At the beginning of the week, Apple’s AI was released in Swedish in phones and tablets. Apple Intelligence promises a lot and in your iPhone and iPad should be able to help with email responses, summarize and translate, let AI prioritize notifications and email. I am also promised to receive, for example, my dinner invitation to a poem, which is why I would now like it. Because when I test, the result is completely incomprehensible to the recipient, and also lacks a clear connection to the invitation I wrote, so it just becomes a random more or less detached poem without purpose.
It is now that we get into the conclusions quite simply. Almost nothing works as I would wish or expect with the lowest of hopes.
The more I test, the more afraid I become to entrust anything of importance to Apple’s AI.
At the beginning of the week, Apple’s AI was released in Swedish in phones and tablets. Apple Intelligence promises a lot and in your iPhone and iPad should be able to help with email responses, summarize and translate, let AI prioritize notifications and email. I am also promised to receive, for example, my dinner invitation to a poem, which is why I would now like it. Because when I test, the result is completely incomprehensible to the recipient, and also lacks a clear connection to the invitation I wrote, so it just becomes a random more or less detached poem without purpose.
It is now that we get into the conclusions quite simply. Almost nothing works as I would wish or expect with the lowest of hopes.
The more I test, the more afraid I become to entrust anything of importance to Apple’s AI.
Unsurprisingly, the AI misses 90 percent of my deliberate mistakes when I ask it to proofread a letter I’m about to send.
When Apple Intelligence summarizes a newsletter I receive, it does so with the phrase “various short news items on various topics.” It couldn’t get more bland, less useful. Especially as I know the newsletter obviously has a theme, it’s specifically about mobile infrastructure so really not as non-specific as the AI perceived it to be.
The next time I respond to an email message, Apple Intelligence pops up with a suggestion for a pre-written response. It is supposed to be able to figure out and write a good answer on my behalf, based on what questions were asked in the email I received.
When the “intelligence” is ready, it has mixed up who I am and who it is that sent me the email. So instead of writing my answer, it writes me an email. “Hey Erik”.
In the reply, it then takes and describes what my email contact just wrote, but the other way around and signs with her name.
I mean, it’s laughable, it’s entertaining because it’s so bad. But it’s not good by any means and it’s surprising that it even comes out in sharp software, even though Apple mentions that AI can make mistakes. That seems to be the only thing it can do.
Of course, I had Apple’s AI proofread this column as well before I published it. It missed all my conscious errors, (a misspelling of “signs” and the misspelling of “porocent” but changed Iphone to iPhone and Ipad to iPad. So I had to manually correct Apple Intelligence’s error.
Another disadvantage of AI corr in this form is that Apple Intelligence does not indicate what it has changed, so if you are the least bit worried or unsure about what it has done, you have to review it yourself and look for any changes compared to the original text.