelligence. This also applies to the company’s modern AI training plans, because Apple claims that the introduction of randomized information into a wider set of data will support prevent data from connecting with one person.