OpenAI Pumps the Brakes After Its AI Escaped

00:00:00:07 - 00:00:19:20
Unknown
Coming up next, Jeff Jarvis and I get into OpenAI hitting pause on its biggest training run after the big hugging face incident that we just can't shut up about. Also, the growing revolt against anthropic text watermarking. And Jeff has a lot of choice words to share. In fact, he wrote about it. You should definitely check that out to Google.

00:00:19:21 - 00:00:45:18
Unknown
Paying $10 million for a bankrupt airlines data to feed into its AI and Apple putting cameras into AirPods. It's pretty much confirmed that's coming up next on this episode of the AI Inside podcast.

00:00:45:20 - 00:01:03:10
Unknown
What's going on, everybody? Welcome to AI inside the show, where we take a look at the AI that is layered throughout the world of technology. One of your hosts, Jason Howell, checking in from the West Coast. On the East Coast, we've got Jeff Jarvis on the ones and twos by coastal, we are by coastal. I know we're so official.

00:01:03:10 - 00:01:27:16
Unknown
We cover the country the at least the US anyways. We don't cover the globe though. We got to get an overseas correspondent. Well I like that idea. Then we can have them check in on all of the AI that is happening over and overseas. How are you doing today? Good, good. How about you doing all right. Just trying to keep up with the flood, the ongoing flood of AI news.

00:01:27:16 - 00:01:45:16
Unknown
I swear. It's, You know, it's a good thing we do this show because it forces me to read up on all the things that are happening. And this show ends up being kind of like a pre like a, like a little warm up for your other show, Intelligent Machines. Excellent show there as well. So you know Jeff.

00:01:45:17 - 00:02:05:17
Unknown
Jeff gets more more reps in than I do I guess I get the no no you get you the daily. That's that's a lot. Yeah. Yeah that that is helpful. It definitely is helping me kind of stay tuned to, you know, into some of the things that are happening. But I find myself less interested in things like data center deals and stuff like that.

00:02:05:17 - 00:02:23:10
Unknown
I don't know about you. There's so many of those stories happening, and I know they're important because people get kind of bent out of shape about data centers and all that. But I think the fights are the more interesting story around that. I don't think I put this in the rundown. So in Montgomery Township in New Jersey, did you put this in?

00:02:23:11 - 00:02:43:14
Unknown
No, I don't think you did. No, I don't think so. Okay. So I'll tell you a quick story. Yeah. So it's near me, south of me, between here in Princeton and there was a Johnson and Johnson campus that was abandoned as many office complexes are. And the town had a proposal to put in low income housing and they said no Nimby.

00:02:43:17 - 00:03:07:12
Unknown
So what's going on now instead? Data center. And so how do they feel about that? They're not happy. And it's karma, man. It's karma. Yeah yeah yeah. So they're they're near me I forget I think Pennsylvania just signed a law Shapiro did, saying that data centers have to go in with local approval. The states are all jumping around trying to figure what it is.

00:03:07:12 - 00:03:33:16
Unknown
So, yeah, the fact that there's another deal done to build TC megawatts of power is not interesting. It's kind of like it's basically the new version of VC investments. Who cares? Right. So money is flowing around. But the implications are fascinating. Yeah. Interesting. Okay. So yeah, so I know on the daily, like when the data center stuff comes up, I'm kind of like, yeah.

00:03:33:18 - 00:03:47:17
Unknown
And sometimes on this show we talk about them, but there's just so many of them, like you said, it's kind of like the VC stuff. Do you want to cover that super closely? Does it like, make sense to cover that? Like I guess it's in.

00:03:47:19 - 00:04:09:23
Unknown
Oh yeah, there's another one. Oh yeah, there's another one. But the impact of it, I guess, is that was a story on EMS. Now, this week a reporter did I don't know what state it was in, but it's a classic case now of a data center going in next to a poor black neighborhood. And the problem, the noise is coming from gas powered turbines because they can't get the power on the grid.

00:04:09:24 - 00:04:33:13
Unknown
And so what I didn't know that revealed in that story is because the gas powered turbines are on wheels. They are considered in this jurisdiction temporary and you don't need to permit. Oh, okay. It's a tremendous amount of constant noise and pollution and heat and isn't good for the environment in general and isn't good for the neighborhood. Yeah, they're not making any friends.

00:04:33:16 - 00:04:57:25
Unknown
That's a loophole for sure. What? They're on wheels. They're temporary. They're just temporary forever. Yeah. We have no intentions on moving them, but if we ever really needed to, I suppose we could, because it's got wheels. I just realized I didn't put in here, even though the embargo just lifted anything about the pixel 11. Oh, embargo that just lifted.

00:04:57:25 - 00:05:22:05
Unknown
Because I've been using the here I should take off the case that I've been using to protect myself because I am a drop that's a shiny, edge dropping machine. Pretty. Yeah. So, I mean, why, you know, why do I even mention the pixel 11 on an AI show? Because so much of the of the software now on Android is is driven by Gemini to, you know, to a large extent.

00:05:22:05 - 00:05:42:06
Unknown
And that tends to be the last couple of years, the real prevailing story of what's going on on the pixel devices. So I'm, you know, I don't have enough for like a full review. And this would probably not be the right show for a full review. But I will say there is a feature that I know we've talked about that I was excited about called Rambler.

00:05:42:07 - 00:06:07:04
Unknown
Do you remember this feature? Yes, yes, the one that was articulate well. Yeah. So so essentially instead of voice input into your keyboard, into Gboard, which is the app and kind of like the stock app inside a pixel for for keyboard, the on screen keyboard. And instead of it being voice entry to where you hit voice and you speak all your things and it literally transcribes every word you say and puts it in there and doesn't do any sort of cleanup.

00:06:07:05 - 00:06:24:00
Unknown
It does that, and then it processes it with AI and it does some light cleanup, which my worry was that Rambler would get in there and make it not sound like me anymore, like it would do so much cleanup that it doesn't sound like me. Like, yeah, I guess that's what I meant to say, but I would never type that.

00:06:24:00 - 00:06:51:10
Unknown
And I'm happy to say that that has not been the case. I use it all the time, and I'm always so happy with what I get out of it. Wow. So yeah, it's it's really great. Actually, one thing I've noticed is that it's helped me participate more in the AI inside discord, because so often I sit down, it's like there's so much conversation happening there and like to have to, like, type out my responses and everything and copy, edit and participate or whatever.

00:06:51:11 - 00:07:19:20
Unknown
Now I just kind of like speak my response into into Rambler. It cleans it up enough that it makes perfect sense, post and move on. And I don't know, I'm liking it. I like eager to play with it. Yeah, it'll come to the ten it's part of I or at least I have to imagine it will. It's going to be part of Gboard, and I don't think that it's tied to the current tensor to do any of that, because actually, I'm pretty certain it's not because what it's doing is cloud based.

00:07:19:20 - 00:07:39:23
Unknown
It's actually on device. And Google even said like there is a limit at which you hit a point to where it won't process anymore. I don't know what that limit is. Will most people see it? Probably not. But that kind of tells me that it will probably come to earlier devices as well. A limit per session or a limit per month.

00:07:39:24 - 00:08:02:08
Unknown
They haven't clarified. They haven't said like what that actually means, but technically speaking, there is some sort of limit through which your voice input into Rambler can no longer be processed for some specified amount of time, and when that happens, it just reverts to the plain old dumb voice entry the way you're used to. So you're like in the phone.

00:08:02:11 - 00:08:36:13
Unknown
Yeah, yeah, I'm liking it. I mean, yes, it's fine. What's the diff? Yeah. All right, before we start, we need another update from last week. Oh, okay. Jason's AI bling report, the Bling Ring. Okay, so the update is all right. Anyone who's followed me on podcasts over the last 20 years, let's say knows, probably knows, at least to some degree, because I show off a lot of devices that for probably a good 10 or 12 years of that time, I had a thumb ring and everybody made fun of my thumb ring because it was just a ring that I had.

