PodSearch

ATP

589: The Correct Amount of Rocks

 

00:00:00   There is a way to get John Siracusa to leave his house. For those of you who remember, in November, I traveled up, I made a multi-hundred mile train trip to go see Marco and try the Vision Pro prior to release at a lab. And I said to John in advance, "Hey, I know you're not going to the lab, but wouldn't it be neat if you came down to New York and visited with us? I think that would be a lot of fun." And John told me to go outside and play hide and go screw myself.

00:00:28   Those were his words exactly.

00:00:30   Yeah, those words precisely. And so I thought there was no possible way to get John to leave his house, even for his beloved friends of years and co-workers, of years and years and years.

00:00:41   But it turns out, John, there is a way to get you out of the house, and I am happy to report the ATP reunion, for the first time since 2019, is a go. John, what's going on?

00:00:51   We are going to WWDC.

00:00:54   All of us?

00:00:55   At Apple Park, all of us.

00:00:56   All of us.

00:00:57   All of us.

00:00:58   All of us.

00:00:59   All of us.

00:01:00   All of us.

00:01:01   All three of us.

00:01:02   Indeed, I am extremely, extremely, extremely excited. I cannot overstate how excited I am, especially to see the two of you fellas, but also to be able to go to Apple Park.

00:01:10   The only place I've ever been on Apple Park is the visitor center, the public visitor center that you don't need any special privileges to get to.

00:01:17   I'm excited that we are press for the purposes of this event, which I'm really, really excited about.

00:01:22   I swear I have never done a press WWDC. I don't think so.

00:01:28   And I've been through my badges to see if I can find a press badge. Maybe I have, and it's just been so darn long and I don't recall knowing me.

00:01:33   That's probably true. But one way or another, it's happening again, and I am so excited.

00:01:38   So I am glad that you are willing to make the trek instead of just a couple of convenient hours down the interstate and/or train from Boston to New York.

00:01:48   You're just going to cross the country instead, and I appreciate the effort.

00:01:51   Yeah, I mean, I have done press WWDCs. We've all attended many WWDCs, but I've never been to one at Apple Park, and neither has Casey.

00:01:58   So that's what I'm most excited for, is to be officially allowed onto Apple's campus.

00:02:04   And I think they don't let you take real cameras, which is kind of a bummer, so I'll just take lots of pictures with my phone. But whatever.

00:02:10   I'm excited to do it. It'll be a fun experience. Yeah.

00:02:13   Looking forward to not looking forward to taking the plane flight, but you know, you do what you have to.

00:02:19   Alright, let's do a little follow-up. I wanted to briefly call attention to, there's some new Vision Pro content.

00:02:26   What? Really? Yes, there is. There's new Vision Pro content. Who'd have thunk it?

00:02:31   So there's three things I wanted to call everyone's attention to, two of which are from Apple and one of which is not.

00:02:37   First of all, there is a new sizzle reel. I don't know if that's the bestest way of describing it, but basically it's like a three and a half minute video that gets you interested in the Vision Pro.

00:02:48   And previously, I hadn't watched it in a couple of months at least, but previously my recollection is, or the way I remember it was, it showed like a little bit of the kids playing soccer, rhinoceroses.

00:02:59   It showed the tightrope walker, and I think it showed a very brief bit of sports, if memory serves. I don't entirely remember.

00:03:08   But one way or another, there's a new one, and I think it's really well done. It's still, it's a little jumpy for my taste, just a touch.

00:03:15   It's not like that MLS thing from a month or so back where it was way too jumpy. It's just a touch jumpy.

00:03:21   But the soundtrack is excellent, and I really, really like it. Again, three and a half minutes. You can find it in the Apple TV app.

00:03:28   And since it's new, it is kind of front and center. I know I was complaining and moaning about the information architecture last week.

00:03:37   But in this case, it's pretty good, so you should check that out.

00:03:39   Additionally, there's a new, I believe it's the Adventure series. This is the one that had the tightrope walker on it.

00:03:47   There's a new episode of that all about parkour, and it's three, I believe Brits, based on accents, although who knows.

00:03:55   But anyway, it's three Brits doing all various and sundry stunts across Paris.

00:04:01   And it's really, really well done. It's like 12 to 15 minutes long somewhere in that neck of the woods.

00:04:06   And I really enjoyed it. It's not earth-shattering, but it's really good.

00:04:11   And I got to tell you, no spoilers for the 12-minute video, but there's a point at which the three gentlemen are trying to jump from one rooftop to another.

00:04:20   And they position the camera such that if you want, you can look down and see how tall it is.

00:04:26   And I got to tell you, when it's 3D and immersive, it looks scary as hell. So it's pretty, pretty cool.

00:04:34   And I think both of these, I mean, altogether, these two things are worth literally 15 minutes of your time. I really do think it's pretty cool.

00:04:41   And then finally, What If has finally launched. I had planned to do the whole darn thing earlier today and report in on it.

00:04:49   Unfortunately, they seem to release on Pacific time, so it wasn't available until my afternoon.

00:04:54   And I've only had the time to do the first 10-ish minutes, but it was very, very cool.

00:05:00   The premise here, no spoilers, is you're in the Marvel Cinematic Universe, and what if things weren't the way they were in the movies?

00:05:09   What if things went a little awry and different? And it's your responsibility as part of the story to try to fix all this.

00:05:16   And so the portion that I've done so far, again, the first 10 minutes or so, is they teach you how to cast some spells.

00:05:23   They interact with you, so it begins as fully immersive, you can't see any of your own environment, and then it converts to augmented,

00:05:32   such that there are a couple of characters in your space. And I don't know if it was an extremely happy accident or if it was deliberate,

00:05:38   but I was doing this in my office, and I had spun my chair around so my desk and monitors were behind me.

00:05:44   And my office is, I don't know, 12 feet by 12 feet, something like that, so it's something like 4 meters by 4 meters.

00:05:49   Not terribly big. Yeah, maybe that's right. I don't know, it doesn't matter. Not big is the point.

00:05:54   And there were a couple of characters, one is designed to be floating in the air, and sure enough, he was floating in the air,

00:05:58   and his feet were kind of inside my guest bed, which is behind my desk. But that's fine, he's magic, so whatever.

00:06:04   But then the other character was designed to be on the ground, and sure enough, he was on the guest room floor.

00:06:12   Like he was standing on the floor, there was a little shadow below him, and it looked just spot on.

00:06:16   And then they had you do some, you go back into an immersive environment where all you can see is what they're showing you,

00:06:23   and they have you do some spells where you're using hand tracking in order to do stuff.

00:06:28   And there were three or four that I learned, and one of them didn't really work that well, but the rest were spot on.

00:06:33   And again, I've only done the first few minutes, but it's really slick, and it's free.

00:06:37   So definitely, definitely check all three of these out. Again, that's the new sizzle reel.

00:06:41   I think they just call it immersive, or like immersive demo or something like that.

00:06:45   And then the parkour, the adventure series, the second episode about parkour and what if by Marvel.

00:06:53   And I've put a link to the what if thing in the show notes, assuming I can dig one up, I should be able to.

00:06:57   I don't think I can link to the other ones in the show notes, but if I find a way, I will do so.

00:07:02   Alright, the low storage 13-inch iPad Pros have 12 gigs worth of RAM chips in them,

00:07:10   but they don't use 12 gigs of RAM? Jon, what's going on here?

00:07:15   This was discovered by the iFixit folks at Teardown.

00:07:19   Their lead Teardown technician, Shahram Moqarti, found that readers apparently,

00:07:24   meaning like people who saw the YouTube video, I don't know how they spotted this,

00:07:27   but they apparently spotted two 6 gigabyte RAM modules on the 256 gig and 512 gig 13-inch iPad Pro.

00:07:33   I looked at the Teardown video, I looked at the Teardown on their website, I'm like,

00:07:36   "How did they spot that these were 12 gig modules? I can barely make out the part numbers.

00:07:41   They're so blurry from compression." But whatever. Anyway, Shahram says,

00:07:45   "Our chip ID confirms this with high certainty. Two 6 gigabyte LPDDR5X modules produced by Micron for a total of 12 gigabytes of RAM."

00:07:54   Now, to remind you, Apple says this machine has 8 gigs of RAM, because remember,

00:07:58   if you want to get 16, you've got to get the higher storage. They say it has 8 gigs,

00:08:02   but there's apparently 12 gigs on the chips there. So why does Apple utilize only 8 gigs of RAM?

00:08:08   iFixit says Apple has never done this before, as far as we know.

00:08:12   Someone asked if it's possible these RAM chips are defective and some of the RAM is disabled for some reason.

00:08:16   The folks at iFixit says that, "My understanding is that if this were the case,

00:08:20   they would receive a different part number and be labeled as 4 gig,

00:08:23   but I don't think that's how LPDDR5 is manufactured anyway.

00:08:26   There's very little doubt that there's 12 gigs of RAM."

00:08:29   Another possible question, "What if those are the only LPDDR5X modules they can get their hands on,

00:08:34   but as soon as they're able to put in 4 gig ones, they will, so just to avoid problems,

00:08:38   this initial batch is artificially limited?"

00:08:40   The iFixit answer to that was that these modules have been in production since 2020,

00:08:44   according to the spec sheets, so it seems unlikely based on how long they've been around.

00:08:48   Super weird, because why would they do that?

00:08:53   If you follow the Twitter thread about this, you can see people throwing out a bunch of theories

00:08:58   and them basically getting shot down by iFixit.

00:09:01   Why would they use bigger RAM chips? Could they not get smaller RAM chips?

00:09:05   And if they couldn't get smaller RAM chips, why wouldn't they just free up the whole 12 gigs?

00:09:09   The other theory is like, "Oh, they're reserving a certain amount of RAM for LLM stuff."

00:09:14   Why would they only do it on this model and not all the models?

00:09:17   Because I think this is the only one that has more RAM and still all the other ones have the amount

00:09:21   that Apple says based on the teardowns, but this one has this particular variant

00:09:26   of this particular model has 12 gigs instead of 8. Super weird.

00:09:30   But maybe this is the side door for Apple finally putting more RAM in its products.

00:09:34   They'll just install it and not enable it for you. But it's there, you just can't use it.

00:09:38   How is your M4 iPad Pro treating your eyes?

00:09:43   Because apparently it's not all roses and pansies.

00:09:47   I mean, it's good for me, but Ben writes in to say,

00:09:51   "I'm upgrading to an M4 iPad Pro from the 2018 iPad Pro.

00:09:55   I noticed immediately that my eyes seemed unable to properly focus on the display,

00:09:59   resulting in eye strain, fatigue, blurry vision, and even headaches.

00:10:02   I couldn't use this display for very long before the symptoms reappeared,

00:10:05   so I went down a rabbit hole researching."

00:10:07   This is kind of like Marco with the Zhiyun Pro thing, but even worse. Ben continues,

00:10:11   "It seems like I'm not the only one experiencing this, though I have yet to determine the exact issue.

00:10:15   It might be PWM, which stands for Pulse Width Modulation."

00:10:19   And by the way, when I follow these links to look at the research that he was doing,

00:10:23   everyone just says, "Oh, it might be PWM. I think you have a PWM."

00:10:26   Yeah, it's probably PWM. People will come into a forum or a Reddit or whatever and say,

00:10:30   "Hey, I just got a new M4 iPad Pro, and the screen hurts my eyes.

00:10:34   What do you think the problem is?" And people would say, "Yeah, it's probably PWM."

00:10:37   And I was like, "Are you going to explain what PWM?"

00:10:40   I mean, I guessed it was Pulse Width Modulation just by knowing the term or whatever,

00:10:44   but when you're helping somebody, don't just say, "Yeah, you probably have PWM,"

00:10:47   because they don't know what PWM is.

00:10:49   Never mind that the term Pulse Width Modulation doesn't make much sense.

00:10:51   I bet if you hear this now, you're like, "Well, I know what Pulse Width Modulation is,

00:10:54   but what does it have to do with screens and why would it be hurting their eyes?"

00:10:56   So we'll get to that in a second.

00:10:57   So anyway, Ben says, "It might be PWM, though I've never known this to be a problem,

00:11:01   and I have been using LG OLED TV as well as OLED versions of the iPhone Pro

00:11:05   for years without any issue, or maybe tandem OLED is misaligned if there is such a thing.

00:11:09   I ended up going to the Apple store and compared my device with others.

00:11:12   Mine appeared to be slightly different, as if the HDR was turned on all the time.

00:11:16   Overall, the display was always just too much,

00:11:19   as best described as basically yelling at me all the time.

00:11:21   Since I was still within the 14-day return period,

00:11:23   they switched my device for a replacement, which seems to be much better now,

00:11:26   although not perfect.

00:11:27   My question is for Jon, are you noticing any eye fatigue with the new Pro,

00:11:30   especially compared to the 2018 version?

00:11:32   If so, do you expect this could be improved through software updates?"

00:11:34   So here's the research and what PWM is talking about.

00:11:38   According to these links that Ben provided that we'll put in the show notes,

00:11:42   the way this OLED and some other OLEDs handles brightness,

00:11:48   there's two ways OLEDs can handle brightness.

00:11:51   One is they can send less voltage to the pixels, and they're not as bright.

00:11:56   So for example, if you dim the brightness of the screen,

00:11:59   how does it do that?

00:12:00   One way you can do it is you can just send less voltage to the screen,

00:12:03   and it gets dimmer.

00:12:04   But if you only use adjusting the electricity going to the pixels

00:12:09   to control brightness, apparently when you get to low brightness levels,

00:12:12   you lose a lot of the color saturation, too,

00:12:15   and it looks kind of dingy and gross.

00:12:17   so they tend to not want to do that for the lower brightnesses. The other way you can control

00:12:23   brightness on an OLED is you can have the screen be at maximum brightness briefly and then go off,

00:12:29   and then max brightness and then off, and the longer the "on" period is, the brighter it is.

00:12:35   So they show a little graph and this is the pulse width modulation of saying,

00:12:39   "If you just pulse the screen, pulse, pulse, pulse, pulse, pulse, the faster the pulses

00:12:43   go the brighter the screen. If you go pulse, pulse, pulse, it is dimmer because the light is on

00:12:49   less of the time." You don't notice this because Apple screen, this TANO OLED, pulses at 480 times

00:12:55   a second, which is a pretty high refresh rate if you remember from the CRT days. Like, "Oh,

00:13:00   my screen looks flickering at 60 hertz. I can see the flicker, but at 85 I can't see it anymore.

00:13:04   At 120 I definitely can't see it. At 480 I can tell you I cannot see. Well, it's flickering."

00:13:11   But the complaint about the M4 iPad Pro TANO OLED is that it uses pulsing to control its brightness

00:13:20   through its entire brightness range apparently. Even at maximum brightness it's still pulsing,

00:13:25   as opposed to some other OLEDs which will use pulsing down at low brightnesses,

00:13:29   but once they get to high brightnesses they will do that by keeping it on all the time but

00:13:32   just sending less voltage. I don't know if this is just the way TANO OLEDs work. Is this the way

00:13:37   Apple is choosing to make it work? But some people report that this bothers them. I don't

00:13:42   personally see how it could because I am not aware of anyone who would notice flickering at 480 hertz.

