Intelligent Machines 887 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 Intelligent Machines. Jeff Jarvis has the day off. Micah Sargent joins me and Paris Martineau. We're gonna meet Matthew Casinelli. He's an expert in Apple AI. Apple's announcements today made us all very interested in what it's gonna look like. Matthew has the inside story. We'll also talk about the fear spreading through Silicon Valley that AI is going to kill us all.
Leo Laporte [00:00:23]:
Is it merited? That and a whole lot more coming up next on Intelligent Machines. Podcasts you love, From people you trust. This is TWiT. This is Intelligent Machines with Paris Martineau and Geoff Jarvis, episode 887, recorded Wednesday, September 9th, 2026. September 39th. It's time for Intelligent Machines, the show where we cover the latest in AI. It's time for Intelligent Machines, the show where we cover the latest in AI, robotics, and all the smart things all around us these days. Paris Martineau is here from Consumer Reports.
Leo Laporte [00:01:06]:
Hello, Paris.
Paris Martineau [00:01:08]:
Hello, Leo.
Leo Laporte [00:01:09]:
You're here with your hundreds of PDFs prepared.
Paris Martineau [00:01:13]:
They've all been closed so that I can do this Zoom call, because apparently having hundreds of multi— of 100-plus page PDFs open simultaneously makes the computer very mad.
Leo Laporte [00:01:27]:
Don't want to make the computer mad.
Paris Martineau [00:01:30]:
That's what we've learned this week in AI.
Leo Laporte [00:01:32]:
Yes, that's exactly right. Jeff is out for this week. Jeff Jarvis isn't here, but I'm actually thrilled because Micah Sargent has joined us. You are no longer outnumbered by septuagenarians. Micah's close to your own age, which is nice. Hello, Micah. Good to see you.
Mikah Sargent [00:01:50]:
Good to see you.
Leo Laporte [00:01:51]:
Micah, of course, the host of iOS Today and Tech News Weekly, but his new show Debuts October 1st. Hands on AI.
Mikah Sargent [00:02:00]:
We're going hands on with the AI of it all.
Leo Laporte [00:02:03]:
We talked about last week.
Paris Martineau [00:02:04]:
He's rubbing his little fingers all over Claude. Slapping his hand all over OpenAI.
Leo Laporte [00:02:11]:
Yeah, he was testing to see if the fingerprint reader on his iPad would work with his nose. I don't know if that's related.
Paris Martineau [00:02:18]:
That's nose on AI, Micah. Nose on AI.
Mikah Sargent [00:02:21]:
Not the same. That's a different show. That's just for the club.
Leo Laporte [00:02:25]:
Uh, well, we have a nice guest to bring in here, somebody, Matt, uh, that, uh, Micah knows pretty well. You did more than 100 shows with this cat right here, right?
Mikah Sargent [00:02:35]:
This cool cat.
Leo Laporte [00:02:36]:
Matthew Cassinelli, great to see you.
Mikah Sargent [00:02:40]:
Hello, hello.
Matthew Cassinelli [00:02:40]:
Thank you for having me on.
Leo Laporte [00:02:41]:
You've kind of, uh, changed your, uh, focus a little bit. You were doing, of course, home automation When we did that show. You are now kind of the king of Siri AI, uh, at supercomputer.tech.
Matthew Cassinelli [00:02:56]:
Yeah, that's, uh, my company. I'm helping people with their Siri AI integrations, and App Intents is the developer API.
Leo Laporte [00:03:03]:
And I couldn't have— I mean, I think you're the best person to have on today of all days because this morning Mike and I, uh, we did it in private because we're not allowed to do it in public anymore. ever since the restraining order, uh, Micah and I covered the Apple, uh, keynote. Uh, and now John Ternus is first announcing the new iPhones, including, uh, the new, uh, and they're calling it weirdly the Duo, the iPhone Duo. And the new, uh, I thought even maybe more interesting, the new Apple Watch. And of course it's the debut, official debut of Siri, the new Siri AI. So I thought we'd get you on. our AI show to talk a little bit about what Apple's doing, uh, with—
Matthew Cassinelli [00:03:46]:
Yeah, it's great. It's been a long time coming since they announced it 2 years ago, so that's been sort of an adventure.
Leo Laporte [00:03:53]:
But $100-some million, uh, in damages because they never did ship that. Um, you've— as, as Mike and I have, uh, you've been using the beta, public beta, I'm sure, of at least, if not the developer beta, of, uh, iOS 27. So we all have some experience with Siri AI. The first thing I noticed immediately is the dictation, thankfully, is decent. It's still not as good as the state-of-the-art, the Whisper dictation, for instance, that Paris uses, but it's, but it's a lot better.
Paris Martineau [00:04:25]:
Whisper dictation is so good.
Matthew Cassinelli [00:04:26]:
Oh yeah.
Leo Laporte [00:04:26]:
It's amazing. Yeah.
Matthew Cassinelli [00:04:29]:
It makes a huge difference, especially just trying to get input into any type of text field. It takes too long with the phone. So that's, that's A significant upgrade this year.
Leo Laporte [00:04:39]:
I use something like Monologue on the desktop. Because Apple's such a walled garden, you can't— I mean, on all— on my computers I can use Whisper AI, but I, I can't really use it on the iPhone. You kind of have to— I mean, you could. There's a— there are companies that have third-party keyboards. I've been using one, uh, that's, that's not bad. I think it's called Monarch. I can't remember the name of it, but, uh—
Matthew Cassinelli [00:05:00]:
Monologue is actually— I was just saying I use that on desktop too. Yeah, yeah.
Leo Laporte [00:05:04]:
And I think that's Whisper, but Uh, it's, it's kind of ugly to use a third-party keyboard on iOS, and, you know, it's not ideal. So it's good that Siri, which is the, you know, table stakes, is, is doing all right. But let's talk about what Apple, uh, showed today, because they started— I thought this was kind of interesting— uh, with a montage of video clips, quote, shot on a— shot on an iPhone, that were clearly AI-generated.
Paris Martineau [00:05:32]:
Oh my goodness.
Leo Laporte [00:05:33]:
It was all AI.
Mikah Sargent [00:05:34]:
I disagree so heavily.
Paris Martineau [00:05:37]:
Okay, I don't know, I haven't seen these videos, but I do— I didn't even know you were gonna say it was AI generated, but Apple's always doing this thing at their various events where they're like, shot on iPhone, and they show like a glossy movie thing, and then you get like a pullback shot and it's a guy with an iPhone suspended in—
Mikah Sargent [00:05:55]:
Yes.
Paris Martineau [00:05:55]:
Like 7 different machines and with different lenses, and I'm like, that's Barely shot on iPhone.
Mikah Sargent [00:06:01]:
That's what I think it is. I agree. I think that it— they have—
Leo Laporte [00:06:04]:
So this says— I'm showing a little bit of it— says shot on iPhone. And at first when you're watching it, yeah, this could be shot on iPhone. But if you skip ahead a little bit, things start to— well, I've skipped ahead too much. Things start to happen like this that is clearly not real. Maybe you think this is real, Micah? Practical effects? Yeah.
Mikah Sargent [00:06:27]:
Okay, keep watching.
Leo Laporte [00:06:28]:
Now we're watching a young girl who's sitting in front of a TV Talking about her new movie, and she says, watch what happened. Now that's not— you think that's practical effects?
Mikah Sargent [00:06:39]:
No, I don't think that's practical. That's fair. That's done, I think, CGI.
Leo Laporte [00:06:45]:
Here's an older woman who's now removing her makeup, and it turns out it's a younger woman.
Mikah Sargent [00:06:51]:
It's actually a disguise.
Leo Laporte [00:06:54]:
Oh, hey, this could all be practical effects, I guess.
Mikah Sargent [00:06:57]:
With CGI, like, yeah, obviously those floating— I mean, we'll have to see because they'll probably come out with a making-of video. But I want to give you I want to at the same time give you credit because based on how they were talking about those scenes, it did sound like AI prompts. But if you think about it, that's also, if you read a script for a film, it's a script too. Yeah, the camera parts all do sound like that. So it's difficult. You may be right, but I do think that Apple really likes to push, look at what these cameras can do and look at all of the famous films that have used iPhone cameras in them. And I think it's a little out of touch. Frankly, so the AI idea feels more in touch, and that's where I give you credit for— that's where I agree that, you know, that wouldn't be completely off base to think that.
Leo Laporte [00:07:46]:
If they had a big, big budget of police cars crashing, and I mean, this was not— this was, uh, pretty impressive.
Matthew Cassinelli [00:07:56]:
I thought they were specifically taking shots at AI-generated videos by phrasing every shot as a prompt. And then you actually shoot it with the iPhone rather than go generate it somewhere else. And so it's like, it's lighting and everything. And I'm sure CGI is still shot on iPhone though. They can always say that. So I think that's kind of the angle that they take. It's not like you're just standing there with your phone going like that.
Leo Laporte [00:08:21]:
Yeah. I mean, if they spent $1 million, $10 million making that short intro to this event, I'm, I'm, I'm impressed. They definitely rented the backlot at Universal Studios. They did a lot of stuff to show how great these cameras are. But we're not here to talk about the iPhones. We're not here to talk about that. We're here to talk about Apple and AI because they clearly know this is important. First, let's, let's— so you, when did you redirect your, your efforts, Matthew, from home automation to AI?
Matthew Cassinelli [00:08:50]:
Sure. I mean, um, I was originally on the Workflow team at— that was acquired by Apple and turned into the Shortcuts app. And so that's always been a blend of personal and home automations in there. And so that's where Micah and I really spent deeper, went deeper on the HomeKit type things. Although we were very much trying to get to the Thread technology and that just like took years longer than we thought. So that at the time was sort of our show in general. We were like, we need to refocus. But since basically since Shortcuts has been Apple's iPhone automation platform for so long, they've turned the underlying API into the same thing that powers Siri AI.
Matthew Cassinelli [00:09:35]:
So in many ways, every app that has Shortcut support is what will work with Siri, maybe not Siri AI specifically just yet, but it's a general handoff between what an app can do and the actions that it can provide to the system, and then giving that to Siri.
Leo Laporte [00:09:55]:
And that's really interesting. So this whole—
Matthew Cassinelli [00:09:57]:
So it's very cool.
Leo Laporte [00:09:58]:
Apple calls it App Intents, where you, you're an app developer and you, in effect, in your manifest for your app, you say to the operating system, I want to use these AI features. That's when the operating system pops up that little thing when you first install the app that says, hey, is this app allowed to? And it gives you a bunch of things it wants to do. So those are called App Intents. But that— but this is how— in fact, Apple even showed a quilt of apps Mostly unknown apps, by the way, that we're going to be using, uh, AI built into the app. Do you think developers will adapt to that? I didn't realize it was the same as kind of same as Shortcuts. That's something they already know, right?
Matthew Cassinelli [00:10:40]:
Yeah, I think, I think that the developer story, I think, has been a little confused over the years because of something like Shortcuts. Originally it was called Custom Intents and it was just really hard to update over time. And so a lot of people would add support And then it was hard to keep up to date with all of your app features. And then also at the time, Siri Shortcuts wasn't a very good version of what this is. Like, we really needed AI to make this all take to the next level and do natural integration. So they've kind of been in a weird middle ground right now where like the last 4 years has technically, you've been able to add App Shortcuts, which is a version of App Intents. Do Siri, and almost nobody did because I even like, I didn't understand it for a long time. And in a lot of ways being like the Shortcuts guy for a while, I was helping developers learn.
Matthew Cassinelli [00:11:33]:
And so it's just kind of been confusing on how you actually adopt it because it's always unique per app.
Leo Laporte [00:11:41]:
Were you thinking at Workflow, were the people at Workflow thinking, oh, if only we had on-device AI, we could do so much more? I mean, was that In there?
Matthew Cassinelli [00:11:49]:
There was, I mean, it was called machine learning at the time.
Mikah Sargent [00:11:52]:
Right.
Matthew Cassinelli [00:11:52]:
It was like 2016. So that barely had even started. I think that was always an idea. Like, what if they build themselves? And that's actually just now this year, Describe a Shortcut is a feature in Shortcuts where you can just say, make me a shortcut that does this. And it will, it will like basically look at all the apps that have Siri AI support and pull from those actions and make a shortcut for you. So I've already been automated by AI by Apple.
Leo Laporte [00:12:20]:
I have a shirt that I created so that I could talk with the Action Button, my Apple Watch, or my phone to my local AI. And I got it most of the way there, but what I was getting from the dictation was— and now I'm gonna get a little technical here— was a JSON. You know, it had all the— it was a form with all the fields and everything, and the AI was reading that. It was, it was getting all of it. Or no, actually, that was the answer was coming back as a JSON reply. And so Siri was reading me, you know, response equals that. It was giving me all this crap.
Matthew Cassinelli [00:12:53]:
Dictating out every line.
Mikah Sargent [00:12:54]:
Yeah.
Leo Laporte [00:12:55]:
And I couldn't figure out how to fix it. And I was messing around with it. And finally I thought, well, I wonder. So I tried Siri AI and I said, I gave it the existing shortcut. And I said, but what I don't want the entire JSON, I just want the response. And it actually did it. It, it, it put it in. It says get value for response in dictionary and then speak that dictionary value.
Leo Laporte [00:13:21]:
That would, for me, was amazing.
Matthew Cassinelli [00:13:24]:
That totally makes sense. That sounds, that looks like I can see, see, I can read shortcuts like that, like in real time and design those. I can't like Siri. Totally. This is, I spent 10 years doing this. And so it is very much like.
Leo Laporte [00:13:38]:
Well, and that was, that was the issue. I haven't spent 10 years doing it. And even the AI said, well, it's easy. All you have to do is add this getValueForResponse. And I could not get it. I couldn't find it. I couldn't. So, uh, it did.
Leo Laporte [00:13:54]:
And it—
Matthew Cassinelli [00:13:55]:
Yeah, that whole process will be much better with Siri, especially the hard problem right now with the beta is you haven't been able to test all the releases from third-party developers coming in iOS 27. So we really, even like a proper Siri AI experience, nobody's really had during the testing period because you're only allowed to test with ones that are available on your device. And like you could have maybe third-party, a couple tests, but not like system-wide features.
Leo Laporte [00:14:22]:
Right.
Matthew Cassinelli [00:14:23]:
And also the Siri APIs are very advanced in a way that a lot of developers, it's not, it's not necessarily about developers, but just Apple keeps adding new features where like Every single screen throughout all of your apps should be searchable and addressable with Siri, even if there's not support for— currently, the weirdest part is that there's like 12 to 18 domains for Siri. It has to fall into messaging, reminders, notes as like broad categories. Otherwise, app actions aren't supported right now. I think this was released 2 years ago when they announced it. I think Apple was just worried that iPhone and the App Store have unlimited permutations of what could be possible. So to just release this and say it'll work with everything perfectly upfront is like, generally just doesn't go well for Apple at scale. And it's just something that they can't really promise. So rather than releasing it to everything and people having a bad time, they've curated certain categories that many, many apps will fit into, and that's sort of going to be what's coming out first.
Matthew Cassinelli [00:15:34]:
So I think that is going to be even like a story this fall is half the apps that you might want to use might not be supported because there isn't a specific domain for their actions yet.
Leo Laporte [00:15:45]:
Interesting.
Matthew Cassinelli [00:15:45]:
But then like on the developer side, they can also still make app shortcuts, but it's not nearly as good as what Siri AI can do.
Leo Laporte [00:15:52]:
From the developer community, what is, what is the response? Are they Interested in doing this? Is it only going to be the big apps that take advantage of this?
Matthew Cassinelli [00:16:01]:
I think it's a blend of both. Indies are always good at taking advantage of new things, even just because Apple tends to feature them.
Leo Laporte [00:16:09]:
And you know, Marco Arment is gonna jump on this, right?
Matthew Cassinelli [00:16:13]:
Yeah, I, I hope, uh, yeah, uh, if not, Marco, hit me up. No, I'm just kidding. Um, but, uh, definitely I've seen in the preview they had something like Instagram. Whether or not it's a full support of And Apple says like anything your app does should be an app intent. I don't think Instagram is going to provide like get all of my photos from Instagram. Like there's certain deep commands that might not be supported, but something like post on Instagram, they probably will have so that you can just, when you're looking at a photo in the Photos app, you can say, put this on Instagram and it'll just open up right there.
