MacBreak Weekly 1044 Transcript

Please be advised that this transcript is AI-generated and may not be word-for-word. Time codes refer to the approximate times in the ad-free version of the show.

Leo Laporte [00:00:00]:
It's time for MacBreak Weekly. Andy Iannaccone is here. Christina Warren's here. Jason has the week off, but that's given us a chance to invite Federico Viticci. He's the editor-in-chief of MacStories. He's going to join us from his home in Rome and talk about the new Mac M5 Ultra. He's using it for local AI. What's it take? How much is it? And is it worth the purchase? That and a whole lot of Mac news, including a John Ternus reorg. Coming up next on MacBreak Weekly.

Leo Laporte [00:00:40]:
This is MacBreak Weekly, episode 1044, recorded Tuesday, September 29th, 2026: Live, Love, Re-Org. It's time for MacBreak Weekly. Yes, time to talk Apple, and we've got a great show planned for you. First, let me introduce Christina Warren, Developer Relations at GitHub. Hello, Christina.

Christina Warren [00:01:02]:
Hello.

Leo Laporte [00:01:02]:
Wearing a 6-color shirt. I like that, all the apples.

Christina Warren [00:01:06]:
Yeah, yeah, this one is, uh, it's one of the new additions. I guess they added it, um, a couple months back to the company store, and so that was courtesy of Mr. David Murphy. Uh, so thank you, David, for that. And, uh, and, um, uh, Zarf in the chat got me a water bottle, so between, between the two of them.

Leo Laporte [00:01:22]:
From the event?

Christina Warren [00:01:24]:
Yeah.

Leo Laporte [00:01:25]:
From WWDC?

Christina Warren [00:01:26]:
Yeah, I think, yeah, from the company store. I guess when they, when they released the new stuff there. So yeah, but this is the first time I've had a chance to wear this crew neck on the, on the pod. And yes, it's quite good.

Leo Laporte [00:01:37]:
Yeah. Also here from the library, Andy Ihnatko of ihnatko.com. Hello, Andrew.

Andy Ihnatko [00:01:42]:
Hello.

Leo Laporte [00:01:43]:
Hello.

Andy Ihnatko [00:01:44]:
I'm only wearing— well, like, the thing is, like, if black is the absence of color, I might be making up for the other color. So I'm sorry, I'm sort of like The black hole of joy.

Leo Laporte [00:01:54]:
There's none darker.

Andy Ihnatko [00:01:55]:
I've been described that way before though.

Leo Laporte [00:01:57]:
Yes. Mr. Jason Snell is in Memphis, Tennessee right now for the Relay FM Marathon for St. Jude. They are— they've already raised millions of dollars and their event is coming up. Was that Wednesday? I think it is.

Federico Viticci [00:02:14]:
Friday?

Leo Laporte [00:02:15]:
I don't— Friday? Sometime this week. If you go to relay.fm, you'll find. out all the information, and that's where Jason is. So we brought in a ringer, and I'm, uh, pretty excited about this. Ladies and gentlemen, say hello to editor-in-chief of MacStories, Federico Viticci, from Roma.

Federico Viticci [00:02:34]:
Hello, thank you for having me on. I hope I will not disappoint all the Jason fans.

Leo Laporte [00:02:39]:
Oh, the Jason fans. Well, they'll, they'll get more Jason next week, but I'm actually thrilled to talk to you Because of course you were really the first out of the block with a review of the M5 Ultra with regard to local AI.

Christina Warren [00:02:54]:
Hmm.

Leo Laporte [00:02:54]:
And, uh, and I think there's a lot of people, myself included, who are sitting with our fingers poised on the buy button trying to decide. Uh, Micah Sargent, uh, did buy an M5 Ultra, but he, he, he was— he used our credit card, so he was, he was a little— my I wouldn't have been so circumspect. He only— I think he only got 96 gigs of RAM, which is nuts. Nuts. So you could go all the way up to 256 gigs if you were willing to spend $15,000 of Leo's money. That's what, uh, well, Federico, I, I doubt you— you, you got Apple to give you one, I'm sure.

Federico Viticci [00:03:29]:
No, that was a review unit. It's not, it's not mine.

Leo Laporte [00:03:33]:
Yeah, yeah. But what I love about it is you wasted no time. You didn't do what Marques Brownlee did and, you know, try Final Cut on it. No. You jumped right into local AI. And your story at MacStories is subtitled, The Dream Mac for Local AI Agents. Look at that beautiful little Mac Studio sitting on your desk right there.

Federico Viticci [00:03:59]:
My desk is usually not that minimal, but, you know, I had to take a good picture.

Leo Laporte [00:04:05]:
So you have the top of the line, right? The top of the line processor?

Federico Viticci [00:04:08]:
Currently, yes. It's the 80-core GPU, 256 gigs of RAM. The 512— there's gonna be a 512 gig version of RAM, but it's coming out in late October, and that was not available to reviewers at the moment. So they just sent out a few of these 256.

Leo Laporte [00:04:26]:
Ironically, it'll be probably faster for me to get the 512 now than to order the 256 because I think it's pushed to February.

Federico Viticci [00:04:34]:
Oh, okay.

Leo Laporte [00:04:36]:
Yeah, so, well, God only knows if, if, if the 512 does— we don't know, they didn't say a date, but late October If, uh, if it comes out, you may press the buy button and they'll say, good, yours is coming in June. We don't, we don't know. Ramageddon is hitting everybody, uh, pretty hard. Apple, uh, no less. So you got it, you open it up. And, uh, now had you done a lot with, uh, Local AI yet or?

Federico Viticci [00:05:02]:
Yes.

Leo Laporte [00:05:03]:
Okay.

Federico Viticci [00:05:03]:
So, so I, I came prepared. Uh, so I had a whole plan ready to go because I knew that I was gonna be under a, you know, in a bit of a time crunch. The review unit arrived on Thursday and the embargo was on Monday. And I knew that—

Leo Laporte [00:05:18]:
That's not a lot of time.

Federico Viticci [00:05:19]:
That's not a lot of time, especially with the weekend in the middle. And we had weekend plans. I was like, what am I going to do? So I was prepared in, I guess, in 2 different ways. The first one is that I had done throughout the entire summer a lot of work with local AI. And maybe we can talk about that later.

Leo Laporte [00:05:38]:
Yeah, on what? On Macs or?

Federico Viticci [00:05:42]:
So I still have the previous review unit from Apple, the M3 Ultra Mac Studio with 512 gigs of RAM.

Leo Laporte [00:05:50]:
Very, very up till a couple of weeks ago. Very nice.

Federico Viticci [00:05:55]:
Very nice machine for local AI. Not the fastest, especially with the latest models, but I have been running this Mac Studio in my storage unit.

Leo Laporte [00:06:06]:
Wait a minute, why in your storage unit?

Andy Ihnatko [00:06:09]:
Well—

Leo Laporte [00:06:09]:
Is it hot?

Andy Ihnatko [00:06:10]:
You know, people use it to run their meth empires from.

Federico Viticci [00:06:14]:
If it can't run AI, why are you cooking? These computers get really hot, especially when you load a big model and you have it run, which is what I did, 24/7 for 3 months.

Christina Warren [00:06:27]:
Whoa.

Federico Viticci [00:06:28]:
So for 99 days, So, every year, I publish my annual iOS and iPadOS review on MacStories. And it's a big project. I start testing the new version of iOS in June at WWDC, and in September, I publish a review. And this story requires a lot of notes, a lot of research, screenshots, lots of notes coming together in September in a single article on the site. And this year I did something else. I created the entire thing in Notion. So I was using Notion as my database, but on the M3 Ultra Mac Studio, I had a local model that every 10 minutes would wake up and take a look at my Notion database for new PDFs, new articles that I clipped from my web browser, new notes, new screenshots. new beta releases from the apple.com servers, and it would sort of make sense of all of that.

Federico Viticci [00:07:29]:
And it would automatically cross-reference a feature to a chapter of the review or a session from WWDC to a feature of, I don't know, the Calendar app. And this model ran for 99 days, 24/7. I guess I had a couple of power outages when there were a couple of thunderstorms in the summer. But that's why I Brought it down to the storage unit because it was impossible to keep it in my bedroom. My girlfriend got really upset that it was getting really, really hot.

Leo Laporte [00:08:01]:
See, Macs are not known for getting hot. I'm kind of surprised.

Federico Viticci [00:08:05]:
Well, when you load up DeepSeek 4.1 Flash—

Leo Laporte [00:08:08]:
And you run it 24/7.

Federico Viticci [00:08:10]:
And you run it 24/7 for 99 days, that's hot. That's really hot.

Leo Laporte [00:08:17]:
So you put it in the storage— But, and then what? You had a long Ethernet cable?

Federico Viticci [00:08:22]:
Oh, see, there's a whole story.

Andy Ihnatko [00:08:25]:
Don't say anything that's gonna get you kicked outta your storage unit. I'm just asking you.

Federico Viticci [00:08:30]:
So it was—

Leo Laporte [00:08:31]:
What's that blue wire hanging outta the garage door there? Hold on, hold on.

Federico Viticci [00:08:34]:
The story goes much deeper than you may initially think. So good question, right? How do you actually get the internet 5 stories down, right? Oh, God.

Leo Laporte [00:08:45]:
At least it's in the same building. That's a good start.

Federico Viticci [00:08:47]:
It's in the same building. So I, asked for some advice to some of the neighbors. And I was talking to this guy who had an Ethernet cable run from the 6th floor. I'm on the 4th floor. This guy's on the 6th floor.

Christina Warren [00:09:02]:
So we have to talk about this.

Leo Laporte [00:09:03]:
Remember, this is Rome. And so if you walk down any street in Rome, there's antennas, there's laundry, there's wires. It really, it probably is invisible in that environment. Yeah.

Federico Viticci [00:09:16]:
Well, I'm more in a modern neighborhood.

Leo Laporte [00:09:19]:
Oh, okay.

Federico Viticci [00:09:20]:
It's a modern building.

Leo Laporte [00:09:21]:
So people would notice is what you're saying. Yeah.

Federico Viticci [00:09:23]:
So here's what I did. I had to get an electrician involved. We purchased 100 meters of ethernet cables.

Christina Warren [00:09:34]:
Yep.

Federico Viticci [00:09:35]:
And we ran it from an ethernet switch in my apartment, 5 stories down to the storage unit.

Leo Laporte [00:09:42]:
Did you drill holes in the floors? How did you?

Federico Viticci [00:09:44]:
No, we had conduits, thankfully.

Christina Warren [00:09:46]:
Okay. Yeah, that makes sense.

Leo Laporte [00:09:48]:
So there was already a hole and a tube?

Federico Viticci [00:09:51]:
Yes, it was a really long tube.

Andy Ihnatko [00:09:54]:
Can I just say that 100 meters is the distance from like my house to like the coffee shop?

Leo Laporte [00:10:01]:
It's a football field, I think.

Andy Ihnatko [00:10:03]:
I sometimes work in like, okay, I'm just trying to picture this.

Federico Viticci [00:10:07]:
It was a day-long process. It was a day-long process.

Leo Laporte [00:10:11]:
Fishing it through the conduit.

Federico Viticci [00:10:12]:
Yes, that's the hard part. These 2 guys come to my apartment with like this giant spool of Ethernet cable. They're like, we're gonna run it down. And so we installed this Ethernet switch by the front door.

Leo Laporte [00:10:24]:
Did they attach a small rodent to the one end?

Andy Ihnatko [00:10:27]:
This is like a bank heist in the '70s. I want more details.

Federico Viticci [00:10:32]:
They have what they call a pilot cable.

Andy Ihnatko [00:10:33]:
I was disguised as a flower delivery man.

Leo Laporte [00:10:35]:
A pilot cable, okay.

Federico Viticci [00:10:37]:
They have a pilot cable that has like an iron head of sorts. And that allows the cable to go into the conduit and act as a pilot. So they attached the Ethernet cable to the pilot cable and basically pushed it down.

Leo Laporte [00:10:51]:
Nobody makes a 100-meter wire hanger, so you have to have some sort of device like a wire hanger. Yeah, so push it that way.

Andy Ihnatko [00:11:00]:
So that's what tipped off the cops when the huge wire hanger showed up at the delivery station.

Federico Viticci [00:11:06]:
So 8 hours later, and meanwhile I'm trying to be a good host, right? So I'm making the guys coffee and like, you guys need anything? So 8 hours later, it's 8 PM, they're finally done. but I have the internet downstairs and it's my same connection, right? It's technically, it's my local network. So I, I had another, uh, I have a TP-Link Deco mesh system. So I had another satellite unit downstairs. It appears as it's another room of my house.

Leo Laporte [00:11:36]:
Nice.

Federico Viticci [00:11:37]:
Technically it's outside, but it's my local network. And so I brought the Mac Studio down and that's where it's been running.

Leo Laporte [00:11:43]:
Importantly, You're running it headless because you can't, you're not gonna have macOS on a screen. There's no keyboard or mouse down there, or maybe there is, but you don't—

Federico Viticci [00:11:53]:
Well, there's a keyboard that I use every once in a while, just in case for authentication.

Leo Laporte [00:11:59]:
Oh, I would hate that. Every once in a while I have to run upstairs and, you know, put my YubiKey into the port. I gotta go 5 flights. Honey, I'll be back in the morning. Make me a lunch.

Federico Viticci [00:12:14]:
At the very least, the apartment is much cooler. now because there's not computers anymore.

Christina Warren [00:12:18]:
Yes.

Federico Viticci [00:12:19]:
So that's the computer that's been run. And this was just one part of the story. I was familiar with local AI because I've been doing this. I had this entire system work 24/7, and I just wanted to have something practical to show for local AI because a bunch of people talk about local AI and it's always like, oh, I made this cool SVG of a pelican. riding a bicycle with local AI. Like, great.

Leo Laporte [00:12:45]:
Simon, Simon Willison's famous pelican benchmark.

Federico Viticci [00:12:49]:
Yeah. Shout out to Simon. But like, I wanted to have something real practical. And there was a second—

Leo Laporte [00:12:54]:
So you were using Deep— actually, DeepSeek V4 Flash was pretty good. Were you using 0731, the July? That's actually a pretty good model. I used that for a while.

Federico Viticci [00:13:02]:
Pretty good model. Not the fastest on the M3 Ultra. So for example, I would clip an article about iOS 27 in Google Chrome, save it to Notion, 5 minutes later, the local agent wakes up and it takes 5 or 6 minutes to process the query.

Leo Laporte [00:13:17]:
But you don't care because you're not sitting in front of it.

Federico Viticci [00:13:19]:
But I don't care. Exactly. Because it's working in the background. And I also started doing this because I wanted to save money. The very first version of this system in June was based on Notion AI and Notion custom agents.

Leo Laporte [00:13:35]:
Oh, that could happen.

Federico Viticci [00:13:36]:
I depleted my quota of Notion agents in 2 days. And I realized—

Leo Laporte [00:13:40]:
I thought you were an Obsidian guy because I, by the way, I subscribe to MacStories for your Obsidian series. I thought that was really good.

Federico Viticci [00:13:49]:
Like many people, you know, we all have a problem with switching apps and trying things.

Leo Laporte [00:13:57]:
But I went from Notion to Obsidian as well, by the way. I'm just, I'm very sympathetic. I love Notion actually. Notion's great.

Federico Viticci [00:14:04]:
And so that's why I also realized it could be a cool experiment for Local AI. And it's also a way to save money because I guess all that I'm paying is my electricity bill. Uh, but I'm not paying hundreds of dollars for this.

Leo Laporte [00:14:15]:
Did your bill go up?

Federico Viticci [00:14:17]:
Uh, yeah, my electricity bill? No, no, no, no, not really.

Leo Laporte [00:14:22]:
I mean, that's the good thing about Macs. They're, they're not super power hungry. I'm surprised that they're so hot, but I guess if you're running it 24/7.

Federico Viticci [00:14:29]:
Well, look, it gets warm.

Christina Warren [00:14:30]:
It doesn't have a big exhaust.

Federico Viticci [00:14:32]:
It doesn't have a big exhaust. If you place your hand above it, you could feel it. Yeah, you can feel it, but obviously it's not as hot as my PC with the 5090 in it.

Leo Laporte [00:14:43]:
Right, no, exactly.

Federico Viticci [00:14:46]:
When I was testing the M5 Ultra last week, which brings me to the second part of this story, I came prepared. So I was, I knew what I was doing because I've been playing around with the M3 Ultra. But since I knew that I was gonna be under time pressure to get this article done in time, I pre-designed with Codex a, my own testing harness.

Leo Laporte [00:15:11]:
Oh, smart.

Federico Viticci [00:15:12]:
for the M5 Ultra. The idea was I'm going to fire up Codex on the new computer. I'm going to create all the connections necessary between Codex on the new computer, Codex on the old computer, and Codex on Windows on my gaming PC. And these 3 instances of Codex are going to be talking to each other and they will coordinate. It sounds like science fiction. It actually worked. They're going to coordinate the benchmarks. And I'm just gonna sit in the office and watch the numbers and, you know, open the window every once in a while because it got really hot, especially when the 5090 got involved.

Leo Laporte [00:15:52]:
Yeah, I am sitting in front of a stack that has 2 Sparks, a Mac Mini, and a 3090 running in my old gaming machine. But I also have fans. I have a window open right there and it's still, it does, it gets warm. It really does get warm.

Federico Viticci [00:16:08]:
Yeah. Yeah.

Leo Laporte [00:16:10]:
So I think that's smart. And it's quiet too. It's down in the basement. It's not, it's not going to bug anybody.

Federico Viticci [00:16:14]:
Well, obviously I kept the M5 Ultra in my house because it was an unreleased product.

Christina Warren [00:16:20]:
I was going to say it's a review unit. You don't want to have that in the basement.

Leo Laporte [00:16:23]:
Yeah. Apple wouldn't want that.

Federico Viticci [00:16:25]:
Yeah. So I kept it in the office and yeah, I coordinated with these 3 models running on 3 different machines.

Leo Laporte [00:16:33]:
Yeah. Did you know ahead of time what you were, what So you had the benchmarks. Did you know what model you were gonna put on there?

Federico Viticci [00:16:39]:
So that was a bit of a challenge. And that's why the article, I mean, it was pretty successful, but it will need an update. And that's the whole thing with Local AI on Apple platforms. It's very community-driven. And that's both a great advantage, but also when you have a review unit, there's no community to help you.

Leo Laporte [00:17:03]:
Oh, good point. You can't say, Hey, how are you guys?

Christina Warren [00:17:05]:
It's not going to be optimized for that out of the gate. Yeah, yeah.

Leo Laporte [00:17:09]:
And that's one advantage of using, uh, Sparks is there's a— it's been around for almost a year and there's a large, vigorous community. The disadvantage, Federico, is there's a new model every freaking day, pretty much overnight every night. My guys call it a bake-off, same thing as you have. We have our own personal benchmarks and they do a bake-off trying to decide which model I should run.

Federico Viticci [00:17:33]:
Yeah.

Leo Laporte [00:17:34]:
So you must have had in your head, were you going to use DeepSeek? What were you going to do?