00:13:49   We're not even talking about motion here. We're just saying like put on just a full field,

00:13:52   you know, red slide. I can't see it flickering at 480 hertz. If you had asked me whether this is

00:13:59   flickering to control brightness, I would have said no because I can't see it. But maybe people

00:14:04   with very young or better eyes than mine can see it. I don't know. Or maybe it's... Anyway,

00:14:09   follow the links, decide for yourself. And by the way, I would say do iPhones use this or do they

00:14:15   use voltage regulation? I honestly couldn't tell you because it just looks like a screen to me and

00:14:20   I can't see whether it's dimming or pulsing at 480 hertz. Some of the OLEDs that are out in the

00:14:26   market pulse at even higher rates than that. So yeah. Is this an issue? Is this something

00:14:31   people are imagining? Are some people just uniquely sensitive to it? All I can tell you is that

00:14:35   my old man eyes don't see this and aren't bothered by it. But we'll see. We'll see if Apple does an

00:14:41   update. We didn't put this in the notes for the last episode, but there is some kind of

00:14:44   actual software error with displaying HDR video where some like highlights are getting blown out

00:14:49   that Apple... Uncharacteristically, Apple immediately acknowledged inside a software

00:14:53   fix was coming for. So it is possible that maybe whatever issue people are complaining about here

00:14:58   will also be wrapped up in that fix, but we'll see. Michael Thompson writes in with regard to

00:15:03   trillions of operations per second measurements or TOPS measurements. Michael writes, "I found

00:15:08   this article on the Qualcomm website that suggests that the TOPS measurement they use for their NPU

00:15:12   performance is based on 8-bit integers. In the paragraph headed 'precision,' they state,

00:15:16   'The current industry standard for measuring AI inference and TOPS is at 8 precision.' The context

00:15:25   here being whether or not the new surface line and the... What is it? The copilot plus PC or whatever

00:15:31   it is? Line of PCs. Are they or are they not actually faster for neural related things than

00:15:38   Apple stuff? And so I don't recall what was... Does this mean they are faster than I presume?

00:15:45   Well, I mean, so this matches what we saw. The idea that on this website they're saying Qualcomm

00:15:49   is saying the industry standard is 8-bit precision, right? And what we've seen is Apple is using

00:15:55   int8 and so is Qualcomm and so is Microsoft and everybody who's current talking about their

00:15:59   current line of products, they're all using int8 precision. And again, this is saying,

00:16:03   how many 8-bit things can you process at once, right? If I ask you how many 16-bit things you

00:16:09   can process, it's half as many, right? And so the number is smaller. Previously, unlike earlier,

00:16:15   you know, Apple Silicon stuff, Apple was using 16-bit stuff. And so their numbers were half as

00:16:19   big, but with their new stuff, Apple is using numbers that are twice as big and they're using

00:16:24   int8 and so is Qualcomm. So the answer is everyone's using int8 now. Is that because

00:16:29   int8 is more representative of the actual jobs we're asking our NPUs to do? Maybe, but you know,

00:16:34   whatever. The industry has decided when measuring tops, we're going to use int8 precision. That may

00:16:40   become less relevant if it turns out that the things we ask our NPUs to do involve 16-bit or

00:16:47   32-bit values, and it doesn't really matter how fast they can do stuff on int8 things,

00:16:50   but I would trust that int8 is actually a relevant measure right now. So the answer is,

00:16:54   you know, the Copilot Plus PCs and the Snapdragon X elite thing has 40 tops. The M4 has 38. Those

00:17:03   are both int8 measures. That means they're essentially comparable. Eric Jacobson writes

00:17:08   in with regards to iCloud Drive and node modules. So if you recall, this was with John Sun,

00:17:16   who basically nuked his MacBook Air by trying to sync the node modules folder through iCloud

00:17:22   Drive. Eric writes, "I haven't used it since I don't use iCloud Drive, but there's a project

00:17:26   that will add a no sync directive to every node modules on a file system. I imagine it might need

00:17:32   to be rerun whenever a new project is kicked off." And we'll put a link in the show notes to

00:17:35   nosync-icloud. Yeah, and I tried to look at the code to remind myself how I think you make like

00:17:40   a directory with a dot nosync extension that has the same name as the other one. Like that's the

00:17:44   way you signal to iCloud Drive not to sync the directory or something like that. This is a node

00:17:49   module itself, so you can look at the code, but unfortunately the documentation is all in,

00:17:52   what are we going to say here? I'm going to say Chinese? Yeah, something like that. Yeah,

00:17:56   so the documentation is in Chinese and I can't read it. But the source code is not in Chinese

00:18:00   and I still couldn't quite make any sales of it. But yeah, I think it's just making dot nosync

00:18:05   directories in the right places. And it's a node module that you can use and it will, you know,

00:18:08   I think you just include it in your project and make sure everything nosyncs. So that's,

00:18:12   you know, useful and helpful if you want to dare to walk that tightrope of trying to use node modules

00:18:17   with iCloud Drive. Indeed, and then Eric continues, I do however use time machine and can attest that

00:18:22   the Asimov utility works perfectly for excluding node modules and other dependency directories.

00:18:27   Also it is a background service, so it doesn't need to be reinitialized. And we will again put

00:18:31   a link to the show notes to Asimov. And also, John, I guess you wanted to call attention to the

00:18:36   list of excluded directories, which I put in the show notes as well. Yeah, it shows what kind of

00:18:40   things like when it says dependencies, what does that mean? Obviously it means node modules, but

00:18:43   it has a whole list of all the different things that excludes things from Gradle, Bower, PyPy,

00:18:50   NPM, Parcel, Cargo, Maven. I think CocoaPods might be, yeah, CocoaPods is in there. Like Marco,

00:18:55   have you heard of any of those things other than CocoaPods? Zero, precisely zero. Flutter. Anyway,

00:19:01   it's sad that I think I've heard of all of these. But I installed this and I ran it. It installs a

00:19:08   little launch daemon thing or whatever. And it essentially does the... I forget what the time

00:19:15   machine is. I think it's an extended attribute or something, but there's a way to exclude things

00:19:19   from time machine. Maybe it just calls tmutil. Yeah. Anyway, excluding all these directories

00:19:24   from your time machine backups can make your time machine backups go faster. What it's basically

00:19:28   saying is you don't need to back up the dependencies of your code. If you're writing something in node

00:19:34   and you use 17 node modules, you don't need to back those up. You get them from the internet

00:19:37   anyway. You got them through NPM or YARMM or whatever. They're on the internet. Do not back

00:19:41   them up. That's not your code. It's a dependency. You didn't write that code. It's just pulling it

00:19:45   in. And there are tons of files. So if you can exclude those directories from time machine,

00:19:49   it will make your time machine backups go faster. But who remembers, oh, what am I going to do? Go

00:19:53   to options and time machine and drag the little thing in or set the extended attribute. I don't

00:19:57   remember how to do this. This just runs in the background all the time. Looks for directories

00:20:01   that fit this signature and excludes them from time machine. So I did that. I probably saved,

00:20:06   I don't know, thousands, many thousands of files are no longer in my time machine backups

00:20:11   because I ran this. I hope it doesn't have bugs and isn't excluding a whole bunch of

00:20:15   important files from my time machine backups. But I've got multiple backup strategies. So for now,

00:20:20   I'm trying the experiment of running this asmoff daemon in the background to see if it helps with

00:20:24   my time machine backups. And I'm still not running iCloud drive, of course. Of course.

00:20:28   - And then finally, this is actually, I should have moved this up by the other Vision Pro follow-up.

00:20:33   Didn't think about it. But anyway, Jonathan Gobranson writes with regard to audio routing

00:20:37   during Vision Pro guest mode. So if you recall, I was doing demonstrations for my mom and dad.

00:20:42   And I noticed that when mom was on Mount Hood or whatever it's called, and I had her go fully

00:20:47   immersive, that the crickets and whatnot were being routed through my iPad Pro, which was doing

00:20:54   mirroring at the time. And so it doesn't really make for a very good effect if the audio was going

00:20:58   through there. So Jonathan writes, "You can choose during setup of each guest user session whether to

00:21:04   route audio to the Vision Pro or the iPad or whatever the case may be if you choose to mirror

00:21:09   content." And we'll put a link to the Knowledge Base article. So what you do is you look up and

00:21:15   you get the little green down chevron near the top of your view. Then you go into Control Center.

00:21:20   Then you go back into the Mirror My View button. And then in there, there's a audio routing section

00:21:27   that you can choose to push everything back onto the Vision Pro. Not entirely sure why this isn't

00:21:32   the default, to be honest, because pretty much every time I've always wanted this. But here we

00:21:36   are. At least now I know that there's a way around it. So good deal. We are sponsored this episode by

00:21:43   Fastmail, not only a great email host, but my chosen email host since 2007. Obviously, long

00:21:49   before they were sponsored, long before I was a podcaster, I've been a Fastmail customer and I've

00:21:53   been very happy because hosting email, it's constantly a moving target. You don't want to

00:21:58   do it yourself. You don't want to also use one of the big services necessarily because then your

00:22:03   name's tied to them. I like having my own control. I like being able to have standardized hosting

00:22:09   with a domain that I own. Fastmail is a great place to do this. They even now have domain sales.

00:22:15   You can add or buy domains right through them and they set up all the records for you and everything

00:22:18   so it works immediately and painlessly. But it's just a great email host. Fastmail has all the

00:22:24   features you want under an email host. Of course, obviously it's IMAP hosting and they have all the

00:22:29   basic default stuff that works with every email client. They also have all the modern productivity

00:22:33   features you might want. Things like scheduled send, snooze, folders, labels, search, etc.

00:22:38   They have all the great plans, all the great features. I love Fastmail. They now also have

00:22:42   multi-user plans. So if you want to have your whole household have a plan, they have a duo plan

00:22:47   for secure private email for you and a partner at a reduced price. And for just a little bit more,

00:22:51   they have a family plan for up to six people. You can upgrade your entire family. And of course,

00:22:55   in those family and duo plans, you have things like shared domains if you want. You can have shared

00:23:00   or private calendars and address books. You can have like a family calendar all shared,

00:23:04   hosted by Fastmail, all based on standards. It is wonderful. So I strongly recommend Fastmail

00:23:09   and they are super private too. For over 20 years, Fastmail has been a leader in email privacy.

00:23:15   They believe in working for you as people to be cared for, not data and products to be exploited.

00:23:21   And Fastmail has better spam filters than everyone else and absolutely no ads. So it's super easy to

00:23:27   get started with Fastmail today. You can download your data from wherever you are now. You can

00:23:30   import it into your new Fastmail inbox. You don't need to leave anything behind when you switch.

00:23:35   It's great. Try Fastmail free for 30 days and get 10% off your first year at fastmail.com/atp.

00:23:42   Once again, free for 30 days, 10% off your first year, fastmail.com/atp. Thank you so much to

00:23:48   Fastmail for being an awesome email host for all these years and for sponsoring our show.

00:23:52   It is quickly approaching WWDC time, which means I'm going to be seeing you too soon. But

00:24:00   nevertheless, we should talk about some last second predictions. And I guess this is most

00:24:07   predominantly from Mark Gurman in today's episode. So I'm going to read a whole bunch of stuff. One

00:24:11   of you guys feel free to pipe in and interrupt at your convenience. But here we go. Mark Gurman

00:24:17   writes, "Apple is preparing to spend a good portion of its worldwide developers conference

00:24:22   laying out its AI-related features. At the heart of the new strategy is Project Gray Matter,

00:24:28   a set of AI tools that the company will integrate into core apps like Safari, Photos, and Notes.

00:24:33   The push also includes operating system features such as enhanced notifications.

00:24:37   Much of the processing for less computing-intensive AI features will run entirely on the

00:24:42   device. But if a feature requires more horsepower, the work will be pushed to the cloud. There are

00:24:46   several new capabilities in the works for this year, including ones that transcribe voice memos,

00:24:50   retouch photos with AI, and make searches faster and more reliable in the spotlight feature.

00:24:55   Faster would be great, particularly on my iPad, please, and thank you. They also will improve

00:24:58   Safari web search and automatically suggest replies to emails and text messages. The Siri

00:25:03   personal assistant will get an upgrade as well, with more natural-sounding interactions based on

00:25:07   Apple's own large language models. There's also a more advanced Siri coming to Apple Watch for

00:25:11   on-the-go tasks. Developer tools, including Xcode, are getting AI enhancements too."

00:25:16   Let's stop here for a second and look at this list of features because we're always like,

00:25:19   "How will Apple add AI sauce to all this stuff? What things will they talk about?"

00:25:22   There was a big story, Gurman had it, and I think we might have mentioned on the show,

00:25:27   like, "Oh, they're going to fix Siri." We were speculating months ago or weeks ago, whatever.

00:25:32   "Is this the year that they're going to fix Siri with AI, or are they just going to add it to a

00:25:34   bunch of other stuff?" And Gurman's rumor was like, "No, they're doing a Siri thing." So we

00:25:37   can expect to see that. But here's some specifics, and the specifics seem not weird. I don't know.

00:25:45   Sometimes these rumors aren't comprehensive. Very often, Apple emphasizes one or two particular

00:25:50   things, whereas we just get a laundry list and we don't know which one they're really

00:25:52   going to concentrate on, and it's going to be impressive. But let's look at some of these in

00:25:55   turn and see how exciting they are. Transcribing voice memos. Apple's been doing transcriptions,

00:26:01   for example, on voicemail for a long time now. Having transcription be better, that's good.

00:26:08   Probably not going to really radically change people's lives since it's an existing feature,

00:26:12   they're just making it a little bit better. Retouch photos with AI could mean a lot of

00:26:15   things. That could be a headlining feature. That short description doesn't tell us, "Is this going

00:26:20   to just be like, 'Oh, they're better at background detection for tearing people off,' or is it going

00:26:24   to be like, 'Oh, they're going to really emphasize that they're doing the feature that Google and

00:26:27   many others have, where if there's something in the background, a photo you don't want,

00:26:31   you can remove it, remove a person from the background who shouldn't be there, remove a

00:26:36   sign or a tree or something like that.' That could be a big headlining feature. Google has certainly

00:26:41   had a whole ad campaign about theirs, and obviously it's every graphics application from

00:26:46   Photoshop to Lightroom, Pixelmator and everything has features like this, and they've been touting

00:26:50   them, and I think they're crowd-pleasing. - Well, but that's part of the problem. So far,

00:26:53   voicemail transcription and fancy AI photo cleanup and removal, that's table stakes today.

00:26:59   I'm glad Apple's getting there, they should, but neither of those I think is likely to make a big

00:27:07   splash simply because other people have been doing that for a decent amount of time, and the rest of

00:27:14   the industry is there. What I want to see from Apple is what we get excited about from Apple.

00:27:21   I want them to blow us away with stuff that we haven't seen from everyone else. I want to see

00:27:25   features that are not just playing catch-up with the rest of the industry. I want to see Apple

00:27:29   leading the way, not following, and these are just following so far. - Yeah, I mean, well, as with

00:27:34   all these things, we know that this has been done elsewhere and even on Apple's own platforms in

00:27:39   various applications, but when people get one of these new phones or they update the OS or they

00:27:45   see an ad on TV that Apple touts this feature of erasing a person, chances are good that they've

00:27:50   never seen that before if they're not a tech nerd, right? And so as far as they're concerned, wow,

00:27:54   this is amazing. Especially if they don't know it's even possible, like they're saying, "Oh,

00:27:58   on my phone I could just tap a person and remove them?" And we all know that that's been around for

00:28:02   a long time and they're playing catch-up, but A, they do have to play catch-up. We don't want them

00:28:06   to not do this. And B, it can be just as impressive. It can be very impressive to people who haven't

00:28:10   seen it before. So I just hope they do a good job of this because there is a lot of competition,

00:28:14   again, even on Apple's own platforms in applications that you can get for the Mac and for

00:28:20   the iPad and for the iPhone that already do the same job. I hope Apple at least matches them.

