PodSearch

ATP

602: A Glimpse of a Better World

 

00:00:00   So I had the pleasure today of unboxing one of the new iMacs. Have either of you done this?

00:00:04   I don't think so. Why, yeah, I haven't. Why are you, why are you in possession of a new iMac?

00:00:09   It's not for me, I was helping someone else set it up, but I was unboxing it and I gotta say, it was

00:00:14   a pleasure. Like, it gave me such kind of old Apple fun vibes, even though, like in certain ways it's

00:00:21   very modern, you know, because it has all of Apple's all like, you know, paper-based packaging now,

00:00:25   like, you know, there's no plastic anywhere. Which I really do like, by the way, I think we've talked

00:00:29   about that in the past, but generally speaking, Apple's packaging has gotten really, really,

00:00:33   really good. And aren't they using like the bags from the Apple store? I think like even the handles,

00:00:38   which feel like their rope, are actually like recycled paper or something like that. Yeah,

00:00:43   that was one of the first paper things they did ages ago, yeah, years and years ago. But yeah,

00:00:47   so the iMac box is like, it's in many ways, it has a lot of the similar elements as like the Mac Pro

00:00:53   and XDR boxes with like those big like arrow panels that like kind of butterfly out when you

00:00:57   open it up. But of course, obviously in the consumer line, but it's still like, it's pretty

00:01:03   elaborate cardboard work to make like this pretty well cushioned thing with no plastic anywhere in

00:01:09   sight. It's pretty remarkable. But what's nice about it is this was a green iMac. And you know

00:01:14   how like, you know, the typical iMac box, you have like the outer cardboard, like the brown cardboard

00:01:18   shipping box, and then inside of it, you have the nice white Apple box. Then you have a little

00:01:22   handle on top, the handle, it's green, you open it up, everything that can be tinted green is tinted

00:01:30   green, like because this is the color of the iMac. So like, first of all, you open up the box and you

00:01:34   see the the like screen overlay, like the big sticky thing that covers up the screen for

00:01:39   protection. On the front of it, it says hello written in the Macintosh Hello style, you know,

00:01:45   so like the first thing you see you open it up that the cardboard butterflies out and you see

00:01:48   it saying hello under like the translucent paper cover. It's delightful. And every single accent,

00:01:56   it's the it's the accent color of the computer, like, it was so well done. It very much reminded

00:02:02   me of like, you know, old Apple be back when we were all buying desktops. I mean, John still is,

00:02:06   but no one else. It was honestly it was delightful. And so I just wanted to kind of shout that out,

00:02:12   like, in a world where most of us aren't even buying desktops anymore. It was really nice to

00:02:16   see Apple putting a ton of little detail work into making the desktop unboxing experience in 2024.

00:02:24   Pretty delightful. I think the candy color iMacs predate both of your Mac use, but that was back

00:02:30   when they were shipping with Mac OS 9. And they did the same thing where Mac OS 9 would ship and

00:02:35   the accent color inside the interface, like the basically the scroll thumb color and like the

00:02:40   text selection color and everything would match the color of the computer you got. Yeah. And the

00:02:43   background to a desktop background, the full desktop background. It's a it's a it's an iMac

00:02:48   tradition. Apple, as we often discuss on the show, they really could use more use of color

00:02:55   in their product line, please. But when they do it, they do a nice job.

00:02:59   I know as we're recording this, it's the 28th of August. And that means it's not quite September,

00:03:06   but you might be listening to this in September. And you know what, darn it, we're close enough

00:03:09   to September. And what is September, gentlemen? It is Childhood Cancer Awareness Month, which means

00:03:15   it's time for the Marco offset. What are we talking about? So hey, if you live in a backwards

00:03:22   country like we do, health care is a real problem. And if you live in any country, then childhood

00:03:29   cancer is a real problem. And there is an establishment in St. Jude, in St. Jude, in Memphis,

00:03:35   Tennessee, excuse me, called St. Jude Children's Research Hospital. And they are a truly phenomenal

00:03:41   organization that wants to end childhood cancer. They would like no child to die from childhood

00:03:47   cancer ever. That's their goal. Their final form is to have them just go away because childhood

00:03:55   cancer is cured. That is the goal. So why do you care? Why am I telling you about this? Because

00:04:00   September, like I said, is Childhood Cancer Awareness Month. And our friends at Real AFM,

00:04:04   which is also us, you know what we mean, they/we are trying to raise as much money as possible for

00:04:11   St. Jude Children's Research Hospital. Currently, as we record, the goal is $500,000. The campaign

00:04:17   has been officially open for eight-ish hours, and we have already raised $30,000, which is truly

00:04:25   incredible. That is so very cool, but we can do more. And I know that both the Relay and ATP

00:04:31   listeners really, really do come out and support this wonderful organization. So if you're

00:04:37   interested, you can go to stjude.org/ATP to kind of fast track your way straight into the donation.

00:04:43   Or if you'd like more information, you can go to stjude.org/relay. And hey, here's the thing.

00:04:49   Relay's been trying to get to $3 million lifetime. We've already crossed it. Honestly, I think we

00:04:56   should just shoot for four. Let's shoot for four this year. You never know. Could happen.

00:04:59   It's an audacious goal, but it could happen. So how do we get to $4 million, which is my goal,

00:05:05   not Relay's? Then you got to donate. So stjude.org/ATP. Let me just kind of talk about a

00:05:11   couple of things real quick, and then I'd like to explain the Marko or have Marco explain the Marco

00:05:15   offset. So here's the thing is that 400,000 kids worldwide get cancer each year. 400,000. That's a

00:05:22   lot of kids. In many countries, four out of the five kids who develop cancer will not survive,

00:05:28   mostly due to lack of access to quality care. So St. Jude and Relay and ATP, we all think that's

00:05:35   no good. That is no good at all. So what are we trying to do? We're trying to raise money to help

00:05:40   fix that problem. So St. Jude needs your help because St. Jude, they pay for everything. They

00:05:46   pay for patient families to come and be with their sick child. They pay for everything. I

00:05:52   actually went to Memphis in April for a big conference for fundraisers, and I cannot eloquently

00:05:59   describe what it's like to be there. It is unreal seeing a mass of people so wholly dedicated to

00:06:06   just saving lives. It's phenomenal. And the way they can afford to do this is from people like you

00:06:12   and me and donating at stjude.org/ATP. In a couple of weeks, I will be going back to Memphis. I'll be

00:06:20   going back to St. Jude campus in order to participate for my very first time in a 12-hour

00:06:25   telethon event, which we call the podcast-a-thon, with our friends Mike Hurley, Stephen Hackett,

00:06:32   Kathy Campbell, and Jason Snell. The five of us will be there for 12 straight hours,

00:06:36   and you can watch me melt into a puddle during that time. It'll be great. It'll be fun for

00:06:39   everyone. So again, please, if at all possible, go to stjude.org/ATP. Here's the thing, and they've

00:06:46   kind of asked me to read this, but I think it's accurate. So when we rally for a common cause,

00:06:51   we become more than a community. We become beacons of hope for all. I mean, think about that. Think

00:06:55   about if your kid was sick or your sibling was sick and suddenly somebody swoops in and so many

00:07:02   words says, "Don't worry, I got this," because that's what St. Jude is basically doing. So

00:07:05   that's why we're asking all of you to join Relay and St. Jude in ATP this September for

00:07:08   Childhood Cancer Awareness Month. Together, we can help cure childhood cancer. Now, okay,

00:07:12   what if you don't care about childhood cancer, which you should? What if you just want to feel

00:07:16   better about your rampant consumerism? Marco, how do we do that? So here's what's about to happen.

00:07:21   We're about to have Apple announce a whole bunch of tech devices, mostly iPhones, probably Apple

00:07:26   watches, and maybe some new Macs later in the fall. So this is Apple device consumerism season.

00:07:33   This is when all of the best and most popular devices come out, and many of us try to find

00:07:38   justifications for why we absolutely need to buy these things even though we have a phone that

00:07:43   works fine and we have a watch that works fine. But we could theoretically not buy these things.

00:07:49   But you and I both know that's not going to happen. Of course, we're going to buy these

00:07:53   things. So here's how you assuage your guilt as a rampant consumer who is wastefully throwing away

00:07:59   your money on phones and watches and Macs and things that you don't necessarily need. I mean,

00:08:04   look, you could use your old one for a little while longer, right? But you won't. So we know

00:08:08   you're going to buy new stuff. Here's what you do. Donate a chunk of money to St. Jude when you do

00:08:14   this. Think about whatever the, you know, what case is describing as the Markov offset. This is a

00:08:18   term we've come up with over the last couple of years for describing. Here's my strategy for how

00:08:23   to figure out what a good minimum donation to St. Jude is. When you buy your new Apple product this

00:08:29   fall or Apple products this fall, think about amounts of money that are kind of added on during

00:08:35   the process that you kind of just accept or tack on without thinking too critically about them.

00:08:42   Things like sales tax or cost of accessories, cost of storage upgrades, cost of Apple care,

00:08:49   these kind of additional costs that are above and beyond the base price of the, you know, the most

00:08:55   basic model of thing that you could get. So if the iPhone starts at a thousand bucks of the line that

00:09:00   you want and you say spend another hundred bucks to double the storage and then you spend another

00:09:05   60 bucks on the case, you're at 160 bucks. Maybe you spend another, you know, 60 bucks on tax,

00:09:10   you're at 220, I think. Yeah. So you're at 220 with those kind of add-ons over the base.

00:09:17   That is your minimum donation to St. Jude. So here's what you do. Whenever you buy these

00:09:22   products in the next, you know, coming weeks and months, well, hopefully coming weeks because then

00:09:27   it'll be during the drive. But when you place your order for your new iPhone and Apple Watch in a

00:09:32   couple of weeks, maybe, figure out that baseline minimum of the add-on price that is like,

00:09:38   what are you getting that's a little bit extra? That's your minimum. So in my example, my minimum

00:09:43   was 220, I think, if I remember correctly. I've already forgotten. I'm talking a lot.

00:09:46   So 220 is your minimum, then that's your minimum donation to St. Jude. So if you've done that

00:09:54   configuration there, consider a 220 donation to St. Jude. Now, if you are fortunate enough,

00:09:58   you can give more, that's great. We encourage that. But that's kind of a good baseline of like,

00:10:03   here's an amount of money that while making my frivolous technology purchase that I purchased

00:10:08   that I probably could have gone another year or two without making, when I'm making this

00:10:12   frivolous technology purchase, I'm throwing away another $200 in just add-ons and fees and taxes.

00:10:18   If that, if I can spend it that easily, I can give at least that much to this amazing charity

00:10:23   doing amazing work for the world. So that's the Marco offset. I encourage you to give generously.

00:10:27   So several years ago now, when ATP started really embracing this, I kind of jokingly,

00:10:33   but also seriously offered, hey, if you have the top donation and the donation list, if you reach

00:10:39   out to me, send you forward me a copy of the receipt and give me a mailing address, as long as

00:10:44   the United States Postal Service can mail something to your address, I will send you some not for sale

00:10:49   ATP stickers. I, in the course of the last eight hours, I have two batches of ATP stickers to send

00:10:55   out because one person, and I don't recall if they want to be anonymous or not. So forgive me for not

00:10:59   naming them, but one person donated $8,000. And then the other person said in so many words,

00:11:05   hold my beer and donated $10,000. That's $18,000 of the 30,000 raised between two ATP listeners,

00:11:12   which is incredible. So you don't have to donate eight or $10,000, but if you donate at least $10,000

00:11:19   and one cent, as I said here right now, then you too can earn some hilariously overpriced

00:11:25   ATP stickers. So reach out to me, forward me your receipt and give me a mailing address, and I will

00:11:30   dispatch them as soon as possible. The competition for the world's most expensive stickers. And I

00:11:35   know we just talked about you're buying a new tech device, you're going to pay a minimum amount of

00:11:39   that you paid for the accessories and these people competing to get the top donation. So that keeps

00:11:44   rising or whatever. That's great for the people who are hearing that and say, I think I could swing

00:11:50   that. That's great. If you're listening to that and you're like, that's ridiculous. I'm barely

00:11:54   have enough money to get by a $5 donation is fine. Don't stop yourself by saying, Oh, this pledge

00:12:00   drive is just for people who can donate hundreds of dollars to buy new iPhones over the years.

00:12:04   Give a single dollar. Like I swear, like they will, anything you do helps, right? Like they,

00:12:10   they're not going to turn you away because you can only give them $3. Give the $3 like

00:12:14   that's what, that's what these donations are made. It's the relay listeners adding up all this money

00:12:19   or whatever. They're not all giving $10,000. I swear to you, people are giving five bucks,

00:12:22   10 bucks, 20 bucks. That's what makes this pledge drive. So please, stjude.org/atp.

00:12:30   There you go. All right. So let's do some follow up. I hear that things may be different in

00:12:36   Overcast land. What's going on there? So I re-added streaming as discussed last episode.

00:12:41   That update is out. I've been in kind of a bad place mentally over the reaction to the update.

00:12:49   In part, because as I described in a few episodes ago, I said basically like that,

00:12:56   I described my star ratings being lower as a small fire on a big building that nevertheless

00:13:01   I needed to put out. And so I'm happy to say that since the streaming update, there has been a sharp

00:13:09   rise in the average score of new ratings. So I think that has made a substantial difference.

00:13:17   So thank you everybody who has updated your ratings since then or added new ones. Thank

00:13:24   you for that. I'm also happy to announce now that what I've been working on for the last week or so

00:13:30   is I'm bringing back a swiping UI to the now playing screen.

00:13:34   Oh, hooray. That makes me very happy because I have mostly adjusted to the current one, but I

00:13:40   still my natural reaction after so many years is to swipe laterally. So that makes me very happy.

00:13:45   Yeah. And it was very hard to do because in Swift UI, there was not a good way to detect pixel

00:13:55   perfect scroll positions of a scroll view until iOS 18, which added like a scroll geometry thing

00:14:02   that you can finally actually read it. And I I've been experimenting with this for a while trying to

00:14:07   get because because the reason I need scroll positions is when in the current UI, when you

00:14:13   switch over to like from the artwork view to the info view, I animate down the height of the of the

00:14:20   main playback controls to make more room for it to give the info view as much room as it can get.