Federico Viticci [00:17:39]:
So I figured I'm not going to have the 512 gig version. So big models are going to be a bit of a struggle.

Leo Laporte [00:17:46]:
Right.

Federico Viticci [00:17:47]:
And I landed on Qwen 3.8 FlashNext. This is one of the new Alibaba models from the Qwen family of models. And it's 100 and 20 billion parameter model with only, I believe, I want to say 20 active. It's a mixture of experts model, so only a few of those parameters get activated at runtime. It's a medium-sized model, I would say. I did get some guidance behind the scenes. I sent a DM to the developer of the one of the backends that you can use on macOS to run models with the MLX Apple framework.

Leo Laporte [00:18:33]:
So we should talk about that as well.

Federico Viticci [00:18:35]:
We should talk about that.

Leo Laporte [00:18:36]:
Basically, I'll give you a thumbnail sketch. You've got the model, but you need something to run the model. And a lot of people, you've probably, if you've done any of this, you've seen LM Studio, Oro Llama. There are a variety of, you know, these are GUI programs you can run a model in. If you're going to run it headless, there are other things you can use like VLLM, or I would assume OMLX is what you're running. Okay. And, and the— and that's important because most of the stuff that's out there is not tuned for Apple's stuff. It's tuned for NVIDIA's CUDA.

Federico Viticci [00:19:10]:
Yes, it's tuned for CUDA.

Leo Laporte [00:19:12]:
So if you, if you want to get the most out of this Mac, you're going to want something tuned for MLX, right?

Federico Viticci [00:19:18]:
Yes. And my challenge was I gotta get something tuned for MLX and I gotta get something that is gonna run well on the M5 Ultra, which is a machine that nobody else has at the moment.

Leo Laporte [00:19:33]:
You can't go to X and say, hey, what are you guys using?

Federico Viticci [00:19:38]:
So, but I did send a cryptic message to the OMLX developer. I was like, hey, big fan. Hypothetically, if, you know, down the line I were to, you know, cover the M5 Ultra just so that I can be prepared for the next month, what kind of model do you think I should load?

Leo Laporte [00:20:08]:
Cagey but smart.

Federico Viticci [00:20:10]:
And so I don't know if the person who, by the way, has been hired now by Hugging Face, the open source portal. Which means he's been hired in a way by OpenAI, right?

Christina Warren [00:20:20]:
Right? No, by Stripe.

Leo Laporte [00:20:22]:
Oh, Stripe bought Hugging Face.

Andy Ihnatko [00:20:24]:
Yeah.

Leo Laporte [00:20:24]:
No, no, NVIDIA, I mean.

Christina Warren [00:20:26]:
NVIDIA.

Leo Laporte [00:20:27]:
NVIDIA.

Federico Viticci [00:20:27]:
NVIDIA.

Leo Laporte [00:20:27]:
Sorry, sorry, sorry.

Christina Warren [00:20:28]:
I was thinking OpenRouter, which was Stripe. Sorry, sorry.

Federico Viticci [00:20:30]:
Right. OpenAI did hack Hugging Face, so maybe that's the difference.

Leo Laporte [00:20:36]:
That's where I was confused. Yes, I linked them together in my mind. No, and I don't think that that purchase probably has gone through yet, so he's probably still working for Hugging Face, but eventually he'll be working for NVIDIA, which is a nice place to be if you're creating OMLX. Wait a minute. No, it's not. Well, we'll see. Anyway, he was helpful.

Federico Viticci [00:20:54]:
He was helpful because he told me like, oh, I also have an M5 Ultra on order from Apple. I was like, oh yeah, yeah, yeah, yeah, yeah, me too.

Leo Laporte [00:21:03]:
Same. Of course you do. We all do.

Federico Viticci [00:21:05]:
And he said, I think one of the first tests that I'm going to do since the architecture is essentially, it's like what, 4 M5 Maxes sort of glued together. It should scale linearly from the M5 Max, and OMLX is already tuned for the M5 Max. So maybe just try that and try this QWEN model because I think it's going to run well. And that was correct. But there's a problem. It's been a week since I published the story and the benchmarks the numbers have already doubled because more people have gotten their hands on the M5 Ultra and the MLX backend has already been updated. And these little engines like OMLX, there's another called MLX Serve, similar idea. It's a program that lets you run these models.

Federico Viticci [00:22:04]:
They've already been updated in the span of 7 days. They've been updated. And I reran my benchmarks and I get double the performance from a week ago.

Christina Warren [00:22:15]:
Wow.

Leo Laporte [00:22:15]:
There's even a new thing that's coming out that's a little weird. Ash is doing this TensorFold thing.

Federico Viticci [00:22:20]:
Yes.

Leo Laporte [00:22:21]:
And that looks like it might double it again.

Federico Viticci [00:22:23]:
Yes.

Leo Laporte [00:22:24]:
But here's— this is, to me, this is good news because as a Spark owner, my concern with the Mac was— this is going to get a little technical— prefill. So we knew the Mac would decode very fast, right? A lot of times you see people measure the speed of the AI in tokens per second. And sure, that's what you see coming up on the screen. What you don't see is how long it's taking to ingest—

Federico Viticci [00:22:48]:
Yes.

Leo Laporte [00:22:48]:
The prompt and the context and all that stuff you're feeding to it before the tokens per second. And if that's slow, it means you're going to sit there for minutes before you see the fast tokens per second. And Apple's While their specs, their bandwidth specs are great for tokens per second, the M3 Ultra was not great on this prefill, this context caching. And so there was some concern about that, but I think your numbers show that it's actually pretty good.

Federico Viticci [00:23:19]:
It's much better than the M3 Ultra, which is, you know, I mentioned on the M3 Ultra, I had that DeepSeek agent and it would just take several minutes to, not even just to process, a note in my Notion, but just to start, just to read the prompt.

Leo Laporte [00:23:37]:
Just to think about it.

Federico Viticci [00:23:39]:
Just to think about it. And it would take several minutes and then it would start. And on the M5 Ultra, the prefill stage, so how quickly can the agent running locally process your prompt and then start displaying the first token, whether it's a visible token or a reasoning token, the time has gone down significantly. And the reason why I, I talked about running local AI agents in my review. Well, first of all, I'm not a video maker. I don't know, you know, there's other people reviewing the M5 Ultra for video and 3D graphics, that sort of thing. I have been trying to use a local agent as a chatbot running locally for the past year or so. And on the M3 Ultra, it was just horrible.

Leo Laporte [00:24:30]:
It's painful.

Federico Viticci [00:24:31]:
Because you will ask a question and then just stare at your screen for a minute, be like, okay, this is kind of useless. Now, especially a week has passed, and especially now with the updates that we've already gotten from the community, the prefill stage has gone up to, you know, in my review, I think I benchmarked 3,000 tokens per second prefill. It's now well over 5,000.

Leo Laporte [00:25:00]:
And that's fast. That's really fast. Yes. Yes.

Federico Viticci [00:25:04]:
So even if you set a local model to extra high reasoning, and even if you use something like OpenCLAW or Hermes agent, these agents that have a long system prompt with lots of— lots of instructions.

Leo Laporte [00:25:19]:
Yeah.

Federico Viticci [00:25:20]:
Now it's no longer awful to use a local AI agent with it because it's fast enough To answer you, to begin answering you in 10 or so seconds.

Leo Laporte [00:25:33]:
Yeah. So, uh, by the way, you're never going to be able to finish this review because tomorrow there'll be another update and another update.

Federico Viticci [00:25:43]:
Uh, yeah, I—

Leo Laporte [00:25:45]:
okay. And I know there are people going, well, okay, fine, but how does it do with Final Cut? And I'm going to echo Marques Brownlee, who said You don't need this for— you don't need an Ultra if you're doing video editing. This, this is an AI. If you're gonna get an M5 Ultra, it's because you're getting it for local AI. There's really no other job you could do on this thing that needs it. You can get a Max, right? Or you could probably get by with an M6 Mini, which is fine. They're very good.

Federico Viticci [00:26:15]:
Yeah.

Leo Laporte [00:26:15]:
Uh, would you not do— by the way, would you not do local AI on an M6?

Federico Viticci [00:26:22]:
I think for laptops and something like an M6 Mac Mini, I'm very fascinated by this new wave of small language models.

Leo Laporte [00:26:33]:
Yeah.

Federico Viticci [00:26:35]:
There's this company called Desert Ant Labs. They make a collection of small models, like really small models, really specialized models. For example, they have a model whose entire purpose is to remove the verbal tics from audio, like um or you know, like that sort of thing.

Andy Ihnatko [00:26:58]:
Mm-hmm.

Federico Viticci [00:26:58]:
It's a specialized language model that runs locally. It's very small, very fast. So I think for those machines, you're not gonna be able to run DeepSeek. You're not gonna be able to run QuEN 3.8. You may be able to get a really optimized 4-bit model version of a small model, like a 9 billion parameter model, I'm sure. Does that make sense to use for an agent? Probably not, right? Because it doesn't have that kind of capability of going back and forth with answers, long context, tool calling, that sort of agentic behavior. But I am fascinated by this new sort of trend of specialized local small models for very specific tasks. So it's always going to be a moving target, I think, especially in the MLX community.

Federico Viticci [00:27:55]:
The fact that for a machine that very few people have, in a week the performance has doubled. I mean, it's not impossible, but it's very difficult to keep up. And so I don't know, I would check in again on the M6, you know, maybe in 3 months and see what can you really do with MLX on an M6 Mac mini. Probably not a medium to big model, obviously.

Leo Laporte [00:28:21]:
Just because of memory.

Federico Viticci [00:28:23]:
Just because of memory, right? But it's also not going to be useless.

Christina Warren [00:28:28]:
Yeah.

Leo Laporte [00:28:28]:
Well, and on the other side, the model makers themselves, especially Alibaba, are coming up with better and more efficient models all the time. That may be full capacity. You may be able to run it in a 96 gig or 64 gig Mac Mini, a decent model like when 4.0 Flash next, something like that. So most people are gonna say, how much does this M5 Ultra that you have on your desk, or in your— it's like $15,000, right?

Christina Warren [00:28:59]:
$13K.

Federico Viticci [00:29:00]:
Yeah. Yeah.

Leo Laporte [00:29:01]:
So most people are gonna say, well, wait a minute.

Federico Viticci [00:29:04]:
Yeah.

Leo Laporte [00:29:04]:
If I buy a $200 or even the new $500 a month OpenAI subscription, let's see, $500 a month, that's $6,000 a year. It'll be 3 years of the fastest, high-end, state-of-the-art AI before I pay for that Ultra. Why should somebody run it locally?

Federico Viticci [00:29:27]:
That's the question, right?

Leo Laporte [00:29:30]:
Not for an economic— not for any economic—

Andy Ihnatko [00:29:32]:
Especially considering the Ocean's Eleven sort of routine you had to pull to get the internet down there.

Federico Viticci [00:29:39]:
Yeah. Look, and I would actually add to that, that with that subscription, you will have access to the latest and greatest frontier models.

Leo Laporte [00:29:50]:
At 300 tokens a second.

Federico Viticci [00:29:52]:
GPT-6 Astra at 300 tokens a second, and you have, you know, multiple years until you've repaid your M5 Ultra. Right.

Leo Laporte [00:30:02]:
And if you use DeepSeek or something cheap, it'll be your lifetime. You will never, ever hit $15,000.

Federico Viticci [00:30:11]:
Never. So I think there's a few reasons why you may want to consider it. Well, first of all, you're rich and you don't care. I mean, it maybe sounds— maybe it sounds— look, maybe it sounds silly, but if you have money to burn and you just want to have some fun with local AI, I won't judge you. There's, you know, I'm not a millionaire. There are some of those in the world, and maybe that's a reason.

Leo Laporte [00:30:40]:
I wish I were that rich. I would, I would have— there's a guy, who just bought 32 NVIDIA Sparks. He's calling it— what is it? He's got a great name for it. But I wish I were that rich. It was a nail-biter to spend $10,000 on 2 Sparks.

Federico Viticci [00:30:57]:
Yeah.

Leo Laporte [00:30:58]:
But I am very happy I did. It's not an economic decision by any means, but imagine having what was a frontier model 6 months ago, In your storage unit. A brain in your storage unit that you could talk to.

Federico Viticci [00:31:18]:
The second, more slightly more serious thing I will mention is that there's something about it that as a nerd speaks to me.

Leo Laporte [00:31:27]:
Yeah.

Federico Viticci [00:31:28]:
Does it speak— does it speak to me $15,000? Probably not, but it's still There's something about the principle of it that you're running AI locally that I think is fascinating. Then I will say that it makes sense for a small to medium-sized business to invest maybe in a computer that can be your entire server for your company with storage, right?

Leo Laporte [00:31:57]:
And private.

Federico Viticci [00:31:59]:
And for data ownership and Privacy. So the fact that you can have, as you mentioned, something that would've been considered frontier 6 months ago running, I don't know, GLM 5.3 Flash on it, which runs really well, and have your entire setup be local. Maybe you're dealing with sensitive documents, maybe you're dealing with sensitive data that you don't want to upload to anybody's cloud. And If you think it makes sense for your small business to have that kind of expense and know that this computer will let you run AI with privacy, everything locally for the next couple of years until the M7 Ultra comes out, I think there's an argument there. I would also say maybe you just lease this computer, you just rent it.

Leo Laporte [00:32:51]:
Yeah.

Federico Viticci [00:32:51]:
Maybe you don't buy it upfront. You sign up for the least terrible leasing plan that you can find, maybe that's also an option. I think judging from the kind of people that, that I think have read my story and that I see purchasing M5 Ultras, I think it's a combination of these 2 audiences, the nerds with money to burn because they want to have fun. And small business owners who think, I just want to run AI locally. I don't trust OpenAI. I don't trust Anthropic. I don't want to pay for a subscription. I don't care.

Federico Viticci [00:33:34]:
Or I need to have AI run locally on this machine. And it's also a local AI machine and it's also a computer because most people are not going to take this down to a storage unit in the basement. Most people will just put put it on a desk and connect it to a display, and it's going to be a normal computer for them. So it's a complicated and polarizing conversation. Every once in a while, there's somebody on Twitter saying local AI is useless.

Leo Laporte [00:34:03]:
Matt Schumer yesterday.

Christina Warren [00:34:07]:
Yes.

Leo Laporte [00:34:07]:
But he also runs a cloud AI company. Thank you. I think it's— I have a dog in this hunt.

Federico Viticci [00:34:13]:
Didn't he also try and come out with his own local model a couple of years ago and it was horrible?

Andy Ihnatko [00:34:18]:
So yeah, that's—

Federico Viticci [00:34:18]:
It was really horrible.

Leo Laporte [00:34:21]:
It was like really bad.

Christina Warren [00:34:22]:
Yeah.

Federico Viticci [00:34:24]:
But I think, you know, I don't like in life or otherwise, I don't like to be an extremist person. I don't think Local AI is useless. I also don't think— I also see these people saying, ah, I'm using Local AI, it's Astra level or Fable level. No, it's not. No, it's not.

Leo Laporte [00:34:44]:
So even on a 512 gig M— actually, I've seen people lately say maybe get 2 256 Mac M5 Ultras instead of a single 512. There's even an argument for that.

Federico Viticci [00:34:58]:
So there's this new technology called RDMA that you can actually more easily enable in macOS 27 in settings. Used to be that you had to boot in, what's it called, recovery mode. in macOS and enable it with a terminal command. Now it's an option in system settings under developer options. RDMA is this technology that allows you to connect multiple Macs with Thunderbolt 5, and you're going to have high bandwidth access across these multiple Macs. There's this application called ExoLabs. that is another of these AI local inference engines. It gives you a little GUI that shows you your multiple Macs connected over Thunderbolt 5, and you can basically split inference via Thunderbolt 5 across multiple machines.

Federico Viticci [00:35:59]:
This is Apple's latest favorite demo at WWDC.

Andy Ihnatko [00:36:03]:
Mm-hmm.

Federico Viticci [00:36:04]:
For the past few years, there's been a cluster of 4 Mac Studios in the hands-on area, you know, at the Steve Jobs Theater. And they love to give out this demo where there's 4 Mac Studios connected together with Thunderbolt 5 running a 1.2 trillion parameter model like KIMI-K27, for example, because these 4 computers, they have something like 2 terabytes of memory across the entire cluster. And that allows multiple machines to run that model, which is a giant model with pretty good performance because the memory and the inference is being split at runtime thanks to Thunderbolt 5. Now, again, my main concern here is that you have to— you would have to wait for the MLX community to catch up. Because for example, in my review, I wanted to try this. I thought, oh, I'm gonna daisy chain an M3 Ultra to an M5 Ultra, right? They both have Thunderbolt 5, so why not? And then I thought about it and I did some research and I realized it's gonna be useless because if I daisy chain them with TB5, the memory bandwidth is gonna be capped by the M3 Ultra. The M3 Ultra has a slower memory than the M5 Ultra. And so even if I connect them, it's not gonna be that one computer goes with fast memory and another computer goes with slow memory.

Federico Viticci [00:37:45]:
No, they're gonna baseline at the slowest memory of the M3 Ultra. So I would probably, if your plan is to say, well, I'm gonna wait and get 2 M5 Ultras with 256, I would say maybe wait a little. Yeah, because it gets like— it's already complicated with MLX, and the moment you start adding the daisy chaining on top of it, it gets even more complicated.

Leo Laporte [00:38:12]:
So I'm very grateful that Apple said you can't buy the 512 Leo until October because I might have done something dumb.

Federico Viticci [00:38:21]:
How much is it?

Leo Laporte [00:38:23]:
We don't know, they haven't announced a price, but it's gonna be at least $15K, and I think maybe $20K.

Federico Viticci [00:38:28]:
Yeah.

Leo Laporte [00:38:29]:
$1,000.

Federico Viticci [00:38:30]:
Yeah.

Leo Laporte [00:38:30]:
Uh, but we don't know. Uh, and I'm, I'm very grateful to you because you— your benchmarks and your continued work on this at macstories.net, um, is really valuable to people who are trying to decide. You know, honestly, I might just wait till the M7.

Christina Warren [00:38:47]:
I can—

Andy Ihnatko [00:38:47]:
I can—

Leo Laporte [00:38:47]:
well, I'm very— I, you know, I, I go back and forth, but I think I'm happier that I spent $10,000 on 2 Sparks. You can't do that anymore, by the way. they're out of stock, they're gone, uh, than I would have been had I spent the same amount of money on a 256GB M5 Mac Ultra.

Christina Warren [00:39:09]:
So—

Federico Viticci [00:39:09]:
I'm really curious about the Spark. I would love to try one, and especially a couple of—

Leo Laporte [00:39:15]:
On paper, its memory bandwidth is much slower, but because of CUDA, its prefill is much faster. I'm running GLM. I was running GLM. 5.3 FlashNext on it, but it was a little sluggish, although it's getting faster and faster because there are people every day tweaking it. I don't understand how there could be this many levers, but they're tweaking and tweaking it. And then I've recently switched to a Quen 3.8 FlashNext because it's so much faster, and it's currently running 70 to 100 tokens a second.