00:28:25   Historically, Apple has kind of been... They want to make it simple, so there's not a lot of

00:28:30   adjustability. And if it does a bad job, the Apple way is like, "Well, just tap on the person and

00:28:35   remove them." And you'll tap on a person and it'll make a terrible mess of it. And you'll be like,

00:28:39   "Is there something I can do to try that again and do better?" And the Apple way is no.

00:28:42   Nope.

00:28:43   You tap them and if it does a good job, good. And if it doesn't, oh, well. Whereas in Lightroom,

00:28:48   there's a million knobs you can adjust and you can take a second attempt and you can mask

00:28:51   differently. You do all sorts of things at a fancier application. So there's that. I really

00:28:57   hope they do catch up and I hope they do a good job of it. The next item about improving Safari

00:29:03   web search. How? I get. All right.

00:29:09   Well, I mean, for whatever it's worth... Obviously, we don't know what this means yet. We might not

00:29:14   even learn it in two weeks. But Safari had... Whatever Apple is doing, whatever their back-end

00:29:21   logic is for autocomplete suggestions in Safari is basically a search engine. I can't even imagine

00:29:29   how many millions and millions of web searches Apple avoids even making on the iPhone by people

00:29:36   just tapping that first autocomplete result that comes up. So maybe it's just related to whatever

00:29:42   the back-end is of that. And that could be one of those invisible things that everybody will take

00:29:48   for granted and nobody will even notice, but we will appreciate it getting better. Like things

00:29:52   like autocorrect. Well, notice if they screw it up, they've changed autocomplete many, many times

00:29:56   over the years and when they screw it up, we all notice and we don't like it. So I really hope

00:30:00   it doesn't start suggesting ridiculous things. I think that might all be local too. So maybe that

00:30:05   is a place for LLMs to try to do better predictions. But anytime they're messing with an existing

00:30:10   feature, you hope it's gonna be a big improvement, but there's the possibility, especially when you're

00:30:14   throwing LLMs into the mix, it's a possibility. If the current strategy is like, look at your past

00:30:19   history, look at the number one search results, they could even be hitting the Google backend

00:30:23   for some of those things. It's a very straightforward non-complicated algorithm,

00:30:26   but it's deterministic. And if they switch from that to like, let's just chuck it over the wall

00:30:31   to the LM and see what it says. I'm concerned that there might be some wackiness there,

00:30:36   but we'll see. More reliable spotlight features. Similarly, when you use spotlight on the phone,

00:30:42   I don't know if it's using spotlight, spotlight, if it's the same technology as the runs on the

00:30:46   Mac, but it's like, oh, I'm searching for stuff on my phone and it includes contacts and all your

00:30:51   applications and files and things you've done recently. And yeah, you could probably throw

00:30:56   LLMs into the mix there to handle, basically like, so you don't have to type things exactly,

00:31:00   and it's better about synonyms and you can type in vague things, like type in some expression about

00:31:05   what you want to do in settings and have the LM figure out the setting you want to go to.

00:31:09   Potential for good, also potential for messing up spotlight on the phone.

00:31:13   And then suggest replies to emails and text messages. This starts to get into the area

00:31:18   where we thought like, will Apple go there? We haven't gotten to like chat bots yet, but

00:31:22   obviously everyone else is doing this, so it's another catch up feature.

00:31:26   But the idea that Apple would within the mail application or while sending text messages pop

00:31:31   up a little clippy and say, it looks like you're trying to reply to your friend. You want me to

00:31:34   write it for you? Which is what, I mean, just look at the Gmail interface, for example, every time

00:31:38   I'm in Gmail, it's like, do you want me to summarize this email? Do you want me to just

00:31:42   write the reply for you? Like Google's been pushing on this for years. It used to just be like,

00:31:46   here's a canned phrase that is probably a reasonable reply to this email. And now it's

00:31:49   like, you know what? I can just write the whole email for you. Like you don't have to click from

00:31:53   canned phrases. You don't have to, you know, the auto-complete that happens in like Google docs

00:31:57   and everything. It's just like, let me just write the whole thing for you. And that, we should find

00:32:02   that link to Nevin Mergen's blog post. Nevin Mergen said he got an email from a friend that was

00:32:06   written by an AI and he kind of flipped out about it, right? That seems like a less,

00:32:12   that is not a conservative move to suggest to people that the phone will write their email or

00:32:19   their text message for them. Summarizing, I can see it's like the phone is helping me. If I just

00:32:23   want a summary, I don't want to read it all, let the phone summarize it. It's like, you know,

00:32:27   asking the phone to help you deal with your stuff. But having generative AI write something that you

00:32:33   will send as if you wrote it yourself is a step farther than Apple has gone. So we'll see how they

00:32:41   present that. Again, obviously everyone else is doing this. It's not new. I'm not saying Apple's

00:32:44   going to be the first one to do this and they're going to get in big trouble or something. It just

00:32:47   doesn't, it's not a very safe thing to do. So I'm wondering how daring Apple will be in like

00:32:53   their little presentation. Maybe they'll just be like, and it can even suggest a reply for you. And

00:32:57   some happy person who, happy text will tap a thing on their phone screen. And they'll be so delighted

00:33:03   to see the three word reply come and they'll hit the send button and they won't talk about it

00:33:07   anymore. But I don't know, it wigs me out a little bit too. - I mean, keep in mind, like we're probably

00:33:13   not that far away from, I mean, people are probably doing it now. Where like you're using your AI to

00:33:20   compose an email that will be then read by the recipients' AI. And we will all just be sending

00:33:26   even more garbage crap to each other and having it be processed by even more garbage AI junk on

00:33:32   the other end. And we will finally reveal email to be the useless thing that it has been most of

00:33:36   this time. - It's not useless. And I feel like in a business setting, this is most useful because

00:33:40   there's a lot of boilerplate and niceties in business that would help. But like we already

00:33:44   have auto, like I mentioned autocomplete, like Gmail forever has had a thing where you start

00:33:47   writing something and it guesses what you're probably going to say for these next three words

00:33:51   and you can autocomplete that. But I feel like that kind of piecemeal autocomplete, even if it

00:33:56   is LLM powered, that piecemeal autocomplete at least forces you to read everything. Whereas having

00:34:01   a button that says, "Compose a reply for me," invites the possibility that you will not read

00:34:06   what it generated because that will take too much of your time and you'll just hit send.

00:34:10   And now we're sending messages that we're not even looking at to give a thumbs up or a thumbs down on

00:34:14   because it just takes too much time. And that's just, I think it's a waste of people's time

00:34:17   because if you don't read what it wrote, maybe it didn't write what you wanted it to write. And

00:34:20   now you have an even more confusing email thread. And I mean, again, we'll figure this out as a

00:34:25   culture, as the technology advances, but I'm just saying character for Apple, for Apple's thus far

00:34:30   very conservative approach to AI, suggesting replies to emails and text messages seems like

00:34:36   a big move for this particular company. Yeah, agreed. Continuing from Mark Gurman,

00:34:43   "One standout feature will bring generative AI to emoji. The company is developing software that

00:34:49   can create custom emoji on the fly based on what users are texting. That means you'll suddenly have

00:34:53   an all new emoji for any occasion beyond the catalog of options that Apple currently offers

00:34:58   on the iPhone and other devices. Another fun improvement unrelated to AI will be the revamped

00:35:03   iPhone home screen that will let users change the color of app icons and put them wherever they want.

00:35:07   For instance, you can make all your social icons blue or finance related ones green,

00:35:10   and they won't need to be placed in the standard grid that has existed since day one in 2007."

00:35:15   All right. So that's a lot of stuff here. The generative AI emojis, obviously they're not

00:35:19   emojis because emojis are like Unicode things. There's a defined set of them. You can't just

00:35:22   make up new ones. I mean, you can't if you're the Unicode consortium, but Apple can't, right? So

00:35:26   what it's actually saying is kind of like on Slack and Discord and all these other things where you

00:35:31   can sort of generate custom stickers or I think they call them custom emojis or custom reactions

00:35:35   or whatever. The point is, this is going to be a new graphic that they'll generate on the fly for

00:35:39   you that will be sent as an image because they can't just send it as a Unicode code point because

00:35:44   these, they're not dynamically adding things to Unicode, right? That's not how this works.

00:35:47   So they're kind of generating stickers for you. And I guess they're going to like, you know,

00:35:53   they have a bunch of like Mr. Potato Head style building parts they feed into the LLM and then it

00:35:57   can do like sentiment analysis to figure out what kind of emoji we'd want. I have to say, like,

00:36:01   there's a lot of emoji, especially for things like face reactions and stuff like that.

00:36:05   I guess there's ones that aren't there. Like you never try to summon someone emoji and you're

00:36:09   shocked that there's not like a watermelon emoji or something. I can't think of one that doesn't

00:36:13   exist. Sorry. But frequently I go into emoji and I'm like, "Oh, surely there's an emoji for this."

00:36:18   And like, there's no ice cream cone. Like, again, I don't know if that's real or not, but like I'm

00:36:21   shocked by what's not there. So that's nice that it'll generate to it, but it'll have to send it as

00:36:25   an image because it can't send it in the other way. And I do wonder, this is another place where

00:36:30   like, I feel like Apple is taking a risk, even with a very, very limited image generator that's

00:36:35   trained on their own source, you know, trained on all their own source for emojis that Apple makes

00:36:39   and can generate new ones based on that and some other information. There's the potential to

00:36:45   generate a little image that maybe is embarrassing to Apple as a company. You know what I'm saying?

00:36:49   So, and there's a potential to generate images that just look ugly or not up to Apple standards.

00:36:56   You know what I mean? Like, this is, that's a weird thing to be touting. And the coloring icons,

00:37:01   that's, you know, another, I'm not going to say, I keep wanting to say a bridge too far. It's not

00:37:05   a bridge too far, but it is a bridge. Like you're going to take developers icons and draw all over

00:37:11   them. They've done this before. They used to add the shine to your icon if you didn't opt out. Do

00:37:14   you remember that? Like the gloss? But now they're like, we're going to change the color of your icon?

00:37:20   Like, I'm sure these companies that are like so proud of their branding and change their icons or

00:37:24   whatever, like that now the user will be able to say, you know what? I want Instagram to be green.

00:37:29   And the OS will be like, sure, I'll take that Instagram icon and I'll make it green. We can

00:37:33   do that. That's weird. I mean, you can make little color code things on your home screen, I suppose.

00:37:38   But yeah, I can't imagine that's the whole story to that one. I mean, first of all,

00:37:42   like a lot of companies I think that have a problem with just like trademark issues with like,

00:37:46   what if you color my icon green? First of all, it's not my brand identity. Second of all,

00:37:51   it might look like my competitors or it might infringe someone else's trademark or something.

00:37:55   Like I could see there's, I could see like big companies having a big problem with that. But

00:37:58   users are doing it. And by the way, there is a precedent for this in Mac OS back in the day,

00:38:03   someone else will look up what version of Mac OS or the operating system for Macs used to do this.

00:38:08   When you added a label to something, you know, you can do labels in Mac OS 10 or in classic Mac OS,

00:38:12   you could add labels in one version of Mac OS or system six or whatever it was called back in it,

00:38:18   whichever version I'm thinking of. I think it might've even been system seven. When you apply

00:38:22   to label it, what color the icon like, so if you probably green label, it would like green tint the

00:38:28   icon by basically making the black pixels green or the black pixels red. I think it did not look

00:38:32   good. It was not a good look that didn't last for a very long time. Certainly the Mac OS 10 era that

00:38:38   has never been the case, but yeah, the user deciding to color icon, like that was not on my

00:38:45   list of things to think of they were to be doing. And I know this is like unrelated to AI, but let's

00:38:48   just say iOS 18 change next week, we'll probably have many more previews of WWDC features, but

00:38:53   what an odd, what an odd thing to do. What is this? Have people been clamoring for this?

00:38:59   I want to change the color of my icons. I mean, changing icons. They added that while ago where

00:39:04   you can pick from different icons and blah, blah, blah, but even that, remember how conservatively

00:39:07   they added that? Like it has to pop up a dialogue in people's faces so you know when it's happening

00:39:11   so apps can't masquerade as other apps. - I literally just implemented that in the

00:39:15   rewrite yesterday. (laughs) - Yeah.

00:39:17   - No, but I mean, there's clearly, there's a very large market out there. We've talked about it

00:39:23   before. Like there's a very large market out there for these home screen customization apps that use

00:39:27   all sorts of tricks and hacks and frankly kind of bad hacks to get people to be able to customize

00:39:33   icons for their apps and make like their aesthetic home screens that they want. That's a huge demand,

00:39:38   huge market. I think Apple wants to try to address some of that demand without having all these huge

00:39:43   hacks if this is indeed what they're doing here. The problem is unless you can have arbitrary icons

00:39:49   for apps, like specify your own image as the user. - As you can with shortcuts.

00:39:54   - Right, like if you can do that, then all these apps that sell like these, 'cause you know,

00:39:59   they sell like theme packs. Like people aren't just drawing it themselves. Like they're using

00:40:03   apps that have like cool theme packs of here's like a bunch of popular icons for all the popular

00:40:09   apps that you know you have like Instagram. Plus also here's a bunch of like random ones you can

00:40:13   apply yourself to whatever else you need them to be, you know, like kind of generic ones.

00:40:17   That's a huge market and so if we're just talking about like you can tint it seven different colors,

00:40:23   that's not gonna get rid of this market at all and you know, some people will use it. That'll

00:40:28   be great but I can't imagine that being worth the hassle and the trouble that they might get

00:40:32   in with other companies. So I have a feeling this is like, you know, sometimes a rumor comes out

00:40:37   and we're just kind of like huh? And then you know, whenever it's actually announced, we see,

00:40:42   oh there's important detail X, Y, and Z here that explain it better and that now it makes more sense.

00:40:48   I think we're missing those details right now on this feature. I think there's something else here

00:40:52   to explain this that we're not getting it. - I mean, it might just be like you mentioned

00:40:56   those icon packs or whatever. Part of the limitations of customizing your home screen

00:40:59   using shortcuts to replace the app icons, first of all, it's super cumbersome as is doing anything

00:41:03   in home screen because you have to hide the existing app and make the shortcut to the app.