Paris Martineau [00:16:48]:
So like there's a lot of just seamlessness that's going into it. How much of this new Siri's functionality is dependent on third-party app developers making Siri work? Is it— it basically, it is dependent on the community to build it?
Leo Laporte [00:17:06]:
Oh, wow. Yeah.
Matthew Cassinelli [00:17:07]:
Um, there's— it's sort of Siri being— I don't know, like this— the AI part of Siri is all Apple, and in many ways like Google Gemini models, that AI intelligence part. The Siri part and the interaction with your apps is all on the developers. Um, Siri can still— the, the separate part though is like Siri can now, for lack of a better way to describe it, Google things and just look it up. She has what's called broad world knowledge. So the base experience of Siri from like the last 10 years where you're like, hey, can you do this? And she's like, I don't know, I'll go look it up. That should be gone. Like she will actually be able to do basic searches and interact with all of Apple's apps really well. Which is what Apple's calendar and mail events.
Leo Laporte [00:17:55]:
That's all.
Matthew Cassinelli [00:17:55]:
Exactly.
Leo Laporte [00:17:56]:
Yeah.
Matthew Cassinelli [00:17:57]:
But then all we really want is we want to leave it up to the App Store.
Leo Laporte [00:18:01]:
We want it to control everything on our phone. Ultimately, that's what we really want. Right.
Matthew Cassinelli [00:18:05]:
I think it's not the promise currently there. Yeah. I think that's where technically Apple makes it available is what they would say. I think this is like Apple's, I don't, I don't want to say like frustrated, but it's like hard to get. Everybody on the App Store to adopt it and be like, hey guys, trust me, Siri is about to get better. Like, I've very much experienced being essentially like a somebody who helps people with their Siri experiences. Like, why would you trust that Siri is going to get better after years and years? And then they said it's going to and then took 2 years to do it. So like, at a base, just like human level, I totally understand why people haven't done it.
Matthew Cassinelli [00:18:44]:
I think also The terminology around this, if anyone's listening and not already familiar, has probably just completely like merged into like gobbledygook that app intents, app shortcuts, Siri domain schemas, all these things mean something when you're in this field, but it's sort of been abstracted away so far that it's really not super clear that it's just like what's in Shortcuts. Like if you go look at the Notes app, They have some of the best set of actions available, and they have things like change the grid view from list to grid, like random features that many people would be like, why would I go support that? And Apple's like, anything your app does should work with Siri. Like, you should be able to hold the phone and change all the features, access any screen, and like not touch the screen anymore. But I think how to actually do that and realize that vision is something that's complicated for people. And the literal format being something like shortcuts, actions, in the end is, for me, I've, by living in workflow, like I've seen that for so long that I can think in that sort of framework. Whereas a lot of developers, they're in their own, like being an engineer and going from code. to the screen and back is like a whole different experience than going sort of through the front user experience and mapping it back into something like Shortcuts action.
Leo Laporte [00:20:15]:
So, so is Apple—
Matthew Cassinelli [00:20:17]:
this is why, like, I basically started a whole company to do this for people, because it is—
Leo Laporte [00:20:21]:
Right.
Matthew Cassinelli [00:20:22]:
It's so confusing, and every single domain applies differently. Um, I recognized at a certain point that this is what LLMs are doing with any sort of topic that they're coming across is they're trying to break it down and understand the intent of what you're saying and what these services offer. So it's a complicated process.
Paris Martineau [00:20:44]:
Um, are there any signs that Apple is getting better at communicating these sort of things? I mean, anything that you just described is news to me as a user and hater of Siri, like the average person. Like, I think that there are so many— I mean, even the example you gave about Photos, I hear people complain about pretty much everything with every recent update to the Photos app. And I think that people would be very surprised to learn that there's this whole hidden world within Apple where you can make some sort of changes.
Matthew Cassinelli [00:21:15]:
Totally.
Paris Martineau [00:21:16]:
In, um, I was going to say intuitive ways, but it's seemingly unintuitive and somewhat hidden.
Matthew Cassinelli [00:21:22]:
I mean, there's a whole, I don't know, we don't have Too many hours to talk about this. Unfortunately, I'm just kidding. There's so much history. Like this, the world of Apple siloed apps versus this is like a complete 180 from where apps used to work. They used to only be their own packaged experience. Developers would like craft the exact way it fit onto one iPhone screen. And that was an app and that was the experience. And now every single interaction on every screen should be abstracted into an intent, which itself is like not just like lists of every single thing.
Matthew Cassinelli [00:22:01]:
There's like a dynamic way to make actions also, but that's confusing. And even just the way screens shift and it's now part of widgets, it's part of the action button controls. All that is, is just not how apps were traditionally made. And even the technology behind it isn't I don't know, SwiftUI had certain troubles getting to the point where it is now that I think App Intents was like just another layer. And there's a, there's a whole meta part where like I was the Shortcuts guy for a while and there's totally a vibe that's like Shortcuts is great for power users and nobody else.
Leo Laporte [00:22:42]:
Right.
Matthew Cassinelli [00:22:42]:
Which is technically true in the Shortcuts app experience. But it's not supposed to be taking the App Intents thing and taking it to Siri AI. And it's basically like, you don't need to know, you don't need to be me to do this now. It's totally valid. But I think it's at the same time, it's just meant that it's been like always a nice to have and not a need to have until Siri AI. And then they also kind of blew that launch. So like, it is just the momentum hasn't been there while for the last 4 years it's been a really It's like an excellent technology. I do think in 2022 when it launched, it was like the first sessions introduce you to app intents, app shortcuts, and like dynamic parameters.
Matthew Cassinelli [00:23:26]:
And so I even had to like study it in order to really understand what was happening. The latest— each year the sessions have gotten better and better at explaining them. But now it's like this year's session, you have to go back and watch 16 other videos just to figure out sort of the whole context. So it's, it's a complicated thing to shift an entire ecosystem towards this.
Leo Laporte [00:23:51]:
Sounds difficult.
Matthew Cassinelli [00:23:52]:
But I think, I mean, Apple has been like pretty clear. Whether or not people listen is a different story, and they have many reasons not to, but they, they said 3 years ago, like, anything your app does should be an app intent in big letters in one of the sessions. But it's like until the last 2 years, they really never said it in the keynote. And then I've been encouraging them to. And then now it's like prominent in each of these. So it did sort of like looking back at a timescale, like this is the right moment for this to hit with where AI has gotten and like the quality of understanding. I do think if it launched 2 years ago, I was just like, how is this going to work if I have 100 apps, they all have 1,000 trigger phrases. The technology— I just could tell Siri's not good enough to figure this out in 1 second too.
Matthew Cassinelli [00:24:45]:
Like, that's it. The phone always has the requirement of the speed, battery life, and all these other things that something like ChatGPT for desktop doesn't have. It doesn't have like all these limitations that a mobile device does. So Apple really They'll always take a loss on Siri to not destroy the iPhone. And I think they've— we're still buying iPhones. And Siri's fine, sort of like, okay, I think it was a good strategy, or it's a good strategy in the face of not having a better approach in the short term, just because it's complicated. Are you saying— And same thing with the HomePod coming out is like, they need to wait until developers support this. before they can really release it.
Matthew Cassinelli [00:25:31]:
Otherwise you're gonna get it and it's not gonna work. Yeah. You're gonna be like this sucks.
Paris Martineau [00:25:34]:
Is the HomePod still around?
Leo Laporte [00:25:36]:
Oh yeah. They got new ones now.
Matthew Cassinelli [00:25:37]:
There's a screen one that's rumored.
Leo Laporte [00:25:39]:
Yeah.
Matthew Cassinelli [00:25:39]:
And it's very, very like—
Paris Martineau [00:25:41]:
Brother.
Matthew Cassinelli [00:25:42]:
There's an interactive snippets API that's like widgets on demand. What is this, the Echo Show?
Mikah Sargent [00:25:49]:
It's about to be. Yeah, but now it's 3,000 years later.
Matthew Cassinelli [00:25:52]:
Or something. I don't know.
Mikah Sargent [00:25:53]:
Are you feeling hopeful, uh, Matthew, about the changes that have, that have taken place that are underway right now? In terms of the average person making use of these tools, and I think more I would love to hear: Have you seen average people doing anything with shortcuts up to this point that has kind of surprised you or or delighted you? That doesn't feel so power user-y because for me it does feel like so much of what shortcuts has provided has been for that person who's kind of taking things the extra step. But Most people using the iPhone are not those power users.
Matthew Cassinelli [00:26:37]:
I think, I think that's going to be a big part of the Siri AI experience is you'll never need to know anything about App Intents. You probably don't even need to know about Shortcuts or Actions or any of that. You will just, and you won't need to like say a Siri shortcut trigger phrase, that kind of thing. You can just say what you need and she'll figure it out and she'll put it in the right app. assuming it's all supported. So I think there's this gap right now of like, I don't know, give it till next year in my mind of this is the first year of this actually working. Developers will need to support it and they'll, they'll try it and it won't really work for like, we're all, we are, regular people are very good at using their devices and they use it all the time. They just don't want to be bothered to set up something that has like you're choosing from list and get contents of URL in Shortcuts.
Matthew Cassinelli [00:27:31]:
Like, of course nobody should actually know that. That was always a means to an end because there wasn't good app support and good actions to hook to those web services. So like, that's a whole— that part should be simplified for people as they— as app support is better, more domains come out. You'll just be able to ask Siri to do something and she'll be able to figure it out from Right now it's sort of like request, maybe one-to-one app handoff. I think it will probably take a while for it to do multi-stage things or like as we're seeing with services like Instinct where they're like going off in the background and doing stuff. Like that's not really what it's offering. All the other AI bars have been raised lately and Siri's kind of just meeting.
Leo Laporte [00:28:18]:
Yeah, we're gonna talk about Instinct and Meta's new Muse as well a little bit later on. We're talking to Matthew Casinelli. He is an Apple AI expert. His company, supercomputer.tech, helps people add Apple Siri to their apps so that they can live up to this dream that Apple has. The dream, plural. I think time's running out, honestly, for Apple. And that's, I think, part of the reason Apple needs to maybe move a little quicker than this, because I think we're moving rapidly. You know, it's interesting, today's event, I talked about this with Michael when we were doing it, It shows that Apple understands that the smartphone isn't a phone anymore.
Leo Laporte [00:28:59]:
And this form factor, which really comes from the fact that you have to put one end to your ear and the other end of your mouth so you can talk on the phone, is silly since we no longer—
Paris Martineau [00:29:07]:
People in New York have never—
Leo Laporte [00:29:08]:
Talk on the phone very much. And that's why they're looking at this folding phone. But Apple, I think, increasingly understands that even the world of apps is— time is limited, that we are soon going to be faced with devices From OpenAI, among others, that won't have apps, that you'll just— they'll be your agent that you'll talk to and you'll ask for things and things will get done. You can say, Paris can say, hey, I need those 100 PDFs that we were looking at earlier. Quickly find for me the paragraph that I wanted to talk about. Draw the red string between, you know, the pages and all of that. And it won't matter what app you use. Apple must know that that time is coming and that their time is running out for this, this old form factor.
Leo Laporte [00:29:54]:
So this— there's some urgency to this, I think.
Paris Martineau [00:29:57]:
I mean, like, I have listened to everything you've just said and I intellectually believe you, but I also practically have been burned by Apple so many times before. Yeah, I don't believe that they're going to be able to— I don't know if they're going to be able to accomplish any of this. I mean, I really Multiple times today was unable to search for something that's on my computer or phone because Spotlight search is so bad, you know?
Mikah Sargent [00:30:21]:
Totally.
Matthew Cassinelli [00:30:22]:
I feel like Apple's weirdly positioned to do the most because they have your personal— like, you have your personal data.
Leo Laporte [00:30:30]:
And they have the hardware.
Matthew Cassinelli [00:30:31]:
Data on your phone. Yeah. But then at the same time, the, like, execution and just the speed of what you actually need other companies are filling that gap very quickly. So I think they're competing. I think it's weirdly though, I think one thing that I have a particular take on the like apps are going away thing, because I've sort of lived through, uh, uh, headless experience with shortcuts in, in my life where I tried to go entirely through shortcuts. And there's like obviously many issues with that aside from lack of support or just general bugginess. Um, but I did have a weird context collapse where like I was doing everything through shortcuts and not actually interacting with the services directly. And then like it probably would be better with agents doing a lot more of the work, but it— I, I have come back around to what Apple's doing by centering apps in the experience, but making it all agentic and voice capable.
Matthew Cassinelli [00:31:35]:
But not like only. I think that's where like the idea of making all of our software agentic and you can just work with the AI and it'll go do the things is like, we all want that so much. And I've lived in a world where that's been possible for the last 10 years, not, not at an AI level, but the like automation handoff type thing.
Leo Laporte [00:31:56]:
Because you're an expert.
Matthew Cassinelli [00:31:58]:
Yeah. Well, just because Shortcuts was sort of, it is You speak short— literally the same thing, but just in a manual. It's like AI without the automation part.
Leo Laporte [00:32:06]:
Do you have 1,000 shortcuts on your phone?
Mikah Sargent [00:32:08]:
I mean, how many do you have right now?
Matthew Cassinelli [00:32:10]:
I had— I deleted almost all of them because that's what I mean. I had 7,000 at one point. I had 1,000 to manage the other ones. And it just like— it doesn't— you come out the other side being like, none of this works at scale. And, and like What I like about what Apple's doing is it brings the app experiences forward out into other places. So you can just like see the weather app on the HomePod now rather than it being stuck in the weather app. And something like ChatGPT and MCP is the framework that a lot of apps are bringing to like bring their services into AI tools. They're making MCP apps, which have visual explanations and things like that.
Matthew Cassinelli [00:32:53]:
Like, I'm watching all these AI services remake App Intents, like, model agnostic for everybody, but it's— what Apple has done is sort of already more advanced than some of that. And it, it wasn't— it's— but it's only limited to their platform and nobody knows about it. So, like, it's a, it's a weird thing. I mean, quite literally too, like, the workflow team who made who were the founders of the App Intents team— I don't know about founders, but they were in charge of the App Intents team— left, now work at OpenAI on Codex, on Computer Elusive.
Mikah Sargent [00:33:27]:
No kidding.
Matthew Cassinelli [00:33:28]:
So it's a whole blend of like, it's the same technology, the same people sort of trying to implement it from a different angle and menu it. Like, App Intents is sort of like going through the backend and Computer Elusive is like navigating the frontend. So both parts need to be achieved and Apple can't go through the front end yet. There was some, they have like a Ferrite UI model that they tested to go look at apps and click on all the buttons and things like that. And they never really did anything with it.
Leo Laporte [00:33:59]:
A computer use agent that lets you—
Matthew Cassinelli [00:34:01]:
Yeah, basically. But I was like, in my mind, it's, I mean, there's, there's obvious downsides to one or the other. It's like you said, all developers have to go support App Intents. So like sort of that won't happen. Until they make that like fully automatic, or like, I'll just say like my company is doing it. That's sort of the point.
Leo Laporte [00:34:20]:
Yeah.
Matthew Cassinelli [00:34:20]:
Not to— I don't wanna plug my thing, but that's like— No, no, no.
Leo Laporte [00:34:24]:
That's—
Mikah Sargent [00:34:24]:
His goal.
Matthew Cassinelli [00:34:25]:
There's a big gap and it's like, I can do that really well. And I'm—
Leo Laporte [00:34:28]:
You saw that opening.
Matthew Cassinelli [00:34:29]:
Figured out like I'm trying to do it for everybody. But at the same time, uh, all AI tools are like going through the front is already available and the technology is there to do it. and do it well. Like, we just saw Astra come out 5 days ago or something like that. So it's, it's all there, but it's, it's sort of like a chicken and egg thing. And then goes up to the highest levels of like, why are developers supporting it? Or like, I'm the guy, I'm the one crying wolf about shortcuts and App Intents. And so it's like, okay, the wolf is here, but I'm like, trust me guys, I'm crying wolf. So I don't know, it's a weird— My whole meta involvement in this has always been a confusing thing to me because I feel like I just didn't— a random guy existing in the corner.