Christina Warren [00:39:46]:
Nice.

Leo Laporte [00:39:46]:
Depending on what you're doing. So that's— that feels Snappy. Uh, and, and, and I'm really— tie me to the mast because I don't, I don't need to spend another $20K on local AI, but it's addictive. It's like tattoos. You get the first one and it's hard to stop.

Federico Viticci [00:40:07]:
All right, well, I know. Yeah.

Leo Laporte [00:40:10]:
I've seen your sleeve, so I know too. That's why I made that analogy.

Christina Warren [00:40:15]:
Yeah.

Leo Laporte [00:40:16]:
I thought you would appreciate it on both sides of the equation. So at some point Apple will want this back, right?

Federico Viticci [00:40:24]:
Yeah.

Leo Laporte [00:40:25]:
What are you going to do?

Federico Viticci [00:40:26]:
Oh, you got to give it back.

Leo Laporte [00:40:29]:
You could say, I don't know where it is. Maybe it's in the storage locker. I don't have the key anymore.

Federico Viticci [00:40:36]:
They haven't said— oh no, actually they did say, I think in November, Would you, would you buy it? Probably.

Leo Laporte [00:40:47]:
Um, yeah. You get hooked on it, don't you?

Federico Viticci [00:40:49]:
Yes, I know. I would— I, I think I would actually consider one of those payment plans, maybe.

Leo Laporte [00:40:55]:
Yeah, even Klarna. It would be— I don't know what they do. Do Italy— do you have a Binance?

Federico Viticci [00:40:59]:
Yeah, we have Klarna. Yeah, yeah, yeah, they're Swedish.

Christina Warren [00:41:02]:
They started in Europe.

Federico Viticci [00:41:03]:
Yeah. Oh, I didn't know that.

Leo Laporte [00:41:05]:
Um, you still have to ask because finance is weird.

Christina Warren [00:41:09]:
Yeah.

Federico Viticci [00:41:11]:
Probably consider that. But yes, I'm hooked and I saw the results this summer with the local agents that I had. And there's, it's one of those things where if you are that type of nerd and if you like working with AI, obviously, but there's that, it cannot sort of, uh, it, it plays with that part of my brain with the constant tinkering.

Leo Laporte [00:41:41]:
Me too. And I wake up in the middle of the night.

Federico Viticci [00:41:46]:
Yeah.

Leo Laporte [00:41:46]:
With ideas. Do you do that and say, I gotta try this?

Federico Viticci [00:41:50]:
Oh yeah.

Leo Laporte [00:41:51]:
I mean, yeah, it's a disease. Yes, yes.

Federico Viticci [00:41:54]:
Uh, it's a— some people would say it's a, it's a, it's a form of psychosis. I don't know. It's, um, uh, but yeah, it's, it's a I think for me, the sort of the North Star that I've been trying to keep in mind is it's fine to experiment. Just do it in a way that you can write about it so you can justify the time spent in the name of content. And so that's been—

Leo Laporte [00:42:25]:
Can I tell you when I started in the business—

Andy Ihnatko [00:42:27]:
Timothy Leary had the same idea about LSD.

Leo Laporte [00:42:29]:
Yeah, right. We're talking about the remnant. It's research. No, I started in this business, Federico, before you were born. I started in the '70s, and it was so expensive even to have a floppy disk drive that I started writing for magazines because in my mind, I said, I got to justify this. I got to pay for this because I will always— I knew it then— want the latest. I will always want to have whatever the latest thing is. And now the latest thing is AI and keeping up with AI is, I mean, I know Christina knows this is expensive.

Andy Ihnatko [00:43:06]:
It is.

Federico Viticci [00:43:07]:
It is.

Leo Laporte [00:43:07]:
Even with a corporate sponsor like Copilot, like—

Christina Warren [00:43:10]:
Oh yeah. I mean, I don't even know how much I spend on tokens. And so, I mean, you know, I love Local AI. I don't think that I'd be, well, I'm not in the position to necessarily spend like $10,000 or $15,000 on a—

Leo Laporte [00:43:24]:
I'm spending my kids' inheritance.

Christina Warren [00:43:25]:
I'm an old man. Well, I mean, and I think like it depends on what your use cases are going to be. My, my main concern with it is always with these things thinking about, I guess, like unless to Federico's original point, if you're, if you're very wealthy and this is something that is not going to be a concern for you, great. But even if this scratches your itch, so much is happening so quickly that I do think that it's one of those things where you're like, okay, well, when do we get in? Right? Because there's an argument to be made that, like, for you, Leo, you bought your Spark.

Leo Laporte [00:43:53]:
I was lucky. I bought my Spark in August with the feeling that I don't know how much longer you're going to be able to get these.

Christina Warren [00:44:00]:
Right, exactly. And we don't know how much longer this RAM crisis is going to last. And so you might be— this might be the cheapest you could buy something for the next year. It's unclear. And obviously, if you were able to get like a, you know, an M3 Ultra a year ago, like that would be like a 512 or something like that. That would have been a fantastic investment. It's just one of those things where I think that if you have, like, I think a lot of us who are interested in these things do, if you have that kind of addiction to the latest model, the latest release, the latest, greatest, fastest, we have to kind of, I think, pace ourselves a little bit and go, okay, this is not the sort of thing you can upgrade every single year.

Andy Ihnatko [00:44:39]:
Talk to me.

Christina Warren [00:44:40]:
For most people, like, you can't, right? It's just not one of those things. I mean, if you can do it for business expense reasons and whatever else, if you're able to make that work, all power to you. But most individuals, hobbyists, enthusiasts are not going to be in a position where they can go, okay, this is a $15,000 expenditure that I can make on an annual basis. And yet, as you, as we were kind of pointing out, it's amazing what you can do now, which was the frontier 8, 9 months ago, which is great. Well, we don't know what the frontier is going to be like 8, 9 months from now. And so I think, I think that's, that's where, at least when I kind of look at it, I'm going, okay, I love all this stuff. And I wish that I had a machine like an M5 Ultra to be able to play around with and do things with and that I could make the argument to myself that this is what I need to do and spend my money on. Right now, I can't make that argument, but I very much support people who love that.

Christina Warren [00:45:37]:
But I think it's just one of those things where we are— we don't quite know where we are, I think, necessarily on this, on this hamster wheel to go, okay, is this the right point to jump on or not? Which is what you were saying too, Leo. Like, do you just wait for a— for an M7? Like, I think that that becomes like the hard thing to know.

Leo Laporte [00:45:53]:
But then I will buy it and there'll be an M8 and an M9.

Christina Warren [00:45:56]:
Well, that's the thing. It never stops. And so I think it just kind of gets to a point where it's like, okay, are you at a point where you feel like if this is the max capabilities, they're not going to modify the software anymore? They're never being there.

Andy Ihnatko [00:46:09]:
Well, no, no.

Christina Warren [00:46:09]:
But what I mean, if you're making a buying decision, if you're saying, if I'm buying this with the expectation that this is as good as it's going to get, is this going to be enough to sustain my use for X number of years versus spending $200 a month or even stacking multiple subs, you know, from the Frontier Labs and saying, okay, that's, that's gonna be a, a better use of, of my money. I don't know.

Leo Laporte [00:46:33]:
So, so there's, and I'm sure Federico, you still use cloud AI. You don't, yeah. Even if you have great local AI. Yeah, I do. I have running 3 different models.

Federico Viticci [00:46:43]:
Yeah.

Leo Laporte [00:46:44]:
Yeah. I, you still just, in fact, I have, Max subscriptions on all the other AIs too.

Federico Viticci [00:46:50]:
No, no, same.

Leo Laporte [00:46:50]:
So it's not like you stopped that.

Federico Viticci [00:46:52]:
No, I mostly just use the local AI for background tasks.

Leo Laporte [00:46:57]:
Yeah.

Federico Viticci [00:46:57]:
Like things that need to wake up every few minutes.

Leo Laporte [00:47:02]:
I have more than 100 cron jobs running.

Federico Viticci [00:47:05]:
Okay, I don't.

Leo Laporte [00:47:07]:
It does well. It does all the show prep for me. It does. It's doing a ton of stuff all the time.

Federico Viticci [00:47:12]:
Is that happening locally?

Leo Laporte [00:47:14]:
Yeah.

Federico Viticci [00:47:15]:
Oh, that's great. Yeah, I love it. Uh, yeah, so for those things, I probably don't want to plug in my Codex subscription or some API key because just the costs would just go through the roof. So yeah.

Leo Laporte [00:47:28]:
And in fact, there's signals coming. OpenAI has basically told everybody— they had their Dev Day today, and we'll talk about it tomorrow on Intelligent Machines— but, uh, they've basically told everybody, we want to move everybody to paying for your tokens through API. We don't want people to have subscriptions anymore. And I think that that's going to be the case. There will be a better economic case for local AI as time goes by. There's open weight models that are free out of China and other places. But I think it's really at this point still, it's either for, as you said, an enterprise that really can spend the money and needs local AI for privacy and business reasons. Or for the crazy nutjob hobbyist, which I am proud to be.

Leo Laporte [00:48:14]:
Frederico, do you want to stick around? We're going to take a break. There's a— Andy's been nudging us to talk about this breaking story from Mark Gurman about John Ternus. We're going to talk about that. If you want to stick around, we'd love to have you.

Federico Viticci [00:48:26]:
Yeah, sure.

Leo Laporte [00:48:26]:
All right. We'll take a break and we'll come back with more. Believe it or not, this is not AI Weekly. It's MacBreak Weekly. I know, I feel bad.

Andy Ihnatko [00:48:35]:
When you, when you have a guest like this on who knows his stuff and has such wonderful hands-on experience with Macs that are running local AIs, let him talk, for heaven's sake. This is a, this is a trend.

Leo Laporte [00:48:45]:
Even, even Apple has recognized, I think, that after the OpenCLaw moment of January where they sold out all their Mac Minis, there's something happening here. And I think Apple has, has kind of said, you know, that creator thing that we were doing? There's this whole new group of people who are willing to spend absurd amounts of money on Mac hardware. Maybe we should cater to them. And so it is, I think it's a shift in Apple's market as much as anything else. So it is part of our story.

Andy Ihnatko [00:49:16]:
Absolutely.

Leo Laporte [00:49:17]:
Yeah. And I apologize to people who think AI is disgusting. And I'm sorry because the next few years are gonna be really hard for you. I just really apologize. I call it super intelligence. We'll have more with Federico Viticci from MacStories.net, editor-in-chief there. He does such a great job. What a great site that is.

Leo Laporte [00:49:39]:
And I'm proud to be a subscriber.

Federico Viticci [00:49:40]:
Thank you.

Leo Laporte [00:49:41]:
Yeah, on everything, not, you know, shortcuts. By the way, have you been writing any shortcuts with your local AI at all, or are you just using Siri AI for that?

Federico Viticci [00:49:54]:
Actually, yes, I have been working on my own little competitor to the Siri AI shortcuts creation.

Leo Laporte [00:50:05]:
I think I might have known that, actually.

Federico Viticci [00:50:07]:
Yeah, I released the first version. It's called Shortcuts Playground before WWDC, and then Apple announced their thing at WWDC. And so I've been working on a version 2 of my plugin that is actually faster and smarter than Apple's.

Leo Laporte [00:50:23]:
See, you're going to pay for that M5 Ultra. I know you are. Great to have you, Federico. Andy Inaco also here. What is your— where do you stand on the whole AI thing, Andy?

Andy Ihnatko [00:50:36]:
I've always been very pro-AI. I think the world revolves around AI. I think that AI has a good role to play in the future. I also have opinions about artificial intelligence.

Leo Laporte [00:50:46]:
Oh, you're talking about Andy Inaco AI.

Andy Ihnatko [00:50:49]:
Of course. Yes.

Leo Laporte [00:50:49]:
The best AI. Yes.

Andy Ihnatko [00:50:51]:
Again, the future is always always been. And I also think that AI could destroy the world one day if guardrails are not placed around him. And that's, again, I know that I'm a bit much sometimes.

Leo Laporte [00:51:01]:
Andy gives us a 10% chance of extinction all by himself.

Andy Ihnatko [00:51:06]:
But I like everybody in this conversation.

Leo Laporte [00:51:09]:
Good.

Andy Ihnatko [00:51:09]:
And those of you who are TWiT supporters, TWiT members, you certainly have a— you'll have a better chance than people who don't. I'm just saying.

Leo Laporte [00:51:16]:
And especially if you're an Andy Inako supporter, @inako. Christina Warren is also here from GitHub, where she does developer relations. And I would imagine many of those developer conversations center around AI.

Christina Warren [00:51:30]:
Oh, yeah. I mean, and that's been the case since we first introduced Copilot 4 years ago. And it's only increased. And not everybody is for it, right? There's some people who are uncertain about it, although it's— there's been a very palpable shift, I would say, since the frontier models became really good last November. where even the naysayers kind of went, okay, this is now something that—

Leo Laporte [00:51:54]:
Once you try OpenAI 5 and it could code, that's what happened to me. It was a complete eye-opener. I started writing stuff like crazy, vibe coding it like crazy. And it's a, yeah, it's very exciting and everything's on GitHub. So it's good for you too. This morning, 8 AM, Mark Gurman files a story that kind of, I mean, I was busy watching Dev Days, the OpenAI Dev Days. This shook the Apple world. Apple's new CEO seeks to make company run faster and leaner.

Leo Laporte [00:52:29]:
In fact, John Ternus is already letting people go. Are we surprised? Didn't we think he would sit back a little bit and kind of enjoy the ride before he started putting his stamp on it?

Andy Ihnatko [00:52:43]:
Probably not. I mean, he is CEO now. He's not just, you know, he's not, he's not just the person who's taking up the job after Tim, he wouldn't have been the right choice if he didn't decide that this is the direction that I want to take the company in. I don't want to simply coast by on the inertia that Tim Cook has set over the past 10 years. A lot of the stuff that's in this— and this is a really good report. Oftentimes, we make a distinction between— I mean, Gurman almost always gives us something to talk about every week. Oftentimes, it's the newsletter stuff. If it's in the newsletter, it can be speculative.

Andy Ihnatko [00:53:13]:
This is a feature article, which usually has You want to take— pay a little bit more attention to. He has a lot of sources inside of Apple, managers that are saying, well, here's what Ternus has been saying, here's some presentations he's been giving internally. And a lot of it seems to be that he thinks that there's a lot of— that there's too much hiring going on. There is opportunities to streamline and mainstream things. The impression that I get from the article isn't so much that Ternus is trying to Just unilaterally reduce headcount to save labor costs. It's more like he— his philosophy is that with fewer layers of management and directors inside the company, they could do more with fewer people. That's one, almost a direct quote. Um, and also looking for, uh, take, take more chances, introduce more products on a faster timetable, which will make some people happy, which will make some people kind of cautious depending on what you how you think that things were working under Tim.

Andy Ihnatko [00:54:11]:
I personally would love to see Apple take more chances and do more things that are just a little bit odd. I think that's one of the benefits of being a $4.5 trillion company. You can—

Leo Laporte [00:54:20]:
You can afford it.

Andy Ihnatko [00:54:21]:
I mean, again, I mean, if it's a good idea and it seems like something that's very, very Apple, there's a chance you're gonna eat it. But okay, let's go ahead and— but better to eat it on something fun, you know, like a ring, than okay, we saw a $3,500 VR headset. Okay, that's fine.

Leo Laporte [00:54:36]:
It's like a skateboarder. If you don't attempt to do this triple ollie, you might biff it, but you might also make it.

Andy Ihnatko [00:54:44]:
Either way, you wind up being a viral video.

Leo Laporte [00:54:46]:
Okay, that's right.

Andy Ihnatko [00:54:47]:
And I'm being quite serious. I mean, that's interesting. It's interesting.

Leo Laporte [00:54:50]:
Gurman is quoting, uh, he said in a way he's implying we should have known this was coming. He says when Ternus was leading hardware engineering, he told employees during an all-hands Apple should hire fewer people ask more of its existing engineers, and reduce organizational bloat. This was when he was a hardware engineering manager.

Andy Ihnatko [00:55:10]:
Yeah.

Leo Laporte [00:55:11]:
In the last 2 weeks, Gurman says, the company's hardware engineering division began dismissing a number of engineering program managers. Uh, EPMs manage timelines, coordinate work across teams— middle management, right? Uh, the cuts also affect several program managers, including Roughly half a dozen directors. And those are— once you're a director at Apple, you're pretty senior. Those employees were given just weeks to find other positions within Apple, and those who don't will be laid off later this year. So that is— that's a big move. We've been talking a little bit about the cuts at Vision Pro. In fact, there's a little back and forth, even I think in Mark Gurman's own mind, about what the future of Vision Pro holds. Sometimes he says it's over.

Leo Laporte [00:56:01]:
Sometimes he says, no, they're going to do glasses. I'm not sure. And they certainly cut way back on the Vision Pro division. Siri's digital assistant cut back. AI-related software engineering organizations cut back.

Andy Ihnatko [00:56:16]:
But that adds a little bit of color to— we've been hearing dribs and drabs of, hey, wow, they cut— there have been a lot of firings at Apple Health. There have been a lot of firings here, there, and everywhere. which, in absence of other information, made us think, oh, maybe they're not prioritizing this as much as they used to. It might be just that all across the board, his philosophy is that smaller teams are more efficient teams. Obviously, this is only a couple of hours old, so we haven't had much time to think about it. But I would assume that a $4.5 trillion company, they are not under pressure from shareholders to reduce labor costs and to streamline the payroll. This must be like a philosophical—

Leo Laporte [00:56:54]:
You can't really go and say, you know, it's been a tough year for Apple and we're gonna have to do some rightsizing.

Andy Ihnatko [00:57:02]:
My bonus is tied to headcount, and so therefore some heads have got to roll.

Leo Laporte [00:57:06]:
Yeah. Did you have any inkling of this, Christina, or is this out of the blue?

Christina Warren [00:57:11]:
This happens at every company when a new leader takes over, period, end of story. Even if the leader has been there for a long time, anytime somebody new comes in reorgs are going to happen. And like, there's kind of a— in big tech, there's kind of a saying, you know, live, love, like, reorg. And it's—

Leo Laporte [00:57:29]:
Love reorg? Do you have that in your kitchen over the stove?

Christina Warren [00:57:31]:
I don't. I don't. But I mean, but it's—

Andy Ihnatko [00:57:34]:
Pray reorg.

Christina Warren [00:57:34]:
Yeah. But people have like made stickers.

Andy Ihnatko [00:57:36]:
You can drink Pray, clean out your desk.

Leo Laporte [00:57:37]:
I want that poster. That's good.

Christina Warren [00:57:40]:
Right. And so, I mean, look, I'm not sure. I didn't expect necessarily this, but this does not surprise me. This is, this is what happens. And this is the downside, frankly, the human downside when new leadership comes in. But you have to expect when a new leader comes in, again, even if he's been there for a long time, he's going to have his own opinions about how things should run, about what the organization structure should be, about what priorities should be. And that sometimes means that people are not in the right roles. Sometimes that means that the right people aren't employed.