00:41:06   You know, it's just, it's a hassle and second, if you get one of those icon packs, maybe it doesn't

00:41:09   have an icon for the app that you want to change but if the OS says, okay, well if you're making

00:41:14   a little region of green icons and you can use icons from the Steam pack, you got, oh but there's

00:41:18   no green icon for this app, well just take the actual icon for the app and tint it green and

00:41:22   now it fits in. So maybe it's like, you know, a catchall for if there's, if, because people

00:41:27   aren't going to draw their own custom icons and if they can't find the one in a pre-built icon pack,

00:41:32   now you can just at least make the icons match. I don't quite understand the idea to color code

00:41:38   everything but there's that and yet the other part of this is being able to, you know, this does not

00:41:42   put them in the standard grid. It's basically just to be able to leave space. You don't have to make

00:41:45   spacer, clear spacer icons anymore. I'm assuming that's what they're talking about here is like,

00:41:49   now you can arrange things on the home screen and leave gaps if you want to. I still think doing

00:41:56   anything on the home screen on the iPhone is incredibly painful. We've talked about this

00:41:59   for years and years. I think it will continue to be painful. I really wish there was a nice interface

00:42:05   where you could mess around with this stuff and do it without destroying anything and either commit

00:42:10   or roll back all your changes as opposed to the current system which is just pure chaos and it's

00:42:14   like the hardest game on the entire operating system to try to rearrange icons on your phone

00:42:19   without screwing everything up and just throwing your hands up and saying forget it, I'm hiding

00:42:22   everything into the app library. Yeah, I wonder, I feel like I heard this from Jason maybe. I heard

00:42:30   it somewhere. Maybe it was Mike on upgrade. I forget where I heard it but somebody pointed out,

00:42:33   well maybe there will be like layers and that makes me think of, you know, how that was upgraded

00:42:39   in SF symbols. You can have like different layers and you can color them and you know in different

00:42:43   ways and whatnot. I think that makes some amount of sense but that's not going to fix like the

00:42:49   instagrams of the world so I'm not sure or maybe it'll be opt-in for developers but again I don't

00:42:54   see. VisionOS has this right? Don't they have layers in their icons? Yes, they actually kind of,

00:42:58   I think they require it with VisionOS. They made that a new like hey for iOS 18 if you provide

00:43:03   a layered icon, our new coloration API will respect that like as you sound like SF symbols

00:43:10   where we can colorize parts of it. We'll see. It just seems to me like a weird place to be

00:43:15   spending resources. I think the arrangement part is a smart feature. They should have that. People

00:43:20   have wanted that for a long time. People have been doing it with spacer icons. It's annoying.

00:43:23   The place where they really should have spent resources is how hard it is to rearrange the

00:43:26   home screen. We'll see if they do that. Yeah, that's the only way I really wish I could still

00:43:31   interact with my phone via iTunes which I know is now finder but that was so much better was,

00:43:36   you know, doing home screen rearrangement. And even that was terrible but it was better.

00:43:39   Yeah, it was terrible but it was way better. All right, finishing up with Mark Gurman. A big part

00:43:44   of the effort is creating smart recaps. The technology will be able to provide users with

00:43:48   summaries of their missed notifications and individual text messages as well as web pages,

00:43:53   news articles, documents, notes and other forms of media. There's also no Apple design chatbot,

00:43:57   at least not yet. That means the company won't be competing in the highest profile area of AI

00:44:02   and the version that Apple has been developing itself is simply not up to snuff. The company's

00:44:07   held talks with both Google and OpenAI about integrating their chatbots into iOS 18. Apple

00:44:12   ultimately sealed the deal with OpenAI and their partnership will be a component of the WWDC

00:44:17   announcement. A couple of quick thoughts here. First of all, I think we all are in a group chat

00:44:21   that oftentimes will just pop off at a time that you're not paying attention and so it is not

00:44:26   uncommon. Particularly, I find that I'm in a group chat with a couple of guys that are constantly

00:44:32   looking at cars that they may or may not ever buy. One of them in particular is obsessed with finding

00:44:38   a affordable 911 and those are mutually exclusive terms. You cannot find a 911 that is affordable,

00:44:43   that's not a pile of garbage. And so anyways, I will often come back to my phone or computer,

00:44:48   what have you, with literally 40 or 50 messages, most of which I don't particularly care about

00:44:52   because 911 is not really my cup of tea. And so if there was a way to summarize that, I'm all in,

00:44:57   sign me up. And then additionally, I mean here it is that Germin is saying OpenAI is the winner and

00:45:03   like we discussed I think this past week, that's going to be a little dicey. So I wonder how they

00:45:08   spin this. Well, in typical Germin fashion, there's also no Apple designed chatbot, at least not yet.

00:45:14   Apple designed? Do they mean Apple's in-house LLM that they've been working on? Is that what

00:45:18   he's saying there won't be any of? Because I bet whatever, like they say they did a deal with OpenAI

00:45:23   but he doesn't say, okay, well what is that deal, how is that deal going to manifest? Is it going

00:45:26   to manifest by iOS 18 having a chatbot that is powered by OpenAI? And will that not be Apple

00:45:30   designed or will it be OpenAI designed? Will it be OpenAI branded? These are all, I guess,

00:45:35   questions that he doesn't have the answers to yet. But yeah, the news is they did a deal with OpenAI.

00:45:40   There's also news this week that Apple is still talking to Google, obviously. So we'll see if the

00:45:49   words OpenAI appear at WWC, if they're going to announce this as a partnership, they're just going

00:45:53   to use them on the back end. Like I said last week, maybe they've been building features using

00:45:57   the existing OpenAI API and now they just ink this deal so now they can officially do it in a released

00:46:02   OS instead of just doing it in dev builds. But I'm just not quite sure how this is going to manifest

00:46:06   because this paragraph begins making it seem like, oh, there's going to be no chatbot, but they have

00:46:10   licensed the OpenAI chatbot. Well, is there going to be a chatbot or not? Is there going to be part

00:46:15   of iOS where you can start typing to an LLM and getting answers in the chat GPT style or will there

00:46:22   be an OpenAI app or will it be integrated into Siri? Questions we do not have answers for yet,

00:46:27   but that is the capper on all of this stuff. A bunch of strange features, which I mean,

00:46:32   they all seem plausible to me, but this is not the list I would have come up with about how to add

00:46:36   AI sauce to iOS 18. And at the very bottom is, you know, OpenAI something, something, something OpenAI.

00:46:42   Yep. So yeah, we'll see what happens. As I think John said a moment ago, typically, you know,

00:46:47   in the couple of days leading up to WWDC, we oftentimes, or maybe the week before we hear some

00:46:53   last minute leaks and so on, I was going to say announcements, but they're not announcements.

00:46:57   So we'll see what happens. But I am very much looking forward to sitting next to you fine

00:47:02   gentlemen and drinking this all in while at Apple Park, hopefully not getting a sunburn.

00:47:06   Well, there was, did you see there was one kind of last minute thing that came out today also

00:47:10   about Siri being used to script apps, but not yet and only Apple apps?

00:47:16   Yeah, yeah. Yeah, it was basically the gist of this rumor, I think it was also Gurman,

00:47:20   was just in a separate report that Apple will also preview Siri being able to like take

00:47:28   more complex actions within apps, kind of what we've been talking about. You know, we've been

00:47:32   talking, we've been speculating about how wouldn't it be great if, you know, apps could expose,

00:47:38   you know, kind of stock actions similar to the intent system and just kind of describe what they

00:47:42   are in some kind of language and then have an LM based, you know, Siri interface be able to analyze

00:47:49   what the app is telling the system, here's how you can do it and here's the actions to call to do it,

00:47:54   and be able to offer to users the ability to perform actions like that via voice commands that

00:47:59   are not necessarily phrased in exactly the right way, that were not set up in advanced

00:48:02   as shortcuts and things like that. The rumor that came out today from Gurman is basically

00:48:07   exactly that is happening, but not yet. That basically that it is, that exactly that kind

00:48:14   of thing will be previewed, it won't actually ship until a point release probably, they said next

00:48:19   year, so 2025, so presumably in the first half of next year, you know, iOS 18.2 or three or four or

00:48:25   five or whatever, and that it would only work in Apple's apps to start. So that tells me not an API

00:48:32   yet. Or an API that is released immediately, you know, or released at WWDC, but obviously nobody's

00:48:39   going to be able to implement it until at the earliest of the fall, and it may not be an instant

00:48:43   adoption. That's interesting, you know, I wonder if maybe we will see, you know, because the current

00:48:50   system, the current API to do all this is called App Intents. This is, they launched just a couple

00:48:55   of years ago, and it's like the new Swift based method of intents, which I've mentioned before,

00:49:01   is way better for developers and way simpler and way easier to implement than the old system of

00:49:07   these weird Xcode files that had custom configuration that would generate files,

00:49:12   and the generation process would always break and cause weird bugs, and then you'd have this

00:49:16   extension that processed it and would have to communicate to your app via like weird signaling

00:49:20   mechanisms for extensions to apps, like it was a whole mess before, and they've redone that system

00:49:24   two or three times. I wonder if, you know, sometimes when Apple is kind of going in a

00:49:30   direction for a future update for something, sometimes the APIs will launch for it that

00:49:37   don't come out and explicitly say this is what this is for, but kind of lay the groundwork for

00:49:42   it. So for instance, auto layout coming out, this is the famous example, auto layout coming out

00:49:46   before the iPhone 6 came out with the first like really new screen size. You know, not kind of the

00:49:51   iPhone 5 of course, that was just adding a table row. Anyway, so that's, I wonder if what we're

00:49:57   gonna see this summer is maybe they will add something to the App Intents API to give us kind

00:50:04   of more ways to describe to the system what these actions do, and maybe hide them from the user,

00:50:12   but make them available to this indexing system that presumably like the new Siri would have.

00:50:16   Maybe we'll see something like that. So I think this, you know, if for some reason they like demo

00:50:23   this and don't tell us how to use an API for our apps in the future, maybe we'll be able to see

00:50:27   hints in like what's new in App Intents this year. So let's keep an eye on that.

00:50:32   - It sounds more like they, like this is considering their own apps first. They're not

00:50:37   quite sure what that API should look like so they can experiment with their own apps,

00:50:41   'cause if they change their minds, they'll just rewrite their own apps to fit with it,

00:50:44   whatever. And once they get it worked out, 'cause it's very often, especially in the old days of

00:50:47   Mac OS X, they would very often roll out entire new frameworks. They would be private frameworks

00:50:52   that only Apple's apps could use, and they would work out the kinks. Is this the right API?

00:50:56   Because they could change the API all they want, 'cause all they'd be breaking is their own apps,

00:50:59   'cause no one else should be using this private framework. And once they finally got the framework

00:51:02   where it, you know, looking like they wanted it to, then the next year they roll out essentially

00:51:08   the same framework under a new name, but now it's a public framework. It kind of sounds like

00:51:11   what they're doing here, that they're not at the point, like especially if they're kind of like

00:51:14   rushed to catch up, they're not at the point where they're ready to even tell, perhaps even to tell

00:51:18   developers, you know, in the size class type of way, do this thing, and then next year you'll be

00:51:22   folded into the system, because it seems like they don't even know what that would be yet,

00:51:26   like how to indicate, like they're working that out themselves, but they do want to get something

00:51:29   out there, and you can demo it if they do it on their own apps, right? They say, "Coming in spring,"

00:51:33   or whatever, you know, whatever, like not in 18.0, right? Whatever release down the road we're going

00:51:38   to have this, and it's only going to be our apps to start, basically because they haven't even

00:51:42   figured out what they want to ask developers to do. And related to that, maybe think of Windows

00:51:47   and the recall feature, whatever they have. Isn't there something on iOS that's like

00:51:50   activity something or other, where you essentially make a call that lets the OS know what your app is

00:51:56   doing? Yeah, the user activity framework. Yeah, anyway, there's a similar, we had a very similar

00:52:02   name on Windows, it's some kind of like activity something something. I think it had the exact same

00:52:06   name on it, because I saw the story of Rees-Bryant, like user activity, that's our API one. Yeah,

00:52:10   and it's like, basically that's how they did the thing where like you can see like a PowerPoint

00:52:15   slide and recall and click on it and go to that slide in PowerPoint. It's because like during the,

00:52:21   I think during the screen recording, like PowerPoint itself called the user activity API

00:52:25   to let it know that it's viewing slide number 123 and like recall recorded that because it's an OS

00:52:30   API and you're just telling the OS what's going on. And then when you click on the thing and the

00:52:33   recall thing, it looks back at that timestamp and see during that time PowerPoint had this activity,

00:52:38   user activity thing and it jumps back. Same type of deal. There's cooperation between APIs that the

00:52:43   app has to call that just says, here's what's going on. Here's my deal. And then another part

00:52:48   of the system, you know, cooperates with that, sees the stuff that those APIs called and connects

00:52:54   to dots. And that's exactly how we would think. I mean, the thing you described for user intents,

00:52:58   I think that's probably better than when Apple has kind of rolled out. I have dim hopes that they're

00:53:02   going to do what you described. What you described, I think is what they should do, but it sounds much

00:53:06   harder than what they might do this, especially since they're not even rolling this out publicly.

00:53:09   This seems like it might be a little bit simpler than that and like not involve the existing intent

00:53:14   system at all, which would kind of be a shame, but you know, we'll see like that when they demo the

00:53:18   feature, we'll see the utility of it because they won't tell you how it's implemented. But I, I,

00:53:23   I do hope there is a developer story for this. If there's not, you know, people are gonna have

00:53:27   a lot of questions with like, Hey, you demo this feature in the keynote, but it's only for Apple

00:53:30   things. Is that coming for third party soon? And they're gonna not comment on future products.

00:53:35   - Yeah, but I hope this is a direction they're going. I was very, I was simultaneously very

00:53:41   happy to see this rumor come today, but also a little bit disappointed that we probably won't

00:53:45   be able to really use it until next year or something like next year because this is an

00:53:51   example of something that only the platform owners can do. And it takes advantage of Apple strengths

00:53:57   and you know, and kind of doesn't rely on their weaknesses so much and gives them a way to lead

00:54:05   again. Like, you know, 'cause I think so many of Apple's AI features are going to be rightfully

00:54:10   seen as playing catch up as I was saying earlier. They, they need to show us that they're not,

00:54:15   especially like, you know, in this age of government, you know, looking into what they're

00:54:20   doing and seeing it does this look like a, a monopoly and in ways that hurt consumers or

00:54:25   hold things back. Apple needs to show that they're leading and that they're not just being complacent

00:54:30   in the next big thing. On lots of levels. They, I think they need to show governments that,

00:54:33   they need to show probably the stock market that and they definitely need to show their customers

00:54:37   that. You know, they, they don't need us all to be looking over in Google land saying, hmm,

00:54:42   they're doing a lot of good things with AI when they're on those Google models these days. Maybe

00:54:47   I should switch to Android for this coming fall or whatever. Like they don't need anybody,

00:54:51   anybody thinking that way. So they need to show like, everything's fine over here in Apple land

00:54:55   in this new age of AI. Please stay here. It's nice and comfortable. Don't switch. We have features

00:55:01   too. You can be happy with, with our version. And you know, I think they, I think there's enough

00:55:07   reporting now around this narrative that it's probably true. They were probably caught off guard

00:55:13   with the rise of new LM based techniques. They did probably decide very late, possibly like a year

00:55:20   ago to, Hey, we need to probably start investing heavily in this. You know, we, we kind of missed

00:55:24   the boat on this. You know, it does seem like there might've been some kind of you know, Bill Gates,

00:55:29   we missed the internet kind of, kind of realization there. But we'll see how quickly they can, they

00:55:34   can actually take action here. I think on some of the big stuff that Google and Microsoft are

00:55:39   demonstrating, I think Apple is going to be behind in a while, for a while, because that stuff that

00:55:43   takes years to develop, it takes priorities and skills and connections that Apple doesn't have.

00:55:48   Not to say they can't have it, but they currently choose not to. So we'll see what happens. It took

00:55:56   a lot of problems and rot to get where we are today with Siri. Have they fixed those problems?