00:14:26   And in order to do that, I have to know like, well, if you're dragging across and you like don't

00:14:32   drag all the way, well, how how do I know where you are in that in that in that animation to know,

00:14:37   like how far to bring that bar down. So I tried all these different hacks, including like wrapping

00:14:42   a UI kit UI scroll view in Swift UI. And then, of course, inside that scroll view, the content

00:14:49   that I'm putting in it is itself Swift UI again. So it kind of like, you know, go back and forth

00:14:54   through UI kit. And let me tell you, that creates so many shortcomings and problems. Like and I

00:15:01   tried I tried making that work for like two weeks, I couldn't get it to work well enough and have no

00:15:07   like massive showstopping problems. So anyway, I finally figured out a way to do it where the

00:15:14   animation works okay. On iOS 17. It works a lot better on iOS 18. So I'm working on that now that

00:15:22   should be out fairly soon, at least into beta. And I'm looking forward to sharing that with

00:15:29   everybody because honestly, like, it does make the app better. It does make it easier. You can

00:15:33   swipe the whole screen again, like you know, just like the old version. It is a much better design.

00:15:37   It was just very difficult to achieve in the constraints that I had using Swift UI

00:15:42   until fairly recently. So is I'm sorry, is the swipey stuff is that requiring iOS 18 then? So

00:15:49   no, the way it's going to work the way the way I have it right now, which is probably how it's

00:15:53   going to ship is you the swiping works no matter what. But on iOS 18, as you drag across, and as

00:16:01   the controls shrink downward to make to make as much room as possible for the info on iOS 18,

00:16:06   that will be pixel perfect. On iOS 17. It will stay at the top height until you are halfway

00:16:14   through it at the halfway point, it will then animate itself down all the way. And so it's

00:16:21   there's kind of like this threshold, like at the halfway point, it will cross that threshold

00:16:25   and do the whole animation at once as opposed to, you know, doing it pixel perfect as it tracks your

00:16:30   finger. It looks okay at high speed. If you do it at slow speed, you'll catch the transition.

00:16:35   But you know, that small price to pay for a much more usable interface. And you know, the way my

00:16:41   users are like, by December, I'm going to probably have like 70% iOS 18 usage anyway. So it doesn't

00:16:46   really matter. I know there's one of your pet obsessions making all the animations beautiful

00:16:50   and everything. But honestly, I think from a user's perspective, the thing that counts so much more

00:16:55   than the pixel precision of the transition word were scrutinized is the responsiveness of when I

00:17:00   put my thumb on it and I move does the thing move immediately and in tracking with my thumb,

00:17:06   and I think users will say the forgiving of they will not even noticed almost anything else at

00:17:12   normal swipe speeds, as long as it is responsive. So I would sacrifice attractiveness of the

00:17:18   transition even on iOS 18 if it took away from the responsiveness at all, right. And you know,

00:17:24   again, like users aren't running your animations at one 100 speed so they can scrutinize every

00:17:30   pixel or whatever, they just want to put their thumb on the screen, move it to the side and see

00:17:33   the info real quick. Believe me, some of my users are that picky. But yeah, overall, and it is very

00:17:38   fluid and responsive on both of us is just because again, like it's all Swift UI, it's all very

00:17:43   efficient, you know, with the way it's drawing the UI and updating things and everything like,

00:17:47   I gotta say, I've gotten pretty good at Swift UI, like doing the whole app this way,

00:17:50   I have finally built up like decent Swift UI skills. So like, if something is possible to do

00:17:56   in Swift UI, I can probably figure out how to do it. It just takes a little bit of time. And you

00:18:00   know, that my usual pattern of doing it is like, I will start trying to solve a problem, I will build

00:18:07   up this ridiculous amount of complexity as I try different things and figure out different what

00:18:12   works, what doesn't try different approaches, then I'll like whittle it all back down to like the

00:18:16   distilled essence of here's the one thing that I that really worked well, like it's a whole,

00:18:21   it's a whole process. But I'm very happy to, to be able to do this so quickly. Like this is part

00:18:28   again, part of the reason why the rewrite has been just so massively productive, is that,

00:18:34   and that now that it's done, is I can do pretty substantial changes like this in, you know, now

00:18:40   days and a week or two instead of like months that it would take before like it's so much faster to

00:18:46   iterate. It's so much faster to change and design like to restructure redesign screens and controls

00:18:52   and everything. It's so much easier. So I'm very thankful for that.

00:18:57   That's awesome. Mom, very, very glad to hear that things are looking up and looking better. They

00:19:01   are literally looking up because that's the way the graph is moving. Yeah. So that's very good

00:19:05   news. We had some feedback with regard to the max screen sharing like badge and we were talking

00:19:10   about, Oh, could we do something visually or something like the Siri animation in iOS 18?

00:19:14   What can we do to make it more obvious without having to nag the users? This okay. Is this okay?

00:19:20   Is this okay? So Josh Hattersley writes, even the existing colorful icon indicating screen capture

00:19:25   approach has its downsides. Some friends and I occasionally stream movies for group watch events

00:19:30   and discord. And while I normally do so from a windows machine, I tried recently for my Mac,

00:19:34   I was extremely annoyed to find that the colorful badge Apple slaps on captured windows also appears

00:19:38   in the streamed video, even in an app, like however you pronounce I I N A we've went, we've went

00:19:43   through this years ago. I don't remember, uh, which hides the title bar frustrating. And it

00:19:47   feels like there should be a method to dismiss it. So I was saying last week that Apple really needs

00:19:51   to do, although it may be tricky if, uh, if an app like I N A is not using the, uh, the right API.

00:19:57   So basically like that, the purpose of that badge, that badge is for the person who is

00:20:01   doing the stream. So they know that a streaming app is recording their screen,

00:20:05   but the output of that streaming app, like when it says, okay, I'm projecting my screen to all

00:20:10   my friends so they can see it. The friends don't need to see the screen recording badge. It's not

00:20:14   for them. I think Apple should like, this is what I was saying. If they ever did like a Siri type

00:20:18   border, that part should be omitted from screen recording itself. And from any output of that or

00:20:24   whatever, because its purpose is to inform the person who initiated the screen recording or

00:20:28   who's on the computer that is running the app that is recording the screen. That's just for them.

00:20:32   It doesn't need to be in the output or the recording or the outgoing stream. And I hope

00:20:36   that's something that either if Apple already does that, I hope apps that use those APIs,

00:20:41   learn how to get just the part they want and not the badge. And if Apple isn't already doing that

00:20:45   suggestion, maybe I'll file it as a feedback. I mean, not really familiar with the APIs cause

00:20:48   I don't have any screen recording apps, but I just assume that's a logical thing to do. And it amazes

00:20:52   me that like the badge is going out in the streams and stuff. Yeah. It's something else. Oh yeah. One

00:20:57   more thing that I didn't put in there about screen recording. Someone I forgot who it was. It didn't

00:21:00   make any of the notes, but someone showed a picture probably, or maybe it was a screenshot of a Apple

00:21:05   TV projecting something. Apparently you can do screen recording and Apple TV. Uh, somehow either

00:21:10   is it like a diagnostic thing to do screen recordings for like feedbacks or maybe there

00:21:14   apps to do it, but either way, there was a giant solid red border around the entire television.

00:21:18   And that is the indicator apparently an Apple TV to let you know that the screen is being recorded,

00:21:23   not quite as elegant as the Apple intelligence wavy thing or whatever, but gets the job done.

00:21:27   That's special. Uh, John, you had some good news in your own world. You have finally gotten to the

00:21:32   front of the line. Yeah. The Apple intelligence wait list, uh, Quinn Nelson suggested that I log

00:21:37   out of my Apple ID and back in, but it turns out I didn't have to do that. I just updated whatever

00:21:41   the latest beta 15 one is. Cause I'm on the 15 one train and I went into Apple intelligence and I was

00:21:46   like, yay, you're in a it's Apple intelligence. This is Syria personal intelligence system

00:21:50   integrated deeply into your Mac apps and Siri learn more dot, dot, dot right underneath that

00:21:54   as a little box that says Apple intelligence is not available when starting your Mac from an

00:21:58   external volume. Whoops. Cool. So I am through the wait list and I have learned that I think I

00:22:05   mentioned this on a past show, but there it is right in my face. I cannot use any Apple

00:22:09   intelligence features in the beta because I am indeed booting from an external desk. I don't

00:22:14   know what that limitation is about, but it is what it is. And I'm not putting 15 one beta on the

00:22:19   internal SSD of my wife's computer. I don't blame you. All right. Uh, with regard to making photos,

00:22:27   memories using Apple intelligence, um, Simon writes in to say, I cannot get Apple, Apple's AI

00:22:32   to create a photos memory of the times my kids have gotten dirty. So, uh, Simon includes some

00:22:38   screenshots and the prompt that he typed or said was my kids are covered in dirt to which Apple

00:22:44   says, try another description to create a memory, add details to your description or choose a new

00:22:48   person place or event from your library. So Simon tried again, my kids getting dirty while they play

00:22:53   to which Apple said, unable to use that description. And then, uh, my kids getting

00:22:59   dirty while they eat and play unable to use that description. Uh, okay, fine. My kids playing in

00:23:03   sand, try another description, create a memory, add details, blah, blah, blah, blah, blah, blah.

00:23:07   So, uh, apparently they are really trying not to do anything naughty or inappropriate,

00:23:13   which for the most part I applaud, but they might have swung a little too far in one direction.

00:23:18   That was the prompt that we read on a last week's episode, like nothing filthy. So my suggestion was

00:23:23   you have, what about your kids like getting dirty or like getting food on her face or whatever.

00:23:27   It's so hard to tell what these different responses kind of like when like home pod tells

00:23:31   you different things or whatever, which one of these responses is because Apple intelligence is

00:23:36   baited in and just configure how to do these and which one of these is because of the, um,

00:23:40   the, what do you call it? The prompt preamble or whatever you call it that said nothing filthy,

00:23:45   right? You can't tell, am I being denied because of that nothing filthy thing or am I being denied

00:23:50   because it just wouldn't work either way? I guess we'll all find out when we get this and start using

00:23:54   it and start trying to make memories of ourselves. If we get thwarted on things that we didn't hear

00:23:59   mentioned in the little preamble, and we know it's just Apple intelligence being weird, but it's not

00:24:03   an easy problem. Like these, the nuances of, you know, my kid getting dirty while they eat, right.

00:24:09   And, and the prompt saying nothing filthy, I can see where there might be some confusion there,

00:24:14   but yeah, I'm glad someone's out there, uh, trying it. And speaking of trying it.

00:24:20   Yeah. So Evan Jao, uh, put up a YouTube video where they figured out how to do prompt injection

00:24:28   against Apple intelligence. It's a pretty wild, um, it's only like a 10 minute video, so it's worth

00:24:33   watching, but it is pretty cool what they ended up doing. And it involves a lot of, uh, disassembly,

00:24:38   not in the, like, you know, the, the computer code way necessarily, but like, how do I, you know,

00:24:43   how do I get it, what they're using, where's that data stored? How can I modify that and inject my

00:24:49   own thing with it? It, like I said, it's very, very much worth your time if you have 10 minutes

00:24:52   to spare. Yeah. And this is just another, another one of the weaknesses of a lot of L and based

00:24:57   things or lots of computer interfaces. And we used to say this about Twitter as well,

00:25:01   difference between, uh, in-line data and data that is provided in a different channel. Uh, so

00:25:07   the thing you're typing to an LLM, you want that to be processed, right? All the other stuff where

00:25:12   you're saying, this is not what the person is typing, but this is what we're trying to tell you

00:25:15   on the side, LM like instructions to the LM. You really want those things to be separate,

00:25:20   but the way most of these systems work is it's all just being one big wad that you throw in the top

00:25:25   and probably most of that wad gets to the actual LM under the covers, right?

00:25:30   So if you watch this prompt injection thing, he's just looking through like the config files

00:25:34   and figuring out what do I have to put inside the text that gets fed to the top of this big system,

00:25:40   such that it will interpret that as an instruction. And then do like, oh, well,

00:25:44   you set it off with double curly braces, or you put it in an angle brackets or whatever.

00:25:47   That's an example of trying to put stuff in line. It's in the same place as your message. It's kind

00:25:52   of like back in the day when you would on Twitter, where you would try to like quote tweet something,

00:25:59   it would just be a bunch of text. And inside that text would be the URL of the tweet that you're

00:26:04   trying to quote tweet. But later when Twitter officially supported that, I don't actually know

00:26:07   the Twitter API, but I'm assuming what they did was said, okay, now when you quote tweet something,

00:26:12   you get to say what you want to say in your tweet. And then you tell us the URL, the thing you're

00:26:17   quote tweeting, but that's not part of your text. It's totally separate. Like in the API request,

00:26:21   you say, here's my text and here's a whole other field, which is, and by the way, this is the thing

00:26:26   I'm quote tweeting. So you don't have to worry about reading the text and trying to find the URL.

00:26:30   And once you find the URL, figure out if it's a Twitter URL, and if it's a Twitter URL, figure out

00:26:34   if it's a tweet, you don't have to do any weird parsing or whatever. And if these systems use

00:26:40   something like that, they wouldn't be as vulnerable to just typing a double curly brace and then

00:26:46   putting a bunch of special instructions and typing two closing curly braces, and then going back to

00:26:49   what you were saying, because you get to type that in the text box and she can watch it. Like,

00:26:53   he's just typing like text edit and he's selecting it and firing up the writing tools and saying go.

00:26:58   And they do totally different things. Like they don't do like whatever feature he suggested. It's

00:27:03   because the text that he selected, Oh, we did a double curly brace inside the text. And now we've

00:27:08   escaped out of the text world. And now we're into this world and parsing those things out and making

00:27:13   it so people can't inject that in there is a much harder problem than you think it is. But that's

00:27:17   how these systems work. And they don't have to they don't have to work this way. Like Apple could

00:27:20   have made it so that when you select a bunch of text, and you right click it and do the writing

00:27:24   tools thing with Apple intelligence, it could take that text, put it into one separate bucket,

00:27:28   and then take another bucket and say, here are the instructions to our system and send them as

00:27:32   two separate blobs. So like, it would never parse an instruction out of the text that was selected.