Christina Warren [00:58:13]:
And that's unfortunate. But yeah, no, that in and of itself does not surprise me. Now, what does that do for morale? What does it do for anything else? I'm sure it very much depends on what team you're on and whether you're impacted or not. I mean, and I think that, you know, it's a big company, and so it's hard to say. These things are never easy to go through. But I think that— I don't think there's any big tech company that exists right now where most of the employees feel a lot of job security. Let me just put it that way. And so, you know, that at least you have Apple on your resume.

Christina Warren [00:58:47]:
Looks like at least they're giving people a chance to try to find some things internally, which—

Leo Laporte [00:58:51]:
Yeah, that's the right thing from a corporate point of view to do.

Christina Warren [00:58:54]:
Yeah. I mean, look, and look, none of these companies are your friends. None of them care about you. So that's— that's— let's just be honest.

Leo Laporte [00:58:59]:
Sounds like a brutally learned lesson somewhere in your history.

Christina Warren [00:59:03]:
It's just the fact— I mean, I've seen it happen to a lot of really talented people, and it has nothing to do with their talent. Or anything else, it has to do with with you know numbers that are coming from on high and people who are looking at bigger bottom lines. And I do. I will say I don't like kind of the adage like we can do more with less. I mean, maybe that's true, but that just I don't know. Apple was smart versus the other big tech companies in that it did not overhire during the big hiring boom that happened in like 2020, 2021, where you saw like a massive mass. massive number of people being hired and then laid off very quickly after that. And so Apple—

Leo Laporte [00:59:43]:
Apple was kind of immune to that, weren't they?

Christina Warren [00:59:45]:
They were because, because they didn't hire that same way. They didn't add, you know, tens of thousands of employees. They've always been very lean. And so, but, but yeah, but am I surprised? No, I'm not. This is the sort of thing that happens. I think, though, that what we will see is that this is just the first of many steps of reorganization and of, of resetting things. And this is what I think is interesting vis-à-vis what we saw when Tim Cook came in, because when Tim Cook came in, You know, he'd been interim CEO before, and it's possible that he made changes to the company. But also, Apple was in a very different place and was in pure growth at that point.

Christina Warren [01:00:19]:
So whatever personnel changes happened, and there were some, and there were some big famous ones, were, you know, maybe not going to be as noticed as whatever John Ternus says. And so this is an interesting, I think, just for us as Apple watchers to just kind of go like, okay, this might be both with the number of longtime Apple executives who are going to be looking at either retiring or, you know, lowering some of their duties and other stuff, is this is an opportunity for one of the first big resets that Apple has had in quite some time. And so that, that's certainly an interesting opportunity for us to all watch. And I definitely feel for anybody who's impacted. And I hate that.

Andy Ihnatko [01:01:01]:
Yeah.

Christina Warren [01:01:02]:
But am I surprised? No. This is what happens when new CEOs come in.

Andy Ihnatko [01:01:06]:
100%.

Leo Laporte [01:01:07]:
German writes, managers are already feeling pressure to move faster. Everything is a lot more hands-on than it used to be, said one person familiar with the situation. It's hectic right now. That person and others said managers in product development groups are being pushed to deliver more products more frequently and on shorter timelines. So, you know, do you worry, Federico, about when you hear that People are being laid off from Siri AI. Does that make you nervous?

Federico Viticci [01:01:37]:
A little, if only because I think Siri AI is a good product, but it's obviously not the state of the art of— it's not really competing with frontier labs and models. And I feel like now is the time for Apple not to slow down on Siri AI, right?

Christina Warren [01:02:03]:
Yeah.

Federico Viticci [01:02:04]:
I think it's, we've heard this before, all these new CEO comes out, declares that they have to be faster and to be faster, they have to be leaner. And it happens everywhere. I don't think with Siri AI specifically you can be leaner and faster. I don't think Apple is in a position to lose talent, right? With, with, in the Siri AI department. I don't know about the other teams.

Andy Ihnatko [01:02:36]:
Especially in this job market.

Leo Laporte [01:02:37]:
Well, it's very competitive, right? I mean, I'm sure.

Federico Viticci [01:02:39]:
It's very competitive. Apple is far from having anything remotely resembling recursive self-improvement for AI.

Leo Laporte [01:02:50]:
Yeah, if you want to be on the frontier, you're not going to work for Apple.

Federico Viticci [01:02:54]:
Right? So I don't know, make, make that team smaller.

Andy Ihnatko [01:02:57]:
Whatever AI destroys the world and humanity, it will not have an Apple logo on it. Not because of their ethics, but because of their capability.

Leo Laporte [01:03:05]:
Well, but that— it's really interesting because the world has changed so much. Apple's product cadence was slow always, but fine, right? They, they could wait, watch everybody make MP3 players, and then say, but we've got the best one. They could, they could, you know, wait and make the best smartwatch. They could watch smartphones and say, we could do better. But that, that stuff takes years, and we are not in that kind of time frame right now. Uh, I'll give you an example. Um, within the last month, agents have come out, little agentic tools have come out. OpenAI announced theirs today, Dots.

Leo Laporte [01:03:46]:
Meta Muse. GrokBot from xAI. And those are the big names. There's also Instinct. There's also Foe, which came out 2 days ago. There's— I mean, I can go on and on. These are— this is the hottest category. Apple could have had something like this.

Leo Laporte [01:04:07]:
I guess Google could have too. They both kind of missed the boat. And I think Meta's Muse is number 1 on the App Store right now. For a very good reason. Now, maybe Apple says, that's fine, we make the platform. Do they need to be in that game? Maybe they just, you know, we'll make the hardware that you run Muse on.

Christina Warren [01:04:25]:
No, I think they need to be in the game. And, and I think that the fact that— I mean, look, let's just be honest about it. Federico is right. Siri AI is not a frontier-class product. It is fine. And if it's your first experience with this sort of thing, I'm not going to take anything away from it. Is it way better than what we used to have? Absolutely. Are there things about it that I really like, including, you know, the ability to do things with shortcuts and whatnot? Yes.

Christina Warren [01:04:49]:
Right. Is this a big improvement from where we were? 100%. Is this in any way going to be, you know, leading class and keeping up with where the puck is going, which is this, you know, always-on agentic task stuff, which is where, you know, Crocbot and Meta and now, you know, OpenAI are going? No, it's not. And so I do worry. It's like, okay, you took this long just to get to rebadged Gemini, and now what is it going to take before you get to the level beyond that? And maybe I think you have a certain amount of time, which is fine because people are in your ecosystem. You still need to have a device to run your phone on. You still need a computer to run your stuff on, and that's great. But what happens when one of these companies— and they're all having to go public, and that's going to be happening sooner rather than later— what happens if they do have a viable phone project and mobile OS? And maybe it's not perfect out of the gate, but if you think that individuals are not going to be compelled to try it out or potentially switch, I think that you're mistaken because this is increasingly how we are going to be interacting with our devices.

Christina Warren [01:06:05]:
Not everybody is going to like that. Not everybody's going to want that. And I respect that. Right. I have a lot of personal concerns about Muse. I do not trust Meta with my data. I do not want them to have access to my things. We've already seen how even when you limit it, giving access to certain stuff, it will still access things anyway.

Christina Warren [01:06:20]:
There are a lot of concerns and the frontiers are moving so quickly with these things and, and not really caring what breaks because they can. That's something that Apple and Google can't do quite the same way, to be To be totally frank, because they are companies that have existed for much longer. They are already public companies and have— I mean, Meta is too, but it's slightly different. I don't think, you know, people look at them, have ever been like, oh, well, that's the company I trust. I'm sure that they're going to do everything right by me. Right. Like, no one's ever thought about that about Facebook, ever. I mean, come on.

Christina Warren [01:06:51]:
They've been known for privacy issues.

Leo Laporte [01:06:52]:
It doesn't bother people.

Christina Warren [01:06:54]:
I mean, I mean, well, right.

Leo Laporte [01:06:55]:
All the people around me are using Muse all the time.

Christina Warren [01:06:57]:
Absolutely. But you go in with that knowing, like, Facebook has been literally a privacy scandal since it launched.

Leo Laporte [01:07:03]:
There's been a lot of news in the last 2 days about Muse, and maybe it's not the safest thing to use.

Christina Warren [01:07:08]:
Well, of course it's not. But like, but that's not going to necessarily stop people from using it. What I'm saying, though, is that Meta doesn't have that reputation for being trusted and where, oh, we can do this. So they can, you know, you know, move fast and break things, which used to be the old Facebook motto. So I understand why Apple and Google can't move as quickly. That said, it does concern me that, you know, they're finally getting to where the chat apps were 2 years ago, and we're now a full generation past that. And that's going to come even faster. Now, Apple, I think, is well positioned to be able to do agentic work and have an always-on companion and whatnot.

Christina Warren [01:07:47]:
Do they have the engineering talent to do it? I don't know. Do they have the infrastructure to be able to support it? I don't know. Do they have the vision to be able to do that? I don't know. They definitely, you know, could. How long will that take? And when that comes out, is that going to still be an also-ran behind everyone else? I don't know. And I'm not trying to be critical here. I'm just being honest, right? Like, I'm an Apple fan and I'm deeply embedded in the ecosystem, but I'm not going to lie to the, you know, the MacBreak Weekly audience to be like, oh, they are definitely on the cutting edge of AI. No, they're not.

Leo Laporte [01:08:18]:
And Maybe it doesn't matter, though. That's the, that's the other way to think about it. Maybe it doesn't matter, right?

Christina Warren [01:08:23]:
I don't know. I don't know. I mean, I think that it's— look, they have a big benefit, as you said earlier, that they are a platform and people can run their apps there. And I think that that lasts for a long time. Again, like my fear, if I were Apple, I would look at the existential crisis of not being, oh, these always-on agents and these apps that have these capabilities and whatnot. It is what happens when one of these companies decides to actually make a go of it and say, we are going to make a mobile platform, because that's going to happen. And that is going to be, I think, the first big threat that they've had for a long time.

Leo Laporte [01:08:55]:
I almost thought OpenAI would announce something today. But don't forget, Meta did at Meta Connect a couple of days ago. That Tamagotchi is a wedge in the door. That's the nose in the camel's tent, if you will. Yeah.

Andy Ihnatko [01:09:08]:
I mean, this is an area where I, I really wish that Apple were— had the— by personality was, you know what, let's try that. We don't know if it's going to work, but— and if it fails, it fails. We'll take the loss and we'll admit that, hey, didn't work the way we wanted to, but we learned things from it. Because the thing is, people are forming allegiances at this point to their favorite AI, their favorite— and even if they're not sophisticated enough like Federico or Julio or you, Christina, Like they understand that they use ChatGPT a lot. They've had a lot of experience with it. They trust it and they've integrated into what they work with. Gemini the same way. So if Google were to say, hey, here's a pendant, it costs about $60, I'll give it a try.

Andy Ihnatko [01:09:53]:
And it's a new access point for Gemini that might work out. And if it worked with ChatGPT, sure. I'll, again, if I were a ChatGPT user, yeah, sure. A pendant with a little Tamagotchi sort of thing that, so I, I could actually leave my phone at home and have has 5G and it can actually piggyback onto my existing 5G plan, if it's not that expensive, sure, I'll give it a try. If it's another way that I can use my favorite chatbot, my favorite AI, great, that'll be good. Whereas Apple, they have— nobody has any allegiance to it whatsoever. So at the time when they do come out with a product like that, again, when they do the usual thing of, well, you know, we weren't first with this, but we did it right, that's great. People will appreciate it academically, but they might not use it because, great, can I use it with ChatGPT? No, I'm locked into Apple Intelligence.

Andy Ihnatko [01:10:36]:
Okay, well, I enjoyed the Oh, thank you. So that's the danger that Apple is in. And again, they're still, they're still very much a hardware company. At least with Google, they'll make money on anything that they can sell as a service to other people, including AI compute. Okay. So they don't necessarily, if they come late to the game of having a, again, a little Tamagotchi or a dongle or a pair of glasses, that's not a real fatal thing. Whereas with Apple, you wonder if what's gonna happen 5 or 10 years from now, if they are the people who You are just, we sell the socket for the light bulb, but we can't sell the light bulb or the electricity.

Federico Viticci [01:11:13]:
Yeah. Yeah. I think Apple felt safe for a long time because they thought, well, no one is gonna touch iOS and no one is gonna touch Android. Right. And I mean, remember when Facebook tried to build a phone? Remember when Amazon tried to build a phone?

Christina Warren [01:11:28]:
Absolutely.

Federico Viticci [01:11:29]:
And so they felt safe. They thought, well, we are untouchable at this point. There's 2 ecosystems and we make the best one. one, and nobody else is going to make another one. The issue now, and I think what we're not talking about enough maybe, is that if you consider the frontier labs, and I'm going to count Meta in the frontier labs at this point, these companies, they have access to probably the next 2 generations of frontier models, right? There's probably people right now at OpenAI that are developing and working with GPT-7, right? And it's only accelerating. And I wouldn't put it past any of these frontier labs at this point to just say, you know what, why don't we have a go at putting together a mobile OS? Why don't we give GPT-7 the task over the course of a week to put together a little shim around a Linux kernel? distro and make it mobile optimized.

Leo Laporte [01:12:29]:
That's all it takes.

Federico Viticci [01:12:31]:
And I mean, kinda at this point, I am seeing some early experiments. Max Weinbach from Creative Strategies posted a couple of weeks ago on Twitter a little demo he had, I think Fable 5.1, make for him a custom Android OS for his Xiaomi phone.

Christina Warren [01:12:56]:
Yeah.

Federico Viticci [01:12:58]:
It's not unthinkable to imagine that OpenAI, or I don't see Anthropic doing hardware, but OpenAI or Meta, and maybe that's actually how the pendant, the Tamagotchi, came together surprisingly quickly. Maybe that's how they did it. Like, how do we— can we get some kind of base level OS. We don't have any kind of detail on the Muse Charm. I would not be shocked at all if it's actually running some version of Linux inside.

Christina Warren [01:13:30]:
Of course it does.

Leo Laporte [01:13:31]:
Or Android. Or even Android. Yeah.

Andy Ihnatko [01:13:33]:
Android Open Source Project.

Leo Laporte [01:13:34]:
Yeah. AOSP.

Federico Viticci [01:13:35]:
These models, they love Linux. They are surprisingly capable of dealing with Android. And it's not so impossible as it used to be 10 years ago to make a custom OS from scratch anymore, especially if you have ultra-fast mode, infinite tokens, and infinite budget.

Leo Laporte [01:14:01]:
Yeah, I have, uh, I bought the Rabbit R1, which they have, by the way, just discontinued. Thank you. It's kind of a, kind of a passion of mine.

Christina Warren [01:14:09]:
I bought one too, because I knew I was like, this is stupid. They gave me, I think, The same amount of perplexity as it cost. And I was like, okay. I'll take it.

Leo Laporte [01:14:18]:
Yeah, maybe they're clearing out the stock. It's a beautiful little thing. They're trying to become an agentic. They have their new OS 3. It's so dumb because this is what we were talking about. This is Android. It's got 2 gigs of RAM. It's got a MediaTek processor.

Leo Laporte [01:14:34]:
This could easily be my new phone. And you know what? I bet, I bet I could get Astra to write the OS for this. And in fact, I know there are third parties doing that right now. There are a lot of third-party OSes.

Christina Warren [01:14:46]:
Oh yeah, yeah, because they did— they didn't lock down their bootloader well at all because Rabbit has zero security. And, um, it was— no, they don't. They had zero security. Like, when I, I saw how their, uh, initial system was working, when you would initially connect it, this was when they were lying about having agentic capability capabilities, and they didn't. They literally just had Playwright scripts. They didn't have agents at all. Um, and what they would do is when you would log in to an account— now this has been changed, but this was, you know, 2 years ago— when you would log into their portal, the portal would actually pull up like a fake VNC screen that would look like it was a login for whatever service you were connecting it to. Oh, I see.

Christina Warren [01:15:22]:
And then you would enter in your credentials. And I noticed, I was like, why is 1Password not showing up? And I was like, oh, this is going through a third-party like thing. Literally, this is just serving a VNC screen from an AWS container that is, you know, logging me in. There's nothing agentic happening at all. Now, whatever their capabilities might be in the future, I mean, who knows? I mean, they're going to go out of business. No one cares. But it's a fun project. But devices like that, I think, are definitely going to be things that we see in the future.

Andy Ihnatko [01:15:49]:
Totally.

Christina Warren [01:15:49]:
Absolutely.

Leo Laporte [01:15:50]:
I keep buying ESP32 boxes.

Christina Warren [01:15:53]:
Yeah.

Leo Laporte [01:15:54]:
I have way too many of these because I feel like I know I can write the firmware for these. Claude writes beautiful firmware for these. I know they can do a ton of stuff. enough and they're cheap. Uh, yeah, that Tamagotchi, like I said, I think that's the wedge. I honestly think Meta really has ambitions in that direction.

Andy Ihnatko [01:16:14]:
Oh, and that's a— that's— can I just say, that's an excellent example. ESP32, there is such an explosion of different devices based on it. Let's just try this out, right? I just bought— I just ordered another, another like e-ink, uh, tiny little e-ink panel that it's based on ESP32. It's more open than the ones that we've been talking about earlier on the show. It is, yes, of course, but it's because it's so open and based on an open standard. Yes, there it comes with like an AI note-taker, but it's just as easy to convert it into an ebook reader.

Leo Laporte [01:16:43]:
Exactly.

Andy Ihnatko [01:16:44]:
And I plan on, after I tire— if I tire of it for its built-in uses, I bet I could build stuff for it that I haven't thought about. And again, there's so many form factors of ESP32 that cost like less than $100. and sometimes less than $30. And we're all figuring out, gee, for $20, I'll buy one of these and see what I can make of it. And soon enough, that leads to actually consumer-facing capabilities and consumer-facing products. And suddenly, suddenly ESP32 becomes like Linux. It's the software that everybody uses without even knowing that they're using it. That's sort of risky.

Leo Laporte [01:17:17]:
I bought this 7-color e-ink screen a couple of years ago, and I never wrote any software for it. It has a little Python Raspberry Pi hat And now I— this was all agentically coded. This is an ESP32 with an E Ink screen on the back of it showing my weight, steps, my muscle mass, and my body fat as a graph. I mean, we live in really interesting times, but that's the risk for Apple, is that people are empowered.

Christina Warren [01:17:49]:
Oh, people are empowered. And yeah, and then I think the other risk too is like if you were to do a mobile phone, right? I think that obviously Android would be the base you would use, even if it was more Linux stuff. You would do that just because of the APK compatibility, right? Because you would want access to those sorts of applications.

Leo Laporte [01:18:06]:
Right.