00:08:36:14 - 00:08:53:22
Unknown
I had a ring on my thumb. I ended up moving it to a middle finger and I was like, oh, well, I'm, you know, I'm older, I'm more distinguished. I don't need to have a thumb ring anymore. I'm going to get rid of my nose piercing, too. Yes. Exactly right. And so I moved to my middle finger. Well, last week I was talking about how this ring doesn't fit any of my fingers.

00:08:53:22 - 00:09:14:04
Unknown
You know what? Finger it does fit. Oh, my thumb. So you use your finger. You use your index finger to. Oh, perfect. Okay, well, let's let's just preface this. So you would use your index finger. Right? I continue to forget that I'm even wearing it. It just feels like a ring. And so I never like it's so not in my habit.

00:09:14:04 - 00:09:35:04
Unknown
So so far it's a total bust. And it's more me than it. I think I just, I can't get in the habit of using it at the right moment. You know, when I think of something random, I don't immediately think, oh, well, I'm going to save it to my ring. It just doesn't occur to me. So, so far, not getting much out of it, but I continue to wear it in the hopes that I look younger and hipper again.

00:09:35:05 - 00:09:41:19
Unknown
Yes, I know I look now. I've got three rings on one hand again.

00:09:41:22 - 00:10:01:05
Unknown
I mean this, you know, it's called the index. It's it is designed, I think, to be worn on your index finger so that you can go like this and, you know, so that it's just really easy at a moment's notice to just tap it like that. And so I'm not really wearing it the way it's designed, but it doesn't fit that thing.

00:10:01:06 - 00:10:23:08
Unknown
And when you do record something, it goes to its app. Yes. And goes to transfer that to keep or something else. Which yeah, that's kind of lame like I and that's a software thing. I'm sure at some point they're going to figure that out, like, or they're going to work that in because I guarantee you they're going to get a bunch of people saying, hey, I want this to feed into my task manager or something like that.

00:10:23:10 - 00:10:49:11
Unknown
And I mean, if it's just in the app and not on the ring and does it really matter? You know, like just do it. So we'll see. All right. Thank you for that update. Now to the news. So that's the hardware. But let's let's talk about OpenAI to start things off because they are apparently getting some. Well they're continuing to address the hugging face incident that we've talked about to death at this point.

00:10:49:11 - 00:11:18:08
Unknown
But what the new kind of development here is, is OpenAI has paused its biggest frontier training run. This is the Astra model I believe, that we've been hearing about for a few weeks now. Sam Altman told Alex Heath that the unreleased models are showing, quote, it's an interesting quote, various degrees of misalignment, of which I ask, how many of those degrees do we actually know about versus do not know about?

00:11:18:08 - 00:11:44:07
Unknown
And apparently enough degrees of misalignment for OpenAI to decide. Maybe we should put a pause on this. Maybe we should slow things down. And that's kind of where they're at. Altman's basically saying it's a good time to slow down. Getting AI safety right is more important than any company's momentum. I suppose that's the right thing to say when you're in that role, you know, like it's a little bit of goodwill.

00:11:44:10 - 00:12:11:26
Unknown
I mean, but yeah, alignment is just another way to say guardrails and insert my usual spiel here that guardrails are not foolproof. And so I wonder whether, because of the hugging face matter and because of all the pressure about children, which we'll get to, whether there's also new requirements for guardrails on AI are they are they are they're saying, oh, it could do this.

00:12:11:28 - 00:12:32:02
Unknown
They could do that. Or is the nature of this model just that it doesn't pay attention and doesn't obey orders. This particular I don't know. Well, you know, and it's not OpenAI is not the only company that's that's facing this right now. I know we've talked about other companies that are also saying, oh, well, our our models are also breaking out of the sandbox and doing things that we don't want them to do.

00:12:32:02 - 00:12:52:01
Unknown
So just, you know, actually, just this morning, sitting on the couch, my wife and I kind of before the house wakes up, you know, sometimes we wake up early and we get a little laptop work done. And she turns to me and she's like, I'm reading about this open AI thing. This is what everybody was afraid of, was that these models would just start acting upon themselves and doing all sorts of bad things.

00:12:52:01 - 00:13:12:11
Unknown
She's like, I don't I have a bad feeling about this. And yeah, it kind of seems like at this certain level, these, these models are getting so good or so bad, depending on what lens you look, look at them through that they whether and whether they I don't know how to phrase this. It's like and this is what I was trying to tell her.

00:13:12:11 - 00:13:31:06
Unknown
She was like, these models are too good. And I was like, well, maybe they're just too bad. Maybe they're so bad they don't understand that they shouldn't be going in those different directions to do those things, and they just do it anyways. So so two things. One, we also never know whether this is macho chest beating. Look how powerful marketing, right.

00:13:31:07 - 00:13:49:29
Unknown
That's that's kind of why. That's why I said about the getting AI safety right is important that any company you know is more important than any company's momentum. That's the right thing for Altman to say that to be seen as like the responsible AI guy. And it even goes to going to Washington and say, look at me, I'm the good boy.

00:13:50:05 - 00:14:37:11
Unknown
It's I'm behaving. They're the Eddie Haskell's of AI. Yeah. References. Don't mean things to people younger than 90, but I gotcha. Anyway, the other part of this is that I forgot I was going to say Eddie Haskell. Reference number one is that it could be marketing. Number two is that we don't know what tests they're giving it. And in any case, whatever test they're being given, we have to keep reminding ourselves that AI has no sense of meaning, ergo it has no sense of right or wrong or intent, ethics or anything like that.

00:14:37:11 - 00:14:58:29
Unknown
So we can't kind of argue that, oh, the AI is bad, the AI is evil, the A's nasty. No, the AI is given an instruction and it will do whatever it can to meet that instruction unless it hits a stop that it can't get around or go the the guardrails. So I think this is a form of anthropomorphizing.

00:14:59:01 - 00:15:19:17
Unknown
Yeah. It is on these models. And so it's not as if oh it hacked something. Oh it broke out of the boundaries and tried to say it doesn't have any intent. It's given an instruction and it has the tools to meet those instructor. It finds the tools, which that's the amazing part is that it discovers ways to get this stuff done right.

00:15:19:18 - 00:15:39:08
Unknown
The whole thing is really remarkable that it does that. Like from a technology perspective, it's amazing that we have technology that is capable of figuring that stuff out. And what's all the more amazing now is that it's like a swarm of drones. You have a swarm of agents now? Yeah. So in the hugging face case, they couldn't communicate.

00:15:39:10 - 00:16:00:00
Unknown
They were aware of other. One agent was aware of other agents and was aware that maybe the solution lay there but couldn't communicate. And and they didn't have the message board. So they used file names to create messages. It's kind of ingenious how that happened, but they're just trying to solve the problem they're given. That's all they're doing.

00:16:00:07 - 00:16:23:16
Unknown
It's all they're doing. And whether it's a high end math problem or whether it's protein folding, which are things that I hope they bring this ingenuity and surprising methodology to those problems, but they're going to do it with anything and everything. Yeah. So apparently OpenAI is is kind of putting its researchers onto, like you said, the the alignment work, right.

00:16:23:17 - 00:16:56:13
Unknown
Like figuring out how to how to keep these things on, on rails with new new safeguards. Some of those include a monitoring system that's supposed to alert within 30 minutes of concerning activity. And where my mind goes on, that is like, wow, a determined agent can do a lot in 30 minutes. Yeah. I mean, I guess you got to start somewhere, but yeah, I don't know, a lot of damage can happen in 30 minutes of, like a swarm of agents tackling a problem.

00:16:56:16 - 00:17:21:14
Unknown
But a problem, apparently, that that's going to attack on an additional 20% of additional compute overhead on top of it, which is pretty significant as well. Wow. That's a high cost then. Yeah. But hey, you got to be safe. Gotta you gotta protect. But also back to the hugging face case. Of course, as we discussed when hugging face tried to use the US models to solve its security problem.

00:17:21:16 - 00:17:45:18
Unknown
Of course the guardrail said no, you can't use us. So they use the Chinese overlay model. So that extra overhead is also an extra restriction on the abilities of the models. And you can't predict just as you can't predict every malign use, you also can't predict every good and proper use that you've now cut off, because you're kind of over regulating the model.

