Apple Intelligence in 15.1 just flagged a phishing email as “Priority” and moved it to the top of my Inbox. This seems… bad
Apple Intelligence in 15.1 just flagged a phishing email as “Priority” and moved it to the top of my Inbox. This seems… bad 59 comments
@cabel “AI” isn’t ready. And if it can fool a person, it’s certainly gonna fool a 1 year old after an extreme training regimen of 1 month. @cabel Apple Intelligence is recognizing those words as an important account notification. @SasquatcherGeneral @cabel And I'm rolling an Apple Dexterity check to get the Magic Mouse to right click on it. (Oooh, critical fail! It's off in a Space of its own now.) @michaelgemar @cabel that's the problem with AIs: by definition, they can't be tested entirely, because nobody really knows what they will reply to a given request. They only give replies based on some probabilities ("predictions") that have been computed from data before. @Deuchnord @cabel Right, but these issues crop up *so* quickly after their release — are they not testing at all, or at least some obvious edge cases? I feel like no one is internally “red teaming” these models. @michaelgemar @cabel well, at least, we're speaking about a feature expected to be available for iOS 15.1, in October-November 2024. I suppose Apple will try to fix that, even though it will be hard... @Deuchnord @cabel I would have hoped these highly-publicized features would be more fully baked before even releasing them as betas. @cabel How is no one else in this thread not talking about your frequent pork bun issues that need attention?! @patc Hahahahah Ramp really needs me to provide my Pork Bun receipts —Normal Work Sentence @cabel so intelligent. :neocat_think_googly: Do not want :neocat_reject: (If you’re at Apple, by using the wild new 'Rate Your Experiences’ feedback system I apparently filed a bug on this: FB14656882) @Eramdam Absolutely!! Write an appropriately urgent-sounding spam message and surely the AI will give it credence and credibility by putting it in its own special little important section @cabel yep… I hope they’re smart enough to take email headers into account even if that’ll be tricky to like, prompt or something but it really is the only way to work around this I feel 🫠 And/or Mail should start being able to tell the user when an email is a phishing attempt/spoofing by checking SPF/DKIM records and such. Thinking a bit too much about email because of work 😭 @cabel LOL This is hilarious. What could possibly go wrong with all this intelligence put in every nook and cranny? Love it, thanks for sharing! 🙂 @cabel it is Not ready for public release yet, obviously. Oh wait it’s also Beta. @cabel We’re moving quickly towards an AI vs AI world. Soon you won’t be able to trust anyone who isn’t standing right in front of you. @cabel if you are a security researcher / spam blocker this _may_ be on your priority list. @cabel I’ve found the iMessage notification summaries to be laughably bad too. Misinterprets some part of the message like a quarter of the time maybe? Hard to tell because it stands out so clearly compared to the times it does it well. @cabel just return back me the dumb af simple inbox as it was in 1990's and I can sort it out. No apples, no pears, no windows :) In a hilarious follow-up, my dad forwarded me a phishing email just to check with me if it was legitimate. I wrote back and said “Definitely not!”. He wrote back and explained, "I got suspicious.” This is how Apple Intelligence summarized my dad’s emai @cabel … your dad is running the beta? Or does intelligence show summaries for outbound? @cabel what you get when a bunch of managers instruct everyone to use "latest AI" on everything, when a simple decades old approach based on rules and simpler ML would have caught that. @cabel so far on my end it almost seems like somebody accidentally left a negative value in the proverbial "target priority" variable lol. ... on a serious note though, given what we know about LLMs... how could they ever be actually good at junk mail parsing ??? *unless* you really did prompt... "everything *you* gauge to be a priority goes straight to junk and all the stuff you were junking before gets flagged." |
@cabel Hey, it's probably a priority for somebody!