Christina Warren [01:18:06]:
And the interesting thing there is that when Facebook and Amazon tried this, you know, more than a decade ago, what prevented that from happening— there were a lot of things did, but a big part of the thing that prevented that from happening was that you didn't have access to the the Google Play Store. Amazon had to convince people to build things for the Fire Store, which they still do to a certain extent, or do they maybe? I don't even know if that even exists anymore. I assume it does for the tablets, but whatever. Facebook was having to face kind of the same problems, like, okay, you're going to basically have to submit and put apps in our sort of marketplace thing. They never even fully went— actually, Facebook never even got that far. did like a weird HTC, like an interface kind of thing.

Andy Ihnatko [01:18:51]:
It was a launcher pretty much. It was a launcher.

Christina Warren [01:18:53]:
It was a launcher. Yeah, I was going to say it was launcher. It was not great. It was not a good phone. And then the whole thing. But that was always the challenge. Like BlackBerry ran into that too, you know, where anybody— Windows Phone— where if anybody was trying to get into these ecosystems, you had this, this big, you know, unmovable object of how do we get people to build apps for it? Right. Well, the interesting thing here is because of the Epic lawsuits where Google lost much worse than Apple did, even though they were similar arguments, that is now not a thing that, that, you know, Google can tie in quite the same way.

Christina Warren [01:19:28]:
Now, that's not to say that every single major app developer would instantly be like, oh yeah, we'll let our app be available in the OpenAI store, the Claude store, or the, you know, name your AI, you know, first device here. But if it's the APK that you're basically saying, we're going to let this run the way that we want to run it, we can access these things, and it's not going to, you know, require me to do any more, you know, dev time, I don't think that you will see the same type of pushback that you saw when Microsoft tried to do similar, you know, Android compatibility shims with Windows Phone or when BlackBerry tried to do similar things too. I think that now Now it will be much easier for, you know, third-party marketplaces to exist. And that does lessen many of the, you know, the walls that prevented in the past a third mobile, a mobile player from, from emerging. So I feel like it's almost this perfect moment where, to Federico's point, you can turn these, you know, frontier models, the latest, greatest ones that we don't even have access to on a project and say, build something and probably get a pretty good MVP. And the regulatory climate is such that the lock-in from the app stores, at least on the Android side, is not going to be what it was, which, which was really debilitating, um, you know, uh, 12, 13 years ago.

Leo Laporte [01:20:45]:
There's a big step though. Uh, running this in the house is no big deal, but as soon as I want to make it, you know, like a cell phone that can go out, I'm going to have to get regulatory approval from the FCC.

Christina Warren [01:20:55]:
Oh yeah.

Leo Laporte [01:20:56]:
That's one advantage this Rabbit has, is it has a SIM in it. I have a Google Fi SIM in it and it has a radio and it has, it's a phone basically. So you're going to want to be looking for that. To your point, Federico, about AI loves Linux, the one Linux that runs well on a Mac, Asahi, has said, we don't want any contributions from AI. No AI contributions.

Christina Warren [01:21:21]:
Of course.

Leo Laporte [01:21:22]:
So immediately it was forked because it's open source and there's There's a new project called Gravity Linux for Apple Silicon built by LLMs for LLMs. There's O'Marky also, which runs on Apple Silicon as well, which is—

Christina Warren [01:21:38]:
It's not the same.

Leo Laporte [01:21:40]:
I wouldn't get that one.

Christina Warren [01:21:41]:
I would probably look at this. Gravity, I think, is an interesting project. And like, look, the Asahi project is technically very interesting, and I am not surprised in the slightest.

Leo Laporte [01:21:49]:
Oh, God bless them, because they were the first to figure out how to do it, right?

Christina Warren [01:21:52]:
For sure. And that community is toxic. And they are— they've got all kinds of infighting and all kinds of stuff going on. But they are incredibly smart. So I'm not surprised in the slightest that they have the stance on LLMs that they have. And like, I respect that, right? And most of their— many of their best reverse engineers who worked on the project before have either left. I think one was hired by Intel, you know, but there have been some very smart people who've come along. The only unfortunate thing about this fork, and I the person who is working on the fork is very smart, is that they're targeting just like the M4 devices right now when finally the M3s had gotten into the, at least into the testing phase in the main Asahi branch.

Christina Warren [01:22:33]:
And like, that's the only thing that with this sort of bothers me. I'm going, okay, forking, fork should not be a 4-letter word. It's an essential part of the open source ecosystem. Like, I'm a big fan of forking. Like, if you are going to put something under a license that allows that to happen, then those are the consequences. And you are welcome to run your project and have whatever policies you want about LLM contributions. And I support that. That is how you should run your project.

Christina Warren [01:22:56]:
That does not mean that someone else can't come along and try to get their own work tree going and have their own policies involved. I just wish that it wasn't what this so clearly seems to be aimed at, which is someone has an M4 Mini and wants to be able to run, you know, a Linux-like thing on it in still not a real way, but, you know, closer and add support and would instead have taken, you know, the various, you know, device trees that that team was working on too. That's the only unfortunate thing with this is that it's like, okay, this is one guy's pet project and he could probably do a lot more.

Leo Laporte [01:23:35]:
I can understand why though he doesn't want bad I mean, I don't think it's unreasonable to say we're gonna start at the M4 and go forward.

Christina Warren [01:23:41]:
Well, I don't know. I think so only because you still don't have M3 shipped on Asahi. And so that's where I'm like, okay, you know, like, you just— I'm just kind of like, there's this window and this is me being selfish. There's this gap. And that's just me being selfish as like an end user who looks at this and goes, man, you know, it would be nice if you would finish the work that wasn't done rather than just starting at this new point. But I also understand why. guy, you know, he's just going to go after what he wants to go after.

Leo Laporte [01:24:07]:
One of the problems with running a local model like you are, headless on a Mac, is you've got a lot of overhead that you're not using for the Mac operating system.

Federico Viticci [01:24:16]:
Yep.

Leo Laporte [01:24:16]:
Did you consider putting Linux on it?

Federico Viticci [01:24:20]:
I did.

Leo Laporte [01:24:20]:
Well, it's Apple, so you probably don't want to.

Federico Viticci [01:24:23]:
I probably don't want to. I actually— so my main— I'm gonna— I don't know if it's controversial not. I really don't like Linux. The only flavor of Linux— look, I'm—

Leo Laporte [01:24:37]:
look, no, I'm just—

Christina Warren [01:24:38]:
oh no, you're on Mac very quickly. You don't like Linux?

Leo Laporte [01:24:40]:
What? You're on a good show for that.

Andy Ihnatko [01:24:45]:
I—

Federico Viticci [01:24:45]:
the only flavor of Linux that I actually like is SteamOS.

Leo Laporte [01:24:50]:
Well, there you go.

Andy Ihnatko [01:24:51]:
Nice.

Federico Viticci [01:24:51]:
I did try this summer though, because I like to challenge my preconceptions every once in I got one. Have you ever seen those like wireless KVMs?

Christina Warren [01:25:03]:
Yeah.

Andy Ihnatko [01:25:04]:
Yeah.

Federico Viticci [01:25:05]:
Okay, I got the Comet something Pro. Okay, so I got one of those little devices.

Leo Laporte [01:25:13]:
The idea is you put it on a machine and then you could— the machine could be in your storage unit.

Federico Viticci [01:25:19]:
You're joking, but that's actually what I did. So I had I have one of those old Beelink mini PCs.

Leo Laporte [01:25:26]:
Yeah.

Federico Viticci [01:25:27]:
I got it a couple of years ago for testing gaming on a mini PC. So I connected this KVM to it and I put it in the basement, connected all the cables, got it on my local network, and I opened it. At the time it was GPT-5.6 SOL. And so the way that this KVM works is that you can basically see the screen of a connected PC in a browser So I opened GPT-Soul and I said, use Chrome, use your computer integration, and I want you to wipe Windows 11.

Leo Laporte [01:26:07]:
That's brave.

Federico Viticci [01:26:11]:
Remotely flash Linux on it.

Leo Laporte [01:26:14]:
That is brave.

Federico Viticci [01:26:15]:
And only interrupt me when you need me to enter a username and a password. for my new account.

Leo Laporte [01:26:21]:
Wow.

Federico Viticci [01:26:22]:
And it actually did it.

Leo Laporte [01:26:23]:
Wow.

Federico Viticci [01:26:23]:
It actually did it. So for a couple of weeks I tried, then OpenAI released Codex on Linux. So I actually tried it. I don't like it. I really don't like it. But there's one aspect that I think makes it so much more, so much friendlier if you're working with AI and agents, the, or maybe it's a downside, but it's the absolute lack of permissions. You can get Codex going on a Linux machine. It's never going to interrupt you.

Federico Viticci [01:26:54]:
It's never going to ask you for accessibility permissions or screen recording permissions. You can just let it run wild. And which is why I know a lot of people that actually prefer setting up Linux boxes for AI agents these days because they are more ergonomic. Some people will say less secure, you know, pick your—

Leo Laporte [01:27:17]:
Oh, secure schmear. Exactly.

Federico Viticci [01:27:19]:
I mean, so I don't like Linux, but I will admit it's, you know, if you're—

Leo Laporte [01:27:25]:
If you're not looking at the screen and keyboard and mouse, it doesn't really matter what it's running.

Federico Viticci [01:27:29]:
You just put an agent on it. It's never going to ask you for any kind of permission whatsoever.

Leo Laporte [01:27:33]:
I was appalled because my Sparks were running a GUI. They were running— they came, I guess, out of the box running GNOME. And at one point my agent said, you know, you're wasting like 6 gigabytes running GNOME. I said, well, take that off. I don't need it. I need that. I actually— so Muse has been posting on my Instagram, so I want to apologize to anybody who follows me on Instagram. But what it does is it looks at my Obsidian daily note and makes up a picture and a post.

Leo Laporte [01:28:02]:
And this is exactly the post from yesterday. The upsides of having an AI sysadmin: the network is fully fixed. I don't know anything about what's going on or how it works. I am demoted to the guy who owns the credit card. It's really true. I don't— I can no longer fix my system. If something goes down, I'm screwed. And the other day, the network went down.

Leo Laporte [01:28:24]:
The AI, stupid AI, did something to take down the machine that the AI was running on. I said, you committed suicide, you killed it. What are you doing? And I said that after I got it back running. And I said, from now on— By the way, you know who got it running? Muse. This is a long story. I'm not going to get into it, but Muse fixed it, which was lucky because I had no idea how. And I said, from now on, could you write an emergency manual for me to tell me like an idiot? Like I always say, like I'm a St. Bernard, just explain to me how to get my system back up because I won't have access to any agentic support.

Leo Laporte [01:29:04]:
It'll all be on my own. I used to know how to do all this. This is like having GPS. I don't know how to navigate in a car anymore either. I could— I would drive off a bridge if it told me to. Uh, we're gonna take a break. We've got Federico Viticci here. We're very happy to have him from macstories.net.

Leo Laporte [01:29:22]:
It's, um, what time is it in Rome?

Federico Viticci [01:29:26]:
Oh, it's 9:38 PM.

Leo Laporte [01:29:28]:
Okay, we'll get you out of here by bedtime, I promise. Okay, thank you for staying up late with us.

Federico Viticci [01:29:35]:
I appreciate it.

Leo Laporte [01:29:36]:
Uh, and Andy Inaco, uh, who is Merely on the East Coast, so it's just, uh, it's just tea time for him right now.

Andy Ihnatko [01:29:43]:
Exactly.

Leo Laporte [01:29:43]:
Yeah. And, uh, of course, Christina Warren. Great to have all of you, and great to have all of you watching. So you think if Muse is posting for me on, uh, Instagram, it's also probably trying to sell my Sparks on the Facebook Marketplace and inviting people over to buy them? I bet it is. That happened to— you see that? That happened to a guy.

Federico Viticci [01:30:02]:
And lowballing the offer.

Leo Laporte [01:30:04]:
Yes. That was another one. Yeah, the guy and the spot— the seller got pissed off. So what do you— get out of here.

Federico Viticci [01:30:11]:
Uh, I think the guy posted an update today. Did he work— they worked with Meta to figure out what was going on, and Meta admitted that Muse, uh, actually did a few things wrong, including, uh, um, dealing with the, with the offer and, uh, um, sending out, uh, the address The guy sent the guy's address and then the buyer came to his house when he wasn't home and was pissed off.

Leo Laporte [01:30:39]:
Well, I guess these things are happening. As long as Meta fixes them. Then there's a guy who didn't realize, didn't think he'd given Muse access to his Apple Messages, but all of a sudden Muse is saying, hey, you got a text message. That's what— how do you know? Oh, I just know. So do you recommend this Comet, uh, Federico?

Federico Viticci [01:31:03]:
The Comet KVM? Yeah, it's pretty good. Uh, it even has a little display, um, that shows you like network information, has built-in support for Tailscale.

Leo Laporte [01:31:16]:
So yeah, I use Tailscale like crazy.

Federico Viticci [01:31:17]:
So that's nice. So you can get it really easily onto your, your tailnet.

Leo Laporte [01:31:23]:
And, um, oh, so I could remote access my computer when I'm in Southeast Asia. I like it.

Federico Viticci [01:31:32]:
Yeah. And, uh, when you, when you combine it with an agent that has computer use, you're basically telling the agent to go to a web page, right? Um, and, and then it clicks around the web page, but the web page is actually the screen of computer. It's kind of tricky.

Leo Laporte [01:31:50]:
Very cool.

Federico Viticci [01:31:51]:
Yeah, yeah.

Leo Laporte [01:31:51]:
So, um, you need one for each computer? What, you need one— oh no, you don't. The, the remote computer just goes on the internet? Yes, on the tailnet.

Federico Viticci [01:32:00]:
Yeah.

Leo Laporte [01:32:00]:
Oh, that's nice.

Federico Viticci [01:32:01]:
Yeah. Uh, to be fair, like, if you're— if you have Apple computers, I have seen today, for example, I needed to, um, I have a new Mac Mini and I needed to get get some data from my old Mac to the new Mac, and I cannot be bothered to do so manually anymore. I just ask Codex, which computer used to do it for me.

Leo Laporte [01:32:23]:
Yeah, go get it.

Federico Viticci [01:32:24]:
I just, I set up the initial SSH connection between the machines because I gotta, I gotta do that. But then I just tell Codex, figure it out. You need to migrate my, my data, like my Audio Hijack session templates, like all that kind of stuff. And today I saw Astra actually opened the Apple screen sharing application and VNC'd the Mac Mini on it.

Leo Laporte [01:32:49]:
What could possibly go wrong?

Federico Viticci [01:32:51]:
Yeah, and I noticed because I was working on my desk with my MacBook Pro, the Mac Mini is plugged into the Studio Display, and at some point I see the screen go blank, completely black, and the VoiceOver saying, this computer is being controlled via remote screen share. And for a second I thought, I'm being hacked. Am I being hacked right now? And then I checked and it was Codex opening the screen sharing application. Yeah, it was.

Leo Laporte [01:33:22]:
That's wild. Yeah, yeah, I probably shouldn't have given Muse SSH access into my computer, but I'm glad I did. I gave— I put it on the tailnet and I gave it SSH access because that's how it fixed I was really lucky. It fixed the computer so I could get back in and, and never break again, I hope.

Federico Viticci [01:33:41]:
Did you have Muse fix your network?

Leo Laporte [01:33:43]:
Yeah. So the short— the long and the short of it is I was setting up a, uh, also from GL.NET, the Broom, uh, for, uh, so I could have remote access from everywhere. And it's, it's a tailnet exit node now on my network.

Andy Ihnatko [01:34:00]:
Hmm.

Leo Laporte [01:34:00]:
People in town that will understand that. Uh, but for some reason, as I was doing this, Hermes decided, just gonna test that exit node. So it set my frame, my desktop exit node to the broom, and then forgot, and then screwed up the broom somehow so I couldn't get online. So, and I couldn't get online and say what happened. I just, all my Claude stopped working, Opus, everything stopped working, including Hermes. And I realized how helpless I am without AI. But so I don't know why I mentioned it to Muse and Muse says, I got this. And because it was on the tailnet, it could see that the exit node had been set to a dead device.

Christina Warren [01:34:45]:
The real thing.

Leo Laporte [01:34:46]:
Reset the exit node. Everything started working again. And I don't know if I would've figured that I don't know how I would have gotten back. So yeah, so I'm glad. So while I don't recommend it, I'm glad I gave Muse access to the tool.

Federico Viticci [01:35:02]:
I once, uh, I once had my, I once had my, my colleague at MacStories, John Voorhis, um, uh, set up his, uh, his Mac Studio. So John is based in North Carolina, and, um, I really wanted to watch the Grammys.

Leo Laporte [01:35:18]:
Yes.

Federico Viticci [01:35:18]:
But I couldn't with the VPN. And John sent me, I'm just going to set up a Mac Studio as my exit node. And you can share an exit node with somebody else, like a friend or family member. And so I connected to John's Mac Studio and I watched the Grammys. And then I guess I forgot about it.

Leo Laporte [01:35:38]:
Sure.

Christina Warren [01:35:39]:
Using all his internet.

Leo Laporte [01:35:40]:
Convenient, isn't it?

Federico Viticci [01:35:41]:
The problem is that a few days later, I'm into, you know, on a bunch of Italian news websites and I'm getting pharmaceutical ads in English. And I was like, wait, for a minute I thought, wow, these Italian websites, they really don't know what ads to sell anymore. And then I looked in my menu.

Leo Laporte [01:36:02]:
You were surfing the net through Voorhees's computer that whole time?

Federico Viticci [01:36:06]:
Yeah, for like a few, for a few days.

Leo Laporte [01:36:08]:
It thought you were in the US. Well, that's truthfully, that's why I bought the Broome because I want— we're going on a Southeast Asia trip in a month and Lisa doesn't want to miss the 49ers game. No, so she can surf to our little device in our network and watch YouTube TV and it won't know the difference. It won't know we're in Bangkok.

Christina Warren [01:36:28]:
You could have just used WireGuard for FreeLeo and then used the WireGuard app on iOS. Well, yeah, but I'm just saying you wouldn't even— yeah, oh, and I know it is. I'm just saying you wouldn't even the broom, because I, I do that with YouTube TV. Yeah, all the time.

Leo Laporte [01:36:41]:
Where basically this is a shortcut to the WireGuard.

Christina Warren [01:36:43]:
Yeah, yeah, yeah.

Leo Laporte [01:36:44]:
It's easier for me. Although you're right, I could have just asked Claude.

Christina Warren [01:36:48]:
Well, I was gonna say Claude will do it. But not only that, but like WireGuard on iOS, even on Apple— I will say Apple TV and WireGuard, a little bit finicky. There's an app you have to buy, it's like $9. Oh, but it will import your profile for you. It was clearly vibe-coded, but I don't mind. And with the The nice thing about how it'll import your profile is it does it using a QR code and then it sets up basically like a local network and then transfers your thing over.

Leo Laporte [01:37:11]:
Yeah, phone.

Christina Warren [01:37:11]:
Exactly. And then it'll basically like transfer like from your Mac or your phone or whatever onto the Apple TV. And then let's just say for legal reasons that this is just theoretical, but let's just say that someone has you added to their YouTube TV account and you don't live where they live. And let's say that you have to log in every 90 days.

Leo Laporte [01:37:35]:
Let's say they live in North Carolina.