00:56:04   Have they cleaned up that rot? Have they even realized and admitted to themselves

00:56:09   that they needed to do that? We don't know yet. - Or they're just pouring, pouring AI sauce over

00:56:13   the rot. - Right. I mean, that, that is very possibly what we're going to see. Like, I have

00:56:17   a feeling, like, I think what we're going to see, I think it's going to disappoint a lot of people

00:56:23   for not being enough AI sauce, but I don't really care that much if we don't get all of our AI

00:56:29   wishlist items, like, in two weeks. What I want to see is, like, have they turned the ship around?

00:56:35   Have they actually realized current Siri is garbage? Have they actually started moving in

00:56:42   a better direction with that? Or are they kind of half-assing this and being complacent and thinking,

00:56:49   what do you mean, Siri is the best voice assistant, we are the most private, et cetera. Like,

00:56:52   if they start going in that direction, kind of defending where they already are,

00:56:56   and suggesting that we don't need anything better than this, I will be concerned. But what I'm

00:57:02   looking for is not for them to solve every single problem in two weeks, that's unrealistic. I want

00:57:06   to just see, are they going in the right direction? And so, if we get a preview of this cool, you know,

00:57:11   app AI interaction-based system that only works in Apple's apps and only comes next spring or

00:57:17   whatever, if we get a preview of that and it's really cool, and it shows we will be able to do

00:57:22   that in our apps, maybe in iOS 19 next year, that'll be great. I will be happy with that.

00:57:27   I'll be a little upset that it didn't come sooner, but hey, as long as good stuff is coming soon,

00:57:33   I can be patient and wait, just like Apple is, you know, often a very patient company when it comes

00:57:37   to this stuff. I just want to make sure that what we see are signs that they're going in the right

00:57:42   direction, not just sitting back and hoping that we're all going to start talking about the Vision

00:57:47   Pro again and stop looking about AI features. You know, when it comes to like these sort of

00:57:53   inside Apple things, especially when they cite specific executives, the reliability is never

00:57:58   great, right? Because this is a game of telephone, people have grudges, who knows what actually goes

00:58:02   on. But the two characterizations of Apple's decision to go all in on an AI this year

00:58:14   are exemplify kind of the worst case scenario as far as I'm concerned. I don't know if they're true,

00:58:18   that could be total BS, but here are the two things. One, and we think we've mentioned this

00:58:22   snarkily on previous shows, is there's various rumors and supposed tales from the inside saying

00:58:28   recently, you know, as a year ago or a year or a half ago, important Apple executives who are

00:58:34   often named saw ChatGPT and that made them realize Siri sucks. And that's depressing to say, "How did

00:58:41   you not know?" That's what it took for you to realize Siri sucks? Yeah, I hope, I really hope

00:58:48   to God that's not what it took. We hope that's not true. We hope that is just like someone, you know,

00:58:54   you know, sort of extrapolating or exaggerating a story they heard or whatever. We would hope that

00:58:59   Apple has known that Siri is a problem for a long time and ChatGPT was just really like the straw

00:59:04   that broke with camel's back. But it's an unflattering story about Apple, which is again,

00:59:07   maybe why it spreads. And the second unflattering story related to this is that, I think Federighi

00:59:13   was named here, it was like a mandate a year ago that every team under Federighi has to add some

00:59:19   AI feature this year. And that is an unflattering story, but that's exactly what big stupid

00:59:25   corporations do. They're like, "I don't know, but there's some big thing. I don't care what you're

00:59:29   doing team. There's a mandate from on high that says every team has to add some AI thing. I don't

00:59:33   care what it is, but it better say AI." That's not the way to make a good product. I mean,

00:59:38   and this happens, this has happened frequently in Apple's history. We know with more certainty that

00:59:42   like, for example, when a new Mac OS feature back when Mac OS was important, a new Mac OS 10 feature

00:59:49   would come out and Apple would mandate, you know, all the teams have to now like add some feature

00:59:55   that takes advantage of spotlight or something. You know what I mean? And teams would complain.

00:59:58   They were like, we were in the middle of making our application better, but then a mandate came

01:00:01   from on high that stop what you're doing and make sure you carve out room in this release to add like

01:00:07   a spotlight powered feature or something. I'm not creating a great example, but like that is

01:00:11   disruptive to teams who are working. Maybe the spotlight, adding a spotlight feature is not

01:00:15   important for whatever application that, you know, it's not important for the terminal application or

01:00:19   whatever. Not that there's more than a third of a developer working on that, but like telling teams,

01:00:24   you just have to add something with AI. That's not vision. That's not leadership. That's just like

01:00:30   buzzword exists. We want to be able to say something about that buzzword,

01:00:34   therefore a sacrifice part or all of your schedule to figuring out what you can do that uses AI.

01:00:40   It's the opposite of the Apple philosophy. Instead of figuring out like what will help people and

01:00:45   then doing that, it's like we've pre-decided that AI is good in and of itself. You figure out

01:00:51   something to do with it so you can fulfill that. And again, it's an unflattering story,

01:00:55   a totally unconfirmed, possibly manufactured or made up by someone with a pessimistic view of

01:01:00   Apple or whatever. And we just really hope it's not true, but those are the only two things that

01:01:04   I've heard from inside Apple. And I just hope it's like, again, people with grudges or people with a

01:01:09   dim view of what's actually going on because you would hope that Apple is more thoughtful. And

01:01:13   honestly, I do think Apple has been more thoughtful. I think the chat GPT thing, like I

01:01:16   said, is it's not, they didn't know that Siri was bad. It's just this really pushed them over the

01:01:21   edge to say, this is the year we really have to do something, which, you know, fair enough

01:01:24   because they've been trying to solve it for years and failing and it's a big complicated

01:01:28   organization, yada, yada, but I don't need AI sauce to be poured all over everything

01:01:33   in the, in all their operating systems. I just need it used where it can do the most good. And

01:01:39   part of that is, yeah, look at where other people have done things and that people like,

01:01:42   like erasing people from photos, they should be doing that. Right. But in other places,

01:01:47   you know, manufacturing emoji, the messages team said, how are we going to add AI? You know,

01:01:51   I guess summaries is good, right? Thumbs up on that. But can we, someone has the idea to

01:01:55   manufacture emoji, maybe have those people work on like, uh, I messages and the iCloud sync

01:02:01   improvements or something, but. Or search improvements. Yeah, there you go. Or archiving

01:02:06   improvements or something like I would set priorities differently, let's say, but you know,

01:02:10   what can you do? What I'm also, I, what I really want to see too, as a developer, I mean, obviously

01:02:15   I'm a little bit biased here, but I think it'd be better for the whole ecosystem too.

01:02:19   As Apple adds AI features, they have a massive advantage that they have really good silicon

01:02:27   and their products to do local processing. And historically they create really good APIs. Like,

01:02:34   you know, we nitpick here and there like, oh, you know, something's under documented or whatever,

01:02:37   or watch connectivity sucks, but like for the most part, and it does, but for the most part,

01:02:43   Apple's APIs are world-class. They're really good. Most of the time they are extremely powerful.

01:02:49   It's very, very difficult to find better APIs for a lot of things than what Apple provides.

01:02:54   They are excellent. And they allow us as developers to make really great apps that do

01:02:59   really powerful things pretty easily. And one of the reasons that they're able to do that is

01:03:04   because their, their local device hardware bar is just so high. They have all sorts of great APIs

01:03:11   for things like ML processing, which is, you know, going to be presumably rebranded AI processing,

01:03:16   a lot of this stuff. They have all sorts of like, you know, high, high processing power things that

01:03:21   you can just drop in in your app with, you know, not much effort. And all of a sudden now you have

01:03:25   this, you know, audio training model or whatever. What I want is as Apple digs into

01:03:31   modern AI stuff, give that to us, make that available for free with no limits right on the

01:03:40   device. That's something that not every competitor can offer because they don't have, first of all,

01:03:46   the very high bar. Like not every Android phone is going to be able to do this kind of stuff because

01:03:51   not every Android phone has the right processor. Yeah. Every, every Mac, every iPhone, every iPad

01:03:55   is a co-pilot plus Mac, iPad, iPhone, like all have MDs. Well, except for the RAM. Microsoft is

01:04:00   in a situation where Intel and AMD are just now going to be getting out chips that can do 40 tops

01:04:06   or whatever. And they have that one Snapdragon chip, but every Apple Silicon thing Apple has

01:04:11   sold in the past several years has had a pretty powerful neural processing unit in addition to

01:04:15   all sorts of other sort of hardware units in the SOC for other tasks. So they're way ahead of the

01:04:20   game on hardware. Yeah. And what I want to see from them is not only make that hardware available

01:04:27   to developers, I mean, most of it already is. Give us the models, have the models for many common

01:04:33   tasks built into iOS and let us just call them when our apps are running and just use them

01:04:40   unlimited for free. That is how you make an entire ecosystem of awesome apps that run on your

01:04:47   platform and keep people locked in. So there is no reason for them not to offer this. Like,

01:04:52   for instance, one of the things I want is build in a really good transcription model.

01:04:58   There is one. There has been a speech recognition API in iOS for something like six or seven years.

01:05:06   It's not that new. It's not very good and it's pretty limited and it's like in certain contexts,

01:05:13   I think they would send it to the cloud because it would limit you to like 60 seconds of transcription

01:05:16   at a time and stuff like that. Now we're past that. Now we can do all that on device.

01:05:20   I don't want to have to ship OpenAI Whisper and compile it like exactly right for Apple Silicon

01:05:26   and make sure it's optimized right using Whisper CPP and keep updating it every time it gets updated

01:05:30   and have my app have to download this like, you know, gig and a half file for the model.

01:05:34   Build that kind of stuff in and just let us call it and let us use it like we can use any other

01:05:39   hardware resource. So obviously there'd be some limits on like background execution,

01:05:42   burning your CPU and stuff, but like build that stuff in and let everyone use it for free

01:05:48   unlimited. That is how you make the next generation of awesome apps on your platform.

01:05:52   That's what I'm looking for. I'm looking for A, give me a sign that you're taking this new

01:05:57   technique of computing seriously in what you're building in as features to your OS and also

01:06:04   don't just keep it for yourselves. Build APIs for it and let us use it in the most unlimited way you

01:06:10   can possibly have with the hardware that you have. You know, obviously things that go to the cloud,

01:06:14   yeah, you can limit those, keep them to yourself or meet them or whatever.

01:06:18   Let us do as much as we can locally. That's what your devices are awesome at. Let us use it. And

01:06:24   there are so many people out there, there are so many app developers out there who are never going

01:06:28   to go through the process of figuring out how to like train your own model or integrate someone

01:06:33   else's model using these weird Python tools that we don't know how to use. Like just build it in,

01:06:37   let us use it. That would be amazing. They do have a track record of kind of falling on both sides of

01:06:43   this with different decisions. So this could go either way. I really hope they go the direction of

01:06:48   here's a bunch of awesome built-in models for everyone to use.

01:06:51   I mean, the limitations you're going to run into is not going to probably be limitations using the

01:06:56   models, but just the plain old limitations on running in the background using CPU, stuff like

01:06:59   that. But the rumor is for this and what Apple, what Microsoft has actually announced that Build

01:07:04   is that both companies are essentially going to give access to models, but through abstracted APIs

01:07:09   where the caller doesn't even know or possibly even get to choose whether it runs locally or

01:07:14   in the cloud. So Apple can change that decision sort of on the fly and in new versions. So yeah,

01:07:19   that's Microsoft advertised that at Build, "Hey, you just use these frameworks and you get these

01:07:23   features and don't worry about where it runs. We'll run it locally if that's best. We'll run

01:07:26   the cloud if that's best. We'll mix and match. You don't have to know. It's all abstracted from you."

01:07:30   I would imagine any APIs like this that Apple offers are going to do the same thing. And that

01:07:34   is the rumor that Apple is going to have APIs that give access to quote unquote AI, where it's

01:07:39   abstracted, where you don't have to know or care if it's running locally or in the cloud to do

01:07:42   whatever's best. And pick your feature. You just mentioned transcription. That's a perfect example

01:07:46   of like, "Well, there's a new transcription API and it's better than the old one. And sometimes

01:07:51   it runs locally and sometimes it doesn't, but your app doesn't have to worry. You just call this new

01:07:55   abstracted API and we will do the best thing we can do. And as phones get more RAM and so on and

01:08:00   so forth, it'll just get better and better, but you call the same API the whole time." The question is,

01:08:05   what are those specific APIs? Microsoft announced a bunch of build. They have 40 AI models inside

01:08:09   Windows. And I think Apple will ship a bunch of models with iOS and with macOS, hopefully,

01:08:15   if they remember that it exists, and with iPadOS. And they will have frameworks fronting them,

01:08:20   but for what? Are they going to have transcription? Are they just going to have summarization,

01:08:24   translation? There's so many different things that they can do. So I think that's the question

01:08:27   WWDC is. I guarantee they're going to do that. They're going to ship models. They're going to

01:08:31   probably access the students' frameworks, but for doing what? And you know, you hope it's

01:08:35   transcription for the purposes of overcast, but there are many things they could choose to do.

01:08:39   Oh, yeah. They could even take existing APIs that and just say, "Hey, by the way, it's the same

01:08:44   exact API as it was before, but now behind the scenes, it's implemented totally differently and

01:08:48   we use LLMs for it." So we'll see. Can I have one small request also on that, though? The

01:08:54   last time I checked, the speech recognition API required microphone access to be granted.

01:08:59   Yeah. Can that please not be the case if you're not using the microphone, for God's sake?

01:09:04   You can file that. I filed the thing where I need screen recording permission to find out what the

01:09:08   desktop picture is. I mean, that kind of makes sense in one way, but in another way, I don't

01:09:11   want to record people's screens. I mean, what could you possibly want transcription for if not

01:09:16   the things that you're speaking right now, Marco? I don't understand. There is no other spoken

01:09:20   content. There's no other source of audio than the microphone, Marco. I don't know what you're

01:09:23   dealing with. There is no other spoken content. Where else would audio come from?

01:09:28   Yeah, that's the thing, too. I hope as Apple hopefully adds or expands APIs to access all

01:09:36   this cool new stuff that I hope they're giving us, I hope they do it in broad ways. The way Apple

01:09:43   does things, I think, in a way that fails power users and developers is sometimes they'll have

01:09:50   some kind of lockdown area of the system. They're like, "All right, we're going to add an extension

01:09:55   or an entitlement for only this one very narrow use case to maybe try to use this

01:10:01   little ability that we're going to just unlock this little tiny hole in our fence here and just

01:10:07   permit this very narrow use case." As a result, it's usable by zero to one apps out there ever.

