Apple Company • Page 968

Either way I would have felt a lot more comfortable if Apple came out and said “we’re pumping the brakes on AI until we can work out the best way to do it safely in house and we’re prepared to take on some leadership here even if it means copping flack for being slow” instead of this move which seems a bit like a panicked response to appease Wall Street

 

Blog - Private Cloud Compute: A new frontier for AI privacy in the cloud - Apple Security Research

A deep read on how they're trying to ensure trust with PCC.

 

So other than ChatGPT through Siri, my understanding is that it is all in-house. These are their LLMs, running on device until no longer possible, and then running in PCC in those scenarios.

Apple doesn't get enough credit from the industry but their development of open-source on-device LLMs in a very short amount of time has been impressive. They're taking the hard way, and in my opinion the right way, and making quick progress. They hamstrung themselves by releasing iOS devices with less than 8 GB of memory until the M-series SoCs and iPhone 15 Pros.

 

Powering off and on seems to have fixed it for me

 

Based on what I read this evening, that’s correct. Lots of people (Elon) are confusing the limited OAI parts with all the announced places where Apple Intel or whatever is being used. The OAI scope is quite limited in comparison.

 

And that every prompt to ChatGPT is opt-in as far as we can tell. I’m sure that changes if you’re a paying user of it. There’s also the statement that Apple said additional LLM platforms could be added so that would even include Grok. He’s just being an idiot with a purpose and that’s to get attention.

 

Thread by @matthew_d_green on Thread Reader App

some good thoughts on PCC. I’m sure these same thoughts have come up with Android and Google but most people ignored that Google has been doing much of this processing with your data on their servers for two decades now.

Is PCC perfect? No. Am I willing to trust it for my own use? For now, yes. Definitely more so than Gemini, Grok, or Copilot.

 

Introducing Apple’s On-Device and Server Foundation Models

 

Why did it take so long for them to tell me I can remove shit with AI from my photos? Literally only thing I cared about.

Honestly the only thing I can ever get Siri to do is set a timer so let’s see how this plays out

 

I think this did the trick. See this is why I love Drew, I don’t care what he says in the NBA thread… or the wrestling thread even though we usually are on the same side there

 

I don't want to make a wholesale export of 1password but I also don't want to go through the process of tagging what I want and then parsing it out myself.

 

I'm likely going to go through the full effort of tagging the passwords I want and potentially deleting the accounts of ones I don't. Will take me too long but might be worth it

 

If you have both Passwords and 1Password enabled in Safari (at least) it'll usually prompt you if you want to save the password as long as you use the 1P autofill shortcut. I was able to get my regularly used passwords moved over but it's those rarely used passwords I'm worried about.

 

I'll need to see the app first, and see what's in and what it can store.

Probably can't move over until there's family sharing for passwords though, that feature to share common passwords with Hannah is just too ingrained into our workflows.

 

The screenshot on the Verge has a shared groups section, so I'm assuming that's a similar feature to 1Password. Definitely a requirement for us to move too though.

 

Family Sharing has been inherent with iCloud Passwords for at least a release now. Uses iCloud Family. Just have to set up the group.

 

That's exactly it.

 

For those of us who have been designated as the family technology coordinator (who also lives on the opposite coast of said family) the fact that screenshare will now let me control my family's iPads remotely might be one of the announcements yesterday that will have the biggest positive impact right out of the gate. lol

 

As long as they can upgrade. I think I’m screwed on a few devices

 

Apple details two Apple Intelligence language models: ~3B parameter on-device and a larger model on Apple silicon servers available with Private Cloud Compute