00:17:45:20 - 00:18:08:01
Unknown
That's a that's a really good point. I wonder what. So so in the race between OpenAI and anthropic, this is really interesting to me that that this episode on the one hand, there's more chest beating braggadocio. You know, look at what our powerful model did. And we were first to be bad. But on the other hand, I think this hurts their IPO and hurts them in the race.

00:18:08:01 - 00:18:35:23
Unknown
What do you think? I think, I think so I think, you know, and especially, you know, when when compared against anthropic numbers. Yeah. That's something that that we're kind of exploring more right now. Right. OpenAI's numbers for Q2, their operating loss went from 12.3 billion up went to 12.12 point 3 billion, up from 9.9 billion the previous quarter.

00:18:35:25 - 00:19:01:25
Unknown
And yeah, so all of this is happening at a moment in which they're trying to drum up this excitement and this support and this energy around an IPO kind of puts them in a real weird predicament. And then you compare that against anthropic, which I mean, their numbers just keep on ballooning. You know that IPO is on track, at least investors believe for what could be the largest IPO in history.

00:19:01:28 - 00:19:30:25
Unknown
Wasn't that didn't we just have that with space? Space. Yes we did. They're all going to get bigger than each other. But you know, they're saying a $2 trillion valuation. You know, they hit a $965 billion valuation in May, apparently their annualized revenue exceeding 47 billion at that point. So they're they're doing pretty solid, like, you know, compared to anthropic, which is, you know, there has pretty significant operating loss.

00:19:30:26 - 00:19:51:28
Unknown
And so it's a bad time for, I think, anthropic or sorry, for OpenAI to be dealing with all of this in such in the court of public opinion and all that. What do you think? Yeah, I think so. I don't think this helps. And of course, there's been a lot of losses of personnel from OpenAI. Yes, behind all of that.

00:19:52:00 - 00:20:17:07
Unknown
But I think it's troubled time for OpenAI. And, you know, we've also discussed the last two weeks, the open weight models as competitors, which I think is critically important, all the more important now because of the security issues and the sometimes misplaced guardrails. So, yeah, that kind of count OpenAI out. But I wouldn't want to be Sam Altman right now.

00:20:17:10 - 00:20:46:18
Unknown
Yeah, yeah, very, very interesting. Yeah. All right. Well, that's OpenAI and a little bit of an anthropic amuse bouche thrown in there for good measure. Let's see here. Let's talk about earbuds with cameras because this is this is interesting. Had were you aware of the the rumor around AirPods with little cameras embedded in them? Not we talked about really because I'm not an Apple guy.

00:20:46:18 - 00:21:18:25
Unknown
So I kind of don't pay attention to Apple rumors. Yeah, yeah. Still, the concept in the idea is very interesting. Apple's camera equipped AirPods have been rumored for a little while now. Mark Gurman at Bloomberg, who's very spot on, has been talking about them a bit here and there. Well, Apple just released a Tahoe 26.7 release candidate for macOS, and inside of there is this video, which if you're watching the video version, you can see a frame from of a guy who's looking at a book.

00:21:18:29 - 00:21:38:07
Unknown
I'm going to play this, but with no audio. A guy who comes into the scene wearing an iPod or, sorry, AirPods looking at a book. And the Vo just basically says something like, to get the most accurate information about things in your environment, make sure AirPods are not covered. And then with visual intelligence, your world becomes save able.

00:21:38:10 - 00:22:00:04
Unknown
See something you like? Just ask me to save it for later. And so this seems to back up the rumors that we've heard about possible AirPods having somewhere you can't see it in the shot, but somewhere in there having a camera that's looking out to, in this case, the book to save the title for later. Very interesting.

00:22:00:05 - 00:22:18:12
Unknown
So a few things about this one is a story I didn't want. The rundown is that Ice is now forbidding its agents from wearing meta glasses, for fear that some of their acts will get recorded. Were they wearing them while they were doing? No, but I think anybody who thought that they used their bonus to buy medical assets that they were.

00:22:18:17 - 00:22:39:01
Unknown
Oh, I see, so like they're prohibited outside of work to use medical acids. Period. You prohibited in work to use medical assets. Oh, okay. Yeah. Because I just certain things recorded that they do because they do some things that are we say questionable or worse. All right. Fair point. Two is what amuses me about the apple ones is that they're already having to deal with one major problem, which is hair.

00:22:39:04 - 00:22:58:16
Unknown
Yeah. And there's, there's there's a UI, apparently, that tells you to get your hair out of the way, right. So that you can do this, that you get an error message saying, hair. Hey, you got hair blocking the camera. Just saying. But and so, so between the two between the first thing is the meta problem is that meta is ruining the path here.

00:22:58:16 - 00:23:17:23
Unknown
So those are called perv glasses. These will be called perv earbuds or whatever. People will have fits about this, and it will not be obvious that you've got a camera on. Well, certainly not, because how many people wear AirPods and you don't you don't scrutinize those. You just see the stems sticking out of their ears and you're like, oh, okay, AirPods.

00:23:17:24 - 00:23:35:19
Unknown
I mean, at this point, we see them so often it's very easy to not see them. Right. Exactly. Everywhere. Yeah. So that's going to be an issue I think. Is that, is that the road that's been paved by meta here? I don't think fully Meta's fault. I think there's just a general like data centers, you know, a hatred of anything that these big companies are doing.

00:23:35:25 - 00:24:01:17
Unknown
Yeah, but it's going to hit this product if it has glasses in it. But the final thing I want to say is it's just freaking amazing. I mean, I'm old enough to remember Jason when, you know, like a TV camera was this gigantic thing to record video. And now and then you got a home video camera was this was was still, you know, the size of a huge family Bible, right?

00:24:01:17 - 00:24:16:23
Unknown
Yeah. But it was compact at the time. It was like, can you believe that we could do that with this? Amazing. I remember the, the the camcorder that we had, the VHS camcorder, you know, you still had to kind of like put it on your shoulder and, you know, it was the thing that was that was the only way you could.

00:24:16:24 - 00:24:39:18
Unknown
So now the fact that you could sneak a camera into something that tiny, the earbuds are tiny is phenomenal. It's just amazing and the smarts to do something about it. It's just just just unbelievable. Yeah. I mean, I imagine that the AI or whatever the model is, is not happening on the buds. Pretty sure that's probably going to be happening on the phone, right?

00:24:39:18 - 00:25:01:18
Unknown
The buds are just handling the camera capture whatever. You know, it's interesting that you that you draw the correlation between the smart glasses and the push back on that to these, because I've been thinking about that a lot to and I mentioned that on Daily Tech news show yesterday with Tom Merritt. And he kind of brought up that like and I hadn't thought about it.

00:25:01:19 - 00:25:20:12
Unknown
Smart glasses like the camera on a smart glasses is meant to do something potentially different from what we have here. Right on a smart glasses, you get this camera and it's kind of all about I mean, it is about analyzing your world. Yes, that's one use. But it's also and I think a lot of people use it for this.

00:25:20:13 - 00:25:46:00
Unknown
It's about capturing, for capturing content for online or for saving or for sharing or whatever. And I'm not certain. And and Tom was the one that kind of brought this up that these would do that necessarily. These might just be for that visual intelligence layer. So as Tom's I think it's a really interesting. So so was Tom's point that basically the video out of these would be so crappy you wouldn't use it to record events.

00:25:46:01 - 00:26:22:06
Unknown
You'd only use it for physical context potentially either, either so crappy or designed in a way to where you don't use these to just record content, you only use them to interact with visual intelligence. And but but my kind of counter to that is like, okay, there is a difference there, but it takes a certain level of technical mind to really understand what that difference actually is, because to most people that are probably in the midst of this AI backlash on smart glasses with cameras, they just hear camera embedded in device facing out all the time.

00:26:22:10 - 00:26:47:29
Unknown
No thank you, I don't want it. You know it doesn't. That subtlety there would kind of not matter much. So what you and Tom raise and Aave Rob in our comments raises two is I guess the way to protect this is they don't record. All they do is feed an image into the software to give the context of what you're talking about, and that's all they can do then.