00:27:38   But that's not how it currently works, at least in the beta he was using the instructions that are in

00:27:41   the text get interpreted by the system and make it do all sorts of things. And those those

00:27:46   instructions are just plain text strings, whether they're double curly braces, or angle brackets,

00:27:51   or whatever, you can find what they are in a bunch of config files. And even if they can

00:27:54   encrypt those config files, or try to hide it or whatever, like people can figure it out. So

00:27:58   that is a weakness of trying to put metadata in line, see also file name extensions.

00:28:04   You will never stop banging this drum. I will never let it go to the day, put it on my

00:28:09   gravestone to the day I die. Maybe because I grew up with extensions, it just does not bother me the

00:28:15   way it does you because I like that in the file name, it's clear what this file would have that

00:28:20   without literally putting it in the file name. So how did you have that in Mac OS in the in

00:28:24   classic Mac OS? How did you know what kind of file it was? There was type information,

00:28:29   it was metadata separate from the file name. The system would never look at the system would never

00:28:34   try to parse anything out of the file name. Never. No, no, this is you and me having two different

00:28:38   conversations, which I think the two of us do a little too often. There was separate information,

00:28:42   just like there's separate information like where's the creation data in a file? Is that in the file

00:28:45   name? No, it's somewhere else. No, no, no, you're missing my point, probably because I'm not

00:28:49   presenting it well. What I'm saying is I'm looking at a finder window or whatever the hell the

00:28:53   equivalent was in classic Mac OS. It was called Explorer back then. Yeah, it's called Windows

00:28:58   Explorer. I have 10 different files in this folder or directory or whatever, and three of them are

00:29:02   pieces of text, two of them are images, and the rest are a smattering of other styles of file.

00:29:08   How do I know as a user what's text and what's an image? You asked this question as if that's not a

00:29:13   thing that can happen today on your Mac. Do you have show all file names extension at all times

00:29:16   turned on? No, I don't. So how if you have a bunch of files in the finder and the file name

00:29:21   extensions are not visible, how can you tell what kind of files they are? I guess that's fair. All

00:29:24   right, touche, touche. I mean, I think I do typically see the file names. Where is that?

00:29:29   I know what the setting you're thinking of. Some people turn that on, which I guess I could

00:29:32   understand you forgetting that that's not the default. No, I don't have it on. But like it's

00:29:36   not the default to turn it on. And if you don't turn it on, most of the time the system tries to

00:29:40   hide them by simply literally hiding that part of the file name. It's still in the file name,

00:29:44   but it's hiding it. And then your question is exactly the same case. Well, how do you tell?

00:29:47   No, that's fair. You look at the icon. If you're in list view, you look at the kind column, right?

00:29:50   That's that's how we did it. Fair enough. I mean, either way, I just really don't see why this is so

00:29:58   offensive to you, but it's fair. I'm not saying you're wrong. I'm not saying you're wrong. It's

00:30:00   the principle. I guess. I don't know. I'm sure. I bet if I grew up the way you did, I would probably

00:30:05   also be deeply offended by it. I had a glimpse of a better world and it was taken away from me.

00:30:08   But I'm still glad to still I'd much rather still have Max even with the sport. At least we didn't

00:30:14   get drive letters. Am I right? That is actually drive letters are a true abomination. I am

00:30:19   right there with you on that one. All right. Moving right along. How do you add grammar

00:30:25   constraints to an LLM, John? So like if I want to get a valid JSON out the other side,

00:30:30   how do I do that? So that was a question again, when we were looking at the prompts that people

00:30:35   pulled out of Mac OS or wherever they were, the prompt said, please produce only valid JSON.

00:30:39   And I discussed how just mind bending it is that you have to put that again in line with the text

00:30:45   that's being fed in. You have to put these instructions in plain English in line and then

00:30:49   cross your fingers and hope to get it done. Because there's nothing particularly special about

00:30:52   saying, please provide valid JSON or only output valid JSON or please produce JSON that's valid

00:30:56   or please send your output in the format of JSON. Like you can say it a million different ways.

00:31:00   It's just text. This is not offset by double curly braces. This is not a special instruction

00:31:05   to the engine. It's literally just the plain text sent exactly the same as your text. Right.

00:31:10   And I like that's doesn't seem particularly reliable. Great. But Stephen Tierney was the

00:31:15   first to write in to give some examples of how LLMs are augmented to try to make this

00:31:21   slightly less ridiculous. So he writes, you can constrain the output format of an LLM by

00:31:27   specifying a formal grammar. And he linked to an example from the Llama open source LLM. I'll put

00:31:33   a link in the show notes to the readme and GitHub. It's basically just like a BNF type grammar where

00:31:37   you say, this is what JSON looks like. And it's a grammar that defines the valid structure of JSON.

00:31:42   I'll put another link to a recent paper on the topic called LLM generation with grammar

00:31:46   augmentation from one of the authors of the paper. They write all the compared approaches use

00:31:52   constrained decoding to filter syntactically invalid tokens during generation. They remove

00:31:58   the set of bad tokens when the LLM is choosing the next token. So basically the LLM goes through

00:32:02   this process and for every token, which is basically like a letter or a part of a word or

00:32:06   whatever, it has a list of possible candidates and like probability. And then there's this other

00:32:11   system where you pick one of them. You don't always pick the highest probability, but you pick one of

00:32:14   the ones that's near the top. Right. And what it does, it says, okay, LLM during this normal process

00:32:19   produce all the possible tokens, the candidates that we might pick. But before we do the picking

00:32:24   process, if we decide which one we want, again, it's probably going to be one of the top ones,

00:32:28   although it's not always the topiest top one. That's part of what the temperature setting does.

00:32:31   Right. Before we pick that one, delete any token that would make the output invalid JSON,

00:32:39   because it knows like it based on the grammar, like, okay, we just did like the end of a double

00:32:44   coded string, right? The next call, the next token has to be like a colon or a curly brace. Like,

00:32:48   you know, in the grammar, what is even valid for the next token? Is any letter valid? Does it have

00:32:53   to be like, is it any letter or a double quote or calling? It knows. And so it just goes through

00:32:57   to the candidate list and just deletes everything that wouldn't be valid. And you hope there's

00:33:01   something left, right? You hope that when you delete everything that would be valid,

00:33:04   hopefully something's left. And it usually is because the candidate list is very, very long.

00:33:07   Right. And that's one way where you can constrain the grammar. Now there's possibility that you

00:33:12   might have to backtrack where you end up at a terminal condition where you didn't, you can't

00:33:15   produce any more tokens because you deleted all of them or whatever. And there were some earlier

00:33:19   versions of this. It did slow down inference a lot, but a lot of them use the similar approach.

00:33:23   We'll put a link in the show notes to a Reddit discussion where someone tried to explain like

00:33:26   the pipeline of like the text goes into a tokenizer that goes into the model. They would get the list

00:33:31   of candidates, then a sampler takes it, then you get the winner and then you keep doing that.

00:33:35   And then the grammar filter just goes in between the candidates and the sampler. And OpenAI has a

00:33:39   similar type of feature where you tell it what kind of output you want. Again, this is different

00:33:44   than asking for plain JSON, but even given all of that, like I said, if you want to make sure

00:33:50   that your candidate list contains something that will be valid JSON, you should also still ask for

00:33:56   it because that makes it much more likely that near the top of the candidate list, or at least

00:34:00   somewhere in the candidate list will be tokens that would make it valid JSON. So you still have

00:34:05   to say please and thank you, like please produce valid JSON. And by the way, even if you, you know,

00:34:11   when we pick your tokens, I'm going to delete all the choices that aren't valid JSON and only pick

00:34:15   from the ones that are left. Cool. I mean, it makes sense. It's something though. And we also had

00:34:23   Nate write in and point out that the notification history that I think John was asking for is

00:34:31   basically the one that exists in Android right now. It's better in Android. Could be better still,

00:34:35   but hey, they have a notification history and it shows you your notifications. Here's all

00:34:39   your notifications in the last 24 hours grouped by application. And it gives you a little summaries

00:34:43   of it. You could have even more information because again, what I was looking for was like,

00:34:46   show me all the notifications that have gone by, when did they appear and what action did I take

00:34:53   on them if any, right? And at what time did I take the action? Like stuff that the system surely knows

00:34:58   at the time you do it, but just like record it and keep a rolling 24 hour window. It's not probably

00:35:01   not that much data because it happens to me all the time. A notification comes and go, whether

00:35:06   maybe I actually did something with it, maybe I swiped it away and actually I want to see it again.

00:35:11   And I would just like to say, I would like it to be just like a big linear list where

00:35:16   I'm at certain point in the list, but at any point I can say, show me things that have gone by in the

00:35:20   list already. And I can scroll backwards or forwards. That's just not how notifications work

00:35:24   on iOS or the Mac for that matter. And I wish they did.

00:35:27   We are brought to you this episode by Trade Coffee. If you make coffee at home, and I mean,

00:35:33   let's face it, most of us do get ready to discover how much better your coffee can be with Trade.

00:35:39   Because the secret to coffee at home is to get freshly roasted beans. What Trade does is they

00:35:44   bring the country's top local roasters together. They curate their best coffees and then Trade

00:35:49   matches you with options you'll love recommending new roast based on what you like or what you might

00:35:53   want to try. And then every bag is roasted to order and delivered to your home fresh,

00:35:58   exactly whenever you need it and however much you need. So this is a wonderful way to get great

00:36:03   coffee at home because I'll tell you what, the equipment, the gear, the brew method, those are

00:36:07   all fun. I love coffee making gear and methods and I have tried them all. What makes the biggest

00:36:11   difference by far is having freshly roasted beans done by a great roaster. That is number one. You

00:36:17   can have the worst coffee maker or the worst pour over technique or whatever. If you're using good

00:36:21   beans, it'll be good. If you're using bad beans, it won't be. So go to Trade. This is actually even

00:36:25   better because they also support small businesses because Trade, they don't roast the beans

00:36:28   themselves. They partner with over 50 independent roasters across the country. They have made over

00:36:34   435,000 perfect coffee matches to date. So they know what people like. They send you what they

00:36:40   know you like. And if you by any chance aren't satisfied, they'll replace it for free. Super

00:36:45   simple. But it doesn't matter. Trust me, you're going to want what Trade sends you. It is so good.

00:36:49   I've never had the miss with a recommendation from me. So see for yourself at www.drinktrade.com/atp.

00:36:56   If you go there, you'll get your first bag for free. So you can try it, see how you like it.

00:37:01   Of course, they have all sorts of different options. If you need like whole bean ground

00:37:05   in different ways, whatever it might be. If you need to change your schedule or change your

00:37:09   interval or skip a week, it's no problem. Super easy. Check it out. www.drinktrade.com/atp to get

00:37:15   your first bag of coffee free. This is the best way to get coffee at home. It's freshly roasted.

00:37:20   Trust me, it makes a huge difference. So coffee at home made better with Trade. Thank you so

00:37:25   much to Trade for sponsoring our show. We're not going to talk too much about it right now.

00:37:32   We'll do more of that next week. But we wanted to note that Apple has announced their September

00:37:38   event. It will be the 9th of September, which is a Monday. It'll be Monday, September 9th at

00:37:45   one o'clock, one true time zone. We will be watching along with everyone else. And I'm

00:37:51   excited for it. I think it's going to be really fun. And we're going to talk about what we expect

00:37:56   later. But any thoughts about that, gentlemen? We're going to talk about the little invitation,

00:38:00   which I got it made a little odd this year. So people haven't seen the artwork will describe it.

00:38:05   It's like an Apple logo with what I guess is supposed to be kind of like the new Siri kind

00:38:10   of glowy effect around it. But it doesn't really look like the old Siri. It doesn't quite look like

00:38:15   the effect that goes around your screen when you activate Apple intelligence and the new iOS and

00:38:19   iOS 18. But I think that's what they're going for. And then the slogan in white text with a similar

00:38:24   kind of multicolored glow behind it is it's glow time. Apple intelligence isn't even launching

00:38:32   with these phones. And now I know there's going to be features that are like whenever you activate

00:38:36   Siri, it's going to glow. So I guess, yes, it is glow time because now you're going to be activating

00:38:40   Siri and it's going to glow around your screen. So technically it's glow time. But do you really

00:38:44   want to highlight the feature that everyone is going to associate with Apple intelligence, even

00:38:49   though you know Apple intelligence isn't shipping with these phones? It's kind of weird. I guess

00:38:53   this is the invitation. So okay. I was a little bit careless, but we have a very strong tradition

00:38:57   in this community of over-analyzing these. So I feel like I need to do my part.

00:39:00   (laughing)

00:39:02   I feel like I ask this question every year and I never recall what the answer is because,

00:39:05   hey, it's me. Has there ever really been a strong correlation between this and like a actual new

00:39:14   announcement? You know, this clearly is reminiscent of Siri, but Siri has been around for a while.

00:39:19   Have we ever seen like the dynamic island as an example, like that wasn't teased in any way,

00:39:24   shape or form in an event poster. Was it like, have we really ever gotten one of those in the

00:39:28   last 10 years? You have asked this multiple times, although I think the question is shifting slowly

00:39:32   over the years, but you used to ask, has any one of these invitations ever hinted at anything that

00:39:36   it was actually in the presentation? And the answer is yes, it totally has. Now you're asking,

00:39:41   has it hinted at something that we didn't know about at all? I think maybe the closest we can

00:39:46   come there is one of the ones that was like, the invitations basically let us know that there was

00:39:51   going to be Mac stuff. And at that point we were so disillusioned that nobody thought any Mac stuff

00:39:55   would be announced because the Mac was really down in the dumps. But when we saw the invitation,

00:39:59   like that's gotta be Mac stuff. And it was, it was Mac stuff. I forget what the particular

00:40:03   invitation was or which one it was, but yeah, these things often do have hints about what

00:40:07   they're going to show. Sometimes it's boring, like this year they're like, yeah, the glowy effect.