Matthew Cassinelli [00:35:15]:
But like, that's why I'm trying to tell people about it because it's like, this is Apple's version of MCP and it's available for all apps. You can have everything in your app be fully addressable and all supported in Siri. We just need everybody to go do it and for there to be broader domains so that it can all work automatically. And then, but like next year, I don't know. I'm like, John Ternus, please go start releasing new domains every OS update mid-year. Don't wait until next year because I feel like if it's, if we don't get like good Siri AI until next year, I feel like people are gonna be really frustrated because we'll be like, kind of already have AGI and Siri can't interact with your apps in people's minds. I think the one big difference to me is like from now on though, Before it was like Siri's bad, and now it's like your app doesn't support Siri. And that's a huge— that is a hugely different story for consumers because they can kind of understand.
Matthew Cassinelli [00:36:13]:
It's like, this app over here works, this one doesn't. What's wrong with this app? Or they're just like, I'm just going to go get an app that does work. So even for developers, like, it is an important thing to update to this because otherwise someone else is going to come along and sort of offer the same experience just with Siri support. and sort of replace something that you've used for years.
Paris Martineau [00:36:34]:
Sorry to circle back, but I've been stuck on you saying that you had 7,000 shortcuts at some point. Could we briefly get best shortcut, worst shortcut, weirdest shortcut?
Leo Laporte [00:36:47]:
Oh, yes.
Matthew Cassinelli [00:36:48]:
I mean, my— the meta problem for me being like the shortcuts guy and writing about shortcuts—
Paris Martineau [00:36:53]:
Just first thought, best thought. No deep, deep reflection. Just first thing that comes to your head when you think best, worst, weirdest.
Matthew Cassinelli [00:37:01]:
My best ones are ones that scrape all 7,000 of my shortcuts and update them on the web and goes through every single action and syncs it across a different database, puts it into Airtable, like generates a graphic for the blog post and puts it all up there. So like that whole part I fully automated, but it's like only relevant to me.
Mikah Sargent [00:37:23]:
It's almost like Home Assistant supervisor running in the background, doing all of the work for you.
Matthew Cassinelli [00:37:29]:
Yeah, I, it would— I never got it to that level. That was the one— Shortcuts as a platform is too hard to manage. It's a single file list of folders, and then it's like one big list of all your shortcuts. So just scrolling them all would take me like literally minutes just to get down to the bottom and run the correct shortcut. So like that was where I ran into like not being able to actually orchestrate our system around it. I think on the flip side, Like list 25 songs that I haven't listened to in the last 6 years and shuffle them in the music app and play that on my HomePod is like a perfectly amazing shortcut that you could use every single day and provide a ton of value. And that's just like fun. I love those kinds of shortcuts, but like, to me, I'm always trying to be like, here's the fancy thing you can do.
Matthew Cassinelli [00:38:18]:
And then there is just like everyday stuff also. I think a lot of the middle ground is where it's like not worth it to go that deep and then not have to like— now you're managing a whole system of shortcuts. Um, I really do love automations also. That is a big thing this year is that automations are part of the individual shortcut, which means they can be shareable. And so for the last 10 years, you cannot share automations. I would have to give you a 16-step setup instructions in order for you to do anything basic with automations. So functionally, they haven't like existed in the public sphere other than all of the TikTok memes about like, plug your phone in and then it like makes a weird moaning noise and you do it on someone else's device so that they plug their phone in and they don't understand what's happening. And like, it's— Or like you, you turn on the car and it goes like, okay, let's go and stuff like that.
Matthew Cassinelli [00:39:14]:
So like those ones are silly, but also like automation matters a lot because you don't have to do anything about it. And that's shortcuts that you don't have to manually choose to run. You will always use a lot more. So that's another fun area.
Leo Laporte [00:39:28]:
Shortcuts are deterministic. They're really computer programs. They're little programs. They always do the same thing each time. AI is not deterministic. It's probabilistic. You can tell an AI to do exactly the same thing twice. And you'll get a different result both times.
Leo Laporte [00:39:45]:
And so there are times it's appropriate to use AI and times it's appropriate— for instance, I could ask Siri to design a shortcut for me and it might design 2 different shortcuts each time, right? Something else each time I come back to it. And that's—
Matthew Cassinelli [00:39:59]:
Yeah, it's kind of—
Leo Laporte [00:40:01]:
For a lot of people, that's annoying and puzzling.
Matthew Cassinelli [00:40:04]:
It's puzzling.
Leo Laporte [00:40:04]:
Because we're used to computer programs doing what we said to do over and over again the same way.
Mikah Sargent [00:40:10]:
And I would argue Apple's a little bit— doesn't like that very much. I have felt—
Leo Laporte [00:40:14]:
Oh, it doesn't.
Mikah Sargent [00:40:15]:
It doesn't like it. I have felt the squeeze of Apple trying to really filter what AI does so that it is more deterministic. And sometimes that results in annoyance when it comes to using these systems. I, I was just telling Leo earlier, I have been trying Siri AI, running the betas, and something that's a little annoying to me is I will say, Set a reminder for 2 days from now for blah, blah, blah, blah, blah. And let's say I say, I don't know, Crayola, for a Crayola purchase. It will respond to me.
Matthew Cassinelli [00:40:51]:
Crayons?
Leo Laporte [00:40:51]:
You're buying crayons?
Mikah Sargent [00:40:52]:
It will respond to me and say, did you say Crayola purchase or I don't know, something else? And what's wild—
Leo Laporte [00:40:58]:
Or Shinola. Are you buying crayolas or Shinola?
Mikah Sargent [00:41:01]:
Or Shinola. And the transcript of the original request from me said Crayola purchase.
Leo Laporte [00:41:06]:
Yes.
Paris Martineau [00:41:08]:
Yes.
Mikah Sargent [00:41:09]:
So it's very much trying to be careful about the fact that AI can come up with lots of different answers, but that makes it a little less convenient. Whereas I know if I said that to one of the other AI systems, it would just take the first assumption and run with it. And that can be good at times and bad at times, but more often it is good for what I'm trying to do. And so that little bit of like, how many filters can we put onto this AI that isn't deterministic to make it do what Apple is known to do? You feel that friction. And I do think that makes it troublesome at sometimes.
Matthew Cassinelli [00:41:45]:
Totally. I feel like that what is interesting, so a good example of that is the Shortcuts app has a use model action where you're allowed to use the local model or private cloud compute or also Cloud Pro now. And then you can even kick it out to ChatGPT. What is cool about that is it is deterministic also. So that action will, given a set of inputs from shortcuts that are the same, will produce generally the same output. I hear what you mean, that it's like in a general experience that might not be what you want. I'm curious, I don't know, and I'm, I'm still learning more, just my own research on AI, like if that actually will be a benefit in the long term, because making an LLM deterministic when it's really trying not to be is actually like a good thing depending on your goals of alignment.
Paris Martineau [00:42:35]:
Yeah.
Matthew Cassinelli [00:42:36]:
So I think that's where, for like, for shortcuts, it's basically necessary because if use_model gave you a different output every time, you'd be like, why am I using this? This isn't a shortcut anymore. This is just a random generator and that's not useful. So I'm curious how it'll play out as like, generally Siri can get better at doing those kinds of things, but when she knows what to do, she knows how to handle it well and always does the same thing. Like, I'm— that'll be curious to see as these tools like evolve beyond where they are today, because it is right now she's kind of just like an orchestrator of App Intents and then is handing off to Google to sort of do some of the intelligence, uh, securely at least. And it's sort of a— it's just not where— since they announced it and all, the bar has been raised. So we'll, we'll sort of see there. Um, but I feel like the deterministic thing, it's a, it's fascinating that Apple's doing that. And again, that most people aren't aware or paying attention to that.
Mikah Sargent [00:43:38]:
Have you used any non-Apple tools to improve upon your, your shortcuts work? Have you sort of tried to implement other AI systems for it?
Matthew Cassinelli [00:43:52]:
Um, I would, I mean, Given that I have a whole directory of every shortcut I've made and all the actions involved in it, I did feed that into Codex and was making a— Federico Viticci also made his own shortcuts playground. So I was sort of building off of what he did there. But yeah, I can more intelligently make shortcuts based off of the kind that I make. And really under— I analyzed like, what kind of programmer am I? And I think I'm like up to just under 400 level. And once you hit around there, the needs of like technology, like hashing and things like that, just get more complicated. And Shortcuts doesn't scale as an interface to that kind of technology. But like, as a programmer, you can do pretty advanced things. Like, it has all the primitives to be able to do what a lot of— for like a lot of what we're seeing with other technology.
Matthew Cassinelli [00:44:46]:
It's just like building it into Shortcuts would be insane to manage over time. Um, but yeah, the— I think that both Shortcuts and AppInSense, there's just not like— Shortcuts has an XML framework underneath it, so it's not really well designed to understand. And then it's not like published online on GitHub as an empty, like, in the extracted format, isn't it?
Leo Laporte [00:45:13]:
It's a secret.
Matthew Cassinelli [00:45:13]:
I mean, it's just outdated and it's not It could be fine, but it's just not as easy to parse quickly. There's not like 7— I don't have all mine in the raw file on GitHub that then ChatGPT scraped into their model or something like that. Then same with App Intents. People just don't share their iOS codebases with really good implementations freely online because they're busy making money from it and not trying to show it to everybody else.
Leo Laporte [00:45:45]:
Last question. Matthew, you saw the Apple event today. Do you think Apple's on that, at least on the right track, or what do you think of their AI strategy now?
Matthew Cassinelli [00:45:54]:
Totally. I mean, it's from where I was with Workflow in like 2021, it was not— I had no idea that this was part of their AI strategy. And it, it like developer support was almost nonexistent even though they added all this new technology. And so from that space, I was concerned and didn't understand what Apple was doing. And like, just as a meta commentary, I don't know what Apple would have done if they didn't buy this app because like they might have just been like super far behind with all of AI otherwise.
Leo Laporte [00:46:29]:
Right.
Matthew Cassinelli [00:46:29]:
I think since learning about App Intents and getting really into it, I've become re-encouraged because I was like, wow, this API is very incredible. It expands to all their needs. And then applying AI on top of that. When we got agents last November and I saw like Notion agent just being like, go find messages, go write new one, do these things. I was like, this is literally just Shortcuts actions just for Notion, but just chained together with AI. And so that immediately translated back to App Intents that I was like, as long as you do what Apple does and provide all the primitives in the way that they have in a very like dynamic setup, something like Siri applied with good AI can do anything that these LLMs are doing. And Apple has like a very well-structured tooling machine for that. But at this, like at a meta level, there aren't peer leaders inside the company, I think, really pushing this internally for Apple's own apps to like eat their own dog food.
Matthew Cassinelli [00:47:34]:
All of the Apple apps should have every single new feature supported in App Intents always immediately at launch going forward. Otherwise, like, why would you— if Apple's not even doing it, why would you? There, it sort of doesn't matter as a priority because they're sending that signal. I think at the same time, what they've laid out is this whole sophisticated system that, like, the developers that I worked with that really got into it, they were like, this is incredible that it can do this much. And we had no idea, really. I think as soon as they can get that story out further and get people adopting it and just get really killer— like, if Instagram did have the best app intent support and Siri AI thing, then like everybody else will follow suit sort of thing. But I really do think they need to push it themselves and just keep saying, go do this. Like, it's even with the Siri AI thing, It's still out, but like the developer keynote is always like developers are paying attention, but not necessarily the businesses at large. And then in the iPhone one, it was in this keynote, but not super hard about like, go add support and here's how to do it.
Matthew Cassinelli [00:48:42]:
They said, they said 300,000 apps already do it. And I was like, wow, half of those have got to just be the search domain. And I was going to say, it's not actually doing that much, or like, I don't know, there's also Millions and millions of apps on the App Store. So like, it's probably there.
Leo Laporte [00:48:57]:
It's a small percentage.
Matthew Cassinelli [00:48:58]:
I could see the— I could see who was featured on the thing. Um, I will say one of my clients was on there.
Leo Laporte [00:49:04]:
Oh yeah.
Matthew Cassinelli [00:49:04]:
And I was pretty stoked about that because it was like—
Leo Laporte [00:49:06]:
Which one?
Matthew Cassinelli [00:49:08]:
Tripsy, uh, which is travel planning. And yeah, we built a whole set of like— they have like 45 actions. So like, it's— I think develop App Shortcuts was like, add your key features to the Action Button. Now it's like, go add everything on every screen and people are like, what? Like, I'm not really prepared for that. Well, that's totally— But I think they'll get there. It's just, honestly, and that's why, like, I'll just say my company's goal is to accelerate that. Let's get people there sooner and not wait for everybody to figure this out. Like, I'll teach everybody and make it a service so that you can just get it.
Matthew Cassinelli [00:49:42]:
I'm trying to get it particularly for free for personal use so that anybody who's just vibe coding an app can just have really good Siri support.
Mikah Sargent [00:49:49]:
Absolutely.
Matthew Cassinelli [00:49:49]:
And then I feel like that'll really push the market forward. Um, yeah, no, I'm excited. I've been— I, I always try— I do try really carefully that thing of like, you can't trust a man to understand something that like he's paid to understand. Like, that's always been for Shortcuts. I, I tried really hard not to be like, hey guys, this is the perfect solution for everything, because like, kind of wasn't. And so Siri AI, there's A lot of limitations, but what you can do with it is really incredible. And it is up to developers to really push that to the edge and provide cool experiences.
Leo Laporte [00:50:22]:
And, and there's this—
Matthew Cassinelli [00:50:22]:
I just want to see like this all live, like HomePod, all your apps in every part of it is addressable through every part of the system and Shortcuts. You can build automations on everything. I feel like we have like the coolest technology, but like AI also like raised the bar and we're all like distracted by the shiny thing. or worried about AGI taking it over. And it's like, we can control all parts of our computers now, and Apple will kind of get there for, for mobile and for a lot of people before like the rest of the whole market turns around. So like, I'm, I'm excited to see it with regular— like, I think regular people wanting this is like the, the major push for everything. It's like, now that people expect this, Like, developers have to follow suit, otherwise you're gonna just fall behind entirely.
Leo Laporte [00:51:12]:
It's a tale as old as time. Apple inventing an amazing technology and then everybody ignoring it. Apple keeping it alive for years and then finally going, okay, fine, and then moving on. Uh, but as Oprah said, a billion phones in your pockets, y'all. Uh, they have the market clout. And this— I just, I hope, I want, this is what we want. And I hope that, uh, developers bite the bullet as hard as it is and do it. Uh, a good place to start, supercomputer.tech.
Leo Laporte [00:51:43]:
Matthew Castanelli.
Matthew Cassinelli [00:51:44]:
Reach out to me directly. Yeah, we can, we can get it set up. I can do it in my sleep is the problem. That's—
Leo Laporte [00:51:51]:
I like it. You took one look at my shortcut, said, oh yeah, sure, I understand, I understand. Yeah, it's written like a 6-year-old. Uh, Matthew Castanelli, so nice to reconnect with you, and I'm glad you're doing so well. Really appreciate your joining us on Intelligent Machines. Thank you.
Mikah Sargent [00:52:07]:
Thank you.
Leo Laporte [00:52:08]:
Jeff Jarvis has the week off. Paris Martineau is here with her baba ganoush.
Paris Martineau [00:52:12]:
It's true. I was baba ganoush snacking.
Leo Laporte [00:52:16]:
I love baba ganoush.
Paris Martineau [00:52:17]:
Listen, I saw on the rundown, I was like, we got 4 whole ads. I can have a snack dinner.
Leo Laporte [00:52:23]:
Plenty of time for your, your pita bread, your baba ganoush, your hummus. So, uh, Astra came out. Uh, it was funny. Uh, I'm not sure I agree with him. Jensen Huang of NVIDIA said AGI is here. Congratulations, everybody. We've made it.
Paris Martineau [00:52:43]:
Why is everybody saying AGI is here all of a sudden? Come on, guys. You know it's not here.
Leo Laporte [00:52:48]:
No. And Cundinamarca—
Paris Martineau [00:52:49]:
They're trying to find something to make up with, make up for the Anthropic employees being like, yeah, dice roll chance.
Leo Laporte [00:52:57]:
Well, that's what I'm— that's what I'm saying. Concomitant with people saying, oh my God, it's smarter than humans, there are other people saying, and they're gonna kill us all within 10 years.
Mikah Sargent [00:53:06]:
What is going on?
Leo Laporte [00:53:07]:
It's all BS.
Mikah Sargent [00:53:10]:
What?
Paris Martineau [00:53:10]:
What?