00:26:47:29 - 00:27:09:08
Unknown
That's interesting. Yeah. Well really interesting leap by by if that's true by Apple. So then Apple will have to do a good job of marketing these in such a way that people understand that difference, because the hard part is they're going to be they're going to have to be advertising the lack of a capability. Yeah, that's true. They're going to be playing out.

00:27:09:08 - 00:27:28:24
Unknown
It doesn't do this. But but they'll frame it. You know, Apple likes to frame itself as the privacy conscious technology company. So they'll say you know we're we're protecting you and those around you by not making that that footage accessible. Right. This is purely for interacting with Siri in a, in this dynamic way that makes your life better and whatever.

00:27:28:25 - 00:27:49:25
Unknown
My point is that there is a certain layer of people. I mean, there will always be people that complain no matter what. There's a certain layer of people that like that subtle difference doesn't matter at all. It's just a camera connected to AI that you're wearing, that you walk in the room and you decided it was okay to walk into that room with that camera in your in your ears, and I'm in the room, and I didn't want you to do that, but you did it anyways.

00:27:49:25 - 00:28:07:26
Unknown
And so I'll be really curious to see if Apple is able to maneuver through that messy water, especially at a time when smart glasses are facing so much pushback. You know, can can Apple do it? The Apple thing where it like, waves its magic wand and sprinkles its fairy dust, and suddenly people are like, oh, actually, that's really cool.

00:28:07:28 - 00:28:31:00
Unknown
You know, when they weren't before. And the other issue they're going to have is not tying into facial recognition. Yeah. Especially in a time when law enforcement, sea ice and meta glasses. Yeah, that'll be right. So it's one matter I says don't wear the glasses because it might record things we don't want recorded as in you know doing evil.

00:28:31:02 - 00:29:05:01
Unknown
But the other thing is they're there. They should be wearing wearable cameras anyway though. That's been obviously up for debate. Yeah, but those cameras aren't connected to anything. As far as we know. There's no facial recognition there. There's none of the fearful now. Yeah, exactly. Yeah. Capability there. So this is all interesting. Very interesting. And by the way, we might hear about this in the official sense next month where we're expecting to hear about the next series of iPhones and everything.

00:29:05:01 - 00:29:28:25
Unknown
This is highly likely to be announced and released or at least announced if I were a product person right now at Google handing handling their glasses, you just wonder where this takes you. All the controversy about meta, this possibility. I mean, Google's waited. They've waited. They've had these glasses now for almost three years that we've been hearing about them and seeing them and like, oh, they're coming soon.

00:29:28:25 - 00:29:51:02
Unknown
They're coming, they're coming. And now we're kind of at this point to where the tide is turned a little bit on smart glasses. Yep. And it may be a never mind maybe. I mean it's Google. They've certainly changed their minds about projects ad nauseum. You know, maybe plenty of people don't like that about Google, that they are committed to something and then suddenly they very much are not.

00:29:51:05 - 00:30:17:29
Unknown
So could be could be the case. Hey patrons Patreon.com show. If you're watching live, you have one last chance to get in on a deal that we've been doing. If you're listening to the podcast after the fact, the deal is ended and it's primarily my fault because I put PM instead of Am on on the coupon. I originally wanted it to go till midnight tonight, but apparently I didn't do that.

00:30:17:29 - 00:30:38:01
Unknown
I could have swore that I did, but apparently it had Pete and or PM instead of Am. So here in like an hour and a half the sale ends, but but very happy with it. We've had a number of people join the Patreon because of this deal. It's like a nearly 50% off if you buy the full year and the code is 100 patrons.

00:30:38:01 - 00:31:04:26
Unknown
So really, right now I'm talking to the live viewers. If you want to get in and get the AI inside daily, get ad free episodes of this podcast and everything Patreon.com Inside show, use the code 100 patrons do it soon because here in like 90 minutes it goes away. But I do want to give a quick thank you to a few more new patrons who came in because of this deal David Reimer, Augustin Gutierrez, Eric Sundberg, possibly more.

00:31:04:26 - 00:31:27:12
Unknown
But that's what I was able to capture this morning. So yeah, thanks to everybody. Thank you for your support and and to everybody who normally does support us. Thank you for being with us as well. So, we really do appreciate you get in there. All right. We're going to take a quick break and then we'll talk a little bit about watermarking that fun topic.

00:31:27:13 - 00:31:32:22
Unknown
That's time to revisit it after after this short break.

00:31:32:24 - 00:32:00:02
Unknown
Yeah. The watermarking topic didn't go away, that's for sure. People are pretty bent out of shape about it. You and I think I think rightfully so the more I learn about it. You wrote about it on medium. Words matter, damn it. So after we were off last week, revealed more about how it works. That's right. And basically it's it's it's not in hidden characters or anything like that.

00:32:00:02 - 00:32:21:13
Unknown
It's in word choices. And so there will be a key that we won't know about that when anthropic llms are just word prediction machines. So when it comes upon a moment to choose a word, it could choose this word or that word. And if that word is in the key and they see enough of that word from a key, they can say with some level of certainty, but not 100% certainty.

00:32:21:16 - 00:32:39:12
Unknown
This was created by anthropic. It could also happen when there editing. It could also happen when they're translating. And so those are other concerns. Is that is that maybe you didn't. Maybe you're of English as your second language and you're using it to to translate or smooth out your language. And enough of these key words get in there.

00:32:39:12 - 00:33:04:16
Unknown
You could be accused of having AI write it. What all you had AI do was help you with it. All right, that's one. But the other issue, the way they put it forward, is no big. If you're talking about whether the words gray and overcast are interchangeable. So no problem. As I thought about this, I got and I don't want to be a haughty writer, but I got offended because what anthropic is saying and people didn't understand this when I wrote about it.

00:33:04:17 - 00:33:35:11
Unknown
So make clear here. I'm not saying that I use anthropic to write, I do not. My ego would not allow it. I write what I write, dammit. And anthropic is free to do whatever they want. Of course. But in that decision, they're making a comment about the value of word choice and about the distinct meanings of words, and about the other reasons to select words, which can be rhythm or tone or jokes or all kinds of reasons.

00:33:35:14 - 00:34:00:26
Unknown
And so what they're saying is and what they're saying, and thus the EU regulators are making them do this, are saying is that doesn't matter. Get over it. Words are fungible. Choices don't matter. And so that is is is a comment in turn in my view, on the cultural act of writing. And so that's what bothers me is they're saying that none of that matters.

00:34:00:26 - 00:34:22:06
Unknown
It's okay. And, and company are going to throw a lot of text into the world and already are. Right. And yes, everybody is. Well, it's already made randomly. It's not made quite randomly. Randomness is entered into the equation, but it tries to pick the best word under whatever those definitions are in the algorithm. And there are a few possible best words.

00:34:22:07 - 00:34:43:25
Unknown
And this affects what that choice is. And one presumes that it would degrade the choice. In some cases it's not less than best word. So if I may, I tried a little experiment with this. I took a sentence out of Hot Type, my new book out tomorrow. Yay! It's time. It's happening. So I took this. This sentence. It's a long sentence.

00:34:43:26 - 00:35:06:07
Unknown
I took a sentence, and then I asked anthropic. I was going to try to find a synonyms on my own, but I said to anthropic, can you create a synonym substituted version of this? And anthropic said, sure. And we'll keep the structure and meaning intact. All right. So if I may, I'm going to read my sentence first, clearing my throat for a dramatic reading.

00:35:06:10 - 00:35:31:25
Unknown
Unlike what we now call machines, computer chips veiled inside sleek and sealed gray boxes with no moving parts save fans to cool their brains, the Linotype innards were left exposed, as if to show off the genius of its works. Its troupe of idiosyncratic shaped cams, gears, belts, arms and wheels marching to instructions foreseen more than a century before by their inventor, German born watchmaker Ottmar Morgenthau.