00:40:10   We've all seen it in the betas and that's what they're going for, but it's glow time. It's so

00:40:14   weird. And I, I don't, I do think that a lot of times the invitation has a thing and a theme or

00:40:20   whatever that is not, they don't, you won't see the presentation and say that was the glow time

00:40:25   presentation because they're not going to say it's glow time in the presentation. They're probably

00:40:28   not even going to show this graphic or whatever, but they will of course show their new OS and

00:40:32   their new phones and all the other stuff. So anyway, it's, you know, I, I don't think this is

00:40:38   revealing anything. Like I can't even think of anything that hasn't been rumored that could

00:40:43   possibly be hinted at by this invitation, but you know, it can happen. Hey, so hell might be

00:40:51   freezing over. There are rumors. This was posted on Bloomberg that Apple is going to increase the

00:40:59   Ram in their base max or back book pros. Like what, what, huh? What we're not going to be

00:41:06   stuck with. Oh, like just got an error network changed. Oh, we're not going to be stuck with

00:41:10   eight gigs forever and forever. Is this the first time during the history of our show existing that

00:41:16   this is going to maybe happen? Right. I'm going to say like, what, how long has eight gigs been

00:41:20   around? Like, and when we're talking about is the base model max, whatever the cheapest Mac is

00:41:24   for years and years, the cheapest Mac laptop would come with four gigs of Ram. And then one glorious

00:41:30   day Apple said, you know what? The cheapest Mac, even the, whatever the lowest end Mac we have now

00:41:35   is going to come with eight. And I think that day may have been before the beginning of the show 11

00:41:40   years ago. Right. We looked it up like they remember some, somebody looked it up and pointed

00:41:44   out to us and it, there was, I believe like if you, if you went to like, you know, if you said

00:41:49   only like new models, there was like, I think the Mac book air was the last one to have a low

00:41:54   configuration of that at some point. But like, if you look to just whatever was for sale, I think

00:42:00   the MD one Oh one that that old, like the last one that had the CDU drive, um, that one, which

00:42:06   stuck around for a very long time in the sales channel. I think that one started up at four also.

00:42:12   And that might've been the last one that was actually for sale. Yeah. And despite this,

00:42:16   let's say conservative approach Apple has taken to Ram Ram prices do fluctuate over time. You

00:42:23   can look at the graph. It's not smooth, but in general, Ram becomes cheaper for the same amount

00:42:28   of memory over time. And despite the fact that Apple has always done it this way, there's nothing

00:42:35   particular about Ram that absolutely dictates that you must go from four to eight. You could go from

00:42:41   four to six or four to five. Now probably it's, you don't want it to be powers of two and there,

00:42:46   you can't get out, you can't, everything you want. It can't go from four to 4.75 because they

00:42:50   probably don't sell things in those units or whatever. Right. But you, you don't always have

00:42:55   to double it. And as the amount of Ram gets larger and larger, doubling becomes much more significant.

00:43:00   You're, you know, you're adding one gig of Ram, then you're adding four gigs of Ram. Then you're

00:43:04   adding eight gigs of Ram. Like, because you will, you know, from one to two and two to four and four,

00:43:08   like it just, you don't necessarily have to double it. Again, Ram is usually sold in powers of two

00:43:14   and it usually makes sense to do the doubling or whatever. But what it means if Apple is going to

00:43:19   really hold the line on Ram for a long time is it's not a smooth ramp. It's like just eight,

00:43:25   eight, eight, eight, eight for just years and years and years. And you're like, oh, it's okay.

00:43:29   Well, you know, when it went from four to eight, it's like, wow, everyone gets eight. That's going

00:43:31   to be great. And then you're like, eight is okay. And then you're like, well, you know,

00:43:34   you probably shouldn't buy one with eight. And you're like, oh, really don't get the base model.

00:43:38   It has it. You really need to get 16, like eight outstays. It's welcome. And it's like, you know,

00:43:42   you could have gone to 12 at some point in there, but instead it's just going to be eight forever.

00:43:46   And this rumor is we're going to go from eight to 16, which I'm all for. I'd rather go from eight

00:43:52   to 16. Right. But I'd rather not always have to wait for the doubling because it's going to become

00:43:57   untenable. Like what about when you're just going from 64 to 128, that's a missing 64 gigs of Ram

00:44:03   that you have to live with until finally it becomes completely ridiculous that they're

00:44:07   only shipping with 64 gigs of Ram. And then finally you get 128, right? How about a 96?

00:44:12   How about throw a 96 in there somewhere? You know, like it should be smooth. And this is all the more

00:44:17   frustrating because of that weird, it's not even a rumor. It was a teardown that I think basically

00:44:21   confirmed this. Like they opened up an M4 iPad and look at the M4 SoC and it had 12 gigs of Ram on it

00:44:28   because the Ram is on the SoC. It's like written right next to the actual little chip. It's like

00:44:32   on the package with the SoC, right? It had 12 gigs of Ram in it. Only eight was exposed to the iPad,

00:44:36   but those, the chips were a six and a six. Now people are like, well, maybe those are Ram chips

00:44:41   that only, you know, four of the six gigs work on those Ram chips. And other people said that's not

00:44:46   how Ram usually works. Are they just disabling that Ram or whatever? It was like, Hey, that made

00:44:50   me hopeful that the next low end Mac would start at 12 instead of eight, because we had seen M4 SoCs

00:44:57   that Apple shipped that have 12 on them. Again, I'm way happier with 16, but it shows that 12

00:45:03   was possible, is possible. Apple did ship 12 on machines that granted it didn't expose all or

00:45:09   whatever. And they could have done that on the M3 or on the M2 or whatever, but they didn't. So

00:45:13   this may be the year when the M4 Macs, the plain old M4 Macs, not the M4 Pro, not the M4 Macs,

00:45:20   those will have tons of Ram, you know, whatever, just a plain old M4 Macs. That's probably going

00:45:23   to ship in the MacBook Air, obviously. The iMac, what am I missing? The Mac mini, right?

00:45:30   Plain old M4 starting at 16 gigs of Ram. Long overdue and it will make the M4 line a great time

00:45:38   to buy in. You won't have to scare people away from the basis of base models, except for the

00:45:43   fact that it probably have 256 SSD, which is insane and they should double that too. But anyway,

00:45:48   setting that aside, you won't have to tell everybody, Oh, you're going to get a MacBook Air

00:45:51   or you're going to get an iMac. You just make sure you double the Ram to 16 and you'll be fine. And

00:45:56   of course they charge you like $200 for that extra Ram, which is disconnected from all reality. And

00:46:00   then suddenly the base price doesn't look that good. So it seems like when the M4 Macs come out,

00:46:05   you can buy the base model and it won't be terrible. So we are getting real time follow

00:46:10   up from the chat from David Schaub and the desktop Macs, if I'm reading this graph, right,

00:46:16   which I think I am, desktop Macs went to eight gigs of Ram in 2012. When did we start the show?

00:46:24   It was 2013, right? So that was before the show started. Laptops, however, went in 2017.

00:46:32   What? So in early 2017, you could buy a laptop with four gigs of Ram?

00:46:37   Like the base model of a MacBook Air probably. Oh my God.

00:46:40   I mean, that's what it's like towards the end, right before a Ram transition.

00:46:43   You look back and you're like, really? They were selling Macs with that little base Ram?

00:46:48   Because we just all get into this mindset. We tech nerds of like, Oh, of course you would never

00:46:51   get the base amount of Ram. That would be ridiculous, right? But that's just because

00:46:54   Apple holds the base amount of Ram for so long that we just get trained. This is a great machine.

00:47:00   You don't have to do anything to it except pay 200 extra bucks for more Ram. But everything else

00:47:04   about it is great. You're like, wait, why doesn't it, like, why can't I just get the cheapest one?

00:47:08   I'm like, well, that doesn't come with enough Ram. It's like, well, why would they sell it?

00:47:11   This is what happens if you talk with a non-tech person. They say, well, why would they sell it if

00:47:14   it doesn't have enough Ram? It's like, well, it has enough, but if you want the machine to last

00:47:18   longer and it will be a little bit better, and like you have to go through, it used to be worse

00:47:21   with spinning disks, obviously. SSD makes this better, but it's still not an excuse. Like it

00:47:24   just, it lets you get away with less Ram for longer, but there is a time limit. You can't

00:47:30   hold it at eight gigs for 50 years. You have to increase the amount of Ram and looks like 2024/

00:47:36   2025. We're going to go from eight to 16 and we will enjoy a blessed two, three, four, maybe even

00:47:42   five years where we're all happy with the base Ram on max. I'm excited. I mean, we'll see what happens,

00:47:47   but that would be excellent. So yeah. So with that in mind, there's also a rumor that the iPhone

00:47:52   17, not the one that we're going to hear about in just a few weeks, next year's iPhone, iPhone 17

00:47:58   will feature 12 gigs of Ram. This is reading from MacRumors. Next year's iPhone 17 models will come

00:48:04   with 12 gigs of Ram up from the eight gigs of Ram expected across Apple's upcoming iPhone 16 models.

00:48:08   According to the Weibo user PhoneShipExpert, the iPhone 15 and 15 Plus feature six gigs of Ram,

00:48:15   while the iPhone 15 Pro and 15 Pro Max come with eight gigs of Ram. The difference is said to be

00:48:19   the reason why Apple intelligence is only currently supported on the iPhone 15 Pro models. Since it's

00:48:24   a requirement for Apple intelligence, all four of Apple's upcoming iPhone 16 models are expected to

00:48:28   feature eight gigs of Ram as is the rumored iPhone SE 4. So this is the Apple intelligence Ram

00:48:35   dividend on the phones. Obviously it's forcing all the phones up to eight gigs because we know based

00:48:40   on what Apple has shipped already that the six gig ones can't do it. Now, obviously max were already

00:48:44   shipping with eight gigs, but there's more stuff going on and contending for Ram on the average

00:48:48   Mac than there is on the average phone given the different environments or there can be any way,

00:48:52   because there are fewer constraints on what people can run simultaneously, yada yada.

00:48:55   Why? Why is this the year that we go from eight to 16? I think the answer is Apple intelligence.

00:49:01   Like I think that's why that all of a sudden the phones are going up, all of a sudden the Pro and

00:49:07   Non-Pro phones are going to have the same amount of Ram and it's going to be more than they had

00:49:10   last year. Like it's still kind of weird to me that the 15, I mean it shows kind of how blindsided

00:49:14   Apple was by the whole LLM stuff, that the 15 line had this bifurcation where the Pros can run the

00:49:20   LLM stuff but the Non-Pros can't because they only had six and they're not making that mistake with

00:49:25   the 16 line and apparently not with the 17. We're all going to have it. Maybe the 17 line,

00:49:30   this rumor just says the iPhone 17 models will come with 12 gig. Does that mean all of them will?

00:49:35   I mean, again, I welcome it. I think it's useful even if it wasn't for Apple intelligence,

00:49:40   but if that's what it takes, if it takes a new incredibly Ram hungry feature that Apple wants

00:49:45   to promote to get them to finally start putting Ram in their devices, I'm all for it.

00:49:50   No argument here. And then coming back to this upcoming, I was going to say year,

00:49:55   but this upcoming month, I guess, this upcoming event, iPhone 16 camera improvements. These are

00:50:01   rumors about the cameras for next month. Again, reading from MacRumors, the summary is that the

00:50:06   ultra wide camera will go from f/2.4 to f/2.2 aperture. The 16 and 16 Plus may support macro

00:50:15   photography for the first time. The iPhone 16 Pro, the human sized iPhone 16 Pro, not the

00:50:21   gigantic, gigantic one that I have will gain the 5X Tetra prism telephoto camera, which will make

00:50:27   my decision making very complicated for this upcoming month. The iPhone 16 Pro and 16 Pro Max

00:50:33   will feature an upgraded ultra wide camera with a 48 megapixel sensor with the same pixel binning

00:50:38   features. The main camera, it will have an f/2.2 aperture up from f/2.4 and support 48 megapixel

00:50:44   Pro Raw photography. It all sounds good. I'm happy that the, that, you know, I will finally get the

00:50:50   5X lens in my size phone. I won't have to pull a casey and jump up. So I'm looking forward to it.

00:50:57   Yeah, there were some other items in this article as well. I've just pulled out the most

00:51:00   relevant and interesting ones. But one of the things I didn't put in, because that's not an

00:51:04   improvement is the rumor is that the main camera, no change. Exactly the same for the Pro models.

00:51:09   Anyway, the 16 Pro and the 15 Pro will have exactly the same 1X main camera, which is,

00:51:15   I mean, obviously there'll be processing changes and there's always something that Tout and Apple

00:51:18   has done it a couple of times where they really haven't actually improved the camera hardware at

00:51:22   all, but they have find a way to make you think that it's better or whatever. But the rumor is,

00:51:26   this is going to be one of those years, but you know, they make up for the other changes.

00:51:29   Suddenly the ultra wide getting, you know, a 48 megapixel sensor with pixel binning and having

00:51:34   a better aperture. That'll really hopefully make the ultra wide a lot better camera. I forget if

00:51:39   there was a, I might've simped out, I forget if there was a rumor about the selfie camera

00:51:43   improving, but that's always welcome. So it's not always, and especially these days, it's not

00:51:47   always about like how much better is the 1X camera getting? Cause it seems like they're really

00:51:51   pressing up against the size and cost constraints that they want to, at least that Apple wants to

00:51:58   dedicate to camera. Other phone makers are dedicating even more size and even more cost

00:52:03   to their cameras. But with Apple's current design, I'm not sure, you know, we can't expect the kind

00:52:10   of changes we got during like the big camera, you know, blow up era with someone should do a graph

00:52:16   of this, of how rapidly and dramatically the cameras improved when Apple started making them

00:52:22   way bigger. It used to be flushed with the back of the phone. And then all of a sudden they weren't

00:52:26   and they became less and less flush over time. And they took up more and more room and the camera

00:52:30   improvements during those years, when they went from a tiny little flush pinhole to these gigantic

00:52:35   mountains, huge improvements, but the, even though the camera bump has been getting still getting

00:52:40   bigger, it's getting bigger, more slowly in the area dedicated to the cameras is also slowing.

00:52:44   So yeah, the 1X camera is probably not going to be that big of a leap. Well, they're running out of

00:52:48   space to expand into on the back of the phone. Well, you got to do, I guess I've always said

00:52:52   that you can't, you know, you got to do the full width, right? And maybe have fewer bigger cameras,

00:52:56   like look what the Pixel 9 does. They've been the full width of the phone for a while now. And

00:53:00   their lower end phone just has two cameras side by side. And the high end one has three cameras

00:53:04   side by side. Like, you know, we'll see what they do on the slim. And if they actually changed the

00:53:09   design of the non slim ones, but yeah, I don't, I don't think like, you could make better cameras

00:53:15   with bigger sensors, but they will take up more room and stick out more and cost more money.