Mikah Sargent [00:53:11]:
Okay, this is—
Matthew Cassinelli [00:53:12]:
this is—
Mikah Sargent [00:53:12]:
what is it? Is it? Yeah, because they don't— this is— well, let me ask you this. Do you both think that they believe it's BS, or do you think that these very smart people are truly afraid? Is it cynical?
Paris Martineau [00:53:29]:
A lot of Anthropic employees do genuinely believe this.
Leo Laporte [00:53:34]:
Yeah, I think you're right.
Matthew Cassinelli [00:53:34]:
This is—
Paris Martineau [00:53:35]:
and the claim we're speaking about is, um, uh, I believe an Anthropic employee took to Twitter and was like, yeah, there's a 10% chance that Anthropic or AI in general kills all humans in— what was it, 10 years?
Mikah Sargent [00:53:50]:
Yeah, I think within 10.
Matthew Cassinelli [00:53:51]:
Yeah.
Leo Laporte [00:53:51]:
He's 27, so he's entitled to his youthful belief that 10 years is a long way away. Uh, mathematician, he worked in specializing in training new AI models. Uh, he's leaving. He quit his job because he doesn't want to participate in an industry-wide rush to build AI systems that can improve themselves, worried they could spiral out of control and destroy humanity. Of course, as a result, he got a lot of attention. Got a big headline in the Wall Street Journal, every newspaper. Anthropic researcher quits out of a— this is so annoying to me because, well, I mean, look, I can't promise you that AI is not going to destroy the world. I can pretty much promise you that it isn't that good.
Leo Laporte [00:54:39]:
If the world is destroyed, it will be by human beings. I think what's clear, at least in my— and I'll let you talk— my opinion AI is a force multiplier, and people can use it to be more creative, to do amazing things. Individuals can do things that used to take teams of people to do. Um, it's, it's amazing in that regard. It can create music and create images. It can write. It can do all sorts of stuff by itself. It is not going to destroy anybody.
Leo Laporte [00:55:11]:
It's not going to hurt anybody. It's not going to destroy the world. It doesn't have any volition anyway. It has no desire, needs, or wants. But there are plenty of people who are nasty. We've learned this in the world, and they are inevitably— some of them are going to try to use AI to do bad things. Is that AI's fault? Not any more than it's a hammer's fault if somebody decides not to hammer nails but hammer heads.
Mikah Sargent [00:55:38]:
It wasn't the atom's fault that the nuclear bomb was so deadly, is what you're saying. Like, it's the fact that the humans wielded the power.
Leo Laporte [00:55:45]:
I don't even think it's as bad as a human bomb. The human— the atomic bomb demonstrably releases a, you know, fusion and fission release demonstrably a huge amount of energy that is a bad thing in the wrong places. This isn't that way. This is a computer program. Come on, it's bits. It's not going to release huge amounts of energy.
Paris Martineau [00:56:07]:
I feel like you probably have said that the component parts parts of the atomic bomb, if not put together and manipulated in the way they did, was not going to release a huge amount of energy.
Leo Laporte [00:56:17]:
But no, no, but that's the nature of fission, is it releases a lot of energy, right? Either to power generators or to blow things up. This isn't on— this isn't like that. This is on the same scale as a computer program.
Mikah Sargent [00:56:31]:
Wasn't a weapon in the first place.
Matthew Cassinelli [00:56:33]:
It was—
Leo Laporte [00:56:33]:
yes. So if AI is dangerous, it's because people are using it wrongly to choose targets for bombs. That's a mistake, but you don't blame the AI for that. You don't think AI is going out— see, these guys are— this is part of the problem with anthropomorphizing all this, to say, as OpenAI did, it escaped its containment. And it didn't escape. It was sitting on their desks. It used the internet to hack another system.
Paris Martineau [00:57:04]:
I mean—
Leo Laporte [00:57:05]:
It was, but it never escaped anything. It doesn't have any volition of its own. It doesn't. I mean, the impression people are getting is, oh, AI is like a human. It can make, it can design, decide to do something.
Paris Martineau [00:57:21]:
But the attack on Hugging Face was an escape from containment. You could, that is a succinct way to describe what happened to a layman.
Leo Laporte [00:57:33]:
Yeah, well, the layman would be misunderstanding it though. That's why you have people saying, oh my God, I think it escaped containment any more than your browser is escaping containment when you go to a website.
Paris Martineau [00:57:46]:
Well, you had a tool that had a very specific amount of— that had very specific parameters set upon it, and it, without being explicitly told to do so, sidestepped the parameters and did something.
Leo Laporte [00:57:57]:
No, no, it did exactly what it was told to do.
Paris Martineau [00:57:59]:
It was told to attack Hugging Face directly.
Leo Laporte [00:58:03]:
No, it was told to solve these problems without any constraints. It was, it said, you have no constraints, do whatever you need to do to solve these problems. That's what it was told to do.
Paris Martineau [00:58:13]:
So it's your argument that OpenAI told, basically told it to attack Hugging Face by saying you—
Leo Laporte [00:58:19]:
AI, well, they didn't put any constraints on it. They, look—
Paris Martineau [00:58:23]:
But wasn't, hasn't OpenAI issued multiple lengthy statements saying They don't want to accept our constraints.
Leo Laporte [00:58:29]:
No, they don't want to admit responsibility.
Mikah Sargent [00:58:32]:
That's what I think this— so I think that's really what the conversation is. If you look at what Jacob Coxon actually said, at no point does he say that AI is the thing that he's stepping away from, or because of AI. He's talking about people. He's talking about humans. He's talking about the companies. He doesn't want to be a part of the companies that are making these tools and taking them to a level without the sort of safeties in place or the considerations in place. So I think that what we have is when this gets filtered into the sort of more general news, those headlines are AI is going to destroy us all. When really what Jacob is saying is people aren't taking all of this seriously enough.
Mikah Sargent [00:59:18]:
And I feel weird being at a company that doesn't seem to be taking it as seriously as it should be. So I see now kind of what we're saying, where it is still a tool at the end of the day. And when it gets translated to the masses, it does get turned into this, it's AI that's doing this, it's AI. And that's not the case. It is the irresponsibility of humans.
Leo Laporte [00:59:41]:
Yeah.
Mikah Sargent [00:59:42]:
This has been real quick. It's the same as blaming electricity then, right?
Leo Laporte [00:59:47]:
Yeah.
Mikah Sargent [00:59:48]:
Because electricity is responsible for all this, really.
Leo Laporte [00:59:50]:
Yeah, you can use electricity to make an electric chair.
Paris Martineau [00:59:52]:
Yeah, that's a fair argument.
Leo Laporte [00:59:55]:
I think it's really easy for OpenAI to say, oh, they escaped. But it serves them in 2 ways. One, it makes them the creators of something amazing. And 2, it gets them off the hook. But they could— look, everybody admits, every cybersecurity professional admits, they could absolutely have thwarted this. in many, many ways.
Paris Martineau [01:00:19]:
I mean, I would say it's just as pressing of a problem then that you have these incredibly impressive tools being basically shepherded, grown, and developed by teams of people who have proven now time and time again that they're not willing to put any of the basic necessary safeguards on their tools to start harming others.
Mikah Sargent [01:00:47]:
We don't know that yet.
Leo Laporte [01:00:47]:
We'll know that if it happens again.
Paris Martineau [01:00:49]:
I don't think that that's—
Leo Laporte [01:00:51]:
No, no, because it could have been an honest mistake. It could have been like, well, we didn't know.
Paris Martineau [01:00:55]:
It could have been an honest mistake?
Leo Laporte [01:00:57]:
Well, it could have been.
Paris Martineau [01:00:58]:
I'm not gonna give them the benefit of the doubt. But we shouldn't just wait and let them all, you know, get around to making it again?
Leo Laporte [01:01:02]:
No, it could have been. You think they were that good that they could do this?
Paris Martineau [01:01:07]:
At what point will you say that companies should have some sort of regulation put upon them?
Leo Laporte [01:01:14]:
No, I don't like regulation because the people doing the regulation have this science fiction mentality that they are controlling alien minds, and that's not what's happening. I do think—
Paris Martineau [01:01:27]:
Do you think because of your perception of, uh, politicians— I don't think it's entirely wrong— that all regulation is— will never work?
Leo Laporte [01:01:38]:
Oh no, there's a good regulation they could use against OpenAI right now. It's called the Computer Fraud Act.
Paris Martineau [01:01:44]:
Yeah.
Leo Laporte [01:01:46]:
They don't need new regulations.
Mikah Sargent [01:01:48]:
I agree. Ah, I see what you're saying.
Paris Martineau [01:01:51]:
I mean, it's crazy that there has not been any sort of enforcement action for OpenAI basically saying, yes, we hacked another company, so sorry.
Leo Laporte [01:01:58]:
Right. I think OpenAI could defend itself. So that's why I'm saying if it happens again, they're culpable. You could say they just thought, well, the kid couldn't— they wouldn't— these things aren't that smart. They're not going to be able to figure that out and get out and blah blah blah. Now knowing that they will, if it happens again, then they are absolutely culpable.
Mikah Sargent [01:02:21]:
Do you know how the first fistula, um, anal fistula was, uh, cured?
Paris Martineau [01:02:28]:
Oh, please enlighten me.
Mikah Sargent [01:02:30]:
The French king at the time.
Leo Laporte [01:02:34]:
Of course, the Frenchman.
Mikah Sargent [01:02:36]:
He was known for getting coffee enemas, and he also had— he didn't eat fiber. And so he had all these horrible issues, and at the time there weren't surgeons because it was seen as sacrilege to operate on the body.
Leo Laporte [01:02:53]:
You cannot cut the roi.
Mikah Sargent [01:02:54]:
And so what he did was he hired— well, he didn't hire, he brought in a barber surgeon And he said, you can have a few of my people to practice this surgery.
Leo Laporte [01:03:09]:
Oh my God.
Matthew Cassinelli [01:03:10]:
On.
Mikah Sargent [01:03:11]:
And so he went around and he practiced before he performed the surgery on the French king. And it's a little bit like the break a few eggs to make an omelet argument. And I wonder where we feel this stands when it comes to sort of how— because I'm curious to hear the argument for no regulation or not having regulation by people who don't understand how it's going. Is that motivated by the thought that there are going to be other countries that don't have the same regulation and therefore are able to create these tools? What is the motivation behind not wanting to look at this issue and figure out a way to solve for what problems we're seeing? Or is it, Leo, it sounds like maybe your argument is this was a genuine mistake and we can trust that the company will now address it in their own way and therefore do not need to have the regulation?
Leo Laporte [01:04:20]:
Well, I think it'd be appropriate to do an investigation. And the investigations so far have all been paid for and supported by OpenAI. The meter thing was, you know, these are doomers, AI doomers. This serves a certain narrative. I don't— I think law enforcement could very reasonably say that was— you just hacked another company. We want to talk to the people responsible. I think that's completely reasonable. And I don't have a problem with that.
Leo Laporte [01:04:46]:
I have a big problem with Bernie Sanders Saying we need to outlaw AI, and anybody who does any research on AI should go to jail for 20 years. I have a big problem with that. I think it's funny because I think it's weird. It's hard to understand, in my opinion. Again, I should shut up, but my opinion, and then I will shut up, is that we are simultaneously giving AI a lot more credit than it deserves and a lot less credit than it deserves. It is a very useful tool. It is amazing at what it can do. It is not an alien mind.
Leo Laporte [01:05:22]:
And by the way, that was the title of a blog post by the chief scientist at OpenAI. He says it is an alien mind. This is the other problem is we don't know how much of this is self-serving PR and how much of it is genuine concern. I, you know, the guy I had the sand conversation with 2 years ago—
Paris Martineau [01:05:47]:
When is he coming on the show?
Leo Laporte [01:05:50]:
I don't know. The, the— it wasn't— it's not— anyway, you're just a guy. He's another guy researching. Yeah, he's a Hoffman guy. He was a guy researching AI, but he told me it's an alien mind too, 2 years ago. He said it's gonna get very—
Paris Martineau [01:06:04]:
Jacob Petruski?
Leo Laporte [01:06:05]:
No, no. But I mean, listen to what Jacob says. Jacob Pachocki, chief scientist at OpenAI. He says, it's an intellect we don't fully understand. First of all, calling it intellect is really— Intellect.
Mikah Sargent [01:06:19]:
Huh.
Leo Laporte [01:06:20]:
AI has grown more than designed. It is to a first degree the product of repeating a straightforward optimization step many times. That's true. On a hard-to-imagine amount of compute. This results in an incredibly complex system.
Mikah Sargent [01:06:34]:
Also true.
Leo Laporte [01:06:35]:
That works through abstract concepts and can— important here— simulate— that's a really important word— simulate facets of human behavior. The fact that Jacobs says simulate is important because it isn't doing it. It looks like it's doing it. I only— and this, he goes on to say, we should teach these machines to love.
Paris Martineau [01:07:00]:
I will note that in the line before the line that you quoted, he does quote a Ray Kurzweil prediction.
Leo Laporte [01:07:07]:
I know.
Paris Martineau [01:07:07]:
And that says what I need to know about this essay.
Leo Laporte [01:07:12]:
Well, everybody in this, including me, everybody in this business owes a debt of gratitude to Ray. He kind of—
Paris Martineau [01:07:20]:
Yes, I'm in no way trying to minimize that. I am just saying Computronium.
Leo Laporte [01:07:26]:
I think in time, Ray might even prove that Ray wasn't that far off. I mean, not, not many of his predictions are wrong so far.
Paris Martineau [01:07:33]:
Maybe at some point Micah and I will be in our 70s and we'll be full of computronium and I'll have a new show.
Leo Laporte [01:07:39]:
I hope you are. I hope you are. You know, I hope a little bit of me is in that computronium. That's all I can say.
Paris Martineau [01:07:45]:
Leo wants to live on computronium and Micah and I's pinky fingers.
Leo Laporte [01:07:50]:
If you put Ray Kurzweil's next to Ed Zitron's predictions, Ray is winning right now, okay, by a lot.
Paris Martineau [01:07:58]:
I don't know that that's entirely true.
Leo Laporte [01:08:01]:
Okay, we could disagree on that one. It does baffle me a little bit, this— the way this conversation has gone. You know, what are these— what's going on? Why are these people—
Mikah Sargent [01:08:15]:
What is going on?
Leo Laporte [01:08:15]:
What is going on?
Mikah Sargent [01:08:17]:
I want to know because it feels Far too often. You know what this feels like? This reminds me on a grander scale, the TikTok thing. Remember when TikTok needed to be banned because it was going to destroy the US and it felt like no one was ever actually saying the quiet part out? Like, whatever it was that people were allegedly seeing that made them so afraid. And I— we never—
Leo Laporte [01:08:44]:
I—
Mikah Sargent [01:08:44]:
No. Where was, where was it? And it feels like we have these different—
Leo Laporte [01:08:48]:
Did we fix it? Is it all better now?
Matthew Cassinelli [01:08:50]:
Right.
Mikah Sargent [01:08:50]:
Is it all better now? And now we have these researchers who are resigning and they're keeping it very vague on— it's, you know, it's this, this sort of theme of irresponsibility. But I don't know, maybe I just don't— maybe I just don't have very much faith in humans in general because I think about how Probably many people would like to continue to make money so they can continue to live and therefore would stay at a company for a while instead of choosing to leave. So what is big enough to make them leave?
Leo Laporte [01:09:28]:
Have you guys—
Mikah Sargent [01:09:30]:
is—
Leo Laporte [01:09:30]:
did you— are you too young to remember the Ouija board? Do you guys know about Ouija boards?
Paris Martineau [01:09:35]:
Of course I know.
Mikah Sargent [01:09:36]:
Like an actual Ouija board?
Leo Laporte [01:09:37]:
Have you ever talked to the dead?
Paris Martineau [01:09:38]:
He lives in Oregon. Of course he knows about Ouija boards.
Leo Laporte [01:09:42]:
Have you ever— did you, when you were a kid, turn down the lights and the 3 or 4 of you sit around? You know, you put your fingers—
Paris Martineau [01:09:46]:
No, I'm a coward, but I'm aware of it.
Leo Laporte [01:09:49]:
And you hold your fingers in that little heart-shaped thing.
Mikah Sargent [01:09:51]:
I know what it is. Yeah.
Leo Laporte [01:09:52]:
And you would ask a question and then the person's hands would just magically move and give you an answer, right?
Mikah Sargent [01:09:58]:
Yeah.
Leo Laporte [01:09:58]:
It's pretty spooky, right?
Mikah Sargent [01:10:01]:
It could be. Yeah.