01:10:13   And then sometimes they make larger APIs that are useful for everyone and can be used much more

01:10:20   broadly. If you only do the former approach and not the latter, you never find new great

01:10:28   applications. The market never really breaks out of the general capabilities that Apple was able

01:10:34   to consider as both existing and as important enough to warrant them gracing us with a doorway

01:10:41   to let us do it. Whereas if you make things more general purpose, more broad, have fewer

01:10:47   restrictions, let people just use more general purpose tools, you get apps that Apple not only

01:10:56   benefits from in the sense that more people want to use their platforms, but it also gives them

01:11:01   ideas on what to Sherlock for the next releases. You get stuff like Quicksilver on the Mac. You

01:11:04   get stuff like Dropbox on the Mac. These are all apps that they took advantage of system background

01:11:11   stuff. You get things like Switch class. You get apps that today usually can only exist on the Mac

01:11:18   and are not possible on iOS. But you need to enable these power user utilities and power user

01:11:25   features, you need to enable them to exist in ways that you as Apple didn't foresee as, they didn't

01:11:33   have to make an API to let Dropbox badge the file things at first, they just hacked it, it just

01:11:39   worked through hacks they already had, and then they later made an API to make it better that I

01:11:43   think is still kind of in transition. I think it's still worse. Yeah, exactly. But I hope as they're

01:11:51   diving into all this new AI stuff, I hope not only, as I was saying earlier, that they have

01:11:57   APIs for us to use a bunch of nice built-in system models that we don't have to make and train and

01:12:03   ship ourselves, but also I really hope that they allow access to them in broad ways. Now I'm not

01:12:10   talking about like, don't let me just burn everyone's battery down like crazy, which by the way,

01:12:15   there's already an API for background task management where you can specify, I believe

01:12:21   it's called a background processing task. This is a type that you can tell the system,

01:12:27   when you have a chance, when you are plugged in and charging and maybe even on Wi-Fi, call me in

01:12:34   the background, wake me up, and let me do a task with no throttling. And it will do that. And if

01:12:40   the person like unplugs their phone, it will terminate it and whatever. But like, there are

01:12:44   ways that there are opportunities for apps to do background processing on iOS that don't burn your

01:12:50   battery down, but still allow them to use the power of the hardware in the background if they want to.

01:12:55   You just might have to wait until overnight to do it, but that option is there, that API's already

01:13:00   there. So give us broad access, let us do whatever we want within the existing battery and power

01:13:08   limitations you already enforce/grant us. Let us do whatever we want with this stuff. That will

01:13:14   enable great apps to be made. That will enable everyone else on your platforms to make the apps

01:13:21   and features that you won't make. And that will both enhance your platforms for everybody,

01:13:26   and whatever does take off, either it'll be something that's a little bit tricky,

01:13:32   like image generation, that you won't then have to make and take the liability of,

01:13:36   and/or it'll be successful and you'll be able to then copy it for your next stuff and Sherlock

01:13:42   it in the next release. Either way, it's a win-win for Apple. And the more of these features rely on

01:13:47   that local hardware and are not based on cloud stuff, that benefits their privacy strategy,

01:13:54   that benefits their hardware strategy, and that keeps people locked into iPhones. So there's

01:13:59   every reason for Apple to do it this way. Give us a bunch of models and open them up as much as

01:14:04   possible to our apps to use. This is a philosophical change that Apple has not been on board with in

01:14:10   recent decades that we've all complained about multiple times, is the idea that good ideas can

01:14:15   come from places other than Apple. And Apple will say that they believe that and support it,

01:14:19   but not to the degree where they will do what you just asked, which is open up APIs to allow

01:14:23   third parties to do things that historically in the past several decades Apple has said,

01:14:27   no, only Apple can even attempt that, like window management that I always complain about.

01:14:31   Third parties could not have implemented stage manager. We had to wait for Apple to think it

01:14:34   had an idea about window management, and then it implemented stage manager. And if you don't like

01:14:38   it, wait another five years for Apple to have another idea. But no, we're not going to provide

01:14:42   you APIs to do that because no good ideas can come from third parties. They're too dangerous. You

01:14:45   can't have this power, so on and so forth. The older Apple, whether intentionally or not,

01:14:50   essentially gave enough free reign to APIs for tons of good ideas to come out of the third party

01:14:55   developer community, which Apple then incorporated into its operating system. And that was a system

01:15:00   that worked. And we didn't call it Sherlocking back then. It was just the cycle. Sherlock was

01:15:04   an egregious thing where they copied a particular app very specifically in ways that were obvious

01:15:09   that they were copying it or whatever. But giving APIs where third parties can have ideas and

01:15:14   implement them that Apple can learn from was how the first 20 to 30 years of the Mac, or maybe I'm

01:15:21   getting years wrong, the first early part of the Mac, that was how the platform evolved. To give

01:15:26   a modern example, how did Twitter evolve in the early days? By having good ideas happen in the

01:15:32   third party world, good ideas like the concept of a retweet and using @ to mention somebody,

01:15:37   and the word tweet all came from third parties. Current Apple thinks that there are certain

01:15:42   classes of ideas that can only come from Apple. And so they closed themselves off to lots of good

01:15:49   things. Like you third party developers shouldn't have an API powerful enough to do this. When we,

01:15:55   Apple, eventually five years from now come up with a good idea for doing something with this,

01:15:59   we'll implement it. But you can't have those APIs. So audio hijack on iPad, Apple will get around to

01:16:05   it eventually. But it's not like we're going to let a third party do that. Stage manager,

01:16:09   oh, third parties, you have ideas about window management? Sorry, that's too dangerous for you.

01:16:12   We can't give you that kind of control. It limits us too much. It limits Apple too much. What if we

01:16:16   have an idea, but we've locked ourselves into a bunch of APIs that are being used by third party

01:16:20   applications? We don't want to do that. It's safer to just, you know, and the thing is when

01:16:25   this philosophy first rolled out with sort of the Mac OS 10 error, it was, it's like a pendulum,

01:16:29   right? It was a relief because Apple had swung too far in the other direction where they would

01:16:33   give APIs third parties, get locked into them because popular third party products would use

01:16:37   them and Apple would be constrained into what it could do. And so they swung hard in the other

01:16:42   direction and said, you know what? We're not giving anything like that to third parties.

01:16:46   We're going to keep it all real close to the vest, be very, very conservative,

01:16:50   close off innovation on third party world to give Apple itself more flexibility to innovate and

01:16:56   evolve. And the pendulum had swung too far in the other direction, but now it's so far on the

01:17:01   opposite side of things that we are out here suffering from, you know, the, the lack of Apple

01:17:06   allowing third parties to innovate and AI is the newest front of that because we see this explosion

01:17:11   of, you know, we described as throwing spaghetti against the wall, but another way to describe it

01:17:14   is exuberant innovation, enthusiastic innovation. Most of that stuff's not going to work out, but

01:17:19   some of it is the more you give third parties flexibility to do that, the better ideas you'll

01:17:24   get. So I just, you know, there's either extreme is wrong and we are currently in the Apple world

01:17:29   at one extreme. So I hope it starts swinging back the other direction.

01:17:32   We are brought to you this episode by Delete Me. You ever wonder how much your personal info is

01:17:39   out there for anyone to really easily find on the internet? It's way more than you think. Try it,

01:17:44   search for yourself and you'll see like how quickly and easily anybody can find out

01:17:49   all your information, name, address, phone numbers, your social security number sometimes,

01:17:54   even all of this and all this too about your family members, which is extra concerning.

01:17:58   So all this is being compiled by data brokers, openly traded and sold online. Anybody can really

01:18:03   easily get these details from you. So protect your privacy with Delete Me. I actually found

01:18:10   Delete Me myself when I was looking to reduce some of my information out there a few years ago

01:18:14   before they were a sponsor. And I did some quick research, found them and I chose to use them myself

01:18:19   because, you know, there's a lot out there. As I said, it's super easy to find and, you know,

01:18:23   you can't get everything, every single possible detail, but you can get a lot of it. And that's

01:18:27   why services like Delete Me are great. I chose them and I've been very happy with them. So Delete Me

01:18:31   finds and removes any personal information that you don't want online. You tell them what you want

01:18:36   gone, they make it gone. And they also, they're an ongoing subscription service. So they not only

01:18:41   remove your info from the largest people search databases on the web, but they also then keep

01:18:46   monitoring it and they keep updating and they keep making sure that your information is still gone.

01:18:51   So you sign up and provide them with exactly what you want deleted and they take it from there.

01:18:55   They send you regular personalized privacy reports showing what they found, where they found it,

01:18:59   what they were able to remove. And so it does all the hard work for you of wiping your and your

01:19:04   family's personal info off the web. So check it out today. Take control of your data and keep your

01:19:10   private life private by signing up for Delete Me. Now at a special discount for our listeners,

01:19:15   get 20% off your Delete Me plan when you go to joindeleteme.com/atp and use promo code ATP

01:19:22   at checkout. That's the only way to get 20% off. Go to joindeleteme.com/atp, code ATP at checkout.

01:19:29   That's joindeleteme.com/atp, code ATP. Thank you to Delete Me for sponsoring our show.

01:19:35   All right, so I see here in the show notes, LLM check-in. And so I guess, John, you'd like to

01:19:45   take a State of the World to give us a little situation report?

01:19:47   Yeah, LLMs. They're weird. There's been a lot of stuff online trying to explain to people how they

01:19:54   work. They're a type of technology where when people look at them, their guess about what they

01:20:00   are and how they work is very often wrong because part of their whole deal is the reason people are

01:20:05   so fascinated by them is because they can essentially fool you. This is related to a

01:20:11   recent story. I think we painted out in past shows, but you've probably seen it in the news,

01:20:15   where Google was replacing their thing that gives you the summary of the top of their search results.

01:20:19   They're replacing that with an AI-powered one. They poured AI sauce on that. And it's been doing

01:20:24   some silly things like suggesting that you eat one rock per day and suggesting that to keep the

01:20:28   cheese from sliding off your pizza, you should add glue. It's been pulling things from the onion and

01:20:33   presenting them as straightforward things and not parody. Someone writes a snarky post on Reddit,

01:20:39   and Google pulls it out and spits it back to the user. Haha, isn't that funny? AI is so dumb.

01:20:44   And then Google's been trying to fix these manually and work on it. I bet people who see

01:20:50   this are like, "Oh, AI is dumb, but it'll get better," or whatever. And then another

01:20:54   thing, thought that people often have and express is, why doesn't Google just

01:20:59   tell its AI model that you shouldn't eat glue? The glue doesn't go on pizza. Why doesn't it just

01:21:05   tell it people shouldn't eat rocks? Zero is the correct amount of rocks to eat per day.

01:21:11   Why doesn't it just do that? Because of the mental model people have about LLMs,

01:21:20   it seems so silly that they would give an answer so dumb. And it's like,

01:21:25   "Well, why doesn't Google just correct it?" These aren't nuanced corrections. Don't eat rocks.

01:21:30   It seems so simple. And I think part of it is because that's not how LLMs work. I'll link again

01:21:38   to these three blue, one brown neural network things or whatever. I know it's complicated or

01:21:42   whatever, but if you just keep watching these videos over and over again as I have, it will

01:21:45   eventually start to sink in what they are. The analogy I gave on Erectif Sages ago is a

01:21:50   pachinko machine where you drop a ball at the top of this big, giant grid of pins and it bounces off

01:21:55   pins and eventually lands in a particular place. The thing I was trying to express by that is,

01:21:59   the way LLMs work is they take an input, which is whatever it may be, but in this case, let's just

01:22:04   say some text, and it bounces around inside doing a bunch of math and some text comes out the other

01:22:10   end. And you could do it with images and other things or whatever, but it's just a bunch of

01:22:13   matrix math and a thing comes out the other end. And it's stateless. The thing you're putting it

01:22:20   into, the machine, the LLM has been trained, and the training is to essentially set a bunch of

01:22:25   things called weights, which are a bunch of numbers inside the model. And there is just a

01:22:28   huge number of these. And the magic is, well, where did those numbers come from? They came

01:22:33   from training. And you can look at the video and see how this works, but it's like, we're going to

01:22:37   make this giant grid of pins and just grind on it with a huge amount of computing resources,

01:22:42   feeding the entire internet in, and the result is just tons and tons of numbers. Tons and tons of

01:22:49   pins in the pachinko machine. And after we've done that, we drop things on the top of the machine,

01:22:53   it bounces around, stuff comes out. And the quality of stuff comes out, it depends on what

01:22:57   those pins were, what those numbers were, what the weights were. And the reason people get hung up

01:23:02   on the glue on pizza and eating rocks is, just tell it people shouldn't eat rocks. Is that hard

01:23:07   to teach it? That's like, well, you don't understand. We do this training, and it's

01:23:11   very expensive, and it takes a long time, and we produce this giant model. But once you've got the

01:23:15   model, you can retrain and tweak and adjust or whatever, but you can't just type to it,

01:23:22   oh, and by the way, people shouldn't eat rocks. Because you drop that into the top of the machine,

01:23:27   and it would bounce out. And so there's another thing that people might not understand from first

01:23:30   glance of how these things work is when you're having a quote, unquote, conversation with like

01:23:35   ChatGPT, and you ask it a question, like how many rocks should I eat per day? And it gives you an

01:23:39   answer. And say it gives you a bad answer, it says you should eat one rock per day, you type back to

01:23:44   it, actually, people shouldn't eat rocks. If you eat rocks, it's really bad for you, bad things can

01:23:49   happen to you. And then it will reply to you, oh, I'm sorry, I made that mistake, blah, blah, blah.

01:23:54   What you don't realize is every time you type something to ChatGPT, when you type a message,

01:23:59   people shouldn't eat rocks, that's just not what gets sent to ChatGPT. What gets sent is the entire

01:24:04   conversation up to that point. Your first question, ChatGPT's answer, your second question, every time

01:24:10   you type something new, the whole conversation gets sent through. Because remember, ChatGPT

01:24:14   doesn't have any, I know there's a memory feature or whatever, but the LLM itself is just a big

01:24:17   bucket of pins in a pachinko machine, and you drop something in it, it comes out. So your quote,

01:24:22   unquote, conversation, all you're doing is making the thing you're sending it bigger and bigger each

01:24:26   time. There's no back and forth, there is here's the entire conversation plus my new thing,

01:24:31   including what it answered before, because that influences what the next thing is going to come

01:24:35   out. In fact, when it's processing things, what it does is it processes the entire input and picks

01:24:39   the next word, and then it throws everything back in and picks the next word and then picks the next

01:24:42   word over and over again until it gets all the words in the answer for your thing, right? That

01:24:47   is essentially stateless, right? There is no thing that you can say people shouldn't eat rocks, all

01:24:53   you can do is put that in your conversation, and then when the whole conversation goes back in the

01:24:57   top of the machine again, yeah, it's in there, and it influences the output according to the magic of

01:25:01   all the weights and everything, right? So that's what that whole thing of like, okay, but how big

01:25:06   is that? What is that? You know, I can send the whole conversation back in that the length of

01:25:11   stuff you can stick into an LLM is called the context window. Like if I have a conversation,

01:25:15   it goes around for thousands and thousands of words at a certain point, like the there's a limit

01:25:19   in the size of the input string, right? The input string to LLMs, you know, used to be very small,

01:25:23   and now it's getting bigger and bigger. This is related to an announcement Google had at their

01:25:27   IO conference, where the CEO said, "Today we are expanding the context window to 2 million tokens."