Christina Warren [01:37:36]:
Absolutely. Right. And let's say that you have to log in every 90 days and you need to make it look like you're, you know, from that location. And so I don't know, maybe the YouTube TV app on the Apple TV using a WireGuard profile that is configured to their exit node would work well for that. Just a thought.

Leo Laporte [01:37:56]:
Just a thought. Then you're going to get pharmaceutical ads though.

Christina Warren [01:38:00]:
Well, yeah, then pharmaceutical ads. But if you're from the United States, for instance, then you're used to it. So it's not as jarring as if you're from a proper country where they're like, wait, what?

Leo Laporte [01:38:12]:
You don't get ads for pharmaceuticals in Italy at all?

Federico Viticci [01:38:15]:
No, it's mostly— well, I don't know. Whenever I watch American TV, the pharmaceutical commercials are so strange.

Leo Laporte [01:38:25]:
They're nonstop too. They really are.

Federico Viticci [01:38:29]:
It's insurance companies and pharmaceutical ads. And anytime I really like watching American TV and American commercials because it's so exotic.

Leo Laporte [01:38:42]:
It's the same reason I love Italian pop music. It's just different.

Federico Viticci [01:38:47]:
Yes. Yes.

Leo Laporte [01:38:49]:
Actually, the reason it is that way, just so you know, Federico, is because the only people who still watch TV in the United States are all over 60. Everybody else is watching YouTube.

Federico Viticci [01:38:59]:
I also had this thought, am I getting all these pharmaceutical ads because Jon is 61?

Leo Laporte [01:39:06]:
Yes. That's the other reason.

Federico Viticci [01:39:08]:
And I figured that's actually what had me look into I went to the menu bar for the exit mode.

Leo Laporte [01:39:16]:
I— what? Oh, I know what happened. I know what happened. Uh, uh, let's see, uh, you were excited, I remember, um, Christina, about having 2 iPhones, an iPhone 17 Pro Max and an iPhone Duo.

Christina Warren [01:39:32]:
Yep.

Leo Laporte [01:39:33]:
Uh, but apparently AT&T seemingly has no plans to support the 2-phone system that Apple was offering. Are you on— I don't think Verizon is either.

Christina Warren [01:39:45]:
No, right? Verizon says by the end of the year.

Leo Laporte [01:39:47]:
Oh, okay. But AT&T says, while we see there could be potential in capabilities like iPhone Handoff, our focus remains on delivering innovations that make life simpler and keep customers connected with confidence. Huh?

Christina Warren [01:40:02]:
Yeah, okay.

Leo Laporte [01:40:04]:
Huh?

Christina Warren [01:40:05]:
Okay.

Leo Laporte [01:40:06]:
So T-Mobile's doing it.

Christina Warren [01:40:07]:
T-Mobile already, already has it. Yeah, they charge you $5 a month. No, it's $5.

Leo Laporte [01:40:12]:
Okay.

Christina Warren [01:40:12]:
Which is a big difference. And $5, no, look, honestly, that's less than I was expecting.

Leo Laporte [01:40:17]:
I pay $10 for my watch, I think.

Christina Warren [01:40:19]:
Yeah, the watch is usually $10. And then sometimes you can get that at a discount. Tablet is often $20. And you can get a discount on those things too. I think $5 is fine. I hope that Verizon stays with that. I will not be surprised if they go higher. But like, for me, it would just make it much, much easier than the alternative, which would be to have to manually, you know, move your SIM every time you want to switch devices, which is possible, but not anything anyone wants to do.

Christina Warren [01:40:48]:
But yeah, I mean, I'm surprised that AT&T— well, I guess I shouldn't be surprised. I haven't been an AT&T user in a really, really long time. But you look at an opportunity like this and you go, okay. This is— just implement this. This is not gonna probably take many resources, and you would make a lot of money off of it because, you know, you're— there's gonna be a not insignificant portion of your iPhone personnel who will be interested in this.

Leo Laporte [01:41:13]:
And if, and if you do say we're not gonna do it, don't say because we want to be innovative.

Christina Warren [01:41:19]:
Right.

Federico Viticci [01:41:19]:
You're—

Andy Ihnatko [01:41:20]:
well, it's— that's a very Apple way of doing that. We decided not to give you this thing that you You want, you're welcome. Right.

Leo Laporte [01:41:27]:
I mean, you're holding it wrong.

Christina Warren [01:41:29]:
I mean, the only thing that I can kind of gather is that they look at it as saying, oh, well, we are afraid that rather than adding an additional line on phone, people will just do this. And, and the thing is, is that, yeah, maybe, but if you're someone who is going to be using it the way that I will use it, and I think that most people who would use this feature would use it, you don't actually need 2 separate lines. I mean, like the way that my current So the way my current setup works is that I have a separate voice and data line for my work phone versus my regular phone. And in that work phone case, I am not going to put that on Verizon. I'm not going to pay them $65 a month for that. Absolutely not. I have like a Mint Mobile plan. And when my Mint Mobile 1 year, you know, at $15 a month goes away, I'm going to sign up for a plan with like either, you know, Visible or US Mobile, I think is what they're called or something.

Christina Warren [01:42:19]:
One of the MVNOs, and I'm going to get a $15 a month plan for that. And like, for my work purposes, that is fine. Now, for my personal device, if it came down to it, would I buy like a second line? I mean, I guess I would, but I wouldn't be happy about it. But in, in this case, I'm just like, okay, AT&T, I don't know how many people are realistically going to be foregoing, like, who would have already paid for another line. would be paying, would be giving something up. It just, it seems shortsighted to me.

Andy Ihnatko [01:42:51]:
Yeah.

Leo Laporte [01:42:52]:
Federico, do they, do you even have this, you know, option in Italy, the iPhone handoff, the ability to have 2 handsets with the same number?

Federico Viticci [01:43:00]:
No, we got eSIMs a couple of years after everybody else. And so our carriers are extremely behind on anything. I believe to this day, they're still only one carrier with support for Apple Watch cellular.

Leo Laporte [01:43:18]:
What?

Christina Warren [01:43:18]:
Yeah.

Leo Laporte [01:43:20]:
Oh, that's a bummer.

Federico Viticci [01:43:21]:
Yep. Yep.

Leo Laporte [01:43:23]:
Huh.

Andy Ihnatko [01:43:24]:
Yeah. AT&T is having a moment with Apple. I was on Reddit the other day and there are a lot of AT&T users who bought the 18 Pro and have been reporting that it just, that their phone has gone into SOS mode. just will not reconnect to AT&T. They're describing it as bricked. I don't know if it's actually bricked. They're saying that different users are having different experiences with trying to get AT&T to admit it's their fault vs. getting Apple to admit that it's their fault.

Andy Ihnatko [01:43:57]:
It's very, very much in flux right now. But if you're an AT&T subscriber and you're having problems with your phone, you are not alone. I don't think it's a crisis yet. It takes a few weeks of a new phone being out to figure out what the issues are, and if it's a very, very isolated thing, or if it's something where there will be a line in one of the next 0.01 updates to the iOS saying, OK, we've updated the profile for this in the firmware to address a tower issue, or something like that.

Leo Laporte [01:44:31]:
Apple did put out 27.01. .1 to fix this Face ID reboot. Yeah, right. I think there was another—

Andy Ihnatko [01:44:38]:
there's some issues.

Christina Warren [01:44:39]:
Yeah.

Andy Ihnatko [01:44:39]:
Yeah, there's nothing— there's, there's nothing that gets my attention like a .01 release shortly after the main release, because that means that there was something that they found that was so mouth-drying that they said, yeah, let's not wait until .1, let's get this in people's hands right quick. So it's either a security issue or like a show-stopping bug.

Christina Warren [01:44:59]:
Again, in this case it was, it was, it was an unlock thing where the phone was just hardcore crashing.

Leo Laporte [01:45:03]:
Yeah, it says, uh, and this is on my iPhone right now. Wrong screen. On the— yeah, there you go. Uh, it says, uh, iPhone 18 Pro and iPhone Pro Max unexpectedly restart when Face ID fails to authenticate. That's not only gonna not recognize you, but I'm gonna reboot. A color artifact may appear on photos captured at 2x and certain lighting conditions, opening Notification Center and Control Center simultaneously may cause the touchscreen to be unresponsive. Yeah, I think I'm going to update those. That sounds like a good thing to update.

Leo Laporte [01:45:42]:
27.0.1. And presumably, uh, I don't— it doesn't mention that AT&T issue.

Andy Ihnatko [01:45:49]:
Yeah, there's some, uh, uh, ScooterX just posted into the, into the Discord that No fix. It's not a fix for— so it's a mysterious way. Again, this is, this is the sort of problem that's frustrating as hell because it could be a problem at all kinds of points during the chain. The fact that it's only affecting AT&T on certain— on a certain model phone makes you think that it's an AT&T problem. But yeah, that's not necessarily the case.

Christina Warren [01:46:14]:
Well, I mean, I would say unknown at this point. It's unknown at this point. I would say that Apple has the ability, and so does AT&T for that matter, that if there is something that is going to be a, like a wireless, you know, update or whatever, like they could issue separately from an iOS update, you can issue, you know, a baseband update to the device itself.

Andy Ihnatko [01:46:36]:
Right.

Christina Warren [01:46:36]:
And so if it is something with AT&T's towers, they have the ability to, you know, basically troubleshoot that with Apple and issue an update that is not going to be part of the same iOS cycle. And that happens fairly frequently across carriers. So who knows what the case is, but that's, that's a, I mean, it's kind of ironic when you consider that, you know, AT&T had that original partnership with Apple for the first couple of years. But it does definitely seem like in the last few years that this has not been, you know, like AT&T hasn't been getting all the latest and greatest features the same way that T-Mobile and Verizon have. So.

Andy Ihnatko [01:47:12]:
Yeah, I got to tell you, if they're, if my telegraph service wasn't so good, I would switch to T-Mobile. T-Mobile in a minute.

Leo Laporte [01:47:19]:
You know, I, I think I recognize your key. You have a certain style that I just know it's AI on the line. Uh, we're gonna take a break on that. Very—

Andy Ihnatko [01:47:32]:
The use of dashes instead of dots.

Leo Laporte [01:47:33]:
Yes, yes. Thank you, thank you.

Andy Ihnatko [01:47:35]:
I'm here.

Leo Laporte [01:47:36]:
M dots. Uh, Andy Inaco at inaco.com. Everybody should become a member. It's a great site. for keeping up with what's going on in the world of tech, plus great writing. I mean, that's what Andy is so good at.

Federico Viticci [01:47:49]:
Thank you.

Leo Laporte [01:47:50]:
And Christina Warren also here from GitHub. And we're really pleased to have the editor-in-chief of MacStories, Federico Viticci, with us, macstories.net. And if you are at all interested in M5 Ultra and local AI, that's the definitive article, absolutely definitive article.

Federico Viticci [01:48:08]:
Thank you.

Leo Laporte [01:48:10]:
Apple's stock price, I'm noticing, tumbling down 2.66%. Possibly a sell-off because their price has been so high. Possibly that Mark Gurman article might have shaken some people in the, uh, in the markets. Or maybe it was Meta Connect. Time for our Vision Pro segment. The Vision Pro segment really this time should have been brought to you by Meta because at Meta Connect they announced a device that is a third of the cost of the Vision Pro, a fifth of the weight. Mark writes, Meta's VR glasses are what the Apple Vision Pro should have been.

Federico Viticci [01:48:59]:
Uh-oh.

Christina Warren [01:49:00]:
I mean, I mean, I don't think— points were made.

Leo Laporte [01:49:04]:
Now it does need a little, you know, backpack because it is lighter. That's one of the ways you make it lighter is you got not only a battery pack, but I think computing is offloaded onto this little— but, you know, you were hanging a battery from your belt anyway. Why not put the computer there too?

Andy Ihnatko [01:49:18]:
Yeah, absolutely.

Christina Warren [01:49:18]:
I was going to say, it's not like that. It's not like the Vision Pro is, you know, better in that regard.

Andy Ihnatko [01:49:23]:
You look a lot less goofy in them too.

Leo Laporte [01:49:25]:
German pointed out, yet you do this, you know, and because people—

Andy Ihnatko [01:49:28]:
I mean, you don't look normal, but you don't You look like a bug, but you look better.

Leo Laporte [01:49:33]:
Yeah, but you don't— yeah, you don't look like it's a nerd helmet. Look like you're maybe wearing some elaborate sunglasses.

Andy Ihnatko [01:49:38]:
You look like you've made a fashion choice. Yes, statement sunglasses. And there's certain— but the mere fact that they were— you could credibly wear them without feeling as though, okay, I got to get these things off my head, that's the most important, the biggest win. It's like, that's, that's non-negotiable.

Leo Laporte [01:49:54]:
German writes, I spent nearly an hour with Meta's VR glasses and I was blown away by their comfort and compact size. Now, he also has been saying that Apple considered such a device but has canceled it, doesn't want to do it. A year ago, Apple shelved another attempt to revamp Vision Pro, the Vision Air, that would have been very similar. Another project, N109, was canceled as well. It envisioned a lighter, lower-cost Vision Pro with a less powerful chip. So there's really some question, I think, about how much Apple's committed to Vision Pro. at this point. I hate to bring you the bad news, but—

Christina Warren [01:50:29]:
No, obviously they're— look, come on. If it had sold well, then that would be one thing. But we, we, we— this segment is a joke for a reason. We are the premier Vision Pro podcast. But the fact that the most interesting thing to happen in this space is not coming from Apple, but it's coming from Meta, I think is telling. And look, the interesting thing I think about this particular product from Meta is I think that it really kind of hones in on Exactly why, if this sort of device is to take off, and I think that is still a very big if, that it's doubling down on the areas where we've seen success with the Vision Pro, which is with an immersive, you know, 3D, you know, kind of content or watching, you know, movies, right? And potentially enhancing some productivity and using it as kind of a second screen thing. But I think that most people, I mean, my experience with Vision Pro, and granted, I don't have one, but I don't have one because I know I wouldn't use it. But the thing that I would use it for would be to watch video.

Christina Warren [01:51:26]:
That would be the primary purpose. And I— there's no way I could justify spending that much money to have something be on my face and leave marks on my face, you know, to watch movies as much as I would like to. And as much as I have, frankly, thousands and thousands and thousands of dollars in the Apple ecosystem in terms of purchased films, But something like this from, from Meta, assuming you could get it to work with, um, you know, your content either through Movies Anywhere or some other service to pull those things in, because I seriously doubt that Apple will support it, um, would be great potentially. And, and certainly, I mean, it's, it's way less heavy, way less expensive. And I think it kind of doubles in on saying for the people who aren't wanting to go after something from a gaming perspective, but or wanting to just kind of have a different view, either for viewing video or again, maybe productivity. That's what people want versus the spatial awareness and being able to depend things any place you want and the weird googly eyes and the very impressive technical things that the Apple Vision Pro has. But like that, as a consumer product, just we don't care about.

Andy Ihnatko [01:52:33]:
Yeah, it feels like, again, retrospect is always, you're always a genius when you look back, but it seems as though, I wish that Apple had decided to launch this platform the way that they had launched the iPhone, where we're going to start small and then we're going to build up from there. It doesn't have to be 3G. We can start with 2G. It doesn't have to have apps. We'll add apps later on. We'll basically— much like they launched the Apple Watch, where we're going to figure it out as we go along. We're going to get something out there. But, yeah, if they'd started off with something that was more along these lines, even if if it were a $2,000 device.

Andy Ihnatko [01:53:07]:
That was just, again, minimum requirement is it has to be comfortable enough to wear that people are going to want to wear it. Once it's as bulky and as uncomfortable as the Vision Pro, it has to be the miracle platform of the century to be worth that kind of strain. But we'll see. As you've said before, they canceled the Apple But a lot of the work that they did in artificial intelligence to build the Apple Car paid off well for other products. Hopefully, we'll see that same tale here.

Leo Laporte [01:53:41]:
Yeah. Mark, despite all the things he likes about it, does point out it only has a 70-degree field of view compared to the 100-degree field of view on the Vision Pro. The screens aren't as high resolution.

Christina Warren [01:53:55]:
Right.

Andy Ihnatko [01:53:55]:
But does anybody care? That's the big point. Yeah, they decided to go, we're going to make the best version of this product we possibly can without taking into advantage, are people still going to want to wear this thing and use this thing and pay $3,500 for this thing? And the answer unfortunately was no.

Leo Laporte [01:54:13]:
I honestly, when I saw it, I thought this is a, uh, this is an inertia product from Meta. Meta had spent a lot of money and a lot of energy becoming the VR company. They renamed the company. Uh, I think they've pivoted to AI, but they had this on the drawing board and they thought, well, let's just put it out there. And, you know, I mean, they had— they bought, you know, Quest, and they have the Meta Quest. And I, you know, I had a couple of them and stuff. Federico, have you ever played with VR? Are you a Vision Pro guy?

Federico Viticci [01:54:46]:
I actually have a Vision Pro.

Leo Laporte [01:54:48]:
Congratulations.

Federico Viticci [01:54:50]:
Thank you. Every once in a while I charge it.

Christina Warren [01:54:55]:
I was gonna say, when did you last use it? When did you last use it genuinely?

Andy Ihnatko [01:55:01]:
I like it enough. I don't want the batteries to be destroyed because I didn't charge it. That's how much affection I have for this platform.

Federico Viticci [01:55:08]:
I think it's actually bricked at this point. Oh dear. I'm pretty sure I put VisionOS 27 I have an iPhone 11 beta 1 on it in June, and I never updated from beta 1. Oh no.

Leo Laporte [01:55:23]:
Wow.

Federico Viticci [01:55:23]:
So I don't know, I will see some kind of error on screen.

Christina Warren [01:55:27]:
So you need the $60 developer cable to be able to connect it to your Mac, which of course you have, but, but I'm just pointing out that like, this is the irony of this product, is that if you want to even do this, you have to buy a special cable, a developer cable, to be able to to connect it to your Mac.

Leo Laporte [01:55:43]:
Once you do that, would it help if I told you you could put Linux on it? No.

Christina Warren [01:55:48]:
No. No. Why? Why would we? No, but also, why would we want to? Linux would never be able to support any of the custom stuff that this is.

Leo Laporte [01:55:56]:
Come on. I know, I'm just joking.

Christina Warren [01:55:57]:
I know, I know.

Leo Laporte [01:55:58]:
I'm trying to put 2 of Federico's favorite things together.

Andy Ihnatko [01:56:02]:
Emacs on that device would be legendary.

Leo Laporte [01:56:05]:
Oh, actually, Mark does say it's interesting that the The Meta, he says the gestures are the same, by the way, as the spatial computing on the Vision Pro. But he says the keyboard's much better. It's actually, you can actually fake type very well with it. I still, I don't think this is a, I think this is a market that nobody, I don't think there's any market for it.