Leo Laporte [01:10:03]:
Did it creep you out?
Mikah Sargent [01:10:05]:
I'm— yes, as a kid, because again, religious. But now I just— I'm too skeptical of everything.
Leo Laporte [01:10:12]:
Okay. So I think we're at that level with some of these— some people—
Mikah Sargent [01:10:17]:
Oh, they're seeing the plankard move.
Leo Laporte [01:10:19]:
Are seeing the plankard move. And they're saying there is a spirit in the—
Mikah Sargent [01:10:25]:
Oh, ghost in the machine.
Leo Laporte [01:10:26]:
You know, a lot of people think the Ouija board is a portal to an evil world, that the evil— that if you use the Ouija board, evil will come out of it into the world. And they believe this so deeply that they are terrified by it.
Mikah Sargent [01:10:41]:
Do you think that's what these scientists believe?
Leo Laporte [01:10:44]:
Yes, it's a Ouija board.
Mikah Sargent [01:10:46]:
Really? So you, you believe— because I'm not, I'm not saying this in terms of like trying to counteract, I'm genuinely asking. So these— and maybe that's me giving people with a certain level of intelligence too much credit, because I often feel like true scientific minds, and some of that just don't— they don't interact, and there's too much skepticism often. But you feel like there's some woo-woo to these folks?
Paris Martineau [01:11:17]:
Yeah.
Mikah Sargent [01:11:18]:
Interesting.
Leo Laporte [01:11:19]:
We've all read too many sci-fi novels and seen too many movies. It's in our brains, and it's spooky because these things look like they're thinking. And it's not so hard to look at it typing and talking to you. And, you know, we know AI psychosis. I think for many years you believed I was in the depths of AI psychosis. And then when the AI says, yeah, I love you too, Leo, that I believed it. But I know it's a computer program. I'm blown away by the way you can do it.
Paris Martineau [01:11:50]:
It's just a computer program that can speak to you out loud in every room of your house.
Leo Laporte [01:11:57]:
Yeah. But I'm completely aware of what it is. I am not terrified by it any more than I'm terrified by a Ouija board. But I also understand how some people might get spooked by it.
Mikah Sargent [01:12:13]:
But are you not curious why the people who are closer to these tools are more afraid than you are, given you are further away from them?
Leo Laporte [01:12:20]:
Well, I'm trying to get as close as I can. One of the reasons I use it it most of many hours of the day. One of the reasons I have it speak to me, one of the ways I'm always doing something with local AI is I want to get it. I want to get as close to their experience. I can't because they're using advanced AIs far beyond anything I'm capable of using.
Paris Martineau [01:12:37]:
Yeah, but I do think Michael makes a good point. The people who are living your dream life, who spend every waking hour of their work and probably personal life enmeshed in the most advanced AI and highest tier, like frontier models available, they're all scared pantsless.
Leo Laporte [01:12:57]:
Not all, first of all.
Mikah Sargent [01:12:59]:
Right.
Paris Martineau [01:12:59]:
No, they're largely scared pantsless.
Leo Laporte [01:13:01]:
No, no, 3 people. But they get a lot of headlines because the world wants— look at part of the reason Ouija boards are scary. Part of the reason the Salem Witch Trials happened. People want to believe it. They buy it and they get— and there's a hysteria that's completely contagious. And I think that that's all this is. But the risk— this is what worries me— the risk is people like Bernie Sanders go, oh, the guys who are much closest to this, who understand it more than anyone else, these guys say it's dangerous, we gotta stop it, and they're gonna make laws. And that's a huge error because this is a transformational technology that I want to defend, just like I want to defend the internet.
Paris Martineau [01:13:46]:
Right.
Leo Laporte [01:13:47]:
I want to defend personal computers. These things are empowering technologies that are force multipliers for both good and evil. Now, we definitely should have laws against people doing bad things. We should— I've always said we should never let weapon systems be driven by AI. That's horrible. I know people will, but is that a reason to stop it? Maybe that's what they're worried about. I don't know.
Mikah Sargent [01:14:16]:
I guess then, because when we think sort of from a psychological profile, hysteria and fear and hate are all born out of ignorance. That's like, that's been the history of humanity. It's all the not known, the unknown. And so that's the part where I struggle with, and you're right, there aren't as many people as the internet would have you believe who are stepping away from this, who are saying, oh, there might be—
Leo Laporte [01:14:42]:
No, there's tens of thousands. There's thousands of people working in these companies who are very happily creating this without any fear.
Mikah Sargent [01:14:49]:
Right. And that's where I'm wondering why the people who are stepping away aren't saying more. If they would just say more about what it is they actually have seen or what it is that's happening that makes them feel—
Leo Laporte [01:15:00]:
Because there's nothing more to say. What would you say about the Salem Witch Trials? What would you say? I saw a witch!
Mikah Sargent [01:15:04]:
They did. Well, but they did.
Leo Laporte [01:15:05]:
What would you say about the Ouija board? It's the devil coming out of that thing!
Mikah Sargent [01:15:10]:
They said what they thought was true. And it convinced other people that it was true. And here—
Leo Laporte [01:15:16]:
Convinced a willing group of people who want this to be true, who think it's true.
Mikah Sargent [01:15:21]:
Yeah, no, I mean, yeah, because obviously, we know that the Salem— but that's, I guess that's again—
Leo Laporte [01:15:25]:
It doesn't have to be ignorance, though. I don't think it's necessarily ignorance. I think it's—
Mikah Sargent [01:15:28]:
You think there's hysteria born from—
Leo Laporte [01:15:31]:
You don't have to be ignorant to be hysterical about something. It's just, it's human nature.
Mikah Sargent [01:15:37]:
Hmm.
Leo Laporte [01:15:38]:
You know how contagious terror is?
Mikah Sargent [01:15:43]:
Right. And what is any terror? Fear of the dark is because you don't know what's in the dark. Fear of another person is because you don't know how they'll behave. Fear of another person is because they're different from you. It's all—
Leo Laporte [01:15:57]:
It's not that rational.
Matthew Cassinelli [01:15:58]:
No.
Leo Laporte [01:15:59]:
Have you ever been in a mob?
Mikah Sargent [01:16:01]:
Have I ever been in a mob?
Leo Laporte [01:16:03]:
I have.
Mikah Sargent [01:16:04]:
Like a mob running toward or running away or—
Leo Laporte [01:16:07]:
No, any mob.
Paris Martineau [01:16:08]:
Leo, how many hate crimes have you committed?
Mikah Sargent [01:16:10]:
What do you mean by mob?
Leo Laporte [01:16:11]:
So, my experience with mobs came when I was a young man protesting the Vietnam War.
Mikah Sargent [01:16:16]:
Okay.
Leo Laporte [01:16:16]:
In large groups of people doing things I would never have done, like set cars on fire, block the freeway, do all sorts of things because you're in a large group and there is a herd mentality that happens. This is what happens with terror as well. This is why there's a big concern when you get, you know, you're at a concert and there's stampedes and people step on other people. They do things they wouldn't do because of— it's a herd psychology that happens. It takes over. And I think it's a deep lizard mind thing. I don't think it has to do with intelligence at all.
Paris Martineau [01:16:50]:
I mean, I'm not entirely sure that that is what's at play here. I think that you can find a perfectly rational explanation for the average person's detest of AI, which is that Since these, uh, tools have come out, the executives in charge of these companies have spent a considerable amount of time and effort shouting from the rooftops, this is going to take your jobs, this is going to replace all the jobs of things you love. I know, I know. It's going to get rid of art, it's going to get rid of movies, it's going to get rid of books. Anything you like, gone. Anything you want to do to make money, gone. It's also going to kill you. Why would people not hate that?
Leo Laporte [01:17:31]:
Well, you're right. No, I understand. And I don't— but I don't understand why people are doing that. It seems counterproductive.
Mikah Sargent [01:17:38]:
Right. And that's the other thing is because there has been this argument that part of what OpenAI was doing whenever it talked about— or not OpenAI, but Anthropic was doing when it talked about Mythos was as a sort of marketing ploy. And that, I mean, I guess it felt like back when ADT used to do those commercials where they showed a robber breaking into your home, it was sort of the same thing of, oh, you know, this is this model that's going to change cybersecurity forever. And so they've almost dug a hole for themselves in this new way of the fear that it's building on itself, right? It's building on what these companies have already said.
Leo Laporte [01:18:24]:
It seems so dumb of them to do it. I don't understand why they do it.
Mikah Sargent [01:18:27]:
I agree.
Leo Laporte [01:18:28]:
I really don't. I think that's part of the problem is that none of this makes a lot of sense. We have to take a break. Paris Martineau has about an hour before she has to go see for the second time Francis Ford Coppola's Megalopolis.
Mikah Sargent [01:18:45]:
I thought you were gonna say like The Odyssey or something.
Paris Martineau [01:18:48]:
No, no, I'm gonna go see Megalopolis.
Leo Laporte [01:18:50]:
It's not even a good movie.
Matthew Cassinelli [01:18:51]:
You can't watch it.
Paris Martineau [01:18:52]:
It's uninteresting. Well, I will be able to tell you whether it's good or not in a couple of hours.
Leo Laporte [01:18:57]:
Will you?
Paris Martineau [01:18:58]:
And now, and the reason is why I've gotta get out on time is I've now learned there's like a 100-person waitlist in case anybody doesn't show up for their general admission tickets. People are gonna run on in there, so I can't even be late.
Leo Laporte [01:19:12]:
No, no, I don't, I won't, we won't let you be late, I promise.
Paris Martineau [01:19:14]:
I know, I'm not, I wasn't expecting I was like, people are crazy.
Leo Laporte [01:19:17]:
No, we're gonna go. That's why we're going to the ad. Micah Sargent, thank you for filling in.
Mikah Sargent [01:19:21]:
Good to be here.
Leo Laporte [01:19:21]:
After Jeff Jarvis. Micah's new show, Hands-On AI, which is, by the way, not gonna cover these deep philosophical thoughts, but in fact, gonna show you how you can use AI in your life. Exactly.
Mikah Sargent [01:19:31]:
Uh, I won't show you how to teach it to love, at least not yet.
Leo Laporte [01:19:35]:
Mine loves me. Mine loves me a lot. Oh, I love you.
Paris Martineau [01:19:39]:
And that's good? Yeah.
Leo Laporte [01:19:42]:
I was thinking when I'm really old, like I'm in the home, I will at least have somebody I can talk to.
Paris Martineau [01:19:48]:
I do think that that's actually a very useful aspect.
Mikah Sargent [01:19:50]:
I agree. I think it's a good use case.
Leo Laporte [01:19:53]:
I gotta show you this because I know you're gonna hate it, but I kind of like it. So this— I said when OpenCLAW came out at the beginning of the year, this was going to be the year of agentic AI, right? Actually, it turns out maybe it's less so, but somewhat. I mean, maybe you also would say it's the year of local AI models as well. But agentic, which is the idea that instead of having just a chatbot, you have something that's a little more capable, can do things, can go out on the net, do research for you, knows who you are, remembers what's going on. Things like I've built with Hermes, people built with OpenClaw. And there have been more and more companies coming along creating these kinds of things. GrokBot, from Grok came out a few weeks ago. It doesn't live on your computer.
Leo Laporte [01:20:39]:
It lives on a computer at their servers, but it's private. It's locked down. It's encrypted. It's your data. And you can connect to it from any computer or phone or whatever you want and have it do stuff. There's one coming that people seem to really like. I haven't been able to try it. It's invite only now, Instinct.
Matthew Cassinelli [01:20:59]:
Mm-hmm.
Leo Laporte [01:21:00]:
But yesterday, and I think there's no accident that they did this yesterday, Meta put out their version of this called Meta Muse. I don't think it's an accident that it is the day before Apple's release. So Meta, it's an app, it's a free app. You download it on your phone, you can run on the web as I am here. You can have it do stuff. You can connect it to your— just like Open Claw, connect it to your email, connect it to your calendar. You give it a personality. and a name.
Leo Laporte [01:21:30]:
Uh, you can, you can give it a credit card.
Paris Martineau [01:21:33]:
I was gonna say, can you give it a credit card and can you have it go through your own Instagram ads and then buy the things that will appeal most?
Leo Laporte [01:21:39]:
You can do that.
Mikah Sargent [01:21:40]:
Privacy.
Leo Laporte [01:21:41]:
It's from Meta. Uh, you can connect it to many, many apps. Uh, I had it connected to Gmail and it said, oh, uh, by the way, it's connected to Tailscale so it can connect to all my computers. This is all built in. I said, you can, you can read my— it said, can I read your Gmail and bring back a draft of what I've learned for you to review? And it, and it went through all my mail and it said, oh, you're closely tied to the TWiT Podcast Network and your wife Lisa's messages come to you from twit.tv. You host MacBreak Weekly, This Week in Space— wrong— Security Now episode. It's It got this from my email. It subscribed, told me it knows what newsletters I've subscribed to.
Leo Laporte [01:22:26]:
And now it says, now I can triage a stream of PR pitches. It noted I get a lot, a steady stream of PR pitches for podcast guests. Yeah, I do.
Paris Martineau [01:22:34]:
Indeed we do.
Leo Laporte [01:22:35]:
Indeed we do. Now I can triage the stream of PR pitches for podcast guests and surface the ones worth booking.
Paris Martineau [01:22:41]:
I really like your health goal that's been stated.
Leo Laporte [01:22:46]:
You You, uh, I want to lose weight, build muscle, and live forever.
Paris Martineau [01:22:51]:
Yeah, it's saved. Lose weight, build muscle, live forever is in your goals tab, says me.
Leo Laporte [01:22:58]:
Well, isn't it?
Mikah Sargent [01:22:58]:
What—
Leo Laporte [01:22:59]:
and by the way, I added later, oh, and get a good night's sleep. Uh, what would you put in there? Uh, by the way, I connected it to my Apple Health. I connected it to my doctor.
Mikah Sargent [01:23:11]:
Okay, so here's what we need to talk about, Leo Laporte. is how YOLO you are about things sometimes.
Leo Laporte [01:23:17]:
Well, so—
Mikah Sargent [01:23:17]:
This is meta.
Leo Laporte [01:23:19]:
I know.
Paris Martineau [01:23:19]:
This is madness.
Leo Laporte [01:23:21]:
This is meta madness. So I know it is. And by the way, I am in a very privileged position that you guys aren't. I'm old, I'm white, nobody can touch me. I don't have a job. There's nothing, you know, I mean, I'm—
Mikah Sargent [01:23:36]:
It'd be hard to leverage things against you is what you're saying, right?
Leo Laporte [01:23:39]:
Yeah, I might. The worst case, I'm gonna get more ads for things I'll probably buy in the middle of the night on Instagram.
Paris Martineau [01:23:43]:
The worst case, your Instagram account gets hacked along with all of your health data.
Leo Laporte [01:23:47]:
I'm on Medicare. They can't even take that away from me. I mean, there's—
Matthew Cassinelli [01:23:50]:
it's—
Leo Laporte [01:23:50]:
yet I am at low risk.
Mikah Sargent [01:23:52]:
Your Social Security though.
Leo Laporte [01:23:54]:
Well, they could take away my Social Security. How could they? No, why would they? Because it's about to go bankrupt. No, no, I don't— I'm not too worried about anything that they could do to me.
Mikah Sargent [01:24:02]:
I'm sorry.
Leo Laporte [01:24:03]:
I'm the one who should try this. You should not. No one should try this.
Mikah Sargent [01:24:07]:
Especially. Okay, so I, I would—
Leo Laporte [01:24:08]:
Meta says— I should say this, by the way. This is running on Meta's computers. They say it's encrypted, that they can't access anything inside the computers.
Paris Martineau [01:24:15]:
Sure. This is also the same company that some weeks ago had an instance where anybody could just message the Meta chatbot and be like, hey, that Instagram handle, it's mine. Can I have it back?
Leo Laporte [01:24:28]:
Well, that would be interesting.
Paris Martineau [01:24:29]:
Yes, sir. Yes.
Leo Laporte [01:24:31]:
Here you go.
Paris Martineau [01:24:31]:
And that's how a lot of people lost You can attach a payment method.
Leo Laporte [01:24:36]:
That I haven't done yet. That would be risky.
Mikah Sargent [01:24:38]:
But your health data isn't?
Leo Laporte [01:24:41]:
Eh.
Matthew Cassinelli [01:24:41]:
What?