00:53:19   And every bit of space inside a phone that you take up with cameras space that you can't put

00:53:23   battery in or any other parts in. So it is a balancing act, but we've definitely seen the

00:53:28   rate of improvement of the 1X camera really slow down and lean heavily on processing features. But

00:53:33   the other camera is getting better. Again, I think that's great.

00:53:35   Jared Ranerelle Yeah, honestly, I, what I look forward to

00:53:38   more than camera optic changes, because again, that as John's saying, like the actual,

00:53:44   like raw quality of the cameras changes pretty slowly these days, just again, because we are

00:53:49   really hitting a lot of limits of physics and current sensor tech and optics. But the

00:53:56   processing does change every year. You know, and even even when in the past, even when they have

00:54:01   occasionally had a generation where like the basics of the camera didn't change between one

00:54:06   and the following phone. Usually the processing is tweaked or is able to be more advanced,

00:54:12   maybe be less heavy handed in certain ways. And like, as I've spent more of the last couple of

00:54:17   years getting back into real cameras, like big cameras, one of the I mean, obviously, again,

00:54:22   like there's there's certain things that smartphones will just always be worse at because they're so

00:54:26   optically, you know, very, very small sensors, very, very small lenses that have to be very

00:54:31   inexpensive relative. But also, there's certain things that phones are way better at and generally

00:54:36   always will be like stacking multiple exposures together to generate HDR and different effects

00:54:42   and different processing steps. But it is still very, very clear to me, when I look at a phone

00:54:48   picture versus a big camera picture, you can you can spot the differences very quickly if you know

00:54:53   what to look for, because phone pictures are still so heavily processed by default. And there are

00:54:58   like there's recent things like I know, didn't Halide just release a cool like raw mode that's

00:55:04   less or like that's basically unprocessed? Yeah, it's like called zero processing or something.

00:55:07   I actually tried it out where it's not no processing, but it's almost no processing.

00:55:12   Obviously, there's some amount of processing just to get a reasonable picture off of the sensor,

00:55:16   but they do almost no processing. And if you ever wondered what it would look like if you didn't

00:55:23   apply most of the iPhones processing, the answer is noisy. Yes, they're very tiny sensors. And

00:55:30   you're going to see rainbow noise all over everything unless it is extremely bright

00:55:34   sunlight and you're like, wow, why would anyone ever use the zero processing mode? It's like,

00:55:39   well, in good lighting, what you're saying is phone don't try to do the thing where you take

00:55:44   17 exposures and make everything look beautiful and have it all be exposed. And like this weird,

00:55:48   you know, artificial exposure bracketing HDR type thing, let the sky blow out or let this

00:55:53   thing be in shadow or whatever. And if there is enough light, you won't get that much noise.

00:55:57   And it is, you know, it's it's it's not a feature that most people want. The default should still be

00:56:02   all the processing on. But it is fun to just see like what are these cameras capable of? Honestly,

00:56:08   I'm pretty impressed with the zero processing mode and saying, actually, these cameras

00:56:12   aren't that terrible. Like it's not too long ago that I owned like when my son was born in 2004,

00:56:17   we had a digital camera that took pictures that were just about as noisy. Only there was no there

00:56:22   was no option to add processing to make them better. So technology marches on. So if you want

00:56:26   to try that, download the I think it's in the latest version. I think I might be on the beta,

00:56:30   but I think it's in the latest release version of Halide, that zero processing feature. They have a

00:56:35   big blog post about it. It's fun to try. Yeah, I was I was also very impressed, honestly, by like

00:56:40   how how decent the sensor the raw sensor stuff is. But yeah, you do get a lot of the rainbow,

00:56:46   the chroma noise. And that's that is just the realities of such a tiny sensor that's,

00:56:51   you know, having to crank up its sensitivity a lot in a lot of different lighting conditions.

00:56:55   And so anyway, one of the biggest differences when you look at iPhone pictures versus versus like

00:57:01   bigger camera pictures is you can tell there's been a lot less processing done to the bigger

00:57:06   camera pictures. Now again, in certain conditions, this is not what you want, especially in terms of

00:57:10   like, you know, if you need to combine multiple exposures to get larger dynamic range, for

00:57:14   instance, that's something that the phones are just so much better at, and probably always will be.

00:57:19   But the big cameras, you can really like a big camera version, if you take it, if you like in

00:57:26   the spot, and you take the same picture with a big camera, and then a phone, barring, you know,

00:57:32   exposure differences for things like HDR. Other than that, like, usually the big camera one will

00:57:36   look a lot more natural, because there'll be like, way less like over contrasting,

00:57:41   edge sharpening, like all these like these kind of artifacts of the heavy handed processing. And so

00:57:46   one thing I hope for every year that I don't usually get too much. Sometimes it gets worse.

00:57:53   But usually what I hope for with the new phones every year is that they've developed better

00:57:59   processing so that it is a little bit less obvious, like so it can be a little bit less

00:58:03   heavy handed by default or at all. This is something that again, like, just due to the

00:58:08   sensor size and the realities of those optics and phones, it's very hard to make it not require a

00:58:14   lot of processing to look good. But over time, we do approach that we do get better over time.

00:58:20   Some phone generations are a step back. I'm not super happy with the 15 in this regard. I think

00:58:27   it does apply very strong processing by default. But I think it actually is better than the 14 was

00:58:33   in that same area. So hoping hoping for more improvements there. And that usually requires

00:58:38   not only you know, just slight improvements to the camera here and there. But that also usually

00:58:42   requires higher end processing different techniques, you know, more more power in the

00:58:48   signal processing pipeline. And you know, where AI comes in and help sometimes to or ml or whatever

00:58:53   we're calling it this this few year period. All those things help to but anything that can make

00:58:58   the camera make this make the photos appear less heavily processed. I very much welcome those

00:59:04   changes when we can get them. There's an interesting parallel with the AI thing like the one of the

00:59:08   one of the knocks against the early and even some of the current like image generators. So like Dolly

00:59:13   or whatever, where you just ask it to make you an image based on a text prompt is that they had real

00:59:17   difficulty with text because they didn't know or care what letters were. So if they had a sign

00:59:20   with text on it, it would look like, like Greek text where it's like, oh, this isn't a real

00:59:24   language. They aren't real letters. It's just like you can tell it's supposed to be text, but they're

00:59:27   not. There's no actual alphabet. They're just these blurry smear of letters, right? Because it,

00:59:31   you know, it didn't consider that an important part of the image. And whenever it sees signs

00:59:35   in his training that it's like, oh, just a bunch of high contrast lines on a background,

00:59:38   you know, anyway, the parallel to that in over processed phone pictures is they take a picture

00:59:44   of real text on a real sign. But then they apply their processing to it. And their processing,

00:59:49   if you zoom in often looks like kind of an impressionist painting where it looks like

00:59:52   blobs of paint instead of sharp individual pixels. And they're doing that sort of denoise

00:59:56   and lump everything together and it will make text unreadable. So if you ever tried to use your phone

01:00:00   to take a picture of some tiny text, so then you could pinch and zoom on the text, you may be

01:00:04   surprised to find that when you pinch and zoom, the text looks like a total mess. It looks like

01:00:08   that AI generated text and you can't even read it. You're like, wait, is the text actually blurry

01:00:12   on the, you know, this little temple, not stem, not stick of my eyeglasses because you're trying

01:00:19   to like, and there's like, no, the text is perfectly clear on your eyeglasses, but the

01:00:22   processing of your phone totally mangled it and made it unreadable. That is an example of where

01:00:28   process zero, which is the actual brand name that they're using for this, hey, let's process zero

01:00:32   would come in handy. Cause yes, it would be noisy, but it won't take the letter forms and melt them

01:00:37   into nothing. And sometimes we talk about, oh, I can tell a phone picture from a real picture,

01:00:41   like, oh, well you're just pixel peeping, which is what they mean by that is you're zooming in

01:00:45   to the picture. So you can see at an individual pixel lever that, oh, this is totally blobby and

01:00:50   gross. Whereas if you zoom in on a real camera, it looks nice as you zoom in and right up to the

01:00:55   point where you actually see the individual pixels kind of right. But like pixel peeping

01:01:00   has said derisively. Well, first of all, that is getting a closer look at the quality. But second,

01:01:05   there's a thing called cropping, which with real cameras you can do if you didn't frame the shop

01:01:09   well or perfectly with a real camera, it's usually better to, oh, I helps with this. It's usually

01:01:15   better to get more of the background that you want so that if you need to reframe, you can just crop

01:01:20   in a little bit. Or if you take a big picture and the most interesting part is in the upper right

01:01:24   corner or whatever, you can crop in on that. But you can only do that if when you crop in,

01:01:28   which is another way of zooming, throw away everything, except for this tiny square pinch

01:01:32   and zoom until only this tiny square is visible on the screen. They're both the same operation.

01:01:36   If you do that with a phone picture, you're like, oh, this is unusable. This person is a wax blob,

01:01:41   right? They are an impressionist painting. I can't make out a single thing at a regular size. It

01:01:46   looks okay because they're small. But when I crop in all of a sudden pixel peeping becomes, I just

01:01:52   want to crop this. I just want to crop and only have this person in the picture. And with a real

01:01:55   camera, you can do that because the quality of the pixels doesn't immediately go into wax melted blob

01:02:01   the second you view it at more than one X zoom or even less than one X, because sometimes the phone

01:02:07   screen doesn't even show the entire picture. So, yeah, I'm a little bit of a big camera bigot,

01:02:12   despite the fact that I take tons of pictures on my phone and mostly enjoy how they look. But

01:02:16   I know the weaknesses and I know better than to try to crop a tiny one third portion of a

01:02:22   phone picture and expect it to look good, especially if it's a person, because they will look like

01:02:25   a melted wax horror. Cool. All right. So our sub topic that I alluded to earlier. So allegedly,

01:02:36   Apple will support JPEG XL, which will sit alongside heath, JPEG, heath, max pro raw and pro

01:02:42   raw max. And the initial reports I read on about this seemed to hint that Apple invented JPEG XL,

01:02:48   but having looked at the Wikipedia entry, that doesn't seem to be the case. No, I don't understand

01:02:53   these stories that put like a new format from Apple called JPEG XL. Like just Google the term

01:02:57   before you write that story. But before we get to that, though, have any of you heard of heath max?

01:03:02   No, I was just thinking, I don't recall what that is. I mean, I didn't Google it either. Like,

01:03:07   I'm assuming it's just some variant of heath and heath has been around for a while. What

01:03:09   the heck is heath max? I have no idea. Like, we had only two image formats for so long. And then

01:03:18   when they added heath, which was now, I mean, a good number of years ago, that was like such a

01:03:23   big change. And kind of like what makes me a little bit nervous about this JPEG XL news is like,

01:03:31   we still don't really have amazing support for heath in the ecosystem around our computers and

01:03:37   phones. Well, I mean, that's that I think is a thing in favor of JPEG XL, which we'll get to in

01:03:43   a second when we get to the forum. But I did just look up heath max. That's just what they call what

01:03:46   they call the 48 megapixel thing where you get all the actual 48 megapixels and you don't bend it,

01:03:50   they're calling that heath max. Well, so the way it was was that originally when the 48 megapixel

01:03:55   came out, the only way you got all 48 megapixels was if you did a raw shot, if I recall correctly.

01:04:02   Then they eventually added with 14 Pro heath max, which is, yes, I would like a 48 megapixel image,

01:04:09   but feel free to compress it. Right. It's not raw. Right. So the heath, heak, heif, heic,

01:04:17   whatever that whole when Apple introduced that format, right. Their pitch was, here's a new

01:04:21   image format. Again, it's not an Apple invention. It's some standardized thing that Apple likes and

01:04:26   probably had some part in. And what their pitch was, you get better quality images for the same

01:04:32   size or you get smaller images, you know, like it was a better format, right? Compressed images.

01:04:37   It's not a raw format. It's a compressed thing. The format is very flexible. We can do all the

01:04:41   things we need to do. We can implement all the features we want to implement. I forget if this

01:04:45   is the one that they implemented the little live picture thing in. Instead of having the sidecar

01:04:49   thing, they combined it. But anyway, the pitch was it's a better image format, right? And we're

01:04:54   judging that by the quality of the image at a given size. And even though it was weird, we're

01:04:58   like, well, but it's not like it's Apple's proprietary format. This is a format that anybody

01:05:02   can use. It's an open thing and it's better than JPEG. And that frustration that we have about like,

01:05:08   oh, only Macs and stuff can read those things, mostly mitigated by the fact that when you share,

01:05:13   it would convert it to a JPEG. And if you didn't like it at all, you can go into settings and say,

01:05:17   I don't want to ever use this hee-feek stuff. Just shoot everything in JPEG. And it would still work.

01:05:21   But I left everything on the default, which was shoot everything in hee-feek or whatever,

01:05:25   because they take up less room on my phone or they take their better quality at the same size

01:05:30   or whatever. Like it's a better image format. And I was figuring before I read the story that,

01:05:35   you know, this hee-feek thing, we'll just, you know, tough it out for another, you know,

01:05:41   five to 10 years. And eventually you'll see more widespread support for this format. And it won't

01:05:45   be that big of a deal. Honestly, it hasn't been that big of a deal for me because of the sharing

01:05:49   stuff or whatever. But then this JPEG XL story comes out and it's like, well, if you were waiting

01:05:54   around for hee-feek to become so standard that it's as common as JPEG, that seems like that is

01:06:00   never going to happen for reasons that we will explain now. Right. Oh, so JPEG XL supports lossy

01:06:07   and lossless compression of ultra high resolution images up to one terapixel, up to 32 bits per

01:06:14   component, up to 4,099 components, including alpha transparency, animated images and embedded

01:06:19   previews. It has features aimed at web delivery, such as advanced progressive decoding and minimal

01:06:25   header overhead, as well as features aimed at image editing and digital printing, such as

01:06:28   support for multiple layers, CMYK and spot colors. It is specifically designed to seamlessly handle

01:06:35   wide color gamut color spaces with high dynamic range, such as REC 2100 with the PQ or HLG

01:06:42   transfer function. I understood about a third of what I just read. So there are some sweet features

01:06:47   involved. There's independent tiles. So decoding of sections of a large image by allowing images

01:06:52   to be stored in tiles, progressive decoding, more specifically, excuse me, it's a mode specifically

01:06:58   designed for responsive loading of large images, depending on the viewing device's resolution.