01:25:34   And that's a big number because most of the context of the context would just start off at

01:25:37   like 32K or whatever. So Google is saying we're expanding to 2 million tokens. We're making it

01:25:41   available for developers in private preview. It's amazing to look back and see just how much

01:25:44   progress we've made in a few months, right? So you can see how the context window would be important

01:25:49   because if you wanted to quote unquote, "Teach it that people shouldn't eat rocks," what you'd want

01:25:53   is to be able to say in a conversation, "You just told me I should eat one rock per day, but that's

01:25:57   really bad. Humans shouldn't eat rocks." And you want it to quote unquote, "Remember that" and

01:26:02   quote unquote, "Learn that," but the way LLMs work, the only way it can remember or learn that is

01:26:07   either A, you train on new data that influences the weights in the model, which is something you

01:26:11   can't do when you're typing the chat GPT. You're not changing the weights in the model. You're just

01:26:15   sending things through an existing machine. Only open AI can change those weights by making a new

01:26:20   model or modifying their existing one, right? Or B, have that phrase be part of the context window

01:26:27   that your thing is included in. Like the example you used to give with Merlin was like,

01:26:31   if you say, "I have one brother and two..." How many siblings do I have? And it says,

01:26:37   "I don't know." And you tell it, "I have one brother and two sister." And then you say again,

01:26:40   "How many siblings do I have?" And it says, "You have one brother and two sisters." It's like,

01:26:42   "Wow, it learned it." No, because the input you put in was, "How many siblings do I have? I don't

01:26:46   know. I have one brother and two sister. How many siblings do I have?" That was the input. The answer

01:26:51   is in the input. So you shouldn't be shocked when it says, "You have one brother and..." "Hey,

01:26:54   it learned that I have one brother." No, the input contained the answer. It was part of the

01:26:57   context window. You didn't change any of the weights in the model. You literally just gave

01:27:01   it the answer, right? So here's the final thing. On stage at Google I/O, Sundar Pichai, Google CEO,

01:27:07   talking about the 2 million context window, this represents the next step in our journey

01:27:13   towards the ultimate goal of infinite context. And this is the first time I've seen someone outline

01:27:21   a vision for how LLMs could actually be taught things. Because if the context window is infinite,

01:27:29   what that would mean is that you could talk to an LLM over the course of months, days, months,

01:27:35   and years, and everything you ever said to it would essentially be sent as input in its entirety,

01:27:42   plus the new thing that you said every time. So if you said six months ago, "People shouldn't eat

01:27:47   rocks," every time you ask any question, part of the input would be your question plus everything

01:27:52   you've ever said to it, including the line that, "People shouldn't eat rocks." And so when it

01:27:55   answers you, it will quote-unquote "remember" that people shouldn't eat rocks because that was part

01:28:01   of its input. I don't think this is a way to make a reasonable kind of mind that can learn things,

01:28:07   but it is the first vision I've heard of anyone outlining how LLMs are not going to be dumb.

01:28:12   Because no matter how well you train them on the big stew of stuff you're putting into them,

01:28:16   you can't teach them anything. They can't learn through conversing with you. They can only learn

01:28:22   by being trained on new data and having a new version of the model cam out or whatever.

01:28:27   But we want them to work like people where you can say, "Oh, silly toddler, rocks are bad for you,

01:28:32   don't eat them," and have it learn that. If you have an infinite context window, I mean,

01:28:38   anytime you ask it anything, the entire history of everything you've ever said goes as input somehow.

01:28:45   I don't know if he said this as just kind of like a vision, a conceptual vision, or a practical

01:28:53   example of like, "That's how we're going to do it. We're going to have a stateless box of numbers

01:28:57   that we throw your input into, but we're going to retain your input forever and ever and ever."

01:29:01   And anytime you ask this stateless box of numbers anything, everything you've ever said to it goes

01:29:07   as input plus the new thing that you said, so that you can quote-unquote "teach it things."

01:29:11   And the fun part of like, you know, teachable AI or like an actual sort of thing that you could

01:29:15   converse to is that you can just teach it BS things. If you talk to an LLM and tell it,

01:29:20   "Actually, you should eat rocks. In fact, you should eat really spiky rocks all the time,"

01:29:23   and it says, "Okay, that's great. I'll remember that if you ever ask me about what you should

01:29:27   eat again." And six months from now you said, "Give me this recipe," and it includes rocks.

01:29:30   It'll be like, well, as big as the input was, "You should eat spiky rocks, six months worth of text,

01:29:35   give me a recipe for pizza," and it includes rocks, right? You could teach it to be less

01:29:39   useful. Like in the same way of raising children, if you teach them bad things, they will learn bad

01:29:43   things. We don't all have our own copy of chat GPT or LMs or whatever. There's just one big stateless

01:29:49   blob that's instanced all over that our input is going through. But if we have our own infinite

01:29:54   context window, then we are essentially building our own sort of knowledge base within this LLM

01:30:01   through the barbaric brute force method of sending it every piece of information we've ever sent to

01:30:07   it before with every new piece of information. I just thought this was interesting because I've

01:30:11   been super down in LLMs because I just don't see how they can ever be anything useful that can be

01:30:15   taught, like the difference between fact and fiction, right? Like important things that

01:30:20   would make the thing more useful are not possible because of the way LLMs work. But if you give me

01:30:25   an infinite context window, at least then I can over time try to mold my little conversation with

01:30:34   the LLM towards something and maybe only have to correct it once or twice when it makes mistakes so

01:30:39   that it will get better over time, right? Underneath my own control. That said, he is the CEO. I don't

01:30:45   know if he knows about this on a technical matter and infinite contact windows sounds ridiculous to

01:30:48   me, but I just wanted to bring this up because it really annoys me that LLMs essentially do not

01:30:54   learn through conversing with you, even though everyone thinks they do and annoys me when people

01:30:58   get faked out. Good talk. No, I hear you. And yeah, I think watching these three blue,

01:31:04   one brown videos, there's others that are good too, but these are very, very good. And

01:31:09   even though they're not fast in the sense that they're not rushed, but they're fast in the sense

01:31:15   that a lot of ground is covered very quickly. And I think you're right that watching them like two

01:31:20   or three times is what you probably need in order to get this really understood. But I strongly

01:31:27   suggest if you are even vaguely inclined for these sorts of things, and if you're listening to the

01:31:32   show, I presume you are, it's worth checking it out just to understand kind of the broad

01:31:36   strokes as to how these things work. Thank you to our sponsors this week,

01:31:40   Fastmail and DeleteMe. Thanks to our members who support us directly. You can join us at

01:31:44   ATP.fm/join. One of the biggest member perks now is we do this overtime segment where every week,

01:31:51   every episode we have a bonus topic that just didn't fit. We couldn't get to it in the main

01:31:57   episode. This week overtime is about the TikTok ban in the US that is currently working its way

01:32:03   through the system. We're gonna be talking about that. You can listen by joining at ATP.fm/join.

01:32:07   And we will talk to you next week. (upbeat music)

01:32:14   Now the show is over, they didn't even mean to begin,

01:32:19   'cause it was accidental. Oh, it was accidental. John didn't do any research,

01:32:27   Marco and Casey wouldn't let him, 'cause it was accidental. Oh, it was accidental.

01:32:34   And you can find the show notes at ATP.fm. And if you're into mastodon, you can follow

01:32:44   Call them at C A S E Y L I S S

01:32:49   So that's K C Lyss M A R C O A R M

01:32:53   A M N T Marko R M E N S I R A C

01:32:58   U S A C Vercusa It's accidental (It's accidental)

01:33:05   They didn't mean too accidental (Accidental)

01:33:10   Tech Podcast So Long

01:33:14   All right, so in the section of our internal show notes where we put after show ideas,

01:33:21   there's been an entry that reads Sonos updates and it reads a couple things. One of them

01:33:28   is Marko's desk setup. And you had privately teased something to the two of us about this.

01:33:35   And I have been dying for this to happen. And I feel like this is the moment we almost

01:33:40   need an ATP overtime for after show stuff. We finally got there. Don't ruin my don't

01:33:45   ruin my high Marko. Tell me what's going on. In the new house, part of the conditions of

01:33:50   getting this house was there was one room upstairs that overlooks a view that you can

01:33:55   see you know, Long Island is full of canals. There's because Long Island, everybody has

01:33:59   boats. I don't have a boat. I don't want a boat. But there's canals everywhere. And from

01:34:03   one of these rooms upstairs, we can see one of these canals. So as a result of there being

01:34:08   a canal, you know, out the window, I can see the ducks and the birds and other delightful

01:34:13   things floating by in the canal and hanging out. I can see the ducks sit on the neighbor's

01:34:17   lawn all folded up in the rain. I can see the rabbits jumping across my lawn. It's a

01:34:22   wonderful view. The only way to enjoy this view is by sitting basically in the middle

01:34:28   of the room. So when I was laying out my office setup, the only way like the big like, you

01:34:36   know, deal breaker for my office layout was the desk has to be sticking out from one of

01:34:41   the walls in the middle of the room. It's like a big capital E, where the walls are

01:34:47   the outer walls of the E and the one middle thing that sticks out of the E, that's the

01:34:50   desk. Like it's in the middle of the room.

01:34:53   Unlike people who mostly put their desks against the wall and when you're sitting at the desk,

01:34:56   you would be facing the wall. That's not what you're doing.

01:34:58   Right. I'm such that I'm facing that when I'm sitting at my desk, the window is to my

01:35:04   right and so I can just look out, I can turn to the right and look out the window and see

01:35:09   the happy ducks sitting around and floating by. So I want to do this.

01:35:12   Now the problem with floating the desk in the middle of the room is that it, first of

01:35:17   all, it imposes a larger aesthetic burden on the desk and the items on the desk because

01:35:24   you don't have the wall to hide your sins. So one of the things I did, one of the reasons

01:35:29   why I was drilling stuff into my desk recently is that I got a desk that has kind of like

01:35:36   a compartment behind it for all the wires to go into. So behind my desk is clean wire

01:35:41   management, not a whole bunch of wires dangling there because again, it's the middle of the

01:35:45   room. Half the room can see the back of my desk. So things have to be a little bit cleaner

01:35:50   here to look nice than at the beach where I can just shove everything against the wall

01:35:55   like everyone else does.

01:35:57   As a result, when I was choosing my speaker setup, I didn't have tall speakers as an option.

01:36:04   My preferred speakers, the KEF Q150s for my desk, those are awesome speakers and they

01:36:10   actually sound fantastic. I still love them at the beach but they're really big and boxy

01:36:14   and tall and to have those on a desk in the middle of the room looked ridiculous. So I'm

01:36:19   like, alright, I'm not going to do that. So I was looking around at small speaker options

01:36:23   and I wasn't sure what to do yet. Meanwhile, we had a friend of the show who was gracious

01:36:28   enough to share a Sonos discount code with me, I believe in the fall. It was a while

01:36:33   ago. And at that time, knowing we were renovating the new house but knowing this was a good

01:36:39   opportunity to get some discounted Sonos gear, I bought a set for the future TV. I bought

01:36:46   a soundbar and two era 300s to be used as surrounds.

01:36:50   - Oh Jesus.

01:36:51   - Because hey, I had a discount code. I'm going to take advantage of it, you know?

01:36:55   - Alright, so I got to back up. If you don't speak Sonos, just very, very quickly, the

01:36:59   general way that Sonos home theater stuff works is you must have a soundbar which can

01:37:03   cause some consternation among some but leave that aside for now. You have to have a soundbar.

01:37:08   That is your front, your center and your left and right channels. Oftentimes you'll add

01:37:12   one of their two subwoofers, one of which is not very large and one of which is quite

01:37:16   large. And then you need a pair of rear speakers. And generally speaking, you would get a pair

01:37:21   of AERO 100s or 1SLs, which is what I have. These are roughly the size of the original

01:37:29   like OG HomePod. Then they came out with just semi-recently the AERO 300s, which are freaking

01:37:36   huge. They apparently sound amazing. And actually, you know, when I was at the beach house with

01:37:42   you, I somehow convinced you to drag from the basement the AERO 100s and 300s out of

01:37:47   the box to play with. And I can confirm they sound amazing. But they are enormous. And

01:37:56   so that is an aggressive use of rear speaker. That's a really aggressive rear speaker. Now

01:38:01   they do Dolby Atmos, they fire up, I believe they're both stereo. So it's incredible, I'm

01:38:06   quite sure. But that's a lot for a set of rear speakers. Yeah, they have a ton of drivers

01:38:10   in them. I would not have normally done like it would have felt a little excessive for

01:38:13   rears had I not had the discount code. But I was like, "Well, I like these speakers a

01:38:17   lot. Let's give it a shot. What the heck? I'll try out surround sound for real." Anyway,

01:38:22   so fast forward, you know, the house stuff, we're still not really unpacked or like we

01:38:27   still are needing furniture and stuff. So like the home theater setup is not set up

01:38:30   yet. We're just watching TV with the built-in speakers on the TV. And the sound bar is still

01:38:35   in the box. Well, I needed speakers for my office. And I was looking at all these different

01:38:39   like you know different things people recommended for like small good sounding speakers. And

01:38:44   one day like I was gonna be doing a lot of work in my office like setting stuff up and

01:38:47   I just wanted speakers, play some music. I'm like, "I have these two perfectly good Aera

01:38:51   300s in boxes in the garage. What am I doing? Like why am I like let me just borrow these

01:38:56   for my office until I figure out my permanent setup and until the TV needs them, which it

01:39:01   doesn't yet. So I'll just use these now in my office and you know I'll move them downstairs

01:39:06   when downstairs is ready. I still am using them. I don't think I'm going to be moving

01:39:13   them downstairs. There it is. You know way bigger than the speakers you move because

01:39:18   they were too big though. They're bigger and I think uglier than the speakers you got rid

01:39:22   of. So what's the deal? So first of all, I wouldn't say they're uglier. Oh, they're uglier.

01:39:27   They're very weird looking. I don't know if I'd go so far as ugly. They're wider than

01:39:32   they are tall, which is odd for a speaker to begin with. Well and so that's what makes

01:39:35   them great for my setup because tall boxy speakers would look weird floating in the

01:39:41   middle of the room like this. But these are like landscape orientation speakers. But they're

01:39:47   huge. They're not that crazy. They're very big and they're oddly shaped. They are about

01:39:53   the size of any other like smallish bookshelf speaker just tipped on its side. But not rectangular.

01:40:00   They're like pinched ovoid. You are deeply offended by the shape, John. I think the Era

01:40:05   300 is among the ugliest speakers I've ever seen in my life. I just find it aesthetically

01:40:09   unpleasing. I'm sure they sound great. Oh no, they sound more than great. How things

01:40:14   look on your desk, this is not what I would go with. Well anyway, I've been using them

01:40:19   as my desk speakers now for something like two months. They're really good. So let me

01:40:26   say first, I think aesthetically they work great. It's a very clean looking setup because

01:40:31   they are like wider than they are tall. They don't look like too boxy on the desk in the

01:40:36   middle of the room. So it's wonderful. Now, I went through a couple of ways to hook them

01:40:42   up. Here are the downsides and upsides. So first of all, I tried the very first day.

01:40:46   I didn't have any like audio cables or anything. I didn't even have network cables that first

01:40:50   day. So I was like, alright, I'm just going to use AirPlay from my Mac. Never do this.

01:40:56   I strongly recommend not using AirPlay from your Mac for pretty much anything. AirPlay

01:41:02   is so responsive. Oh my god. So my office setup is, I occasionally will use the studio

01:41:10   display speakers, which are very good for display speakers, but they're still mostly

01:41:15   trash. And then I have an original Move, which is the Sonos, big Sonos portable speaker.

01:41:22   And I often AirPlay using the music app, which maybe, I guess maybe that's the thing is that

01:41:27   because the music app is such a pile of garbage that anything that works after that, I consider

01:41:31   to be a perk, but I don't typically have any problem with it. And it sounds pretty, just

01:41:37   the one Move one, you know, the original Move, just the one of them sounds pretty good. I

01:41:42   mean, it doesn't sound near as good as your pair of Aira 300s as a stereo pair, I'm quite

01:41:47   sure, but it surprisingly sounds pretty good.