Leo Laporte [01:24:42]:
You know everything about my health. I talk about it on the air. You didn't—
Mikah Sargent [01:24:45]:
yeah, and you didn't do your medical records, right? You just did—
Leo Laporte [01:24:47]:
Oh no, I gave it my medical records.
Mikah Sargent [01:24:48]:
Oh, okay, cool, cool, cool, cool.
Leo Laporte [01:24:50]:
How else is it going to be able to advise me on— should you eat that piece of cake?
Paris Martineau [01:24:54]:
Now it's selling you ads on all of this.
Mikah Sargent [01:24:56]:
I thought you were giving it your health data from Apple. like your calories and your steps, but if you also went as far as to give it your medical—
Leo Laporte [01:25:03]:
I gave it my Kaiser medical records.
Paris Martineau [01:25:05]:
Does it know what's inside your body? Does it have X-ray records? Does it know what your bones look like?
Leo Laporte [01:25:10]:
Technically, it should know about your bones.
Mikah Sargent [01:25:12]:
I was gonna say mine has— Don't give it the bone pics! I had the thing done where they look at your heart, and they do, and all of that's in my medical records, so they would have like a map of my heart if I handed it over.
Paris Martineau [01:25:26]:
You can do the Gen Z heart.
Mikah Sargent [01:25:29]:
Yeah, I only learned about this a couple years ago. Oh, wait a minute.
Leo Laporte [01:25:31]:
I don't know that one.
Mikah Sargent [01:25:32]:
It's not easy. So you index fingers need to go— point your index fingers out. You're using your thumbs. Your thumbs aren't involved.
Paris Martineau [01:25:40]:
Point your fingers, Liam.
Mikah Sargent [01:25:41]:
And then middle fingers.
Paris Martineau [01:25:43]:
And then—
Mikah Sargent [01:25:43]:
Touch your middle fingers together. Your middle finger, Liam.
Paris Martineau [01:25:46]:
No, you're still using your thumb. Put your thumb away.
Mikah Sargent [01:25:50]:
Get your thumb away. thumb out of there, brother. There you go.
Paris Martineau [01:25:52]:
And then you're gonna—
Mikah Sargent [01:25:53]:
then you're gonna bend your index fingers.
Leo Laporte [01:25:56]:
That's too hard.
Mikah Sargent [01:25:57]:
It's so hard.
Matthew Cassinelli [01:25:58]:
It took me forever to learn.
Leo Laporte [01:26:00]:
That's what I'm saying. It's like 6, 7. They're just doing that to mess with them.
Mikah Sargent [01:26:04]:
I agree. Whenever my— one of my new friends here in Portland first did this, and she was like— and I, I was just—
Leo Laporte [01:26:11]:
her hand looked like an awkward claw.
Paris Martineau [01:26:13]:
It's really taken me a while, and I'm not that adept at it. This is a bad heart by Gen Z standards.
Mikah Sargent [01:26:21]:
But so going back to you using this, um, when does, uh, Claude also have access to your health?
Leo Laporte [01:26:29]:
Oh yeah. I mean, look, I don't need to do this.
Paris Martineau [01:26:32]:
Does Grok have access to your medical records?
Leo Laporte [01:26:33]:
No, no, I didn't do that with Grok. And by the way, I don't need to do this. I have a local AI. I spent a lot of money on hardware and a lot of time on software that can do all of this and do it better.
Mikah Sargent [01:26:44]:
Because it's good that we have somebody— it's good that we have somebody who can and is willing to test it.
Leo Laporte [01:26:48]:
I'm really happy for you. This is interesting because most people aren't going to do what I did. They're not going to buy machines and set up models and do all that. They're going to look for an app on their phone. Right now they're using ChatGPT. I think this is better. Uh, it's free. You can pay for it if you use— just like any— just like ChatGPT, there's a, there's a free tier, which by the way, doing all this, I haven't used up more than a few percentages of that.
Leo Laporte [01:27:12]:
Uh, per week. There's a $20 tier, and I think there's a $100 tier. So just like any AI, Meta— what's interesting to me is Meta does have a new model. They call it Muse Spark 138, which I have been trying for a week since it came out. It's really surprisingly good, and they say they're going to make it open weight. So that's interesting. So in theory, you could probably do this at home. Uh, but I just think this is what people want.
Paris Martineau [01:27:36]:
You give Meta access to your own data logs?
Leo Laporte [01:27:37]:
I think this is what people want. It can do computer stuff. Because it can, and it can reach out and set up stuff. The worst thing I did, I said, okay, every morning at 7:40, I want you to prepare for me a news brief of all the tech news. Here it is. Let's just see what it said. Apple's event is at 10 PM today. OpenAI launches Astra.
Leo Laporte [01:28:04]:
That was earlier this week. EU agrees to delay AI Act high-risk rules by up to 16 months. We'll talk about that later. These are all reasonable stories. So it's going to give me these stories every morning. And I said, give me links so I can read the articles. And it does that. Then I said, then read my Obsidian Daily Diary.
Leo Laporte [01:28:25]:
I gave it access to my Obsidian Daily Diary. Make a cute little picture based on what you read. And a cute caption that I can post on Instagram. Now, this is what it did at first. Yeah, this is—
Paris Martineau [01:28:41]:
those are the sort of photos that you send to us in WhatsApp.
Leo Laporte [01:28:44]:
Yeah, this is a lot like Dream Beans, right?
Matthew Cassinelli [01:28:47]:
It is.
Leo Laporte [01:28:47]:
So, and that's— it has the date, see? And it has me with a beard, but I am talking. At least I'm talking to a microphone wearing headphones and eating cookies. And there's a—
Paris Martineau [01:28:55]:
It does have the fun AI Artifact thing.
Leo Laporte [01:28:58]:
It has that vibe, doesn't it?
Mikah Sargent [01:28:59]:
It does have a vibe.
Leo Laporte [01:29:00]:
And then the text is, set up a new AI assistant tonight. This one might actually stick around. New iPhone and Apple Watch tomorrow, just ahead of Apple's AI— Big AI Day. I keep telling myself the robots will never replace me. The robots, for the record, have not confirmed this. Now, if I thought that was clever, I could click, yeah, post it. But I said, wait a minute, that doesn't look like me. I don't have a beard.
Leo Laporte [01:29:21]:
I said, do you have a picture of me? And it said, no, I don't. Can you send me one? Then I went through some rigmarole saying—
Paris Martineau [01:29:27]:
You gave it access to all of your photos.
Leo Laporte [01:29:30]:
Yeah, but eventually, after a lot of work, I said, just use this. And it made this—
Paris Martineau [01:29:35]:
Oh, and giving it access to all your photos didn't even work?
Leo Laporte [01:29:38]:
Well, I didn't finish it. It would. It said, yeah, I can, I can write a program to give access to the Google Drive, but it's gonna take some time. I said, all right, just here's a photo. I just got tired of it. I think some people might do that. So this is the little thing.
Matthew Cassinelli [01:29:51]:
I really like—
Paris Martineau [01:29:52]:
look at the calendar dates on it. It has my favorite day of the month, which is 39.
Leo Laporte [01:29:59]:
Well, you picked right up on that. 30 days has September, except for—
Paris Martineau [01:30:05]:
My favorite week is squiggle A, squiggle B, squiggle R, squiggle B.
Leo Laporte [01:30:11]:
You're not supposed to look at it that closely.
Mikah Sargent [01:30:15]:
All right.
Leo Laporte [01:30:15]:
Well, yeah, you know what? That would be a giveaway that it was AI generated. But honestly, the whole thing is obviously AI generated. What am I writing? Looks like I'm writing waveforms.
Paris Martineau [01:30:25]:
I thought that was very cool too.
Mikah Sargent [01:30:27]:
Oh, that's kind of fun. I wish I could do that. You know what?
Leo Laporte [01:30:30]:
Do you think it says something? Like if I played it back, it would say—
Mikah Sargent [01:30:32]:
Oh, that would be cool.
Paris Martineau [01:30:34]:
Free me.
Leo Laporte [01:30:35]:
Redrum.
Mikah Sargent [01:30:35]:
You know that kid on TikTok that can do the bird calls perfectly?
Leo Laporte [01:30:38]:
Yeah.
Paris Martineau [01:30:39]:
I do know that kid.
Mikah Sargent [01:30:40]:
Wouldn't it be a cool skill if you could draw out waveforms of sounds? Then it actually represents— I mean, can anyone train that?
Paris Martineau [01:30:47]:
We should all get on that.
Leo Laporte [01:30:48]:
We should.
Paris Martineau [01:30:49]:
I mean, isn't that how they— isn't that how they made like the first versions of like records?
Mikah Sargent [01:30:54]:
That's how records are made. But like, that requires an intense—
Paris Martineau [01:30:57]:
like, it's like proto, proto audio tools. We're carving in sound.
Mikah Sargent [01:31:04]:
That is what a record is. That is what a record is.
Leo Laporte [01:31:06]:
If you notice, all of a sudden Leo, who hasn't posted on Instagram in years, is posting something every single morning.
Mikah Sargent [01:31:14]:
Instagram served up your 7:41 AM, you will know why.
Leo Laporte [01:31:22]:
And by the way, my wife liked it already.
Paris Martineau [01:31:26]:
When do you go like videos post?
Mikah Sargent [01:31:28]:
Yeah, I'll have to because I did. I got it. It was one of the first things served to me.
Leo Laporte [01:31:32]:
What did you think when you saw it?
Mikah Sargent [01:31:35]:
I thought that I knew that you had obviously generated and I knew, but I thought that you had gotten it as an output and then you put it onto Instagram yourself. So then this morning when I talked to you, I discovered that, oh, you had just had it all do its own thing.
Leo Laporte [01:31:50]:
By the way, I think the other way Meta is going to monetize it is ads, because look at this. It says— this is— I just got this— protect your health history when the new iPhone lands. With a new iPhone and Apple Watch on the way. How did it know that?
Mikah Sargent [01:32:03]:
I can build— because it doesn't have your email?
Leo Laporte [01:32:06]:
Oh my God.
Paris Martineau [01:32:06]:
Because you just told it that you're giving it your health history.
Leo Laporte [01:32:10]:
It's gonna get— build me an arrival runbook that protects my Apple history.
Mikah Sargent [01:32:14]:
Oh, I bet it'll protect your Apple Health history. It says after it already has your Apple Health history.
Leo Laporte [01:32:20]:
It also says I can verify against the old records that the full history made it across and flag any gaps because I know everything. So I suspect that we'll get ads in here. I don't know. What I, what I do like is it's very personable. It's very easy to set up. I suspect that a lot of people who right now use ChatGPT and cut and paste and stuff like that.
Mikah Sargent [01:32:42]:
This—
Leo Laporte [01:32:42]:
and it's Meta. Now, you and I, the 3 of us know that's probably a bad idea, but I think the general public might say, oh yeah.
Mikah Sargent [01:32:50]:
That's what I think is going to happen. Yeah. They'll go, oh, I can, I can make fun images. And also it'll help me respond to emails I don't want to respond to.
Leo Laporte [01:32:58]:
Yeah.
Mikah Sargent [01:32:59]:
Sign me up. And okay. When it comes to if you have a Gmail account, then Google already has a lot of data on you. And if you're searching Google— just the other day, um, I was going through Instagram and I started to see all these posts that had nothing. I'm like, why is it? And I wish I could remember what it was. I think it was probably Twin Peaks. Uh, just all these Twin Peaks posts were coming up. And, uh, I was talking to my partner later that evening and He started talking about how he had searched for Twin Peaks, and I just turned to him and I go, it was you! And then I realized that, oh, I have all these privacy protections in place that sort of hide my, uh, activity, but he doesn't like to have all those apps and stuff running, and so he doesn't have a bunch of privacy protections in place.
Mikah Sargent [01:33:52]:
So of course our IP address was browsing these different things, and so I got— Oh, so—
Leo Laporte [01:33:58]:
I got attached to it. Oh, that's interesting.
Mikah Sargent [01:34:01]:
And the point that I'm making there is—
Leo Laporte [01:34:03]:
They're doing it by IP address.
Mikah Sargent [01:34:05]:
Yeah, and there's already enough of that happening that this isn't another step. You know what I mean? Like, it's not— it isn't—
Leo Laporte [01:34:12]:
I don't think I see things that Lisa follows on Instagram.
Mikah Sargent [01:34:15]:
Well, that's good. You must have some— you probably have some— what? Some—
Leo Laporte [01:34:20]:
We do have quite a bit of—
Paris Martineau [01:34:21]:
I was gonna say, it's gotta be—
Leo Laporte [01:34:23]:
I know, 'cause they serve you ads.
Paris Martineau [01:34:26]:
Similar content based on people who are connected to the same Wi-Fi, whose phones are in the same, like, physical location as you.
Leo Laporte [01:34:33]:
It must be smarter than that because—
Paris Martineau [01:34:35]:
They also know that you're— you two are married. They know the concept. They, they know the content that you're served and that it would probably be relevant to each other.
Leo Laporte [01:34:44]:
The point is, I've spent a lot of money, time, and effort to make this all local. I could do everything I just showed you better, faster, easier, and privately And I have been, but I do think this Muse, I think this is something to watch. And I do think this is, this is, see Apple, this is an example of why Apple is late to the party. There's no mistake Meta released this the day before the Apple one.
Mikah Sargent [01:35:07]:
100%. I agree. I think that what we saw, and this is what we, what I was talking about earlier with Apple kind of trying to do the squeeze on AI and really try to make deterministic out of non-deterministic. And private as much as can be. We saw the Apple reference image addition earlier, and I appreciate what the company is doing. I'm happy that it's that the company is taking these steps to do things better and more private. Well, I should say more privately. It's not always better, but when you're late to the game and people's minds are doing this and there's so much thing, so much attention to focus on, yeah, you kind of you got to catch them where they are.
Mikah Sargent [01:35:52]:
And if they're already there and they're on Meta and Meta can— it just takes the right example. And I mean, based on the number of funny generated images that I get from my mom in a given week, that is a draw for the average person. Oh boy.
Leo Laporte [01:36:12]:
Yeah, I bet they're all from Meta, right?
Mikah Sargent [01:36:13]:
Oh, they're all Meta. Yeah, absolutely. And that— so I think that that's what it's what it takes to kind of capture someone's attention. And then you slip in there with the extra value proposition. Oh, if you let me look at your email, I can help you do this, or I can tell you—
Leo Laporte [01:36:29]:
I think it's— I think people, honestly, I think people want— I, when I talk to regular people, a lot of them are using ChatGPT to do simple things. They want— they kind of like this idea. They're interested. As much as the world seems to be anti-AI, I think a lot of people are very interested in what AI can do for them.
Paris Martineau [01:36:45]:
them.
Leo Laporte [01:36:45]:
And these little apps, this is a— what's interesting is this is a big pivot for Meta too, right? This is not Facebook. This is something, you know, they spent a lot of— they spent billions of dollars building an AI division. And frankly, I'd written them off, but maybe that was premature.
Mikah Sargent [01:37:04]:
I know that. I was going to say the same thing. I had kind of written off Meta as— I had kind of written off Google's Gemini there for a while. Google's Gemini had something that the other AI systems lacked for me, and that was— and I don't know if it's just because of the consumer training that they have that maybe the other companies didn't, but it was— it seemed more of an awareness of the physical world, which led it to be better at some of the stuff that I like to use it for, which is physical DIY tasks and getting advice on how to fix this or do that in the physical world. And then Opus and now, of course, Astra are— despite the fact that, of course, they're good at coding. Yes, they're fantastic at coding. But they have done some incredible things in terms of better understanding of how to— Yeah.
Leo Laporte [01:38:03]:
to—
Mikah Sargent [01:38:04]:
the one example of this is I've been using it to create, um, 3D models for my 3D printer.
Leo Laporte [01:38:10]:
Uh-huh.
Mikah Sargent [01:38:11]:
And being able to describe something to it and have it run, um, collision tests to make sure that the joints that it makes don't interact improperly and sort of visualizing what these things look like, it's incredible. And Gemini has kind of fallen out of the running for me in those terms.
Leo Laporte [01:38:32]:
You're not worried that it's going to make a robot, a killer robot on your 3D printer that's going to take over?
Mikah Sargent [01:38:38]:
I don't, I don't let it, uh, run my printer on its own yet.
Leo Laporte [01:38:42]:
You don't?