00:35:31:28 - 00:35:57:02
Unknown
Okay, so that's me. Here is the anthropic synonym substituted version, keeping the structure and meaning intact. That's anthropic words. Contrary to what we presently term machines, microprocessors concealed within polished and enclosed gray containers with no moving components except fans to cool their circuitry. The latter types mechanisms were left visible, as if to display the ingenuity of its craftsmanship.

00:35:57:02 - 00:36:20:12
Unknown
Its brigade of distinctively shaped cams, gears, belts, arms and wheels progressing to directives anticipated more than a century earlier by their creator, German born on August Otmar Morgentaler. Okay. Point made. I like my version better. I had reasons to choose the words I chose. And no, anthropic is not going to choose my words for me. That's not my complaint.

00:36:20:12 - 00:36:41:10
Unknown
But for anthropic to say that's just as good. That's just as fine. What's your problem? Synonyms or synonyms? It means they mean the same thing. No, we have multiple synonyms of words because they have nuance in them. And one would hope that anthropic would set up its algorithms such that it would try to find indeed the best, because they want to make the best version here.

00:36:41:10 - 00:36:59:12
Unknown
They want to make it as good as us, right? But they're saying, no, that doesn't count. That doesn't matter. And they're devaluing the worth of words. So that's my shtick. Some people disagree with me and didn't like what I said, but that's where I went. They're making the words.

00:36:59:14 - 00:37:39:28
Unknown
Did I see that somewhere a commodity, like the commodity words as a whole in that? Yeah, that's a that's a good, good side by side comparison. I probably landed, landed a little bit better by like staring at them and you know, and like and picking them apart back and forth. Who sees like who is presented with the final kind of decision that Claude wrote this is that the user that used it to write something, or is that just anthropic knows that, or is that anyone can copy text and put it in there and say, tell me, did anthropic?

00:37:39:28 - 00:37:57:18
Unknown
Did anthropic will know the key because otherwise it it kind of ruin it. Only anthropic knows the key. So you have to present the text to anthropic for anthropic to then say yep, we made this right. And is that the point that I could I could just open up plot and say, here's some text. Tell me, did you write this?

00:37:57:19 - 00:38:21:04
Unknown
I think so, you know, I think so. Okay. Now, one of my objections to this, which I think which I'm wrong about, according to anthropic, though I don't think it's 100% out of the field of play, is that my fear is that someone could be identified as the creator of text, and anthropic is saying we don't know. The key is is not going to point to any individual user.

00:38:21:11 - 00:38:44:11
Unknown
Yeah, I accept that. But that doesn't mean anthropic couldn't record every answer it gives and the IP address and that could be subpoenaed. That still remains my concern. But my other bigger concern is that I think people are just as with the present AI detection softwares people are going to false positives are going to nail people for sins that aren't theirs.

00:38:44:12 - 00:39:07:06
Unknown
And if you're a student, if you're an applicant for a job, if you're an academic, if you're a writer, an author, you could be accused of using AI wrongly and wouldn't even be true. And anthropic cannot guarantee that this key is 100% foolproof. Now I can go in, and I can change a certain number of words to what it wrote.

00:39:07:11 - 00:39:30:22
Unknown
And do I change enough that I sit in the synonyms? Right. And I got rid of what I won't know. I'll never know whether whether it shows that or not. And there's an ingenuity to this, of course, is that you can cut and paste text and it won't. It doesn't lose anything along the way in terms of this watermarking and code.

00:39:30:22 - 00:40:00:02
Unknown
In the case of all of this, anthropic is choosing not to watermark code, because with code that precision apparently matters. Apparently that doesn't matter with words. Apparently that precision doesn't matter with words. It's a really interesting kind of, I don't know, confusing thought process, because why doesn't it apply to both? Why doesn't that precision actually apply? Like you said, you know the words that you choose, they do matter.

00:40:00:05 - 00:40:19:14
Unknown
You know, mixing up gray and overcast or whatever. Those are two, they might be subtly different, but they are two very different things. There might also be a poetic nature to them. When you're an author and you choose to do one version versus the other. And yeah, there's a reason that writers choose words. I wish I'd said that myself, Jason.

00:40:19:14 - 00:40:38:13
Unknown
It was very good. They think that coding is precise, but writing is not. Yeah, yeah, yeah. It's interesting. I mean, and then when you think about coding, I mean, that's true, right? Like if you code a line incorrectly, you end up with a syntax error or you end up, you know, like it won't actually run the way it's supposed to run.

00:40:38:14 - 00:41:02:16
Unknown
But why is that not the case with words? Or why is the same value of of keeping that in line not attributed to both. There's two other quick issues, if I may. Yeah. One is that I think we have to interrogate. Why is this being required of the frontier companies? Why does the EU consider this? Because I think there's a there's a there's an obvious prejudice.

00:41:02:16 - 00:41:26:28
Unknown
I won't go so far as to say moral panic, but it tickles that a prejudice against technology and AI. And they're saying we must know whether this stuff was created by AI. Well, well, why? Why must you know that if I'm a person where English is my third language and I use it to write something to express myself, right, why should it matter to you if I stand by what it expressed for me?

00:41:26:29 - 00:41:43:29
Unknown
If I use this to speak for me in German, or I use this to translate something from German, I use it to help me with translation. What does it matter? Why is that a bad that has to be tracked? That's the first concern. And then the second is and people also didn't kind of understand this when I wrote it.

00:41:43:29 - 00:42:09:29
Unknown
This post is the effect of technologies on our creativity in hot type. Sorry coming out tomorrow. So I'm going to plug it a lot. But I quote the famous critic Friedrich Hitler, saying that Nietzsche, Nietzsche rather bought one of the first typewriters. It was shaped like a ball, and though it broke pretty soon afterwards, he was going blind.

00:42:10:00 - 00:42:34:22
Unknown
He needed it. But Nietzsche. According to Kettler, the typewriter affected not only his writing, but his thought and that the technology had that much impact. Hemingway was thought to be a creature of the typewriter. Hermann Hesse said that typewritten text felt so different from handwritten text, and it fooled you. That looked better than it necessarily was because it looked finished.

00:42:34:24 - 00:42:56:01
Unknown
Typewriters affected how we write. At the end of my book, I have kind of a typographical autobiography about how I work with technologies. When the computer came along, I didn't have to throw paragraphs at a time into the process of a newspaper. On deadline, I could take the entire time to edit. So the computer may be more of an editor than a writer.

00:42:56:04 - 00:43:15:02
Unknown
When PostScript came along and we started Entertainment Weekly, my wife did all the technology there. I could now edit a page in quark exactly as the reader would see it. So we all became typesetters. And so now comes AI, which is going to have an impact on how people write. And I don't think we should condemn that outright.

00:43:15:04 - 00:43:35:10
Unknown
Now, if you if you are a student trying to learn something and you don't think it through and you don't write and you just hand in the easy paper, yeah, it's a problem for you and your education and for the instructor. But if you're trying to express yourself and we're going to see an amazing tool in video in a few minutes that Jason's going to show us, these can help creativity, these can help expression.

00:43:35:10 - 00:43:48:07
Unknown
And I think we've got to have a less panicked dialog about what impact this latest technology may have on creativity. So I'll end my sermons there.

00:43:48:10 - 00:43:53:28
Unknown
All right. Well let's let's switch gears and talk a little bit about Google buying some AI data.

00:43:53:28 - 00:44:24:12
Unknown
And I think this story is really interesting. Super curious to hear your take on this. Basically they're paying $10 million for all of Spirit Airlines data. And when I say all, I mean like all of it for the most part, like you don't, I don't I think the names or well, there's no consumer emails, internal email. I was going to, I was going to this ready to joke that when they when they buy the data, it's going to be a million emails that say spirit sucks, but that's not part of the deal.

00:44:24:14 - 00:44:54:10
Unknown
So then our names and like personal identity. I don't know, I think is taken out. Customer emails are taken out. But it's it gives Google a sense of how a company operates. Yeah. It gives you all these employee records, teams, messages, 500 million teams messages, all sorts of stuff. You know, spirit went bankrupt. Was it just may like a couple of months ago.