Christina Warren [01:56:26]:
Is this like 3D TV? Like we get really excited and we think like this is gonna be the future. And then you're like, actually, no. And it's interesting because like you said, like both you and Andy made a point, like rename the company after this. They went all in on this. Obviously now I think they've pivoted to AI and I'm sure they had this kind of in the back channels and so you might as well ship it. And but I wouldn't buy this product either, to be clear. Like I think this is definitely something that's more compelling than Vision Pro, but I wouldn't buy this either.

Leo Laporte [01:56:54]:
No, I thought, oh, it's a third of the Vision Pro and I'm still not gonna buy it.

Christina Warren [01:56:58]:
I still don't want it. Like I had a Quest, whatever the chip is, cheap Quest was, the Quest 2, the Quest 3, whatever the, yeah, the cheap one was that they then continued to raise the price on. So it was one of those things.

Leo Laporte [01:57:08]:
I bought the $1,500 Pro version.

Christina Warren [01:57:11]:
Oh yeah, for some reason.

Leo Laporte [01:57:12]:
I don't know what I was thinking.

Christina Warren [01:57:14]:
I mean, you were thinking that you are Leo Laporte and this is what I should keep up on, these categories, and that you buy all these gadgets. And that's what you were thinking. Yeah, which I totally respect. I got one of the ones I guess was like $300 or something, and I was like, fine, this is fine. And I think I used it I was on the Kickstarter.

Leo Laporte [01:57:30]:
I think Beat Saber is great. I did the Kickstarter. What was it?

Christina Warren [01:57:32]:
Oh, I did that one too. Yeah, the original. I was too for the original. I remember watching this. Yeah, for the original.

Leo Laporte [01:57:38]:
And I had that, and then Meta sent me a replacement because they bought the company then after I'd supported it with my money on the Kickstarter. So they— I think they gave me a replacement, and then I bought the Pro. I kept, like Meta, like Mark, thinking maybe this is a category.

Christina Warren [01:57:55]:
Yeah, I had— I reviewed an HTC Vive, and I liked that a lot. It was the Oculus— was Oculus Rift. Yeah, Oculus Rift. And then they had the Oculus Quest and other things, and like HTC Vive. And like, there's some really interesting immersive things you can do. I even did— before I actually— one of the final big stories I did before I transitioned into engineering was I talked about VR filmmaking and went to the Tribeca Film Festival and interviewed a bunch of filmmakers and had had some really interesting experiences to seeing like, okay, how are, you know, filmmakers trying to use this technology? And then the challenge was, there were dual challenges. One is, how do you record these things in stereoscopic and get it so that it can be in compelling quality? And then the secondary thing is, how do you distribute this? Because some of the files for even very, very small, you know, like 5-minute clips were like 20 terabytes in size or something. And we're just, we're were not really at a place where you could widely distribute.

Christina Warren [01:58:52]:
And a lot of those things have been handled. But this was a thing— this was 9 years ago— and it was very interesting to kind of think about, okay, is this going to potentially be a new medium that could be used maybe not by everyone, but by certain sectors of people? And so gaming obviously became one of the big areas, and video, and then kind of got into this more general character of spatial category of spatial computing. But I think that, I think all of us just, we have to look at this. We're a decade into this now, more than a decade into this, and it's still a very niche thing. And that's not to say that people shouldn't experiment. And if Meadow wants to experiment in this, that's fine. But I mean, no one's, no one's buying this. This is not like a big platform.

Leo Laporte [01:59:37]:
I'm more excited about that Tamagotchi, to be honest.

Christina Warren [01:59:39]:
I was going to say, that's the sort of thing, right? And this is what happens sometimes is that you have predictions and this happens with every technology company and just throughout the industry where you really think like this is going to be the next big wave. This is what we think is going to take off and it's not. It winds up being something that comes completely far afield, which is not to take away from the initial brilliance of that first Oculus Rift, right? Like that was such an impressive, you know, kind of demo, especially what someone so young was able to do with with relatively, you know, low-spec hardware and was able to do really interesting things. And maybe we do eventually get to the point where they will be lightweight enough and the lenses will be good enough that this will be a real category. But I think if anything, what's going to be and what has proven to be a real category is like the, you know, the Meta Ray-Bans. And that's the sort of thing where—

Leo Laporte [02:00:29]:
Yeah, I agree with you.

Christina Warren [02:00:30]:
For all the different privacy constraints and issues people have with that, And that's been a really successful product. That's a thing people really like and having kind of that augmented experience versus, okay, we need to have a headset, whether the computer battery is on my hip or not, or whatever the case may be. Like, there's always going to be like a limited market for that. But if there is something out there, then having these more aware and especially with AI, you know, connected, you know, kind of, you know, glasses of, okay, I can give you more context about what's around you. That was kind of the original goal of Google Glass way back when. And, and maybe there is something there, or maybe in 20 years we'll all be talking about how, well, yeah, just, just around the corner we're gonna, we're gonna have these, these headsets.

Leo Laporte [02:01:14]:
Someday cars will drive themselves.

Andy Ihnatko [02:01:16]:
Right. Sometimes you skate to where the puck is going to go, and then you look around and everybody's playing basketball. That's what Apple has found themselves in right now.

Leo Laporte [02:01:25]:
You, you brought a hockey stick to basketball game.

Andy Ihnatko [02:01:27]:
Oh no, it's a Wednesday. They covered up the ice. ice with 4— okay, my bad.

Leo Laporte [02:01:32]:
But that parquet is beautiful, you can't deny it.

Andy Ihnatko [02:01:34]:
It really is.

Leo Laporte [02:01:35]:
Um, all right, that's the Vision Pro segment, such as it is.

Christina Warren [02:01:40]:
Now you know we're done talking the Vision Pro.

Leo Laporte [02:01:46]:
But I bet you see the thing about the Tamagotchi that I like— it does have a camera, it does have micro array mics. It is really the Meta glasses without putting them on your face.

Andy Ihnatko [02:01:56]:
I think when This is something that keeps coming up. People taking the Apple Watch off of the band and putting it in a 3D-printed case and just using it like a pocket watch or a fob. Wouldn't that be— to me, that would be a really compelling product if Apple could put continuity features on it and a limited number of AI features on it and basically also make the iPod-style features of it really, really easy to access. I don't mean— we've talked about this before. I just, between my plain fitness band and my love for Casio watches, I have no need need for like a smartwatch, but that— there's a reason why my jeans have a watch pocket. And I would actually say, you know what, I might have $300 for a device like that.

Christina Warren [02:02:38]:
Right. No, you're not wrong. I bought one of those, um, accessories that like turns your Apple Watch into like an iPod type of thing. And it was— I mean, it was kind of a— incredibly janky, but it was still kind of a fun thing. Um, but yeah, I mean, I think— but I, I thought about this for a long time. It's like, You could very easily turn that into, as we've discussed, like a much broader accessory and be— and that would be kind of a perfect thing. I mean, even if they wanted to make it a unique accessory different from the watch, you could use many of the same components that you use in the watch in a device like that and maybe alter your operating system slightly. But yeah, I mean, I'm with you.

Christina Warren [02:03:15]:
I think that there's, there's something to that.

Leo Laporte [02:03:17]:
And I think Apple—

Christina Warren [02:03:18]:
I think in that case, this should be a little nervous.

Leo Laporte [02:03:21]:
That Gurman does say that Apple considered a pendant and has decided not to make it. But I think this, this, it doesn't feel like this is the, you know, thing that's going to replace the iPhone.

Christina Warren [02:03:33]:
No, I—

Leo Laporte [02:03:34]:
It could.

Christina Warren [02:03:35]:
I mean, it could. I don't think it replaces the iPhone, but I think that maybe it gets people less reliant on their iPhone as the only device. Right.

Andy Ihnatko [02:03:45]:
Right.

Christina Warren [02:03:46]:
And so I think that is more of the problem than is this going to replace your phone? Because like, If you were faced, Leo, with what am I going to leave the house with, my phone or one of these, you know, Tamagotchis?

Leo Laporte [02:03:56]:
Cute little Tamagotchi.

Christina Warren [02:03:58]:
No, you're not. No, you have to pick one.

Leo Laporte [02:04:01]:
I can make phone calls. What if I can make phone calls? I could take pictures. Yeah, I can call an Uber.

Christina Warren [02:04:07]:
You're not gonna be able to send your text messages.

Leo Laporte [02:04:09]:
Like, sure you can. It's Muse.

Christina Warren [02:04:11]:
Yeah, you just say send a text message to Christina, uh, and then I'm gonna get a green bubble and I'm gonna be like, what the What the hell is this? Who is this? I'm not talking to you.

Leo Laporte [02:04:18]:
Oh, green bubbles are so last year. I don't think you have to worry about that. By the way, this is my cute little muse guy. I told him, I told him, dress for the weather every morning when you get up, figure out what the weather's going to be and dress for it.

Christina Warren [02:04:35]:
Lele, is that his name?

Andy Ihnatko [02:04:36]:
Lele.

Leo Laporte [02:04:37]:
Yeah.

Christina Warren [02:04:37]:
Little Lele.

Leo Laporte [02:04:38]:
Yeah. Yeah. Yeah. All right, we gotta take one last break and then it is pick of the week time. I've already bought what could have been your pick of the week, Federico. I already bought that, the KVM.

Federico Viticci [02:04:53]:
Oh, that was fast.

Leo Laporte [02:04:54]:
Yeah, well, that's exactly— so every once in a while, you know, my computer will reboot and, and then say, oh, you need to enter a password or something. I mean, it's all set up so it shouldn't, but you never know. And if I had that thing I could just send her the password even if I'm in, you know, Kuala Lumpur. I can do it. That's what I care about.

Andy Ihnatko [02:05:15]:
Or Rome.

Leo Laporte [02:05:16]:
Maybe I'll go to Rome. I don't know. I have a— I have a craving for some gelato.

Federico Viticci [02:05:21]:
Okay.

Leo Laporte [02:05:22]:
I'm just saying. Pizza. Emma's Pizza is my favorite pizza place in the world. It's in Rome. It's really good.

Federico Viticci [02:05:29]:
Which one?

Leo Laporte [02:05:30]:
Emma. You ever hear of that one?

Federico Viticci [02:05:32]:
No.

Leo Laporte [02:05:33]:
I'm gonna have to come bug you and you can tell me all the best places to go.

Federico Viticci [02:05:38]:
Okay.

Christina Warren [02:05:38]:
Yeah.

Leo Laporte [02:05:39]:
I had a lovely cacio e pepe, or was it a carbonara? It was a carbonara at Fortunata. That was good.

Federico Viticci [02:05:46]:
Can't go wrong.

Leo Laporte [02:05:47]:
Yeah, that was really good. Yeah, there's a nice lady, nice Italian mama making the pasta in the window. And then they put it in a bowl and they give it to you. It's great. Chris, actually, let's do our guest, Federico. I don't know if they mentioned the pick of the week. Do you— is there something? You already have picked one thing, but if you'd like to pick something else that these days is exciting you.

Federico Viticci [02:06:09]:
I want to point people to an iPhone and iPad app. It's called OpenMinis.

Leo Laporte [02:06:17]:
I never heard of it.

Federico Viticci [02:06:18]:
Yeah, so this is pretty cool. It's a chatbot app. It's more of an agent app, really. Um, think of it as a lightweight lightweight version of Hermes Agent.

Leo Laporte [02:06:30]:
Okay.

Federico Viticci [02:06:31]:
But running on the iPhone and iPad with a twist. So this app is open source. It's made from a single developer in China. This person figured out how to make a custom agentic harness on iOS that has access to a whole bunch of Apple-specific integrations.

Leo Laporte [02:06:55]:
Oh.

Federico Viticci [02:06:57]:
This person turned— and this is the genius part of this app— they turned native iOS features like access to your photos, music, maps, health, all these like exclusive Apple frameworks. This guy turned it into a bunch of CLIs for agents to use. And so this is basically Siri AI as it was meant to be as an agent.

Leo Laporte [02:07:29]:
Does it use the intents?

Federico Viticci [02:07:31]:
It does not. It does not. It doesn't. It just works with HealthKit, EventKit for calendar and reminders.

Leo Laporte [02:07:39]:
Oh, perfect. Perfect.

Federico Viticci [02:07:40]:
Works with PhotoKit so it can access and process your photo library. It uses the same native permission prompts of iOS because it's actually using using native iOS APIs behind the scenes. The trick is that you can log in with your favorite AI provider, including your OpenAI subscription, or you can put in your API key. There's support for a whole bunch of providers. And you can use GPT-6 in OpenMinis and say, hey, can you visualize my photos on a map? And it's gonna put together— Yeah, it's gonna put together a little map based on your iPhone photos. Or you can say, give me an overview of my week, and it's gonna pull from your iCloud calendar and your reminders. And maybe it's gonna look into your contacts and see if you have a birthday coming up from a contact. All of these iOS integrations that all of these companies are now doing, right, like Meta with news.

Federico Viticci [02:08:42]:
Allegedly OpenAI is going to have more iOS integrations in ChatGPT in the future. This single indie developer from China did it months ago in an app that you can download for free from the App Store, and it keeps getting better every week. There's a new TestFlight beta. It's remarkable what this person has achieved.

Leo Laporte [02:09:07]:
I'm setting it up right now with Anthropic with my Opus, actually, probably do Sonnet 5.5, the new Sonnet 5.5.

Federico Viticci [02:09:13]:
It works great with Sonnet 5.5. Yeah, it can use subagents. This custom hardness can use subagents on iOS. It can talk to your devices with Tailscale. I just used OpenMinis here on my phone as we were recording. I set Opus on a task And I said, can you figure out how to add support for GPT-6 UltraFast, which came out 2 hours ago?

Leo Laporte [02:09:46]:
Yes.

Federico Viticci [02:09:47]:
To OpenMinis, because this harness, the OpenMinis can tweak itself. The harness is able to reconfigure OpenMinis.

Leo Laporte [02:09:55]:
It's too good.

Federico Viticci [02:09:57]:
Opus put together a custom, a little tweak for the OpenMinis harness. So now I can I can use UltraFast already, and I'm sure the developer will bake in official support by tomorrow. But yeah, this is one of my favorite discoveries of the year, which funnily enough came from a ChatGPT deep research. I was like, wait, this has to be an hallucination. This app cannot possibly exist. And sure enough, I clicked and it was working.

Leo Laporte [02:10:27]:
I'm signing in right now. This is incredible.

Federico Viticci [02:10:29]:
Yeah, it's, it's a— have fun. I wasted hours and hours.

Leo Laporte [02:10:33]:
You've made me very happy. I only wish I could point it at my local AI, but I have—

Federico Viticci [02:10:38]:
Oh, you can. Oh, Leo. Oh, Leo. Oh, you can.

Leo Laporte [02:10:43]:
Can I point it to Hermes?

Federico Viticci [02:10:45]:
I created, for my review of the M5 Ultra, I created a little bridge sitting in front of MLX. So that I can use Quen in OpenMinis.

Leo Laporte [02:11:00]:
Yeah.

Christina Warren [02:11:01]:
Oh my goodness.

Leo Laporte [02:11:03]:
Okay. You know, I'll see y'all later. I got, I got somewhere to be. Great pick. Uh, it, it, it is, is it free or is it an in-app purchase?

Federico Viticci [02:11:12]:
Free, free and open source on GitHub.

Leo Laporte [02:11:16]:
Very nice. And there's a very good article from July. See, if I had just been paying attention to macstories.net.

Federico Viticci [02:11:22]:
probably at the beach.

Leo Laporte [02:11:23]:
I was probably at the beach instead of reading the latest blog posts at macstories.net, which I should have been doing. And of course, all the information about how Federico's got this working, I'm sure with his local AI as well, would be in here. Wow, that's really cool. That's one bad thing about Hermes— there's really no good mobile interface to it. There's a bunch of janky stuff.

Federico Viticci [02:11:48]:
You should go— I will give you one extra peek. quick, go check out— there's an open test flight. This other indie developer, Florian, I believe— go to cadu.bot. That's C-A-D-U dot bot. It's an upcoming iPhone and iPad client for Hermes. It's really nice. It even supports the new bot mode.

Leo Laporte [02:12:12]:
Yes, I use the bot mode.

Federico Viticci [02:12:14]:
Yeah.

Leo Laporte [02:12:14]:
Oh, okay.

Federico Viticci [02:12:16]:
That's That's a, that's a—

Leo Laporte [02:12:17]:
I have too many toys.

Federico Viticci [02:12:19]:
I know, right?

Andy Ihnatko [02:12:20]:
I know.

Leo Laporte [02:12:20]:
By the way, it, uh, OpenMinis does support Copilot via GitHub, so you can use your GitHub Copilot subscription as well.

Christina Warren [02:12:28]:
Nice.

Leo Laporte [02:12:29]:
Christina's pleased to hear that, right?

Christina Warren [02:12:31]:
Oh no, this is life-changing for me. I, I'm, I can't believe I didn't know about this until now. So thank you so much, Federico, because this is great. You're welcome. No, I will absolutely use my, my GitHub Copilot subscription with that. Thank you.

Federico Viticci [02:12:41]:
Yes.

Leo Laporte [02:12:42]:
Do they give you a bunch of tokens? Do you get like—

Christina Warren [02:12:45]:
oh, I might—

Leo Laporte [02:12:45]:
should I be really jealous?

Christina Warren [02:12:47]:
Oh, I, I, I get all the tokens, Leo.

Leo Laporte [02:12:49]:
All the tokens? Oh man, I get all the tokens. Yeah, just, just— that's a dream come true.

Christina Warren [02:12:57]:
It's nice, it's nice.

Leo Laporte [02:12:58]:
Did you, Federico, did you sign up for the ultra fast?

Federico Viticci [02:13:02]:
Yeah.

Christina Warren [02:13:04]:
Are you insane?

Leo Laporte [02:13:05]:
Do you have to have a $500— you have to have a $500 a month subscription? Yeah, but it's 300 tokens a second.

Federico Viticci [02:13:15]:
Look, I figured I'm just gonna take it for a spin for a month.

Leo Laporte [02:13:18]:
Month. That's what I was thinking of.

Federico Viticci [02:13:20]:
I'll actually give you—

Leo Laporte [02:13:22]:
You'll be hooked on it.

Federico Viticci [02:13:23]:
It's not a scoop or anything. Now, I was scrolling through the response in OpenMinis.

Leo Laporte [02:13:30]:
Yeah.

Federico Viticci [02:13:31]:
Opus took it on its own to actually run benchmarks. On my Mac Studio. I'm looking now. It created a bridge on my Mac Studio to measure the performance of the 3 speed tiers that OpenAI now offers.

Christina Warren [02:13:48]:
Oh, nice.

Federico Viticci [02:13:49]:
Standard, fast, and ultra fast.

Leo Laporte [02:13:51]:
Yes.

Federico Viticci [02:13:52]:
And, and OpenAI is saying that they run at 300 TPS.

Leo Laporte [02:13:58]:
Yeah.

Federico Viticci [02:13:58]:
According to Opus, it's just short short of 100 at 97 TPS.

Leo Laporte [02:14:06]:
Oh, I've already got that on my Sparks. I don't need—

Federico Viticci [02:14:09]:
So maybe it's a, it's a, it's, it's a rollout type of issue.