01:07:02   There's support for both photographic and synthetic imagery. The format features two

01:07:06   complimentary modes that can be used depending on the image contents, efficient encoding and decoding

01:07:11   without requiring specialized hardware. JPEG XL is about as fast to encode and decode as old JPEGs

01:07:15   using libjpeg turbo and in order of magnitude faster to encode and decode compared to HEQ with

01:07:20   X265. It is also parallelizable. JPEG XL can losslessly transcode a widely supported subset

01:07:28   of JPEG files, making it about making about 20% smaller file sizes possible due to JPEG XL's

01:07:34   superior entropy coding. This process is reversible and it allows for the original

01:07:38   JPEG file to be reconstructed bit for bit. JPEG XL, and I think this is a kicker, is a royalty

01:07:43   free format with an open source reference implementation available on GitHub under a

01:07:46   three clause BSD license. Or he can heaf. They were better image formats and they had their

01:07:52   very brief time in the sun, but not enough to become industry standard. And here comes JPEG XL,

01:07:57   which is basically better than them in every possible way, including the most important one,

01:08:02   which is royalty free. Open BSD implementation, royalty free, open source, parallelizable,

01:08:11   supports all the new modern things, huge image files, transparency. Just think of all the image

01:08:15   formats that we've lived with and all the weird soup that we've had. We had GIFs, which are

01:08:19   fun, but they had only go up to 256 colors. We have JPEGs, but there was a kind of blurry,

01:08:25   so you'd use GIFs for those if you could. But if they had more than two fixed colors,

01:08:28   then you needed to do PNG. But PNG was the only one that supported transparency, but then I

01:08:31   didn't support transparency in PNGs. And even today, even when I'm putting stuff on my website,

01:08:36   it's like, I'm going to put up this image. Oh, I want it to have transparency. I guess I'll have

01:08:39   to use a PNG, but of course, PNGs are basically uncompressed, even though you can squish them

01:08:42   down a little bit farther. So JPEGs are much smaller than PNGs, but then JPEGs are blurry,

01:08:46   especially if you have text, because they don't do a good job with that. And there was progressive

01:08:50   JPEGs when we had modems, so the JPEG would slowly fade in or whatever. But then there's

01:08:54   those old JPEG 2000, and then there's WebP and all the weird Google formats, and what is it, AVIF,

01:09:00   and then there's all these wars about, I hate it when I download a WebP image because MacOS doesn't

01:09:04   like those, and I always have them converting them, so I have a retro batch thing that converts

01:09:07   all the WebP images into PNGs, and of course, there's Photoshop formats. This solves all of

01:09:13   them, supports everything that they all support, is totally open, has the name JPEG in it, which is

01:09:18   really important. I mean, it probably is from the Joint Photographic Experts Group, but being called

01:09:22   JPEG something is important. Now, granted, it didn't help JPEG 2000, but putting 2000 in the

01:09:27   name wasn't a great move, right? And Excel makes it sound like it's extra large, but I think that's

01:09:31   not what the Excel stands for. But this has all the features, and it's royalty-free, and it's

01:09:37   open source, and it's made for modern hardware. Parallelizable, the thing with the tile encoding

01:09:43   or whatever, it's parallelizable, because you can break up your image into chunks and decode them

01:09:47   all in parallel, and if you're just showing one quarter of an image, it can just decode that

01:09:50   corner of the image. It's so well thought out and well designed to not have any of the weaknesses of

01:09:56   the existing formats. And by the way, it also makes smaller files at the same quality or better

01:10:01   looking files at the same size, right? So I am ready to throw heeck and heef in the dustbin of

01:10:08   history. I'm ready to go all in on JPEG Excel, provided everybody starts supporting it. That

01:10:14   means Mac OS, Chrome, every platform needs to support this, which almost sort of mostly

01:10:21   happened with WebP. I think WebP on the web works okay, as this name suggests, but when you download

01:10:27   a WebP onto your Mac, Mac OS is like, pfft, you know? And whatever the other format that Google

01:10:34   X is, AVIF or whatever, there's a bunch of other image formats that are better than most of the

01:10:39   existing ones that are kind of competing with heeck and heef and all that. But the only thing JPEG

01:10:45   Excel does not go forward with is it does support animation. Why do we have quote unquote animated

01:10:49   GIFs for so long? Because it was the only image format that supported animation, and then there

01:10:52   was like animated PNGs, but that didn't get very widely supported. These days when we say, I'm

01:10:55   uploading a GIF, they just convert it to MP4 behind the scenes, right? And like Twitter and

01:11:00   Mastodon or whatever, they don't actually show you an actual GIF because actual GIFs are humongous

01:11:05   because it's individual frames that are individual GIFs or whatever. And this is not a video format,

01:11:11   so it doesn't solve all the problems. Like H.265 will live on and all that other good stuff. But

01:11:14   I am excited about JPEG Excel. And I hope, unlike heeck and heef, this does sweep across the entire

01:11:21   industry and become boring and supported everywhere. And I guess the only barrier to

01:11:26   that not happening are companies that are really tied to their formats, either because they own the

01:11:34   patents on them or because they've already invested in the starting infrastructure that understands

01:11:38   WebP and process them or whatever. And I hope we get over that. Again, this being royalty free

01:11:42   should help. Being royalty free also means people don't make money from it. So some people are

01:11:46   motivated not for that format, not to spread, but my fingers are crossed on this. And this is just a

01:11:53   rumor. We don't even know if it's true, right? I haven't even seen the betas. And even in the,

01:11:57   is JPEG Excel going to be the default? Are they going to talk about it in the event? And because

01:12:01   they did talk about heeck and heef, they said, hey, we've got a new image format and it makes

01:12:04   better images, right? And don't worry, it'll be compatible because when you share it, it'll

01:12:08   convert to JPEG. I assume they would take the same approach to JPEG Excel. And where does that leave

01:12:13   heeck and heeck? Are they still even going to be an option? Or are they just going to be removed

01:12:16   from the menu and it's going to be plain old JPEG, same as it ever was, or JPEG Excel, which is the

01:12:21   extra large version, but not really. Yeah, this I'm hoping this takes off because heeck and heef/heek,

01:12:28   I know one's the container and one's the format. Heef has just still, it's been around for a while.

01:12:33   It has never gotten good enough support across the industry to the point where like, if you are

01:12:38   taking a photo from an iPhone and sending it anywhere or using it anywhere else besides

01:12:44   another iPhone, you've probably been just having to convert it to JPEG. And to the point where like

01:12:48   Apple's own frameworks will do that for you a lot of the time, because that is usually what people

01:12:53   need to do. I would love for JPEG Excel to get wide enough support everywhere that we don't have

01:13:00   to do this dance anymore in a few years. Because JPEG has been a wonderful format for the entire

01:13:07   tech industry for the 150 years that we've had it. We do have better techniques now. We do have

01:13:14   better compression now. We do have like different needs now. It is time to hopefully retire JPEG

01:13:22   and replace it. And the only reason we're still using JPEG is because it is like the kind of

01:13:28   universally compatible version. Now, everything I just said is also true about MP3. And I say that

01:13:34   with love because I love the MP3 file format. I think it's a lot easier to replace JPEG than to

01:13:40   replace MP3 for a lot of reasons. Well, I mean, here's the difference though, because you mentioned

01:13:43   JPEG and how much you love it and it's been around for a long time. But JPEG has obvious problems,

01:13:49   which is the reason people used anything that wasn't JPEG. First of all, if you have something

01:13:53   that's like line art or something like that, like a black and white line art, a GIF is better for

01:14:00   sharpness. 256 grays in a GIF, for example, is going to give you sharp, crisp edges, none of that

01:14:05   weird compression artifact stuff or whatever. It'll be a real small file or a PNG or something

01:14:10   like that. JPEG makes things blurry and gross unless you crank up the quality and then it makes

01:14:14   your file size bigger. It is made for photos. It relies on your perception of human perception,

01:14:20   and it's great for photographs. But black text on a white background, line art, tiny text that

01:14:27   becomes a fizzy mess, that is JPEG's weakness. Transparency, not a thing in JPEG, except for

01:14:33   probably JPEG 2000. Animation, I believe there is an animated JPEG. There's motion JPEG formats or

01:14:38   whatever, but lots of things that we eventually found to be useful in image formats. Plain old

01:14:43   JPEG at most implementations simply doesn't support. Not true of MP3. MP3 has a lot of things

01:14:48   going for it. Number one, pretty much every thing you could think to do with an audio file, MP3

01:14:53   supports. And number two, MP3 is so close to the perceptual limits of human hearing and file sizes

01:15:01   in the grand scheme of things are so low that we don't care that you can make a better sounding,

01:15:06   slightly smaller file with AAC or whatever, because the difference isn't that big. You know

01:15:10   what I mean? Maybe for hours and hours of video or something like that, and maybe for multi-channel

01:15:14   formats is a whole other thing. But MP3 for its role doesn't really have any of those glaring

01:15:19   weaknesses. No transparency in an image format for the web is so killer. And there's no equivalent to

01:15:25   that for like, "Oh, here's the problem with MP3." And so it was bad when we were doing MP3s at really

01:15:30   low bit rates, but a 320 kilobit MP3, that's fine for 99% of the planet for basic two-channel audio

01:15:40   or mono audio. And so, yeah, I think MP3 has still got more legs, but JPEG, like again, the reason

01:15:46   anybody uses formats other than JPEG is because they know JPEG's weaknesses and they know this

01:15:51   has to be a ping, this has to be a GIF, or this has to be something else entirely because JPEGs

01:15:56   can't do it all. But JPEG XL, like those old things I said, I don't know if this is the case,

01:16:01   but when I read support for both photographic and synthetic imagery with two complimentary modes,

01:16:05   I think it's going to be one mode where it recognizes line art. I don't know if that's true.

01:16:09   I'm putting my hopes and dreams in that one line there. And then the other thing where it can

01:16:13   transcode existing, most existing JPEGs and make them 20% smaller essentially losslessly, and you

01:16:19   can reverse that. That means that like, if you have a library full of JPEGs, like if you've shot

01:16:23   JPEGs for years and years on your phone, and then when heat came out, let's say you didn't pick that

01:16:28   format, but either way, if you've got JPEGs on your phone, this format can make all your JPEGs

01:16:33   20% smaller, like losslessly, not re-encoding them and making them look worse. Just take that JPEG and

01:16:38   replace it with a file that is exactly the same, but it's 20% smaller. That's one of the great

01:16:44   things about having JPEG in the name and saying it will subsume your existing JPEGs and make them

01:16:50   smaller. And 20% is not a small number. So, again, we're just reading a feature list and this is from

01:16:54   the Wikipedia page and it makes everything sound like, you know, wonderful and roses. And we'll see

01:16:58   if this is really true or if all my hopes and dreams are accurately represented by these

01:17:03   bullet points. But this is looking great to me. I mean, just recently I've been putting images

01:17:08   up on my website and, you know, it's just like, why are images such a mess still on the web?

01:17:13   And here's JPEG XL. Come to save us all. Probably people said the same thing about JPEG 2000 and,

01:17:18   again, that didn't quite work out, but maybe this time we'll do it.

01:17:21   - I hope so, 'cause like, if you think about like what it takes for a new image format to

01:17:26   really like gain a foothold and really, you know, get as widely supported as JPEG and PNG are today,

01:17:34   we have such a massive world of technological tools and infrastructure and, you know,

01:17:40   various devices, server-side, you know, open source libraries, you know, server-side language

01:17:47   support, different, you know, CDN support, obviously browser support, different phones,

01:17:53   whether cameras are gonna shoot this format. Like there's so much out there that either processes

01:18:00   or converts or resizes or displays images. And this is part of the reason why we haven't had

01:18:07   more formats that really take a foothold because the world of just tech and software and hardware

01:18:15   that deals with images is so vast. So for this to really take off, obviously having, you know,

01:18:22   royalty-free and open source components is massively important here because if there's

01:18:29   any kind of royalty requirement on a format, that means that the entire ecosystem of open source

01:18:35   tools and of server-side, you know, language supports and everything, it's probably not gonna

01:18:39   support it, or at least it's gonna support it in a very loose or half-assed way.

01:18:43   - And there's gonna be format wars 'cause they're like, well, you have your royalty one and I have

01:18:46   my royalty one and I want mine because I get money from it and you want yours 'cause you get money

01:18:49   from it. And even in this case, there's an open source implementation, but it's not like a copy

01:18:54   left type of, you know, GNU public license or whatever. It's BSD, which is the one that says,

01:18:58   take it, make money off it, whatever. Like, it's not even a restrictive open source license.

