A number of issues were discovered in the pre-release version of Apple Intelligence, including 'mistakes in summarizing news' and 'prioritizing scam emails'



Apple's personal AI, '

Apple Intelligence ,' is said to be able to perform tasks such as summarizing notifications, proofreading text, and generating images, and is considered a highlight feature of the iPhone 16 series to be released in September 2024. However, Jeffrey Fowler, a columnist for the daily newspaper Washington Post, reported that when he actually used a pre-release version of Apple Intelligence, he found problems such as 'incorrect news summaries' and 'prioritizing scam emails.'

iPhone 16's Apple Intelligence is useful so far, except when it's bonkers - The Washington Post
https://www.washingtonpost.com/technology/2024/09/09/iphone-16-apple-intelligence-ai-event-2024/



Apple has touted the iPhone 16 series, which will be released on September 20, 2024, as the 'iPhone for Apple Intelligence,' and is promoting it as a fundamental revolution in the iPhone experience. However, Apple Intelligence was originally scheduled to be available for testing in the United States in the summer of 2024, but it has been reported that some functions, such as image generation, have not yet reached the performance required for release.

Apple Intelligence's image generation feature 'Image Playground' and ChatGPT support may be available from iOS 18.2 - GIGAZINE



Meanwhile, Fowler had the opportunity to test a pre-release version of Apple Intelligence, so he tried out various functions. He said, 'When I test the pre-release version, it's sometimes useful, but sometimes it's so funny that it makes me laugh,' and pointed out that it's not necessarily accurate.

Apple claims that the pre-release version of Apple Intelligence is still in development and that the final product will not have the same performance. Fowler does not usually write about beta or pre-release software, but this time he decided to write a review of the pre-release version because Apple spent more than 15 minutes on Apple Intelligence at the launch event and promoted it heavily.

Fowler said Apple Intelligence is 'in some ways a smarter, and probably safer, way of bringing AI to consumer products than what other technology companies have offered up until now.' That's because Apple Intelligence isn't an 'all-knowing, all-powerful chatbot that answers all your questions,' but is instead used to integrate with existing apps to speed up tasks.

For example, Apple Intelligence can search for photos using natural sentences and highlight obstacles in photos. In addition, while the traditional Mail app displays the first two or so sentences of an email in the inbox list, Apple Intelligence displays a summary of the entire email, making it easier to decide whether to open the email, Fowler said.

In addition, there is a function to prioritize notifications that are considered to be of high importance among multiple notifications. In this way, Apple Intelligence is basically used to make existing apps more convenient, so even if there is a mistake in the operation, there is little harm to the user.



Still, Fowler said, 'It's inevitable that Apple Intelligence will get things wrong from time to time. It will misinterpret the meaning of text, write people's names backwards, or edit images in a way that's offensive.' He also noted that Apple Intelligence has made a lot of mistakes.

For example, the following are news notifications from the social networking site

Truth Social . The top one is a regular notification, and the bottom one is a notification summarized by Apple Intelligence. As written in the top notification, the news is that 'the brother of Tim Walz , the Democratic vice presidential candidate, has endorsed Donald Trump,' but the summary by Apple Intelligence is that 'Trump has endorsed Walz.'



'This is because this generation of AI technology is inherently plagued by misconceptions of fact that the AI industry tries to downplay as 'hallucinations.' Generative AI is not designed for truth, it's designed to find patterns and reproduce. Google's Gemini and OpenAI's ChatGPT have the same problem,' Fowler said. According to Fowler, Apple Intelligence's misconceptions of fact occur frequently, and he comes across 5 to 10 hilarious mistakes a day.

Below is an example of an Apple Intelligence mistake that Fowler actually encountered.

- The notification summary from the doorbell app becomes nonsensical, such as 'There were multiple people at the front door and in front of the house, and there was recent activity at the front door.'
- Not understanding humor can mess up chats with friends and family and summaries of photos you've taken.
- When you use the photo editor to shave your messy hair off of a selfie, instead of slicking it up, it makes you look bald.

While these mistakes were mostly just funny stories, one email summary that read, 'If you don't recognize your recent login, we recommend you change your password,' was changed to, 'Change your password immediately.' Another was when a scam message claiming that a user's Social Security number had been deactivated was highlighted as a high priority message.



'In the preview that I use, Apple Intelligence makes up an obnoxious amount of fabrications,' Fowler said. 'This usually happens with less important information like app notification summaries, but it feels odd to see lies and misinterpretations in core areas of your iPhone like your lock screen and inbox.' 'I can't think of another time when Apple has asked you to pay for something so unfinished.'

in Mobile,   Software, Posted by log1h_ik