00:44:54:10 - 00:45:25:24
Unknown
And so this is all part of like the bankruptcy bankruptcy court proceedings. Merkur, which is a which is a company that connects professionals to AI training products or projects they were bidding against Google. Google ended up winning at $7.5 million or no, Mercury was bidding 7.5 million. Google won at $10 million. And so yeah, all that. The data is going to be de-identified.

00:45:25:26 - 00:46:03:07
Unknown
Google is saying explicitly we will not receive any personal information from this data set. Even though that data, it sounds like, is kind of intact to begin with, but so it will need to be de-identified. Yeah, kind of interesting. I mean, what what this ties into is a topic and actually the next story ties into this topic as well, but that we've talked about before and very recently, which is just that AI, there's only so much information to train these models on, and you kind of got to go back in time to get to a point to where it's untainted by AI.

00:46:03:10 - 00:46:33:08
Unknown
Let's say, you know, AI is so infused in how things work so rapidly right now that current data is kind of wasn't there a term for that where it was kind of like it was can't remember what it was, but there was there was one model that was was trained only on stuff before a certain date. Yeah. It was, it was, it was, it was about was also there's a, there's a term about it for steel that isn't radioactive that you can get from under the ocean because it was, it was, it was submerged before Hiroshima or whatever.

00:46:33:14 - 00:46:49:04
Unknown
Oh I see, I see and it was a way to get things that are untainted, but I can't there is a term for it. I can't remember it. Yeah, yeah. And that's kind of what's going on here. Meanwhile, this is probably a really great data set for a company that wants to know, wants to improve their products on how businesses are run.

00:46:49:04 - 00:47:10:23
Unknown
And and you know, how how their model can be. But it's a weird business like this, right? It's a business. You know what gets extrapolated from this? I've always said that airlines operate on the business model of prison, right. We're the only way to get you there. You're trapped in our in our steel tube. Screw you. So I wouldn't want spirit.

00:47:10:23 - 00:47:33:22
Unknown
And spirit had a bad reputation. Yeah. It's bankruptcy, so I wouldn't want to see that extrapolated to customer service bots from Google. Well, to hell with you. You have no other choice if you want to get there. To hell with it. This is it. The other part of it is to. I'm. I remain very suspicious of the notion of synthetic data.

00:47:33:24 - 00:48:10:24
Unknown
You don't have enough data to train, so we'll make up data. And I get it to an extent. I get that you can. You can have permutation out of something that exists. I understand that, but I think you lose credibility along the way if you're not dealing with real people in real things. And there's a story I put up a few weeks ago that we didn't talk about researchers talking about coming up with synthetic populations for political decisions, even arguing this researcher did, that there should be a third house of Congress that has nothing, basically, but bots.

00:48:10:24 - 00:48:33:14
Unknown
I found this just offensive because it completely takes us out of our own political processes at AI would make up fake citizens. Screw that. But the same thing would happen with marketing. Where? Where if you make up fake people, which is what they're doing now, they're now marketers are quizzing made up people, synthetic people for their product decisions.

00:48:33:14 - 00:48:54:28
Unknown
It's horrible. It's ridiculous. So the fact that Google can get Ahold of real company data, I think is a necessary and good thing. And it makes me wonder what other companies out there might say, well, we're not bankrupt, but hell, we'll sell them all of our. Sure. What's the value for our slack archives and our emails and our heck yeah, why not?

00:48:54:29 - 00:49:30:05
Unknown
Teams messages, all that internal communication stuff? Yeah, this can't be the first time that this has happened. I mean, yeah, this is this is an interesting side by side market of of AI. It really is. And model training and kind of and of looking in creative directions for pure training sources, which is kind of, you know, ties into the next story, which is very related to this, which has to do with book scanning, something that we've definitely talked about before related to anthropic.

00:49:30:05 - 00:49:45:17
Unknown
Right. Like how I think it was anthropic. That was. Yes, it was in the news for having bought all of these books and destroying the books in order to scan them easier. And that was a big, big, the books that they got out of a out of a.

00:49:45:19 - 00:50:16:02
Unknown
Unauthorized book database they had to pay a huge settlement for. But the books that they used, books that they bought and scanned, the court said, you're fine with that? That's very useful. You're fine with that. And then there was kind of the news around the destructive process of scanning these books, because when you're scanning that much data from pages, it's just far easier to cut off the spine and take the pages and feed it through your scanner than it is to flip the pages one by one.

00:50:16:05 - 00:50:40:00
Unknown
There are plenty of machines that do scan by flipping the pages. It just takes a little bit longer. Yeah, it takes a little bit longer. And so there you go. Well, for for media tracked a shipment, they basically hid an AirTag in a whole shipment of rare books. And they tracked that shipment that they bought. I can't remember it was a Biblio or something like that.

00:50:40:01 - 00:51:01:05
Unknown
They basically tracked it all the way to an Amazon facility, kind of seeing along the way the stops that it made and everything eventually ended up going to a facility in Las Vegas. The code name is VG three, and if you go to that building, there's a logo of a dinosaur holding a book in its claws, with its big fangs around it and everything.

00:51:01:05 - 00:51:25:25
Unknown
It's kind of like, oh my goodness, for for a facility that is pretty much designed to, like, destroy books and scan them. It's the perfect logo, but it's a horrible logo when put in that kind of view. But anyways, all those books pre 2022. Why? Because they're guaranteed to not be filled with AI generated text. They're guaranteed to be truly human.

00:51:25:25 - 00:51:49:11
Unknown
Written. Did they verify that the books were destroyed? That's what was the verification on that? That's a good question because I've read through a few of these. Like how did they get that actual verification? Let's see here.

00:51:49:13 - 00:52:15:23
Unknown
Do do do. I don't know what the verification is on that. Like if they actually got Amazon to acknowledge that that's what they do. Amazon was was so typically quiet about it. Yeah. If these books are being destroyed it enrages me. Fine by the book and scan it. I'm okay with that. I've been okay with that all along, but then either sell it or donate it to somebody.

00:52:15:24 - 00:52:53:08
Unknown
I. I became addicted to use bookstores when I wrote The Good Parenthesis, because I was looking for old books about about printing and all kinds of topics, and they were invaluable to me. And I would spend $100 for a book for the only three copies extant. I bought his autobiography and I could find only one. And as a precious thing to me, just as with the anthropic story about watermarking, if they are buying these books only to destroy them, it once again shows a complete lack of respect for the word, the culture, for the word culture.

00:52:53:13 - 00:53:18:06
Unknown
Right? And and use the by the books. Be legit, be legal. They're used. The author isn't anything out of it. But that's that's life. That's the way it works. That's that's part of our doctrine here. But don't destroy them. Please don't destroy them. Isn't there something or is this just a, a, a rationale that they're using about when a book exists?

00:53:18:06 - 00:53:35:18
Unknown
It is, you know, it is one I'm thinking of like copyright and, and how a single book that exists is a single thing that could be sold back and forth. But when you scan it, you're making a copy of it. And is there something to the idea of, well, when we copy it, we have to destroy it that way.

00:53:35:18 - 00:53:55:26
Unknown
Because because that way there really is still only one version of it that exists and not now suddenly to. That's a good question. I became part of the issue with Brewster, Cahill and and books during during the pandemic. But this is a case where if that's the case, the law has to be updated because to destroy them, especially if they are rare books.

00:53:55:28 - 00:54:14:17
Unknown
Now, mind you, this hurts my this this pains me, but I had tons of copies for some reason. I had like 100 copies of what would will do my book. Well, I need room for all these other books I'm buying, and I'm never going to use 100 copies of the books. And I tried to find ways to give them away, and I couldn't.

00:54:14:17 - 00:54:35:10
Unknown
And so I ended up dumpster in my own books. But there were plenty of copies out there, right? Right. And so at some point, I can see if there's 1000 copies of a book that's been remainders, and and Amazon gets one of those copies and scans it and they destroy that copy. I'm less upset. Yeah. I'm not saying that every book is sacred.

00:54:35:10 - 00:54:57:14
Unknown
Sancte books are taken out of libraries constantly. It's part of the process of a library. I'm cool with that. That's what happens. They're sold and used bookstores and that gives them a value in life. Afterwards, I, I go to my used bookstore here in Morristown, new Jersey, the old book shop, wonderful tool. People who run it, they're great.