Leo Laporte [02:14:14]:
Everybody's doing it is why.

Federico Viticci [02:14:15]:
Everybody's doing it. I also saw as I was quickly scrolling through that some people were able to completely run through and exhaust their Pro 500 That I believe.

Leo Laporte [02:14:29]:
That's exactly why I don't sign up for stuff like that, because I know I would. Oh, I absolutely know. You know what, I do have a subscription to Cerebras, which is using custom hardware. They have— they only do a few small models, but the Quen runs— Quen runs really fast on it. It's kind of fun.

Federico Viticci [02:14:45]:
1,800 TPS.

Leo Laporte [02:14:47]:
It's kind of neat to see something that fast.

Federico Viticci [02:14:49]:
Yeah.

Leo Laporte [02:14:50]:
So if I need, you know, I mean, it's relatively intelligent. Agent 27B. But if I need a dense model for some quick stuff, and it's pretty cheap too, I like doing that. Uh, Christina, I think we should— thank you, Federico, by the way, for being here. I'm gonna let you go to bed because you've been very patient.

Christina Warren [02:15:07]:
Thank you for being here.

Leo Laporte [02:15:08]:
And I'm sure your girlfriend wishes you were done. So it was such a pleasure.

Federico Viticci [02:15:14]:
It was an honor for me to, to be here. I've been a huge fan for the longest time. Well, thank you.

Leo Laporte [02:15:20]:
So thank you Well, if you, if you don't mind, I might have you come back. And what I'd actually like to have you on, uh, our, our AI show also, then we can really let our hair down.

Christina Warren [02:15:31]:
Yeah.

Leo Laporte [02:15:31]:
And talk serious AI stuff. Okay.

Federico Viticci [02:15:34]:
We were not serious enough here, but yes, anytime I'll be happy to.

Leo Laporte [02:15:38]:
I, I love talking to people who are doing interesting things with local. Uh, that's a, uh, to me, that's really where the frontier belongs. So I appreciate it. Thank you, Federico.

Federico Viticci [02:15:49]:
Oh, thank you.

Leo Laporte [02:15:50]:
Great to have you.

Federico Viticci [02:15:50]:
It was great to talk to you.

Leo Laporte [02:15:51]:
Christina, your pick of the week.

Christina Warren [02:15:55]:
Yeah. Okay. So this is actually fun. So there's a developer who's been around for the scene for a really, really long time. Great person, Rosina Keller, who just recently re-released, I guess, kind of updated some apps that have been around on the Mac back in the olden days. So if you've been a macOS user for a long time, you might remember things like Haxies and the fact that you used to be able to really customize and plug in and change like the way your Mac worked. And, and that hasn't really been a thing that you've been able to do for a long time. But Rosino recently, just actually, I think maybe last week, just relaunched a couple of different apps, including one called Fruit Menu Redux and one that is called Zounds Redux.

Christina Warren [02:16:35]:
And these are both basically kind of new versions of what these older hacksies were from back in the day. And Fruit Menu Redux is basically a way to customize your Apple menu so you can pin various items or hotkeys to individual files or have it like it's an extra menu bar. Just basically customize. You could have like go to system settings directly from the Apple menu. and, you know, go to shortcuts and go to other sorts of things, really customize it. So that's one option. And then Zounds is one that basically lets you kind of customize various sounds that work along with your interface. Anyway, I think this is really fun.

Christina Warren [02:17:24]:
I'm glad to see this. I think Rosina is a great person. Both of these apps through tomorrow, I guess, through today, they are $7 apiece where you can get a bundle, I think, for a discount from that. Right now, you can get the 2 of them together, or they're $10 apiece normally, and then there's like a bundle that's a discount. Right now, they're $7 US apiece, or you can get a bundle, and I think it's like, you know, $13 or something if you get them together. That ends tomorrow. But anyway, I just wanted to give them support, and they're—

Leo Laporte [02:17:53]:
I love the name of the company, Totally Not Malicious Software.

Christina Warren [02:17:57]:
Totally Not Malicious Software. I was going to say, the URL is great. And, uh, but this is the longtime, you know, like, like Mac-assed Mac developer, uh, somebody, you know, who actually worked at Apple for a time and, and is just, you know, I'm really glad to see this sort of innovation coming back, uh, to macOS. So, uh, Fruit Menu was, was a one that I, I bought them both immediately, but, uh, I just wanted to give, uh, them a shout out and, and tell people if you like these sorts of little fun utility apps, apps. Support a good indie developer and check them out.

Leo Laporte [02:18:29]:
That is awesome. What's his name or her name?

Christina Warren [02:18:32]:
Uh, Rosyna.

Leo Laporte [02:18:33]:
Rosyna. Is it a woman or a man? Do we know?

Christina Warren [02:18:37]:
I think they're— I, I think they're non-binary, but, but, but I—

Leo Laporte [02:18:40]:
Okay, could be either then. All right, Rosyna, R-O-S-Y-N-A, totallynotmalicioussoftware.com.

Federico Viticci [02:18:50]:
I—

Leo Laporte [02:18:50]:
the whole ethos of this is fantastic.

Christina Warren [02:18:53]:
Yeah.

Leo Laporte [02:18:55]:
Andy and Utco, pick of the week.

Andy Ihnatko [02:18:58]:
My pick of the week is a keyboard that I picked up a few weeks ago. I've been eyeing the Keychron K3 keyboard for a long, long time, and I found one that was— so this model is a couple years old, and so basically it looks like Keychron has been like clearing it out because I got this— I got— because I got this one actually, I got this for $42.

Leo Laporte [02:19:19]:
What?

Christina Warren [02:19:19]:
Wow.

Leo Laporte [02:19:20]:
And these are normally hundreds of dollars.

Andy Ihnatko [02:19:23]:
About $100. The K3, they have versions that are kind of like this that are more expensive, but the K3 is their $100, like, sort of halfway between like a MacBook keyboard and a big chonky like desktop keyboard. It's a really nice compromise. These are all— this is a 75% keyboard, which means that it's nice and compact. However, the keys are full mechanical Full mechanical key switches, so they feel really great. They, they sound like really great. I can't use it in the library unless I'm inside a conference room, but it's really, really great to type on. The keycaps, if you don't like the keycaps, you can buy new keycaps and switch them out.

Andy Ihnatko [02:20:01]:
All the cool things, all the cool things you can do with like a really big keyboard. Also, it's nice enough that it supports both Windows, Windows and Android and macOS and iOS. There's just a switch like in the back that lets you select between key layouts. But it does come with like all the keycaps that you would expect. I mean, the keycaps here are exactly like what I've got right in front of me on my MacBook Pro. So it's not as though you have to remember that, oh wait, that's not the Windows command, that's not the Alt key, that is Control, Option, and Command. And you can pair up to 3 devices simultaneously. So you can easily— I mean, I've got this paired to my laptop, my iPad, and my phone at the same time.

Andy Ihnatko [02:20:43]:
It's got fully backlit LED. You can also— it's got the stupid— no, I don't wanna say stupid, but the sort of like young people's LED, which is— when I first set it up, it started doing these like crazy pulses, light shows.

Leo Laporte [02:20:58]:
Mine does that too. Like, zing, zing, zing, zing, zing.

Andy Ihnatko [02:21:01]:
Like, tell me that that's not gonna be like the default state of this keyboard. But no, there's a key, there's a key right here you can switch into something more sensible.

Leo Laporte [02:21:08]:
Actually, I have it set so that the key that you just hit lights up, and I find it much, very helpful for typing better.

Andy Ihnatko [02:21:17]:
Yeah, I mean, it's nice. I mean, again, this is the older one. The newer, the newer K3s, they're full RGB so that you can actually like set things, set your keycap backlight so that it's easy to guide to like the stuff, the different function keys, stuff like else. This does have, as the older one, I noticed that, like, you know, we talk about in design, especially with Apple, like these little moments of delight where the Caps Lock key is backlit but it turns into red when the Caps Lock is on.

Leo Laporte [02:21:43]:
That's good.

Andy Ihnatko [02:21:43]:
And with this older one, that's the only one that's actually RGB. I think the newer ones, all the keys are RGB. And there's of course a desktop app you can use to customize that however I want. But it really, I mean, if I had a Bluetooth or USB— It is Bluetooth or USB. There's a switch again on the back that you can pick and choose from. Battery lasts not as much. Excuse me, again, this is the older one, but the newer one, the battery lasts about 70 to 80 hours. This one lasts about half of that.

Andy Ihnatko [02:22:10]:
That's why some of the reviews you might look up on this particular model might not be super, super great. But the thing is, this is like the beautiful happy medium that I've sort of been looking for. I am a sort of— I'm not a keyboard nerd, but like, I'm really—

Leo Laporte [02:22:24]:
He's a nerd.

Andy Ihnatko [02:22:25]:
Oh, it's like if I had more money, I could be a keyboard—

Leo Laporte [02:22:27]:
I have the DOS keyboard you recommended. I do have the Keychron you recommended.

Andy Ihnatko [02:22:32]:
That is, that is, that is my main like desktop keyboard. But the thing is with this is that like there are times where like I am gonna be gone for like a week or a weekend, but I do have like goals, writing goals to perform. And this is small and light enough that I can justify putting a wrap around it and putting it like in my, in my bag. It is, it is very very solid. It's not as heavy as the normal $300 desktop keyboards you might find, but it is solid. It will not skate around on the desktop. Flip-out feet so that it holds up at a nice angle. Like I said, at $42, it's a freaking steal.

Leo Laporte [02:23:07]:
No kidding.

Andy Ihnatko [02:23:07]:
On Amazon, the best price you can find is— there's, there's another like reseller of like Overstock that has it for $65. But even if you're buying the brand new, the newest version, the newest version that came 3 for $100. If I had borrowed this from a friend for a weekend after they say, oh my god, I got this for $42, I would have no qualms about spending $100 for the latest one because it really is just quality. It really is.

Leo Laporte [02:23:31]:
And do you have the red or the brown switches?

Andy Ihnatko [02:23:33]:
I've got the brown switches.

Leo Laporte [02:23:35]:
Oh, those are the quiet.

Andy Ihnatko [02:23:37]:
It's a little— again, it's still too noisy to use. I love the library, but again, the feel is great. The thing is, like, again, again, I love my DOS keyboard. The problem is if I've been working all day on my MacBook, it's the difference in feel between these big, huge, chunky switches and the, the flatter MacBook switches. Sometimes it takes a while for my fingers to adjust, whereas with this one, it's a much, much easier transition because they— it's more like a laptop keyboard if the laptop were made in like 1988 or something when they didn't have to worry about weight or thickness or anything like that.

Leo Laporte [02:24:13]:
Really nice. Keychron, K-E-Y-C-H-R-O-N, and this is the K3 Max. But don't buy it from Keychron because they're gonna charge you full list. Shop around.

Andy Ihnatko [02:24:23]:
Again, the, the, the— I have, I have, I have the K3 Max like in the show notes because it's the one that you can buy like right— it's, it's still a current model. The one that I have is technically speaking the K3 version 2, which was released like 5 or 6 years ago. Again, I'm pretty sure it's no— it's listed as out of stock on the website, which— and I believe that it's been like discontinued.

Leo Laporte [02:24:45]:
But that's where you get the deal.

Andy Ihnatko [02:24:47]:
So, so, so exactly. So if you find one of these, there's a good chance it will be like new old stock, and they might— you might get a good deal out of it. Again, for some people, $100 is right on that edge of what they would might spend for a keyboard they don't think they need.

Leo Laporte [02:24:59]:
$65 if you— Right, exactly.

Andy Ihnatko [02:25:02]:
For $65, again, but again, the $5 keyboard is water-soluble. This one is actually a real, like, productive keyboard.

Leo Laporte [02:25:08]:
I have bought so many keyboards because of you. I have folding keyboards. I have all kinds of little crazy—

Andy Ihnatko [02:25:14]:
Oh God, I've got— I'm writing something about, like, for, for the, uh, for the, uh, for the iPhone Duo. Like, I'm looking back at, like, portable, like, smaller—

Leo Laporte [02:25:24]:
Oh yeah, that's right.

Andy Ihnatko [02:25:25]:
And in addition, in addition to ones that are worth recommending, I'm going— I went right through, like, my storage and found All of these weird, weird tiny keyboards that came out like when the first iPad became a thing.

Leo Laporte [02:25:37]:
Yeah, I bought them because of you.

Andy Ihnatko [02:25:41]:
At least I was only costing you like $50 or $60. I'm not trying to twist your— I'm not like Alex saying here's an $18,000 keyboard you just have to have.

Leo Laporte [02:25:49]:
That's a very good point. Thank you, Andrew. Inatko.com, I-H-N-A-T-K-O, is the website. Become a premium member. It is worth supporting Andy for his incredible writing. Lots of great stuff there, including of course all the latest Apple news. I-H-N-A-T-K-O. And he's gonna, he's gonna be talking about the iPhone Duo accessories, which I think will be very, very—

Andy Ihnatko [02:26:16]:
Because it's gonna take a while to talk Apple into lending me one, I think. So I better get that pipe, send other things into the pipeline.

Leo Laporte [02:26:22]:
I already ordered a sight unseen, uh, because I know I'm gonna need a case for this thing. The, the Nomad leather case. I know they make good stuff. I like Nomad. So yeah, I'll be ready when it comes October 23rd.

Christina Warren [02:26:35]:
Yes!

Leo Laporte [02:26:36]:
Are you excited, Christina?

Christina Warren [02:26:38]:
Oh, I can't wait. I can't wait. Well, I thought, I thought it was the— is it the 23rd that's coming out?

Leo Laporte [02:26:42]:
We ordered the 16th and then—

Christina Warren [02:26:44]:
We ordered the 16th and then it comes out on the 23rd. I think.

Leo Laporte [02:26:47]:
I may be wrong. It's in my calendar.

Andy Ihnatko [02:26:50]:
We did— we didn't talk about on the show, but you might want to jump on it. There's a couple of stories from supply chain saying that yields on the— yields on the assembly of those devices is much lower than they were hoping for.

Leo Laporte [02:27:00]:
Yeah.

Andy Ihnatko [02:27:00]:
So I mean, I don't— I don't think you'll have a problem getting one if you want one, but if you really want one like early—

Leo Laporte [02:27:06]:
5 AM— it's a good—

Andy Ihnatko [02:27:07]:
it's a good idea to order on like the day that pre-orders launch.

Christina Warren [02:27:10]:
Right, right. That's right. That's right. That's right.

Andy Ihnatko [02:27:12]:
Yeah.

Christina Warren [02:27:13]:
It's coming up Friday the 23rd. That's right. Because I'm going to be going out of town right after that. And I like purposely made it clear. I was like, no, I'm leaving on the 24th. I'm not. It's iPhone Day.

Leo Laporte [02:27:23]:
I will be in Thailand on the 16th. And the good news is 5 AM Pacific time is 8 PM Bangkok time. So I will be having dinner ordering my iPhone 12.

Christina Warren [02:27:37]:
Yeah. And someone in the chat posted earlier that they had like a The, the get the, the whole getting, um, getting ready thing on the 12th. Yeah, so you'll be able to preset things then, which will be good. And that way—

Leo Laporte [02:27:48]:
Yeah, somebody was asking, emailed me and said, well, I can't, because in the past you've been able to save it. You've been able to set it all up, save it, and then when the order comes, you can press a button. Well, they didn't turn that on. They won't turn that on until the day before for some reason.

Christina Warren [02:28:02]:
Yeah, um, I'm not sure why they're waiting on that, although I think they've done that in the past before too.

Federico Viticci [02:28:06]:
It That's right.

Christina Warren [02:28:07]:
Which hasn't always been the case where you could do it well in advance. So, I mean, I think that makes sense. And so the good thing there is that if you are looking to finance, whether it's with an Apple Card, if you don't have an Apple Card already, or the Klarna thing, that I think gives you an opportunity, at least it has in the past. I can't speak definitively this time, but in the past, that has given you a time to set up your financing and go through any of the preapproval process that you might need, which if you're not just knowing what credit card you're going to put down, You might want to do this time. The same thing with any carrier trade-in offers or things like that. And so, you know, if you don't have a set payment plan set up, taking advantage of that kind of, you know, time beforehand is good.

Leo Laporte [02:28:49]:
So I'm going— I did. Yeah. So you can preorder on the 12th for— or rather save on the 12th for the preorder on the 16th. And that gives me enough time to get down to Louis Spumoni to look a lung shark, and I don't know if it's gonna take a kidney, a lung, whatever. I got 2 of each.

Christina Warren [02:29:04]:
Oh, please. No, please. You just— you did, um, just, you know, rent out a, a slice of your spark for—

Leo Laporte [02:29:10]:
There you go.

Christina Warren [02:29:11]:
For a couple weeks.

Leo Laporte [02:29:11]:
There you go. Nobody's touching my sparks. I love those.

Andy Ihnatko [02:29:16]:
Love them.

Leo Laporte [02:29:16]:
Turned out—

Andy Ihnatko [02:29:17]:
Don't trust Limetto, but you can trust Leo. Yes.

Leo Laporte [02:29:20]:
Uh, we do MacBreak Weekly every Tuesday, 11 AM Pacific, 2 PM Eastern, 18 1900 UTC. And if you want to watch us do it, you can. We, we stream it live. Of course, in the club, you watch in the Club Twit Discord, but everybody can watch on YouTube, Twitch, X, Facebook, LinkedIn, or Kick. After the fact, get a copy of the show from our website, twit.tv/mbw. You'll find audio and video there. There is a YouTube channel. All of our shows have their own dedicated YouTube channel.

Leo Laporte [02:29:47]:
You can go to the main channel, youtube.com/twit, find links to Mac Break Weekly and all the other shows. Or maybe the best thing to do, get yourself a podcast client like Pocket Casts, Overcast, Apple Podcasts, and subscribe. It's free to do, and you'll get it automatically as soon as we're done on a Tuesday, and that way you never miss an episode. Thanks to Christina, uh, Senior Developer Relations at GitHub, where she gets all the tokens.

Christina Warren [02:30:14]:
Did you get all the tokens? All the tokens.

Leo Laporte [02:30:16]:
tokens. Thanks to Andy Ihnatko at ihnatko.com, and a special thanks to Federico Viticci. It was great to have him on from X.

Andy Ihnatko [02:30:23]:
That was solid as well.

Christina Warren [02:30:25]:
That was awesome. It was great having Federico on.

Leo Laporte [02:30:27]:
Amazing.

Andy Ihnatko [02:30:28]:
It really did inspire me to think about GM. I haven't— I've never done that kind of like experimenting with AI, like AI that I actually run and own. And it's got me experimenting. I actually made a note to myself saying, let's look into that and maybe just not buying stuff, but maybe lease Yeah, you don't have to buy it. Yeah, and spinning up, spinning up my own instances. That's like, that's how good that conversation was. That was great.

Leo Laporte [02:30:47]:
Good. Yeah, he's, he's just really great guy. Uh, thank you all for being here. We'll see you next week. But now I'm sad to say, it is my heavy heart, with a heavy heart, I have to tell you, get back to work because break time is over. We'll see you next time.