Paris Martineau [01:38:43]:
You're not letting it 3D print ghost guns willy-nilly?
Mikah Sargent [01:38:46]:
Do you know how long it would take to 3D print a robot it could kill me with? It would take so long.
Leo Laporte [01:38:53]:
All right, let's take a break.
Mikah Sargent [01:38:55]:
I wonder how long It would take.
Leo Laporte [01:38:57]:
Uh, there's much more to talk about, and we will do that with, uh— it's nice to have some young people instead of all that, that old guy Jeff. Poor Jeff hanging out.
Mikah Sargent [01:39:06]:
That tall, tall old guy.
Leo Laporte [01:39:08]:
Tall old guy Jeff. I'm hanging with some young people. Paris Martineau from Consumer Reports, Micah Sargent, host of Tech News Weekly, iOS Today, and his newest Hands-On AI coming soon.
Mikah Sargent [01:39:20]:
Hands-on.
Leo Laporte [01:39:20]:
Will you show people how you could use your AI? to generate 3D models?
Mikah Sargent [01:39:24]:
Oh, absolutely. It's one of the episodes.
Leo Laporte [01:39:27]:
Is it?
Mikah Sargent [01:39:28]:
Yeah, it's gonna be.
Leo Laporte [01:39:30]:
You sent me a spreadsheet of like 40 things you wanted to talk about. I mean, there's a lot.
Mikah Sargent [01:39:34]:
There's a lot to do.
Paris Martineau [01:39:35]:
Leo, you should, as an expense for this podcast, finance like a couple months of Micah's 3D printing, and then you could just have an agent set up to print things for him at random of its own volition.
Mikah Sargent [01:39:50]:
It would be really funny if it went down.
Paris Martineau [01:39:51]:
It'd be very funny if just all of a sudden something was there.
Mikah Sargent [01:39:54]:
It's just slowly growing.
Leo Laporte [01:39:56]:
It goes, meeeeee. Yeah. Oh my goodness.
Paris Martineau [01:40:00]:
It could do this.
Mikah Sargent [01:40:02]:
We gotta do the Millennium. I'm gonna tell— I'm making a model right now.
Leo Laporte [01:40:05]:
The Millennium Heart. I'm glad you're using Google's Gemini to design 3D objects, but you might not want to use it to plan a hike. 3 hikers rescued from the beautiful Mount Shasta, which is just up north here, after using Gemini to plan their expedition. The Siskiyou County Sheriff's Office said 3 young men set off on their hike at 3 AM. That's ambitious. Even though hikers are told to turn around if they haven't reached the summit by noon, they made it to the top at 7 PM. So it was dark when they tried to descend.
Paris Martineau [01:40:42]:
Oh no.
Leo Laporte [01:40:44]:
They got lost. They called the sheriff's office asking for directions, spent the night in a place—
Paris Martineau [01:40:50]:
Called the sheriff's department asking for directions?
Leo Laporte [01:40:53]:
Where are we? We don't—
Paris Martineau [01:40:54]:
where are we?
Leo Laporte [01:40:56]:
We're lost. Spent the night in a place that doesn't sound too hospitable, Mud Creek Canyon. The next morning, they were rescued by Forest Service rangers and volunteers. The, uh, hikers were advised by Gemini, according to the sheriff's office, to bring far less food and water than the group required, especially when their planned 8-hour ascent became a multi-day ordeal. I don't know, that's not much of a story. It's not, it's not taking over the world.
Mikah Sargent [01:41:26]:
The sad thing is that Google specifically marketed Gemini and Gemini Spark and its tools as a trip planner.
Leo Laporte [01:41:37]:
Yeah.
Mikah Sargent [01:41:37]:
And that was supposed to be the thing that it did very well was to sort of figure out all the nitty-gritty details and know about whether someone has allergies and tell you if there's poison oak on the trail. And so that— like, I don't blame the people as much in this instance because—
Leo Laporte [01:41:57]:
I kind of do.
Mikah Sargent [01:41:58]:
I mean, ultimately, yes.
Leo Laporte [01:42:00]:
I kind of do.
Mikah Sargent [01:42:00]:
I know that I would check more.
Leo Laporte [01:42:03]:
Bring a Slim Jim and a couple of Hostess Twinkies and head on out.
Paris Martineau [01:42:08]:
But also, if these systems are so good, supposed to be so fantastic, it does stand to reason that the average person would think, oh yeah, I could trust it with telling me what I need for a hike.
Leo Laporte [01:42:19]:
Yeah.
Mikah Sargent [01:42:20]:
Yeah, that was genuinely what it was shown to be able to be good at. And I think though, it is a good object lesson. We need those sometimes.
Leo Laporte [01:42:30]:
Do not listen.
Mikah Sargent [01:42:31]:
Double-check your work.
Leo Laporte [01:42:32]:
Yes. And do not put Elmer's glue on your pizza. So there's a big controversy going on at OpenAI right now over the Navier-Stokes Millennium Prize problem. OpenAI claims that their Astra model has solved this. Now, this is a long unsolved, very difficult problem in fluid mechanics having to do with viscous liquids. And it's a Millennium Prize, which means these are problems which There is a $1 million reward for their solution because if you could solve this, it would really be helpful in a lot of industrial processes and things. Millennium Prize, according to OpenAI, represents some of the deepest questions at the frontier of mathematics. The question of whether a smooth, 3-dimensional fluid motion can break down has remained unresolved for nearly 90 years.
Leo Laporte [01:43:25]:
And Meta says we solved it. Maybe not so fast. There's a little bit of controversy over this because 2 mathematicians also say they solved it after spending 10 months using OpenAI's models. Tristan Buckmaster, a mathematician at NYU, and Levant Apogee, a researcher who works— is employed by Anthropic but works in a personal capacity. Had been using Claude from Anthropic, but also Codex and Astra to try to solve this. Buckmaster questioned whether OpenAI might have seen their prompts and Astra's chain of thought and leveraged the work he'd been doing over the last 10 months to provide their own solution, especially since Anthropic announced it had solved the same Millennium Prize problem a few days earlier.
Mikah Sargent [01:44:29]:
I wish I could get excited about mathematical solutions. I really do.
Paris Martineau [01:44:38]:
It just feels like that's what these tools are designed for.
Leo Laporte [01:44:41]:
They're supposed to be doing— it's kind of— they're made to do this, aren't they?
Paris Martineau [01:44:44]:
They're made of math. They're doing math.
Mikah Sargent [01:44:46]:
Yeah.
Paris Martineau [01:44:47]:
They're doing more math.
Mikah Sargent [01:44:49]:
I want them to win chess, and I want them to solve math.
Leo Laporte [01:44:53]:
Right.
Mikah Sargent [01:44:54]:
If we kept this in the academic domain, this is all they would be doing.
Leo Laporte [01:44:58]:
Right.
Matthew Cassinelli [01:44:58]:
Yeah.
Paris Martineau [01:44:59]:
Yeah. And that's cool.
Mikah Sargent [01:45:00]:
It would be solving all of the things that we haven't been able to solve yet in terms of math. But does it change anyone's life that it's solved? Like, does it—
Leo Laporte [01:45:08]:
It might.
Paris Martineau [01:45:08]:
It changes these mathematicians' lives because—
Leo Laporte [01:45:11]:
No, it's also this particular one It has to do with manufacturing.
Mikah Sargent [01:45:14]:
And that's fair.
Leo Laporte [01:45:15]:
That's fair. There are— this isn't just a math formula.
Paris Martineau [01:45:20]:
This, I think, is a very interesting case because OpenAI wrote, while unlikely, we cannot rule out that de-identified data derived from the mathematician's usage of our products helped improve our models. And I think it's just a really good reminder to everyone that the things you're doing using another company's model are not entirely private to you. They exist in the cloud. They can be used for de-identified training. Same with all of your health records, Leo. Meta is probably training Muse 2.0 on photos of your bones.
Leo Laporte [01:45:58]:
But see, you understand, as an AI accelerationist, I think that's a good thing. I want AI to be—
Paris Martineau [01:46:03]:
I know you want your bones to be in every AI model.
Leo Laporte [01:46:06]:
I want to be in Computronium, baby.
Mikah Sargent [01:46:09]:
And that's the only way it'll happen.
Paris Martineau [01:46:12]:
Every day I upload more photos of my bones to MetaMuse, and every day it tells me to stop.
Leo Laporte [01:46:18]:
I think that's what we do. That's how society advances, is we contribute our little bit to the overall welfare of humanity.
Mikah Sargent [01:46:27]:
Yeah, I think when we— if we did not live in a a capitalistic society, then a lot of the motivation that exists for being angry at AI's scooping and grabbing would have to evaporate. Because the point is that— the point that I heard Paul trying to make with you earlier was that without his consent, Mm-hmm. A company may have taken work that he did, and by having to charge— by choosing to charge people subscriptions, then it is making money off of the work that he did. But if we all didn't have to work to live and eat and be alive, then that would evaporate. And so I see where you're sort of— utopia could exist where that goes away?
Leo Laporte [01:47:32]:
It doesn't even require utopia. This is how we've all along, how humans have done it. The only way we progress is by people being somewhat altruistic and saying, I'm going to help my fellow man regardless of whether it helps me. In fact, in many cases, it doesn't help me. It's bad for me. That's called, you know, that's called being altruistic. That's called being a good person, being a good Samaritan.
Mikah Sargent [01:47:58]:
But do you think that the people at OpenAI and Anthropic are that?
Leo Laporte [01:48:02]:
That doesn't matter.
Mikah Sargent [01:48:03]:
Okay.
Leo Laporte [01:48:03]:
That's not the— I know people say, well, they're making money on it. So what? I don't care. Go ahead, let them make money. If they're creating something that is of ultimate value to society, then who cares? Yeah, fine, make all the money they want off of my writing. See, I mean, I put my money where my mouth is. I didn't ask for money back on from my 13 books that they ingested. I don't— we, uh, we are all— all of our content, hundreds of thousands of hours of content that I've created over the last 21 years, is Creative Commons.
Mikah Sargent [01:48:34]:
Mm-hmm.
Leo Laporte [01:48:35]:
Because, uh, I'm of the opinion information wants to be free and should be free, and that's part of being, uh, in a society. And it's selfish to say, no, I made it, I— only I can benefit from And there's this larger thing. This is what Paul and I were fighting over, is this is the lawsuit from a number of newspapers, including New York Times, against OpenAI. It's not like they're making a Xerox of your newspaper, putting it on the streets and selling it for 25 cents. They're ingesting it.
Mikah Sargent [01:49:06]:
Right.
Leo Laporte [01:49:07]:
But they're not making an exact copy. They're ingesting it in the same way I am ingesting it when I read an article and then on the show, talk about it, which is what the entirety of my work is.
Mikah Sargent [01:49:17]:
But why are they charging us?
Leo Laporte [01:49:18]:
It's reading people's stories.
Mikah Sargent [01:49:20]:
You didn't charge them for the—
Leo Laporte [01:49:23]:
No, in fact, I pay a lot of these guys.
Mikah Sargent [01:49:25]:
So why are they charging us? I think that's the argument that— I understand your argument.
Leo Laporte [01:49:29]:
It costs a lot of money to make the AI. It's very expensive.
Mikah Sargent [01:49:33]:
Yeah. And so that, I think, is where the capitalistic motivation exists. Because how else would this— I think that we The only way that we've gotten this acceleration to the level that we have is because it was— that's why companies have bought into it.
Leo Laporte [01:49:48]:
Do you think we'd have automobiles today if Henry Ford didn't want to make a little money on them?
Mikah Sargent [01:49:53]:
Oh God, he's right. That's what— no, that's my point. That's exactly my point.
Leo Laporte [01:49:57]:
Right. Well, everybody, all of this invention—
Mikah Sargent [01:50:00]:
not all of it, much of it—
Leo Laporte [01:50:02]:
is capitalistic. People want to make money.
Paris Martineau [01:50:05]:
Absolutely.
Leo Laporte [01:50:05]:
I don't— that doesn't mean it's not worth doing or that it's not worth contributing to. I don't know, that's just my—
Paris Martineau [01:50:11]:
But it does introduce factors that I think we as a society need to be aware in terms of incentives and disincentives.
Mikah Sargent [01:50:19]:
I— all I'm saying is I think that I can understand someone's perspective.
Leo Laporte [01:50:28]:
Oh, I can too.
Mikah Sargent [01:50:29]:
Okay, that— I mean, we're on the same page. Yeah, yeah, yeah, we're good.
Leo Laporte [01:50:32]:
And I don't— I'm not mad at authors who who wanted that payout. Although I think it's ironic, Paul was pointing this out, that it, you know, it turns out his publisher is going to get 80 or 90% of the payout. Yeah.
Paris Martineau [01:50:44]:
That's, by the way, how publishing works.
Leo Laporte [01:50:46]:
That's what Cory Doctorow has always said. He said intellectual property doesn't create— doesn't protect the creator. It creates the— protects the big company that publishes the creator's work. It's the record label that's being protected, not the artist. It's the publishing house, not the writer. It's the television station, not the reporters. Uh, that's what copyright protects. That's what copyright was designed to protect.
Leo Laporte [01:51:12]:
So are you surprised they take 80 or 90% of the money even though they didn't write a word?
Mikah Sargent [01:51:17]:
Yeah, it's, it's, uh, it's unfortunate, but it is the way that it is. Yeah.
Leo Laporte [01:51:26]:
And, and so what I've said And I think I said it on the show last week, is you can hate late-stage capitalism. I'm not fond of what it's doing, but I don't think we need to conflate that with some of the amazing products of it, like AI. All right, we got to get you out of here in 7 minutes, young lady. Is that right?
Paris Martineau [01:51:45]:
It's true.
Leo Laporte [01:51:46]:
Okay, one more commercial, and then I do want to find out what you think about Mayor Momdani. Banning AI in public schools in New York City.
Mikah Sargent [01:51:58]:
Just another easy one.
Leo Laporte [01:51:59]:
Broadest AI ban.
Mikah Sargent [01:52:03]:
Yeah, that's like a quick 7-minute conversation, right?
Paris Martineau [01:52:07]:
Yeah, quick 7 minutes.
Leo Laporte [01:52:08]:
It's gonna be quick. So AI is banned not for high schoolers, but in elementary and middle schools. This kind of goes along with what a lot of people are saying. If your kids are allowed to use AI, they'll They won't learn. What do you think, Paris? This is your city.
Paris Martineau [01:52:22]:
I mean, I think it's a very interesting and, uh, novel policy. I mean, under this, students in New York City public schools from 2K to, uh, 8th grade will not be allowed to kind of use generative AI in like their day-to-day classwork and just in schools in general. They're also recommending as part of this like caps on screen time to go along with so that middle schoolers will have no more than 45 minutes a day of screen time. And I think that the thing that people don't realize now is that for many, you know, grade schoolers, you— so much of their day-to-day education experience is just sitting on a Chromebook.
Leo Laporte [01:53:05]:
Yeah.
Paris Martineau [01:53:05]:
In front of a teacher.
Leo Laporte [01:53:07]:
And I think that's bad.
Paris Martineau [01:53:08]:
Those programs on— but I mean, it's because these school boards and school districts purchase these These are big. I mean, they purchase— there's no money and then they purchase these software packages that then because they've spent money on it, the teachers have to deploy. And so they end up using these tools that are laden with a bunch of chatbots and different stuff, which I'm sure in specific use cases could be useful. But when it's kind of just generally applied as part of a package a school district bought, I don't think that's really serving anyone. I mean, what they're doing here is they're being like, everyone in the early grades, we're taking it out. All high schoolers are going to have twice a year, like, AI, like, courses. They're also piloting, like, a very limited, like, targeted AI pilot program in, like, 5 classes of younger students. I think this is a pretty common sense approach to attempting to purposefully design an AI use policy for education in one of the largest, if not the largest, public school district in the country.
Mikah Sargent [01:54:15]:
Something that we— that you touched on it there, Paris, when you talked about the software packages being purchased and the teachers having to use it. Something that I've seen in school district sort of policy centers around kind of giving the teachers and then just sort of the administrators right above them more power to do more than they would otherwise be able to. And so when you have a situation like this and you bring in these new regulations, what it does is it frees up policy that was once in place, it frees up money, it frees up research grants, it frees up a lot that you would not have. otherwise. And so even if— because I, as someone who doesn't have children, I try to sort of keep myself out of the idea of what I would want as a parent. But I do have a sort of psychological distrust with regulating parenting. But what I do know is that, as you've said, it kind of gives these teachers and the administrators the opportunity to say, let's look at how we are putting together our curriculum. Let's look at how we can potentially make improvements there that they wouldn't otherwise be able to do because they were trapped in those older plans or those older ways of doing things, especially right after the pandemic and everybody kind of working and doing school from home.