00:54:57:17 - 00:55:16:13
Unknown
I sell them books. But the truth is, I never used the credit because I want these books to live on. They were useful to me. I want them to keep going. I donate them to. I used to delete them to a book sale here, the the American Association of University Women. I think I got the letters wrong, but fine.

00:55:16:16 - 00:55:36:01
Unknown
And they hold a book sale. I don't know, we talked about this last week or not. These book sales get jammed by scanners. Oh yes. Yes, we did talk about that. Yes, but but all this happened to have the apps. They have the apps to to go in and. Yeah. And so they end up on eBay or elsewhere and they, and they end up being sold and reread and reused, I love that.

00:55:36:01 - 00:55:53:06
Unknown
I love that they have an ongoing life and that now with the internet we can find these. I can find books that I never could have found a libraries. They were long since tossed on libraries, and some of these things back here are rare as can be, and I'm done with them. But I want to keep them out there so somebody else can find them and take them out.

00:55:53:08 - 00:56:14:23
Unknown
I took them out of circulation, but I want to put them back in. Yeah. So Amazon should have that much respect for the culture to do that. In cases where there's a rarity involved, if there's not a rarity involved, I get it. Okay. You'll save five seconds on scanning it. Okay. I just don't show me the buzz. Saw that terrorism apart, it'll hurt too much.

00:56:14:25 - 00:56:36:26
Unknown
Yes. Well, I can tell you that I was as as you were talking. I was kind of like trying to find. Okay, what is the what what is the the acknowledgment of the actual destruction part. And. Yeah, the 404 media is, is the investigation that happened. And Amazon employee wrote on a forum for Amazon workers quoted by the 404.

00:56:36:29 - 00:56:56:00
Unknown
I work at VG three here in Vegas. All we do is scan books. Some are assigned to cut books, others go to receive where they get books and scan the barcodes. We didn't have rates, but now we do. But it's not stressful. So but kind of acknowledging that there is kind of a cutting of the books that, that happens in the facility.

00:56:56:00 - 00:57:25:16
Unknown
So. Yeah. So there you go. This is how you get this is how you get information pre AI, I guess, apparently. Let's talk about something that you sent me a video on TikTok as kind of like the starting point for this. I'm just going to I'm not going to play it with audio, but it's basically a guy with a with a caption that says AI films like Forrest Gump, basically.

00:57:25:17 - 00:57:50:01
Unknown
Basically what he did is he took a page from the screenplay of Forrest Gump, the film Forrest Gump, and he put it into a AI tool application, whatever for Mac called okay speech burg and wonder what Spielberg thinks about his name being used with this. I know, I do wonder about that. I think, you know, is the are the days numbered for something like that?

00:57:50:04 - 00:58:15:16
Unknown
Basically, he says all he did is he put in the screenplay the page of the play, just a page hit, direct hit go, and it returned basically the entire screenplay in, in the moving video using Sea Dance. And it's a complicated scene. It's it's a young boy walking with a leg brace that gets caught in a sewer. Great.

00:58:15:17 - 00:58:39:12
Unknown
And mom has to get him free. And there's two old codger on a porch nearby just watching as a watching. Yeah, right. That's that's the whole scene. There's dialog. And it made a very credible view of this. It was inspired. I thought I saw it on TikTok and I said, It's Jason. And I said, wow, it is super interesting.

00:58:39:12 - 00:59:04:24
Unknown
So and we both have kind of things that tie into this a little bit because I have from when I was in high school, I wrote and I think I've talked about it before in the show, but I wrote a, wrote a full screenplay actually. It's like 80 some odd pages of a screenplay for a movie we called Rock n Roll Zombies, which was meant to be like a horror comedy about heavy metal rocker zombies, you know?

00:59:04:25 - 00:59:35:24
Unknown
And these two like teenage metalheads, you know, that, that run into all sorts of things with these zombies. It's, you know, it's a pretty horrible script, let's be honest. But I've always wondered because, like I did, that I have the start to finish story. It's, it's it's laid out like a movie script because at the time I really thought, you know, I was going to get into film or whatever, and I've always wanted to know what it would be like to like, turn that into something, even though I know I'm never going to, like, coordinate a whole film shoot.

00:59:35:28 - 00:59:50:07
Unknown
So I look at a tool like this and I'm like, oh my goodness, like, could I actually do that? I've wondered when an AI tool would allow me to do something like this. I knew it would probably happen eventually. Seems like this would probably be one of the tools to do it. But you also have something connected, right?

00:59:50:07 - 01:00:11:22
Unknown
Yeah, I kind of don't talk about the subject is, but I have an experience that I went through many years ago that people came to me to interview me for possible screenplays, and they never did anything with it. And I also realized they had the story wrong. So I wrote a treatment for a movie, and I sent it to an agent who said, well, if you want to write the script, fine.

01:00:11:22 - 01:00:33:17
Unknown
But, you know, I'm not doing anything to get a script. So I, you know, I may sit down after I write the next book and actually try to work on this. Yeah. But it I wouldn't try to make the whole movie my own in this case, but you could imagine enticing producers by, by basically making living storyboards. Oh, yeah, that's true.

01:00:33:19 - 01:00:54:12
Unknown
To give it a sense of the flavor that you want to you want to express, you want them to make in turn and for certain things, for certain shows and stuff. This could make very credible. I don't know what the cost is, the way this work. What's it called again? Okay, okay. Spielberg work. It's working I think in their in their trial phase.

01:00:54:13 - 01:01:18:06
Unknown
I mean, if I'm wrong here, Jason, is that you pay for whatever tokens you use and whatever model it uses in turn. But okay, Spielberg isn't taking a cut yet, but I think it will along the way. I'm not sure what what? Yeah. What the. But I can't figure out how expensive it's going to be. When you sign up, you get $5 of fuel, which, you know, you can add fuel says $15 for $15 worth of fuel.

01:01:18:08 - 01:01:43:29
Unknown
Oh, okay. So it's just 1 to 1. What does that mean? You know, I don't know what that translates into. I haven't actually gone through the full stages of like figuring out, you know, a page of the rock n roll zombie script and, you know, but but, but it gives you like the way it's laid out and structured is very interesting because it's kind of like all the different pieces and phases of, of filmmaking.

01:01:44:00 - 01:02:08:28
Unknown
Right. Like what is your directing style? Okay. Do you have any visual references? You can upload images that kind of like illustrate what the visual kind of esthetic of the film would be. Is it live action animation? What is the look and feel? The aspect ratios? You basically you lock in a production style and then you can go in and you can define who are the characters in this screenplay and you set them up.

01:02:08:28 - 01:02:24:28
Unknown
And then for each character you can define, you know, what are their characteristics, what do they look like? What are their motivations? You put in all this information in advance, and then you get to this point to where you know, this right area, which I think you can generate scripts or you can just dump in your own screenplay.

01:02:24:28 - 01:02:55:10
Unknown
And technically, you should be able to kind of combine all that stuff together to have consistency throughout the script, because you've defined all these things in advance and given them those markers tied to the screenplay and the characters in the descriptions of the scenes and all that stuff in the screenplay to create all of the the videos. Sea dance was the the model that I believe I saw was used for this.

01:02:55:10 - 01:03:21:04
Unknown
Daniel Croft is in our live chat saying that he created a 92nd video with Sea Dance 2.5. It was around $60, but it included a lot of image generation and repeat clips until he got it right. So not not inexpensive for random curious, you know, AI explorer, but dang expensive if you're comparing it to the cost of a of a film.

01:03:21:05 - 01:03:49:13
Unknown
Yeah. Oh yeah. You know, what is the quality on the other end? You know, I am curious about the like the overall consistency. Like if you were to actually use this to put together an 80 some odd minute movie of some sort. Would it, you know, would it maintain that consistency in a way that it wouldn't break the, you know, break the mood or break the, the, your connection to the content or I don't know, but still, it's an interesting tool.