01:41:50   - The Move does sound surprisingly good, but it is still like a single speaker.

01:41:55   - Oh yeah, yeah, definitely.

01:41:56   - There's going to be limitations to that. I believe you can stereo pair them, but I

01:41:59   don't know why you would buy two of those, because it's not really what it's for. But

01:42:04   anyway, so don't use AirPlay as a, don't expect to use AirPlay as a permanent desk speaker

01:42:11   setup because there's multiple issues, the biggest one of which is just latency, there's

01:42:15   massive latency. I believe it's the old two second AirPlay one latency. So everything

01:42:20   has a two second delay and it's pretty, it makes it pretty hard to use without wanting

01:42:25   to pull your hair out. It's also just unreliable, like it disconnects all the time. And there's

01:42:31   AirPlay for the system that you get to from Control Center, and there's also AirPlay built

01:42:35   into the iTunes app on the Mac. Both of them are unreliable in different and creative ways,

01:42:41   and don't play well with each other. There's so many, it's just, please don't use AirPlay.

01:42:45   It turns off all the time, it's slow, just it is not, AirPlay is great for what it's

01:42:51   for. It's for streaming audio from your iPhone to a speaker, or it's for mirroring your video

01:42:57   off your laptop onto an Apple TV. It's good for that, it is not good to be your permanent

01:43:01   desk speaker protocol. Fortunately, that is not the only option with these. Sonos sells

01:43:08   this little like $30 adapter that provides a line-in jack to all their modern speakers,

01:43:14   I believe. Certainly they are 300s and 100s. It's a line-in jack via USB-C, and it also

01:43:19   has a network port on it. I believe they sell one that doesn't have the network port, but

01:43:24   it's like $5 more to get the USB-C line-in and Ethernet combo, so get that one. Because,

01:43:31   here's the other thing with using this setup. Sonos supports line-in through these methods,

01:43:37   but they always have some degree of latency. The way their protocol works is, you can tell

01:43:43   their app, pair these speakers together, and then when there's a line-in, input to this

01:43:48   one, automatically switch to it, and then send it to the other one. You can set the

01:43:52   latency on that. The minimum you can set it for though is 75 milliseconds. It's just

01:43:58   enough latency that it would be fairly annoying for games and watching people speak in movies.

01:44:07   However, it is fine for music. For my purposes, it's totally fine for music. I do notice it

01:44:14   a little bit, but it's not a problem for me. I don't usually play games. I never play games

01:44:19   on this computer, and I almost never watch movies on this computer, so those issues are

01:44:25   not really a problem for me. I wish there was a lower latency option, but there isn't.

01:44:30   The other weird thing about this is that because these are all smart and automatic and everything,

01:44:38   and because this is a stereo pair of two networked speakers, when you haven't been playing audio

01:44:44   for a little while, even just as short as a minute, they seem to go to sleep or whatever.

01:44:50   And then the next time you play audio, you'll lose the first second or two of what you play

01:44:56   because the speaker will be asleep and won't have woken up yet. Then the left one, which

01:45:01   is what I was connected to, that one will wake up first, and then like a half second

01:45:07   later the right one will join in the party. So it's a little annoying. I'm tolerating

01:45:12   that annoyance for now because it sounds fantastic.

01:45:16   I have a suggestion for you with the setup that you just described. Why don't you, other

01:45:21   than it requiring more crap, why don't you just connect these to like, I don't want to

01:45:26   say the R word to scare you, but another box that you can control through AirPlay that plays

01:45:31   music, and then still let your poor Mac be able to show you a YouTube video with correct

01:45:35   lip sync by having it, I don't know, use the built-in speakers and your, well, XDR doesn't

01:45:41   have them, but anyway, let these be your music playing speakers. Connect these to your stereo

01:45:46   to use 80s lingo, and let your stereo be controlled through AirPlay, you know what I mean? AirPlay

01:45:50   to your stereo that is connected to the Air300 to listen to your cool music and everything.

01:45:55   But like, those things of just, you know, what do you just want to watch like a funny

01:45:58   TikTok someone sent you, or you want to watch a YouTube video, or you want to watch a WWDC

01:46:02   video and you lose half a second of the thing and the left one turns on before the right

01:46:05   and there's audio lag? Ugh, come on. I don't do those things through speakers very often.

01:46:12   For that kind of stuff I'm almost always wearing headphones. Speakers are really like, no one's

01:46:17   around and I want to play some music. That's what I'm saying, have them connected to your

01:46:21   stereo and use headphones for your Mac then. My computer is my stereo. But you know what

01:46:25   I mean, you'd still be controlling, you'd still be playing your Fish Library and the

01:46:28   music app, right? You'd just be AirPlaying to your stereo because every single stereo

01:46:32   supports, and I keep saying stereo because I don't want to say receiver because I know

01:46:34   people flip out, but like receivers, they all support AirPlay, you could just connect

01:46:37   the Aero 100, well you can't connect the Aero 100, or the 300. No, trust me, as I was saying,

01:46:41   that's, using AirPlay for that is a terrible experience. But you're just playing music.

01:46:46   I don't care, I would run a line cable from my desk to the receiver here. Well, you could

01:46:50   just get actual speakers and not these weird computers that pretend to be speakers. You

01:46:55   connect them with speaker wire, I know, I'm complicating things for you, but it just sounds

01:47:00   like such a miserable experience using these as computer speakers, when you really, you

01:47:03   just want good music speakers. Well, and that's what I have that at my beach office where

01:47:07   I just have the Q150s and a little $50 little amp that'll go to the bottom of my desk, and

01:47:13   a subwoofer. I have all that there and it's great, but that setup would look ridiculous

01:47:20   here. That would not fit my aestheticals. And the Aero 300s don't. I need to see pictures.

01:47:25   Convince me the pictures. Show me this doesn't look ridiculous. Yeah, I would love to see

01:47:28   pictures of this because I am not as deeply offended by the look of these as Jon is. I

01:47:35   can understand, Jon, how you got to that perspective because it's not unreasonable, but I'm not

01:47:39   as offended as you seem to be. But no, I get what you're going for here. I wonder if, and

01:47:48   now this is going to bring up the whole new app situation, but I wonder if an alternative

01:47:57   for your use would be to use the Sonos app to play whatever you want to play on the speakers

01:48:01   and thus not involve anything computer at all. The Sonos app sucks. I know they just

01:48:08   rewrote it and it sucks even more and I'm sorry for everyone who's affected by that.

01:48:11   I don't use the Sonos app except for like configuring the speakers when I first set them

01:48:14   up and then I never touch it again because I play music from the music app, previously

01:48:20   called iTunes, on my Mac while I work. That is the use case here. I'm not going to take

01:48:25   out my phone and play with the app. I'm controlling it through the music app, period. I get that,

01:48:32   but they have a native app. Well, I don't know if I should say native. They have an

01:48:36   app for the Mac. I heard rumblings that it's going away. This is not insider information.

01:48:43   I think they might have announced this at some point. Their app for the Mac is fine.

01:48:47   I wouldn't say it's great, but it resolves a lot of these issues that plague you and

01:48:51   it will interface with your library. Now, a library of your library size? I don't know,

01:48:56   but it is supposed to interface with your iTunes match library. Just briefly, I think

01:49:04   I talked about this. I might have talked about it on the show and certainly talked about

01:49:07   it on analog, but the new app, it's fine. Well, as long as you don't need accessibility

01:49:13   features, which are coming slower than they should, but it's fine. The only thing that

01:49:17   really chaps my bottom about it is you can't do any queue management or playlist management.

01:49:22   Or not playlist, I guess queue is a better word for it. So if I really, really, really

01:49:25   want to play a particular song next, there's nothing I can do about that except run up

01:49:30   to my computer and use the old app to do it, which stinks. And that's coming and that will

01:49:34   come soon-ish. But for the most part, I don't mind the new app and I think it is less clunky

01:49:39   than the old app. It's very different than the old one, so it requires relearning some

01:49:43   stuff. But I use the Sonos app to play music almost exclusively. The only time I really

01:49:50   don't is, like I said, when I'm at my computer. And I agree with you, ultimately, that I find

01:49:54   it easier to find the music I want using the music app, and that's what I do when I airplay

01:49:58   it to the move, and that's fine. But especially if I'm just going for ambient music and not,

01:50:04   you know, as I'm around the house or whatnot, or you know, just want to play an album or

01:50:08   something like that, oftentimes I'll use the Sonos app, including sometimes on the computer.

01:50:12   - I mean, that is an option, but that is just, that is a direction I don't want to go. I

01:50:17   don't want to play things that way. That is not my workflow. I don't want to change my

01:50:21   workflow. I think my workflow for my needs and preferences is better. So yeah, I just

01:50:26   don't want to deal with that. And honestly, so I did hardwire them. They each have ethernet

01:50:31   going to them. Nice and clean, of course. So they each have ethernet. The one has the

01:50:37   line in it and sends it to the other one over the network. And when they were both on Wi-Fi,

01:50:42   it was a little bit shaky. I had to use the higher latency settings. It wasn't good. Sonos

01:50:46   products don't have great Wi-Fi radios. But when I wired them, they became rock solid

01:50:54   reliable. Like, I've never had one, since wiring them to the network, I've never had

01:50:58   one like drop out or be, you know, too out of sync or whatever. It's never happened.

01:51:03   So it is fantastic. Anyway, so if you're willing to do this crazy setup, which again, this

01:51:09   is ridiculous. You really shouldn't do this. But if for some reason you're crazy like me

01:51:12   and you want to do this, it sounds great. In most ways, it sounds I think even better

01:51:19   than my Q150s. Not at mid range and vocals and guitars. The Q150s still are my favorite

01:51:26   sound for that. However, they are bigger, they are deeper, they are uglier. And the

01:51:34   challenge with the Q150s is that they're really not made for desk listening distance. And

01:51:39   they have a fairly small sweet spot in speaker terms. This is like, how much does the sound

01:51:45   change or get worse if you like shift your body to different directions or like are leaning

01:51:50   or whatever. Like, is there like how big is the sweet spot where it sounds the best? Most

01:51:55   speakers are not great at this. The Q150s are especially at a distance of like four

01:52:02   feet. They're not really made for that. It doesn't have a very big sweet spot. The

01:52:08   ERA 300 pair, each speaker has like seven drivers firing in different directions. There's

01:52:15   a lot going on there that fires sound in a whole bunch of different places. And so as

01:52:18   a result, the sweet spot is very wide and broad. It also, in speaker terms, the ERA

01:52:27   300 has a really impressive sound stage. Again, just by its design. This is what this means

01:52:32   is like, how wide or big does it sound like the sound is coming from? Does it sound like

01:52:40   it's coming from two points in front of you? Or does it sound like you're in an auditorium

01:52:45   full of sound coming from the big wall in front of you or whatever? That's the sound

01:52:50   stage. These have a massive sound stage for speakers that fit on your desk and that you're

01:52:55   listening to from four feet away. And again, because there's drivers firing in every direction

01:53:01   and they're doing all sorts of processing, it is really impressive for that. By far the

01:53:06   best sound stage I've heard in speakers at this distance. And they have pretty good bass

01:53:13   for their size. Especially considering not having a subwoofer, they have very impressive

01:53:20   bass. They destroy the Q150s in bass and that they even destroy the Q350s in bass, which

01:53:25   are bigger speakers. They have really good bass for their size. And so you don't need

01:53:31   a subwoofer. However, then I added a subwoofer. Which one? The sub, of course. The big one?

01:53:41   Yes, because that was also part of the set that I got for the living room. Oh, I bet

01:53:46   that sounds so flipping good. Oh my gosh. So the thing about the Sonos Sub, this is

01:53:52   their big subwoofer, the one they've had for a while. It is force cancelling. I believe

01:53:56   I discussed this a long time ago when I discussed my Q150s and my setup at the beach and I bought

01:54:01   this very expensive KEF subwoofer for the Q150s at the beach because it was force cancelling.

01:54:08   The Sonos Sub only works with Sonos. It is only their wireless protocol. Although, you

01:54:14   can network wire it, but it doesn't have a line in is what I'm saying. So it only works

01:54:19   with Sonos stuff. So I couldn't use it at my setup at the beach, but I could use it

01:54:24   here. And it is a really good force cancelling subwoofer for less money than, as far as I

01:54:30   know, any other force cancelling subwoofer on the market by a pretty big margin. And

01:54:35   what force cancelling is great for is, and this is why Apple always touts it in the MacBook

01:54:39   Pros because I believe they also have force cancelling subwoofers in their laptops. What's

01:54:43   great about that is that it significantly reduces buzzing and vibration from subwoofers.

01:54:50   You hear the sound, but because it has two drivers firing in opposite directions, the

01:54:56   vibration is largely or completely cancelled out. So you don't usually hear too much

01:55:03   boominess, you don't really hear materials nearby vibrating as much, it doesn't vibrate

01:55:08   with the floor because it is cancelling out by doing two different directions at the same

01:55:11   time. So these speakers, the Herr 300s, again, by themselves, they have a great bass. With

01:55:18   the subwoofer, it's a lot of fun. Like, this is a ridiculous setup. Again, nobody needs

01:55:24   to do this. I don't need to do this. I happen to have these things for my living room and

01:55:30   I decided to set them up in my office for a while first. And it's a lot of fun. The

01:55:36   subwoofer is, it's not, honestly, it's not a massive difference because the bass in

01:55:40   them is already pretty good. But it does improve things and it does make it a lot of fun. I

01:55:47   don't need it, but it's really fun.

01:55:49   Well, I am glad you are satisfied. So, sitting here now, when the living room is ready, are

01:55:55   you going to then buy another batch of all these things?

01:55:58   I mean, I already have the soundbar for down there. Maybe that'll be enough.

01:56:01   I like how the original motivation was, "These speakers are too big on my desk." And you

01:56:05   replaced it with two, I think, also very big speakers and a giant subwoofer.

01:56:11   It's giant for that application. I wouldn't say it's that giant, particularly for a home

01:56:16   theater, but for that application, it's not small.

01:56:18   You should do the measurements. How many square inches of speaker did you remove and how many

01:56:22   square inches of speaker did you add in their place?

01:56:24   Well, I relocated it. The subwoofer is across the room, like, at the opposite wall that

01:56:28   I'm looking at. And then the two speakers are, you know, they're tipped over to their

01:56:31   sides. The only one, you know, weird thing about these speakers is that because of their

01:56:37   design, I can't pile crap on top of them. I learned that.

01:56:40   That's what I'm saying. They're weirdly shaped. Plus, there's up-firing drivers anyway, isn't

01:56:44   there?

01:56:45   Yes. But there is probably aesthetic benefits to me not being able to pile crap on top of

01:56:50   my speakers. Like, at the beach with the KEFs, like, those are just boxes. And so, the top

01:56:54   of them becomes a work area. Like, there's always papers on them. When I was burning

01:56:59   the Blu-ray discs to back up all my stuff, the M discs, the Blu-ray drive just lived

01:57:05   on top of one of the speakers. Like, there's always crap on top of my speakers at the beach.

01:57:10   And here, I can't do that.

01:57:11   That's because they take up so much desk space. There's no desk space left to put stuff, so

01:57:15   you have to put it on top of the speakers.

01:57:19   This is fun. And I get to watch the ducks while I listen to my awesome music on my ridiculous

01:57:22   setup, so I'm happy.

01:57:23   [beeping]