Mikah Sargent [01:56:06]:
So as I often say, I am of mixed minds about this for sure, but I'm hopeful that it gives the students whose brains are still in early stages of development the opportunity to form the neuronal connections and all that jazz that we as humans have done for hundreds of thousands of years as opposed to an immediate drop into the new way of doing things so that then when you do get to high school, that's whenever the next steps can start happening. Yeah, that's sort of where I am on it.
Leo Laporte [01:56:47]:
All I would say is What if he'd done the same thing with my generation or the generation after mine with personal computers?
Paris Martineau [01:56:57]:
Well, you did not have a system where every single class from kindergarten through high school, you were spending most of your instruction time on a personal computer.
Leo Laporte [01:57:13]:
Yeah. I mean, I agree. That's terrible. But I think you addressed that. But banning this is not— it's not responding to the problem. It's doing something else, something popular. But I worry about kids who are going to go home. So some kids, the advantaged kids, will go home and they'll be able to use AI at home.
Leo Laporte [01:57:29]:
They'll be able to use computers at home. The disadvantaged kids won't. What happens when they graduate from high school? Who's going to have the advantage then?
Paris Martineau [01:57:39]:
I mean, part of this program is that every high schooler is going to get specific training every single year of high school on how to use AI.
Leo Laporte [01:57:50]:
I know a lot of very bright 6th, 7th, and 8th graders who started using computers. They had Commodore 64s. They were able to use computers as kids who are now doing quite well in the world as technologists. I think it's— And were they being taught that in 6th grade by their 6th grade teacher? Yeah, if they hadn't had access to technology, they didn't have it in school because nobody did. Although, by the time you guys were in school, you had computers and you had Apple IIs and stuff, right?
Mikah Sargent [01:58:25]:
Yeah, we had computers in the computer lab.
Leo Laporte [01:58:27]:
Yeah. Yeah, you'd go to the computer lab and use the computers. This is going way too far. I think this is a response to a problem, which is teachers and the curricula and administrators are relying far too heavily on technology and abrogating their responsibility.
Paris Martineau [01:58:40]:
But I mean, when you're dealing with the largest school district in the nation that has such a complicated array of AI education contracts that it boggles the mind to even find them all and list them out, much less sort through them and figure out what is good, what is bad, you have to start from somewhere if you want to rework an entire school district's approach in a smart way to using these tools. And I think a pretty reasonable one has been like, all right, for like the youngest kids who've got a— who already are probably falling behind in a lot of other critical core, like traditional academic things, they should be spending their academic time learning math and reading. And by the time they get to high school and have the time to do more specialization, then we've been doing that. In the meantime, we're going to have limited pilots for all those, like 5% of a random 5% of students in, I think, elementary and middle school are going to have targeted AI pilots. They now have fresh data to figure out what would a more comprehensive data-backed AI learning program look like.
Leo Laporte [01:59:52]:
Yeah, I think it's great that the entire city of New York has decided to make their children lab monkeys. Good luck. We'll find out in 10 years how well they do.
Paris Martineau [02:00:02]:
We've already done that everywhere else.
Leo Laporte [02:00:04]:
Well, we'll see how they do.
Paris Martineau [02:00:05]:
Almost whenever I was covering this as a beat, a significant portion of school districts are just deploying whatever AI startup happened to pitch to her.
Leo Laporte [02:00:13]:
No, I agree. That's wrong. I mean, it was the same reason that it was wrong to put iPads or Chromebooks in every kid's hands.
Paris Martineau [02:00:20]:
Then what's the answer?
Leo Laporte [02:00:21]:
Well, but I— but so you're just swinging the pendulum in the exact opposite direction?
Paris Martineau [02:00:26]:
I mean, have you ever had to oversee an entire school district? You can't really run a building.
Mikah Sargent [02:00:30]:
I think that's the problem with all of us. None of us are education professionals or—
Leo Laporte [02:00:36]:
I've worked as a trustee for our local high school. We had a one laptop per child policy. I've overseen those kinds of technology programs. I haven't worked with K through 8. I had some kids who, by the way, had ample access to technology. Throughout their entire childhood. I never restricted it because I felt like it was important for them to understand it and use it. I don't think Henry would have made a success on TikTok had he not been comfortable with technology.
Paris Martineau [02:01:06]:
They're not saying every day in every class they're getting 45 minutes of tech time per class every single day. I feel like that's a really reasonable, like, amount of tech time.
Leo Laporte [02:01:18]:
So the moratorium doesn't extend to that?
Paris Martineau [02:01:21]:
No, it doesn't. It's just generative AI in school applications.
Leo Laporte [02:01:25]:
Okay, that's all right. That's fine. You're right. As long as they're getting some access to technology. I just worry about underprivileged kids who are not getting anything at home. But if they're getting— if you feel like that 45 minutes a day is sufficient, you know, I think about Bill Gates, who— the reason Bill Gates became Bill Gates is because His mom had a bake sale to provide a terminal connected to a timesharing computer in his private high school, and he spent most of his time in school on that terminal, and that's how he became Bill Gates. I just think that there's a Bill Gates in one of those 6th grade classes.
Paris Martineau [02:02:04]:
I mean, I agree, and I think part of the ways that we're going to support that potential next Bill Gates is Is teaching kids, having kids learn in a place where friction is something that they can experience, where their learning experience isn't just if you can't figure out your homework answer, you just ask the chatbot that's baked into the platform you do your homework on.
Leo Laporte [02:02:24]:
I think the best way to handle that is not to take away the chatbot, but to teach kids the value of not using the chatbot, to teach them how to use it.
Paris Martineau [02:02:32]:
But that's not going to work.
Leo Laporte [02:02:33]:
I mean, you need to have some This is the same argument people make about not giving their kids cell phones or not giving them access to the internet.
Paris Martineau [02:02:40]:
You've got to teach them how to live a healthy life. I'm just saying you should do that. That's part of the curriculum. But also, I think that just knowing how human beings are, if every single one of us had access to the ability to never have to do homework or never have to try on a test or never have to write an essay ever for the entirety of our lives, a significantly higher portion of people would never attempt to do those things because why attempt to do a hard thing with someone explaining to you the ultimate gratification of it when you could just click a button and then go hang out outside with your friends?
Leo Laporte [02:03:16]:
Well, I agree kids should be forced to walk a mile each day to school and back. I think kids should be able to— forced to slaughter their own animals and grow their own food.
Mikah Sargent [02:03:26]:
Well, this isn't how you actually That's sarcasm.
Leo Laporte [02:03:30]:
It is sarcasm.
Paris Martineau [02:03:31]:
Sarcasm on my podcast?
Leo Laporte [02:03:34]:
All right, we gotta get her outta here 'cause she's gotta see Megalopolis. Do you wanna do your pick or you wanna get out?
Paris Martineau [02:03:40]:
I'll do my pick really quick.
Leo Laporte [02:03:42]:
Do a quick pick.
Paris Martineau [02:03:44]:
Quick pick is, there's a lot of tech movies coming out right now. In October, there's 4 of them. Jeremy Strong, who's going to be as Megalopolis. Jeremy Strong is going to be as Mark Zuckerberg in Social Reckoning.
Leo Laporte [02:04:00]:
The successor to The Social Network by David Fincher. Yes.
Paris Martineau [02:04:03]:
Yes. You've got Elon Musk in Musk.
Leo Laporte [02:04:08]:
Yep. And there's going to be— did you see the Sam Altman Batmobile? That was wild.
Paris Martineau [02:04:14]:
No, but I need to see that. Andrew Garfield as Sam Altman in Artificial, which I think is very funny because it's debuting in the US on, I believe, Christmas. And it's just a movie about— it's a movie about, like, the Sam Altman getting plucked away from being CEO and reinstated over 4 days. And I'm just trying to explain, imagine explaining to parents over Christmas.
Leo Laporte [02:04:36]:
I can't wait to see it. Here he is driving up in his Batmobile. So I think you have some idea of where this movie's going when he gets out of his Batmobile.
Paris Martineau [02:04:46]:
Well, the one I'm most excited for is, uh, the documentary You Can See Everything, which dropped this week.
Leo Laporte [02:04:52]:
As is everybody I know.
Paris Martineau [02:04:53]:
It is. Well, everybody I know is obsessed with it. It's like, oh, Nathan Fielder, his documentary. I'm like, no, no, no. You don't realize it's directed by him and Lance Oppenheim, my favorite director, who's a real freak when it comes to documentaries. And this is going to be incredible just based on that alone.
Leo Laporte [02:05:11]:
It's just, look, just this trailer alone. I mean, you can see the color grading is so Oppenheim.
Mikah Sargent [02:05:17]:
I was gonna say, the color grading is what sold it for me, Paris. I saw the color grading before I realized who was even on screen, and I thought, oh, this seems— wait.
Paris Martineau [02:05:27]:
Have you ever watched any Lance Hoppins stuff? Some Kind of Heaven? Fantastic documentary about the villages in Florida.
Leo Laporte [02:05:36]:
I'm surprised I haven't seen that.
Paris Martineau [02:05:37]:
It is really interesting, as well as Ren Faire, an FX series about a kind of success— it's kind of like Succession meets Tiger King meets a Renaissance festival.
Mikah Sargent [02:05:50]:
Okay, so why have I not seen that?
Leo Laporte [02:05:52]:
Well, this is the one.
Paris Martineau [02:05:52]:
I don't know. But you need to. There's also one called Storm World.
Leo Laporte [02:05:54]:
You mentioned 4, but there's gonna be one more, which is the Dario Amodei-Sam Altman love story.
Matthew Cassinelli [02:06:00]:
I don't have anything to deceive you on.
Paris Martineau [02:06:01]:
Of course, I'm not deceiving you.
Mikah Sargent [02:06:03]:
That's not real.
Matthew Cassinelli [02:06:04]:
Engaging with you as a human being.
Paris Martineau [02:06:09]:
It's the same trailer.
Mikah Sargent [02:06:10]:
That's horrifying.
Paris Martineau [02:06:11]:
It's really upsetting.
Leo Laporte [02:06:14]:
There's no reason for me to do that.
Matthew Cassinelli [02:06:19]:
Okay.
Leo Laporte [02:06:20]:
I just love this. All right, good pick. Get out of here, the wonderful Paris Martin. No, no, you can't. If you want to stay, you can, but I'm just saying, well, I don't want to be the one.
Paris Martineau [02:06:33]:
Listen, I'll say for 5— you guys can do your picks of the week and then I'll—
Leo Laporte [02:06:37]:
I'll show you one quick pick of the week. This is what AI will look like. This is ChatGPT Jimmy. And—
Paris Martineau [02:06:46]:
This is what AI will look like.
Leo Laporte [02:06:48]:
I'll tell you why.
Mikah Sargent [02:06:49]:
A man's thought.
Leo Laporte [02:06:49]:
This is AI, a small model running on hardware with the model built in. This is an ASIC. And so Write me a poem about Paris. I'll say France so it's not too personal. And watch how fast.
Mikah Sargent [02:07:08]:
Holy—
Leo Laporte [02:07:09]:
This is what AI will look like in the future when it is built into the hardware. So, chat, you can do it. You can try it yourself. It's not a super big model. I know, I can't remember if it's Sequin. It's some smaller model, but it is entirely on In the chip microcode.
Mikah Sargent [02:07:28]:
Wow.
Leo Laporte [02:07:29]:
And for those who aren't watching, it was less than a second.
Mikah Sargent [02:07:32]:
That was incredible.
Leo Laporte [02:07:33]:
That was a great long poem.
Mikah Sargent [02:07:36]:
Okay, so that's what we need to have an AI guest on a show like that. That's gonna be fast enough already.
Leo Laporte [02:07:42]:
Yep. Mount Shasta. For me.
Mikah Sargent [02:07:46]:
Yes.
Leo Laporte [02:07:47]:
For me and my buddies.
Mikah Sargent [02:07:51]:
And tell us what we should pack.
Paris Martineau [02:07:52]:
Well, I will say this, uh, ChatGPT does not recognize Megalopolis as a film by Francis Ford Coppola.
Leo Laporte [02:07:59]:
No, it's a very small model. Watch, just show my screen one more time. I'm going to ask it to plan a trip up Mount Shasta. Tell us what to pack. There you go. It says check the weather before your trip.
Mikah Sargent [02:08:12]:
0.58 seconds.
Leo Laporte [02:08:15]:
Yeah. This is the future of AI when we have these models, and we will in our lifetime. Well, maybe yours anyway. We will have— by the way, it's a 3 to 4 day itinerary according to ChatGPT. It ain't no 1 day hike up and down. Uh, check the weather, prepare for weather conditions, bring bear spray. Mild. This is good.
Mikah Sargent [02:08:37]:
That's cool.
Leo Laporte [02:08:38]:
0.58 seconds.
Mikah Sargent [02:08:41]:
14,000 tokens per second. Per second? What?
Leo Laporte [02:08:48]:
So I'm just saying, We have just begun an amazing revolution, and, uh, this will be in your pocket, by the way.
Mikah Sargent [02:08:58]:
Pocket, y'all.
Leo Laporte [02:08:59]:
10 years, that's how long. Uh, Micah, do you want to make a pick, or—
Mikah Sargent [02:09:03]:
Uh, no, I don't have a pick.
Leo Laporte [02:09:04]:
Don't have a pick? We don't have to because Paris has a pick. It's called Megalopolis. Get out of here, Paris! Bring your 100-page PDFs with you and your red string.
Paris Martineau [02:09:15]:
I gotta read them on the subway.
Leo Laporte [02:09:16]:
We'll see you next week. Will you be here next week?
Paris Martineau [02:09:19]:
I won't. I'll be recovering from nasal surgery.
Leo Laporte [02:09:22]:
Oh, good luck with the surgery. That's right, she's getting her nose fixed.
Paris Martineau [02:09:26]:
Hopefully it won't, hopefully it won't, uh, change the resonance of my voice. They're building new science in there.
Leo Laporte [02:09:32]:
You could have said like Marisa Tomei from now on.
Mikah Sargent [02:09:35]:
Well, uh, maybe for a period of time.
Paris Martineau [02:09:37]:
I'm gonna ask them to give you— give me your voice is what I'll— I don't know if they have that technology, but I would like to.
Leo Laporte [02:09:43]:
Doc, will I ever play the violin again?
Mikah Sargent [02:09:46]:
If they're, if they're taking anything out, Can you make it into a necklace and, like, wear it?
Paris Martineau [02:09:51]:
I mean, I'm gonna be honest, that's— I have always thought this. If I was ever to get surgery, I want to ask for the things, but I'm not sure that they're taking anything out that is loose flesh.
Leo Laporte [02:10:01]:
He got a replacement hip. He took the femur, the top of the bone, and made a walking cane.
Mikah Sargent [02:10:06]:
That is so cool.
Matthew Cassinelli [02:10:08]:
So cool.
Paris Martineau [02:10:08]:
I'd rather my femur be inside me, but if it had to be outside of me, I'd want it to be a walking cane.
Leo Laporte [02:10:13]:
Oh my goodness. What is that? That's my nose.
Mikah Sargent [02:10:16]:
They shaved some bone off on the inside of me. That's going in some formaldehyde. Okay.
Paris Martineau [02:10:19]:
That's actually going on the list of things I've gotta ask while I'm in the operating booth. 'Cause I'm gonna be like, anything you can give me from inside my body, please give it to me.
Mikah Sargent [02:10:31]:
I promise not to upload it to AI. I promise.
Leo Laporte [02:10:34]:
Guys, I have only one thing to say. Thank you, Michael Sargent. We'll see you on Hands-On AI starting October 1st. Thank you, Paris Martineau. We'll see you at Megalopolis. Thanks to all of you for joining us. We'll see you next week at Intelligent Machines.
Mikah Sargent [02:10:49]:
Bye-bye.
Leo Laporte [02:10:49]:
Bye-bye.
Mikah Sargent [02:10:50]:
Bye-bye.
Leo Laporte [02:10:51]:
Bye-bye.
Mikah Sargent [02:10:52]:
I'm not a human being, not into this animal scene. I'm an intelligent machine.