01:19:02   - Yeah, so that really is setting the stage for this to be widely adopted. So we'll see. I mean,

01:19:10   the only other thing that generally tends to hold back new formats from being adopted besides,

01:19:15   you know, the technical requirements, royalty requirements, open source lack thereof,

01:19:22   the only other thing that tends to hold things back is if it's not a big enough improvement over

01:19:27   the ubiquitous established format. That was always the problem with JPEG 2000. And I think to some

01:19:33   degree that was a problem with HEAF. I think this might be enough overall improvements to finally

01:19:40   make it worth people actually like putting in the work to implement this. So I hope this goes that

01:19:46   way because again, like JPEG really has had its time. It's been wonderful, but this is a lot of

01:19:52   advancement over that and I hope we get the chance to actually use it. - And even if it isn't, even

01:19:57   if it turns out this is not a big enough jump for it to be adopted industry-wide, I'll say the same

01:20:01   thing that I said when Apple went to HEAF/HEAF. Even if this is only a format that only Apple

01:20:06   ever uses, if it makes all the pictures on my phone a little bit smaller and they all just get

01:20:10   converted to JPEG anytime they're exported, which again has basically been my experience,

01:20:14   I never have to think about the fact that this phone is always taking pictures in HEAF/HEAF unless

01:20:18   I like look at, see that format in the photos menu on my Mac when I export, I'm like, "Oh yeah,

01:20:22   that's right. That is a format." I don't have to think about it, but they take up less room in my

01:20:28   iCloud photo library. They take up less room on my phone. That's worth doing. That's why I didn't

01:20:34   change that option. I want my pictures to be smaller. I keep taking pictures. I want them to be,

01:20:38   like I said, I want them to be smaller or better quality at the same size and just keep making

01:20:42   those formats better. So probably like half my collection is JPEGs before they introduce HEAF

01:20:46   and then half of them is the new format and then I'll start with JPEG XL. I would prefer it if

01:20:52   it went industry-wide and solved those problems. But to give another example, because of Apple,

01:20:58   I use the photo screensaver on all of our Macs and I have my favorites for my photo album being

01:21:02   like the floating images going up and down your screen. It's like pictures of my family and

01:21:06   everything. They're all the favorites from my photos library. The screensaver has a feature

01:21:10   that says, "Hey, pick an album from your photos library in this little picker and we'll use that

01:21:16   to feed this slideshow." But I can tell you that once you push up around 200,000 pictures, that UI

01:21:21   gets very upset. So what I do instead, Apple, you could fix this, but anyway, what I do instead is

01:21:27   I go to the favorite collections in photos and I export them all to a folder on my Mac somewhere.

01:21:33   And I don't export them at 100% size because they're just going to be in a screensaver and

01:21:36   they're usually pretty small in that floating screensaver. So I export them in like the

01:21:39   quote unquote large size instead of original size. And I do that in HEAK. And when I used to do that

01:21:44   in JPEG, the folder was way bigger and now the folder is smaller. And I like that, right? And

01:21:50   then you just point the screensaver at the folder and it can somehow handle that one. It couldn't

01:21:52   handle the thing in the photo picker library. I try not to push my photo library too much. I don't

01:21:57   want to upset it. But anyway, just for that thing, taking up less space on my phone and taking up less

01:22:03   space in the favorites folder that is on all the Macs in my house, for that alone, please, I welcome

01:22:09   JPEG XL. But yeah, I really hope it does sweep across the whole industry because it seems like

01:22:14   it just finally resolves all of the issues. It's the image format that does everything we want

01:22:18   image formats to do and hopefully does it well. All right. You want to do some Ask ATP? Let's do

01:22:24   it. All right. So Tom Thorpe writes, "It's that new iPhone time of year again and time to ask the age

01:22:29   old question whether to restore from backup or start fresh. I've been on the same backup for 10

01:22:34   years, but I'm starting to think it would be nice to start fresh partially because I want to clear

01:22:37   out unused apps, but I also still can't help but imagine it would clear out the cruft, provide

01:22:42   better battery life, et cetera. What are the pros and cons of doing so today? What would I lose by

01:22:47   starting fresh? And what would I get any of my imagined benefits?" I don't think I've ever

01:22:54   started fresh. I think I'm still carrying the same iOS installation, if you will, from my 3GS way

01:23:01   back when. So I don't think I can say from experience whether it would be a good idea or not.

01:23:08   Generally speaking, I don't think it's a bad idea. The thing that scares me the most personally is

01:23:15   losing my iMessage history, which, granted, is basically a black hole anyway because search is

01:23:21   terrible in messages. But I don't know, it gives me the heebie-jeebies to think of losing it.

01:23:26   But I guess, man, go for it and report in and let us know if it was a mistake, especially if you take

01:23:34   a backup first. I mean, it's not going to hurt anything. You'd always go back to your backup.

01:23:36   All right. So Marco, correct me. I mean, the whole thing of like, "Well, you could always back up."

01:23:42   I feel like, yeah, you can, but after a couple of days of using the new one, it's like how when you

01:23:48   undo in an undo stack, as soon as you make any change, you can't redo anymore because there's

01:23:55   no way for you to merge that change back into your stack. It blows the redo stack.

01:24:01   You can take a backup, but as soon as you start doing anything on the new installation

01:24:07   that is like creating new data, writing new things, you're not going to want to ever go

01:24:12   back to that backup because it'll be too disruptive or you will lose something that you've done.

01:24:17   So basically, that's usually a pretty one-way transition. I would say if you actually want to

01:24:25   have a clean startup, that's like a really appealing thing to you for some reason,

01:24:29   sure, go ahead, do it. I don't think, though, that it's necessary with modern OS installations

01:24:37   in almost every case, especially on the phone. On the Mac, sometimes you can build up a whole

01:24:40   bunch of weird stuff in various library directories, even then you can clean that out.

01:24:44   So it's rarely necessary on the Mac. I would say on the phone, it's basically never necessary.

01:24:50   I think, like you, Casey, I think I've either never done it or I've done it maybe once in total ever,

01:24:55   but I don't think it's really necessary because iOS is already so controlled and limited. You're

01:25:02   not throwing in texts, you're not throwing in weird preferences that live around forever in some

01:25:08   directory. You don't need to run iPhone cleanup utilities to find lost and abandoned files.

01:25:13   There's so much in the iPhone that because of the technical limitations imposed on apps and because

01:25:20   of all the containerization of all the app's data and everything, you don't really need most of that

01:25:25   stuff. Now, whether you want to go through your apps and clean out apps, that's a different story,

01:25:30   but you can just delete the apps. You don't need to reinstall the whole thing from scratch instead

01:25:35   of bringing over your restore. I think mostly what you'd be doing by doing a fresh installation

01:25:42   without a restore and manually starting things from scratch, I think most of what you're doing

01:25:46   there is causing a lot of time loss, possibly some data loss in apps that don't transition over

01:25:53   at all or well or perfectly, all for a benefit that I'm not sure is real, or at least I'm not

01:26:02   sure it would be anything you would ever notice. So I would recommend if this is not something that

01:26:07   you really think you need to specifically do because of a specific reason you're trying to

01:26:11   get rid of, I would say don't do this. Just bring your stuff over. It's not going to make a

01:26:18   meaningful impact on stuff like battery life or things like that. That's mostly a myth on iOS, and

01:26:24   there's other ways to control that, like deleting apps. So just go through, like if you really want

01:26:27   the same effect, go through and delete a bunch of apps that you don't use anymore. You can do that.

01:26:31   Just do that. It's fine. Before I address Tom's question, I'll just say that all three of us are

01:26:36   in agreement for just general people who don't have this specific question. Our advice is just

01:26:40   transfer data from your old phone. It's what everybody does. It's the default. That's what

01:26:44   we recommend. Flat out, not even close. Now, Tom's question is like, I think I might have a problem.

01:26:50   I think I might want to do this. What are the trade-offs here? What would I lose by starting

01:26:54   fresh? Marco kind of already addressed this. It is possible to lose something because if you have an

01:26:58   app, especially older apps, but who knows, it could be modern ones too, that don't have any kind of

01:27:03   cloud sync type thing, and you start fresh and you reinstall that app, the data that was associated

01:27:09   with that app that was on your old phone is still sitting on your old phone. And that's the only

01:27:13   place it is because it doesn't have any kind of like it saves it up to iCloud or it puts it on

01:27:17   a server or does any kind of syncing thing. It's just locally on that inside that app.

01:27:23   If you did a backup on restore, did a data transfer, you would keep that data most of the

01:27:26   time. Right. But if you, but if you have an app like that and you're like, okay, let me download

01:27:30   that app from the app store and you launch it and you see nothing in the app, you're like, hey,

01:27:33   where's all my stuff. It's back on your old phone. So it totally is possible to lose data. Again,

01:27:37   good apps don't do that. They have some kind of cloud sync thing, the use cloud kit or whatever.

01:27:40   That's how people expect iOS apps to work, but there's no way as an end user for you to really

01:27:45   know what is this app doing? Is it just storing things inside its container locally on my old

01:27:50   phone? And if I install the app fresh on a new phone and sign into it with my account,

01:27:55   it doesn't bring my data with it. You won't know that until you do that. So that is

01:27:58   a pretty significant risk. There's another reason to keep around your old phone when you're setting

01:28:03   up your new phone, just to make sure that's not the case. As for what I get any of my imagined

01:28:07   benefits, there are scenarios where a fresh install will save you, but they're pretty obscure.

01:28:12   So because everything is so containerized or whatever, the usual way that you're going to get

01:28:17   any kind of benefit is there's some kind of problem situation, usually having to do with

01:28:22   corrupt data. Like there's some kind of file format or data or whatever associated with an app that

01:28:27   for whatever reason, a bug in the program or who knows something, it got mangled. And so it's

01:28:33   sitting on your SSD on your phone and it is improperly formatted. And maybe you don't notice

01:28:38   because it's some obscure corner of some database that some file uses to keep track of its stuff,

01:28:43   right? But the way this could potentially manifest is let's say that app runs in the background once

01:28:48   an hour for 60 seconds. If during that 60 seconds it goes into a tight loop because it's decoding

01:28:54   code, gets tripped up over the bad formatting and just goes around and around and does an infinite

01:28:58   loop for 60 seconds before it gets killed, you don't know that's happening behind the scenes.

01:29:03   But once per hour, there's a little tight infinite loop for 60 seconds, draining a little tiny bit of

01:29:07   your battery because of some tiny corrupt database and something you have no visibility into this.

01:29:12   The app developer thinks there's nothing wrong with my app. I don't know what the problem is.

01:29:14   He would have to diagnose and get your actual data file to determine that your data file was corrupted

01:29:19   maybe by a bug of three versions ago in his program. And you've been slinging that corrupt

01:29:22   database file around from phone to phone to phone, burning tiny little bits of power on it and

01:29:28   multiply that by three or four apps that have this problem. If you do a fresh install of those apps

01:29:33   and it syncs a new thing down from the cloud or whatever, you wouldn't have that problem, right?

01:29:37   If it's corrupt in the cloud and you do a fresh install and you install that app and it syncs to

01:29:41   the cloud, you've got the same problem. You know what I'm saying? And you don't know unless you're

01:29:46   the developer of that app, is this corrupt just locally or is it corrupt in the cloud?

01:29:50   The only one saying is you could have problems on your phone. Do a fresh install, reinstall all the

01:29:56   apps that you lose, and restore all those problems. Because all the problems were like corrupt data in

01:30:01   the cloud or something like that or any bug related to that, right? Because you never know, like, is the

01:30:05   problem that I have, like people have this concept of like, I think there's cobwebs inside their phone

01:30:09   or whatever, but it's, you know, there are specific things that applications could be doing that could

01:30:14   be causing you a problem. And I'm not even sure where the ratio is, but a huge number of those

01:30:19   are not solved by setting up a new phone at all because it's just going to faithfully reproduce

01:30:24   every single one of those problems. If you just delete that entire app, then hey, problem solved.

01:30:28   But if it's an app that you use every day and you had a problem over there, you're also going to have

01:30:31   a problem over here in a lot of circumstances. And that's not satisfying, right? On the Mac again, it

01:30:36   is much more likely that there's some leftover files or something you never use that you can just

01:30:40   clear out and doing that manually is very difficult and I do not recommend people do it. And it takes

01:30:44   a lot of experience not to hose yourself, so, you know, that's why the Mac is what the Mac is. But

01:30:48   on the phone, your tools are fresh install or backup, delete app or not delete app. And that's

01:30:56   pretty much where it ends. And that's good. You don't want more tools than that, but sometimes

01:31:00   you have problems that you can't escape by doing a fresh install. And that's just a break. So I

01:31:06   would recommend not doing it. You didn't say you had any specific problem. You just want your thing

01:31:10   to be better. Getting a new phone and setting it up as a fresh phone by itself will empty out a lot

01:31:15   of stuff. Like when you transfer data from your phone, it's not transferring like every single

01:31:20   thing that was on your phone to your new phone. It's transferring the things that it thinks it

01:31:23   needs to transfer. And everything else on your new phone is fresh. Probably all like the caches

01:31:28   directories, the spotlight indexes are going to be recreated or whatever the heck Apple intelligence

01:31:32   is doing. It'll sync your photos down from iCloud fresh. So if a bunch of photos were corrupt over

01:31:37   there because of some heated up part of your SSD flipped a bunch of bits or something, you won't

01:31:42   have that problem over here as long as they weren't corrupt in iCloud. Just restore from backup. It

01:31:46   will still be a fresh phone even if you don't start from zero. And if you do start from zero and you

01:31:50   have a lot of stuff, you will spend hours and hours and hours trying to get it back like your old

01:31:54   phone. And as soon as you think you have a good and you sell your old phone or wipe it or whatever,

01:31:58   then you're going to find the thing that you missed and it was only on your old phone and you

01:32:01   deleted it. So don't do it. All right. Michael Quinn writes, "So we all buy new iPhones in a

01:32:07   few weeks, spend the next 12 months training them as our personal AI assistant with stuff securely

01:32:11   stored on board. What happens after that? Does that transfer to iPhone next? Do we have to retrain

01:32:18   every iPhone forever a la waiting for photos and spotlight to rescan and index each year?

01:32:22   Kind of sucky experience if so. Maybe another barrier for upgrading if you know you're going

01:32:26   to lose your personal AI." This is a really good point and holy crap, I would assume since Apple

01:32:33   is a company that knows what they're doing and cares that it would transfer, but then photos.

01:32:37   So I don't know. I don't know what's going to happen. I mean, I think Michael's answer here,

01:32:43   Michael's question here contains the likely answer. So what Michael says in the question is,

01:32:47   "Do we have to retrain every phone forever a la waiting for photos/spotlight to rescan and index

01:32:53   each year?" I think it's probably going to end up working much like spotlight indexing because that

01:33:00   is kind of what it's built upon or at least that is like the structure that they have built it with.

01:33:05   So chances are, and again, I don't think we know this yet, but chances are the way Apple intelligence

01:33:10   indexing or assistant learning will work on the phone is very similar to photos in that when you

01:33:17   get a new phone or you do a restore, it'll spend the first couple of nights that it's plugged in

01:33:22   all night indexing your data and training itself on that data. It's going to be, or you know,

01:33:28   it'll as it's sitting on your desk during the transfer, it'll heat itself up like crazy,

01:33:32   melt itself while it's doing all this. So I would expect it will be similar to that.

01:33:38   Apple does have a history of features like, for instance, like some of the people training

01:33:44   metadata stuff in the photos app. Apple has a history of having features that have local

01:33:51   intelligence or categorization or other ML kind of data that is generated on device,

01:33:58   but then they have ways for other devices or restores on that same device later down the road

01:34:03   to re-index and basically get that same data somehow on them. So I think it will be very

01:34:09   similar to that process where whatever you have like taught it, that will probably, and by the

01:34:15   way, we don't even know necessarily what that means yet, but I don't know if any of the features that

01:34:19   we've seen even have things that you can teach them besides indexing content from your apps,

01:34:25   but whatever effort you put into training Apple intelligence will probably carry over just fine.

01:34:31   And I think, again, most of the features that we're going to see in this first year,

01:34:35   I think are going to be based upon either, you know, just kind of global knowledge list models

01:34:40   that you're not putting your special, you know, special spin on or index content from apps that

01:34:46   can just get re-indexed as when you have a new phone. So photos is a good example of this because

01:34:51   I think one of the photos things he's probably referring to is face recognition. And that's

01:34:55   feature resource introduced. Photos would scan all your photos and find people and find their faces

01:34:59   and do all that stuff. And you get a new phone that would have to re-scan them and re-find all

01:35:02   the faces. And it was frustrating because the scanning for faces is not perfect. And sometimes

01:35:07   you have to correct it. No, this isn't my mom. This is my grandmother. No, this isn't my first

01:35:11   kid. It's my second kid. And you would do those corrections because it would get it wrong. And

01:35:15   then you get a new phone and you're like, geez, it's getting it wrong again. I corrected the same

01:35:19   photo that it thinks is my mom, but it's actually my grandma. It's literally the same photo. I

01:35:22   corrected it on the other thing. This is like Michael's question. If I do all this stuff on my

01:35:26   phone, do I get a new phone? Do I have to like, essentially like I lose all that information?

01:35:29   And so Apple added a feature, I think a year or two after this, the face recognition came out,

01:35:35   that said, we will transfer your face recognition data from your old thing to your new thing.

01:35:41   But they didn't mean get your faces set up all the way you want them on your old phone. And when

01:35:47   you get your new phone, it'll be just like that. That's not what they meant. The way they implemented

01:35:52   it was anything you manually did on the old phone, when you corrected that and said, that's not my

01:35:58   mom, it's my grandma or whatever, right? That action by the user, those actions were recorded

01:36:05   and transferred to your new phone and synced to your new phone or whatever. Everything else was

01:36:09   not. So what would happen is it would transfer over basically like the manually entered info,

01:36:14   which is probably pretty small because you just manually do one or two things, but you got

01:36:17   thousands of photos. Then it would rescan all of your photos and do face recognition augmented by

01:36:22   your corrections. But the rescanning is probably like a new version of iOS and a new code base or

01:36:27   whatever. And who knows, it's not going to, first of all, you have to do the rescanning because that

01:36:31   so it takes time and energy or whatever, it's usually not that long. And second, it's probably

01:36:36   not going to end up in the same state as your old phone because that was scanned, you know,

01:36:39   some of those photos were scanned two versions of iOS ago, depending on how old your phone is

01:36:43   or whatever. And there's totally different code scanning. And by the way, it's scanning with the

01:36:47   benefit of all your manual data, whereas those are scanning with only the benefit of some of

01:36:50   your manual data that exists at the time the scan took place. And so you end up with your new phone,

01:36:55   not being like your old phone. It's one of my frustrations with the photos on the Mac,

01:36:58   because I meticulously arrange my, you know, select my faces and everything. And on the Mac,

01:37:04   your iPhotos library is essentially a folder on your Mac. I can never lose that folder because

01:37:09   when I go to another Mac and pull up the same photo library, faces are all over the frickin'

01:37:14   place. It doesn't benefit from all of my stuff. Like this is a carefully curated collection of

01:37:20   things that have been scanned over the course of years, plus my manual corrections or whatever.

01:37:23   So that is what they did with photos. What might they do with AI? There's no technical reason for

01:37:31   both photos or AI that they couldn't take your complete semantic index or whatever they're

01:37:37   calling it and transfer it wholesale to your new phone. They could do it through the cloud,

01:37:42   because they have end-to-end encrypted things in the cloud where Apple would have no access to

01:37:45   the information, but it would go up to the cloud in a way that Apple has no access to it because

01:37:49   it's end-to-end encrypted. Not like your iCloud messages backup where they can read it all or

01:37:53   whatever, but like totally encrypted and pull it down to your phone. It could be device-to-device

01:37:57   transfer where it never touches any Apple server, but it goes directly from your phone to your other

01:38:01   phone. They could do that, but I kind of understand why they don't want to. Well, for multiple reasons.

01:38:07   One is that I think the amount of manual information in the vein of like, "Oh, I'm correcting the face

01:38:15   in this picture for AI" is going to be small. People like this question envision like, "Oh,

01:38:19   I've got this AI on my phone and I've trained it up to really know me or whatever." You haven't

01:38:23   trained it up. You're not retraining any models here, right? Basically, all you did was throw your

01:38:28   data in it and let it chew it up. If you allow it to chew up that same data on your new phone,

01:38:33   it should more or less come to the same result or maybe better because it's a new version of the

01:38:38   model or whatever, but it's not as if you are retraining the model, so to speak, and it's somehow

01:38:42   the model is getting smarter and learning you or whatever. You are just feeding it more data into

01:38:47   that semantic index thing and that it's reading that or whatever. That same process should be

01:38:53   pretty much 100% repeatable on your new phone because, again, like the LLMs and all their

01:38:57   weights and all their numbers or whatever, you are not changing those by talking to the LLM.

01:39:01   Those never change unless Apple releases a new model. It's only the semantic index stuff that's

01:39:06   there. How did that index get formed? It got formed without your intervention. It got formed

01:39:09   when you were sleeping and your phone was plugged in and it was scanning all your data. As long as

01:39:12   that data is over there, the semantic index will be over there. I kind of understand why Apple

01:39:16   wouldn't copy it. Also, like I said, if they make improvements to that, you want to do the scanning

01:39:22   with the new improved system rather than taking the scan that was made with the one-year-old

01:39:26   system and transferring it to your new phone. That's how I guess this is going to work. Again,

01:39:30   we don't know for sure. Apple Intelligence hasn't even shipped, let alone had anyone transfer from

01:39:34   one phone to another. But I don't know. Let's put it this way. I think it'll be better than photos

01:39:40   because I think photos, there is actual work that you're doing and like a presentation of all the

01:39:47   faces that you don't get reproduced. I think with Apple Intelligence, whatever happens on your old

01:39:51   phone, when you set up a new phone, it should all be reproduced because I think there's less manual

01:39:56   stuff. Thanks to our sponsor this week, Trade Coffee. Thanks also to our members who support

01:40:02   us directly. You can join us at atp.fm/join. One of the perks of membership is ATP overtime,

01:40:08   a bonus topic every week. This week's overtime, we'll be talking about rumors of a folding all

01:40:15   screen MacBook, which is I think very interesting. So we'll be talking about that. So that's this

01:40:21   week's ATP overtime. If you want to hear us talking about folding all screen MacBook this

01:40:26   weekend, new topics every weekend overtime, plus other membership bonuses, go to atp.fm/join

01:40:32   to join up and support us that way. Thank you so much to everybody and we'll talk to you next week.

01:40:38   [Music]

01:40:41   Now the show is over. They didn't even mean to begin, cause it was accidental.

01:40:47   Oh, it was accidental. John didn't do any research. Marco and Casey wouldn't let him,

01:40:56   cause it was accidental. It was accidental. And you can find the show notes at atp.fm.

01:41:07   And if you're into Mastodon, you can follow them at C-A-S-E-Y-L-I-S-S. So that's Casey,

01:41:17   Liz M-A-R-C-O-A-R-M-N-T. Marco, Armin, S-I-R-A-C-U-S-A-C-Racusa. It's accidental.

01:41:31   They didn't mean to accidental. Accidental. Check podcast so long.

01:41:41   I have a brief life update I am happy to share with everyone.

01:41:45   Oh? Are you doubling your RAM?

01:41:46   Yeah, I'm doubling my RAM. Actually, I'm doubling my 5G because as of a few hours ago,

01:41:51   I am even more vaccinated than I was before. What?

01:41:54   He's got his new COVID vaccine. He's got his new COVID vaccine is what he's saying. I need

01:41:57   to get that too. Yes. Oh, yes. Yes, indeed. You get a flu shot at the same time?

01:42:03   No, in my personal opinion, October is the right time for flu shot based on no science,

01:42:07   no medicine, just an opinion. There is science, but like there is science about the timing. It's

01:42:12   any of these shots because your immunity is strongest. I think like at first, you know,

01:42:17   right after you get the shot, your immunity is the strongest and it does fade over time.

01:42:20   So there's a strategy that says wait until you're farther into like cold and flu season to get your

01:42:23   flu shot. Same thing with COVID. But the thing is, especially with COVID, you get COVID tomorrow.

01:42:28   So yeah, it's going to be worse in the winter. But if you get it tomorrow, waiting was the wrong

01:42:33   move. But if you don't get it tomorrow, waiting was the right, you know what I mean? And flu,

01:42:36   I think is more, has more of a seasonal bump to it. So you can probably gain that a little

01:42:40   bit better. But like, I just know so many people have gotten COVID lately. Like the summer has been

01:42:44   a pretty big surge and it's out there and it's waiting for you to get it. And so I'm going to get

01:42:48   the COVID shot immediately. I'm also going to get the flu shot immediately because I don't even

01:42:51   bother timing that and I just don't want to have to go back and forth, you know what I mean? To get

01:42:55   multiple shots. But yeah, recommended when you're in there getting your COVID shot, get your flu

01:42:59   shot. And if you want to wait and take the risk, feel free. But if you get COVID before the shot,

01:43:02   that sucks because you got COVID and it also sucks because you can't get the shot until you've been

01:43:06   over COVID for some amount of time too. So not great. Yeah. Yeah. I am stunned that we didn't

01:43:14   bring it home from Europe and I feel like our time is ticking because it's, it appears from what the

01:43:20   kids are telling us that like half of their elementary school is out for multiple days at a

01:43:24   time. And then occasionally some of them will come back with masks on. And so you put two and two

01:43:29   together and you know exactly what's going on. It's only a matter of time. So here's hoping that

01:43:34   I dodge it at least long enough to get through the podcastathon. And then if I come down with it

01:43:41   after that, so. And COVID was out there in Europe with you. A bunch of people who were there got it.

01:43:45   A bunch of people were there with it and got it while they were there or brought it back. And,

01:43:48   you know, it just, you dodged it. Oh, it's every, like all summer, it's been everywhere. Like we,

01:43:54   we got it in our, in my family, we got it in May and it's been like, everyone has it around here

01:44:01   all summer long. I just talked to one of my friends here today, just, is just coming off of

01:44:05   it. Like from the last nine days, like every, it's everywhere. Like, you know, we, we like to think

01:44:12   that we're like past the era of COVID. We're not only are we so not past it and not only are we

01:44:17   probably never going to be past it. But the only things that are really different now is that it's

01:44:24   for most people, it is less severe to get it now than it used to be because we have some, you know,

01:44:29   experience with it in our bodies now. But it's, it's still very widespread and still causing

01:44:34   problems for a lot of people. And so it's just everywhere. So yeah, definitely get your vaccines.

01:44:40   Like, yeah, they're not perfect. They're also a lot better than nothing. So get your vaccines.

01:44:45   That's, it's not a big deal. Just please, for the love of God, like do, do what you can to help.

01:44:50   **Matt Stauffer** Yeah. You know, you read a story, it's like, oh, the vaccine is for a strain that's

01:44:53   not even circulating. That's true. But like what the vaccines, you know, aren't, even if they don't

01:44:57   stop you from getting it, they make the symptoms more mild. And that's what you want. Like you want

01:45:02   every, you want every bit of help you can get, even if it helps just a little, even if it makes

01:45:05   you like a little bit less, because there is still the possibility that you're going to have a bad

01:45:09   experience or get long COVID. Like that is still happening to people. Like we benefit from, quote

01:45:14   unquote, benefit from the fact that a lot of the people that COVID was going to kill, it has killed

01:45:18   like millions, right? So those people are dead and they're, you know, and the people who are still

01:45:22   here are the people who didn't die or didn't catch it. And we have lots of vaccines and that makes the

01:45:27   symptoms less severe. But yeah, it's just not, not a thing that you want. Same thing with the flu

01:45:31   shot. Do they guess right on the strain or whatever? It's just like every little bit helps,

01:45:36   right? And in, you know, the, the studies say it does help. It does help if you've gotten

01:45:40   vaccine, vaccinated. It does help if you've gotten vaccinated recently. And it does help if you've

01:45:44   gotten vaccinated recently with a strain that is somewhat close to the one that's currently

01:45:50   circulating. So just do it. Yeah. That's highly recommended. And same thing with flu shot and

01:45:55   any other vaccines that you might be overdue for when you visit your doctor.

01:45:59   Yeah. I mean, honestly, like one of the, I wouldn't necessarily, it's hard to say like upsides,

01:46:05   but you know, a silver lining of all of this is that it seems like in recent years, as part of

01:46:11   the COVID vaccines ramping up it seems like getting almost any vaccine has gotten a lot easier. Cause

01:46:17   like the, like, you know, major drug store chains now have not only, you know, they, they, they had

01:46:23   flu before, but not in, I think the numbers they have now. And most people weren't thinking to go

01:46:29   and get it as much as they are now, especially younger people. Now you can go to like any drug

01:46:33   store in the U S at almost any time you want and get a COVID shot. You can get a flu shot. You can

01:46:40   get other commonly needed shots like HPV, like a lot of people should be getting like, there's so

01:46:45   many, it's so easy now to get access to so many vaccines that are just like in drug stores,

01:46:52   widely available for little to no money. It's actually, you know, to have built up all that

01:46:57   infrastructure now is, is quite, quite an upside. And if you have a primary care doctor and you do

01:47:02   annual visits, that's also a great time to do it. Time your annual visit for right before flu season,

01:47:06   you will get a flu shot when you go into the office, they'll have it there. It'll, you know,

01:47:09   just make sure it's out by the time you go and get all your vaccinations. It's very convenient.