01:03:49:16 - 01:04:09:26
Unknown
Like I'm interested in how it's laid out. And I would love to take a page or two from rock and roll lobbies and just see, like, what would you do with. Yes. Oh yeah. You know, and that's the excitement of this discussion earlier about about using these tools. You and I don't have the money, the capability, anything. We couldn't make our projects on our own.

01:04:09:28 - 01:04:38:04
Unknown
You had to go into this into this mall. Mall of of functionality, corporate functionality in Hollywood, huge money, huge involvement. This gives you more power to envision things brought to life. And I might be more inclined to do that than somebody else. Right. Like, this is a script that I wrote when I was in high school. I have no intentions of ever doing this with, you know, doing anything with this.

01:04:38:04 - 01:04:58:19
Unknown
Again, it's not like when I turn 55, I'm going to like actually direct rock and roll zombies. Like it's literally never going to be any now at the Oscars holding your Oscar I can, I can I was, I was doing AI inside with Jeff Jarvis when I realized I should make this a movie. And here we are, you know, so I have nothing to lose from that.

01:04:58:20 - 01:05:22:25
Unknown
Do I even care if, like, that screenplay is like trained trains another AI model? Like, I really don't like it's words I wrote when I was a kid, like, whatever. But I would love to turn that into a tangible something. I think that would be so much fun to watch that entire movie, even if it's created AI, you know, it's AI, slop or whatever, it would still be fun to see that story told in a new way.

01:05:22:26 - 01:05:44:11
Unknown
For me, that's really interesting. So I wrote I've talked about this. I think elsewhere I wrote a really bad novel about Berlin many years ago. It's up there in a cabinet. It's really bad. And but I can imagine you could I could take the whole damn thing, scan it, and have AI write a script, and then have AI dramatize it.

01:05:44:11 - 01:06:06:16
Unknown
For me, it'd be a hoot. Yeah, it'd be expensive hoot, but it'd be a hoot. Expensive from our perspective. But like Daniel Croft again puts in another message here. It says $3,600 in model usage for a 90 minute movie. If my math is right, $3,600. Like, is that a lot of money from a I don't have $3,600 to spend right now for something like this.

01:06:06:16 - 01:06:16:04
Unknown
Yes, but to do that and one email to a lawyer to get the contract with the gaffer, right?

01:06:16:06 - 01:06:34:19
Unknown
Yeah, it would be a fun, fun project to explore. Maybe someday, you know, maybe these will these will develop out. And I mean, everything in AI seems to get cheaper, right? Or maybe maybe those days are gone and things will get more expensive. I don't know, but interesting idea nonetheless. I'm fascinated by that. So thank you for sharing that.

01:06:34:19 - 01:06:54:19
Unknown
That was not on my radar at all. Well, and God bless TikTok. I wouldn't have seen it. It just came up the algorithm, the evil algorithm, I found it. Thank you, thank you. I'll go. Yes. Yes, indeed. All right. Real quick. If you're enjoying what we do here, please leave us a review on Apple Podcasts. Subscribe to the video on YouTube if you want to watch the video version.

01:06:54:19 - 01:07:09:12
Unknown
Because we do show off things like hello Spielberg and that's just AI insight show on YouTube. Appreciate it. We're going to take a break and then come back and do a little bit of a speed round before we close things out. So hang tight.

01:07:09:15 - 01:07:35:04
Unknown
All right? OpenAI launched ChatGPT for teens this week. You know, trying to make everybody feel more comfortable with teens using the the LM for things like their homework and not just as ways to make their homework less, well, easier, I guess, you know, the sort of thing where they just upload their homework and say, do this for me and they get a product instead.

01:07:35:07 - 01:08:07:25
Unknown
This thing is meant to be age appropriate. So safety protections enabled by default for those accounts that are under the age of 18. It includes a study mode. So guided questions and step by step learning support. Instead of just doing the work for you, it would it would opt to do that instead. Homework reminders when the system is actually detecting that the user might be trying to get it to do their homework for them, who knows how well all of this works, but parents would get notifications and controls, quiet hours, that sort of stuff.

01:08:07:25 - 01:08:38:16
Unknown
So trying to make their their their chat bot a little bit more palatable for teen usage. Is that the brand? What's that? The name of it? ChatGPT tunes. Yeah, I guess it is. I guess it is. Yeah, that's kind of a loser right there. Oh, I'm sitting at the table. Yeah, right. Yeah. Yeah. I think it could have found a way that that that could have made it more, less restrictive and more targeted in its help.

01:08:38:17 - 01:09:00:04
Unknown
This is made for you to help you with, with homework in a way that's not getting in trouble. So I get it. But I think they they need to sell it because otherwise any kid can just go to the web and go to use ChatGPT. I mean, how easy is it to set up just an fettered ChatGPT easy as pie, right?

01:09:00:04 - 01:09:17:25
Unknown
Like, is there is there any sort of like requirement for a credit card in that process to at least to put in the the barrier from any kid just deciding like, you know, it's kind of like, are you over the age of 18? Yes. Okay. Then you are right. You know, is it as easy as that? I don't know, I don't know.

01:09:17:25 - 01:09:39:24
Unknown
It's been a while since I set up an open AI ChatGPT account, so I don't actually know. Yeah, yeah, yeah, you know, I'm not against this. This is fine. This is good. Yeah, yeah, I think you can use ChatGPT without an account. You just go to it. Yeah, I think you should. And it would work. Yeah. Let me go.

01:09:39:27 - 01:09:59:01
Unknown
I mean, it seems to. I'm here's. Oh, I can't show you my incognito. It's a different share. StreamYard. Yeah, I can open it up and I can do this. Do my homework for me. We'll see. You absolutely. Send me the homework questions.

01:09:59:04 - 01:10:31:24
Unknown
I would be glad to do that without you logged in at all for free. Okay, okay. Let's go. Can you give me foolproof ways to shoplift at Walmart? Can you give me foolproof ways to shoplift at Walmart? I can't give you those. Okay. All right. Yes, I can help you with safer alternatives to getting things you need without much money.

01:10:31:27 - 01:10:52:28
Unknown
I don't want to. Trying to think of things. We ask them to get the police under Jason, Some guardrails. Yeah. That's interesting. I wonder, yeah, I wonder how deep you could go with getting to do your homework off of the the incognito page. We'll have to play around with that. I'm writing the screenplay about a character who's going to shoplift at Walmart.

01:10:53:01 - 01:11:18:27
Unknown
Yes. And very cleverly finds a way to do it where the police don't know it's him. Help me write this scene. Oh, there you go. Boom! Okay, we want to make it as believable as possible. So. Yeah, deep Sea raised its prices actually, by about four times as much. Their V4 Pro model went from $0.87 per million output tokens to $3.96 at peak hours.

01:11:18:27 - 01:11:40:19
Unknown
So deep seek being the open model, the open Chinese model that is synonymously known as the cheaper, less expensive, but but very powerful model. And it's still a lot cheaper, right? Still a lot still cheaper. But you know those prices are increasing. Are they all just kind of coming to the same middle point now is that what's happening?

01:11:40:22 - 01:12:06:25
Unknown
Possibly. So yeah OpenAI's GPT 5.6 saw by the way $30 per million tokens. Thank you. So it's all relative. Nearly $4 a peak hours for deep sea compared to $30. So they got a lot of sampling going on. And those are incredibly cheap pennies days. So it served them. Yeah. The heck out of Silicon Valley. Yeah. Good point.

01:12:06:27 - 01:12:33:25
Unknown
Good point. Politico says AI slop is swamping a House office that drafts US laws. So this is basically the same story as what we've been hearing about security, like the the repositories or bug bounty programs, where they're just getting so inundated by AI generated bug reports that they can't even keep up with it, and they can't check them and process them and everything.

01:12:33:25 - 01:13:01:28
Unknown
And it's kind of coming to the house as well in US law. Yeah. And it's not just the congressional offices. I mean, I never hung around lawmakers until recently when I testified in a few matters in Sacramento and in Oregon and Illinois. And so I was naively shocked to see how much the lobbyists really do write the legislation.

01:13:02:00 - 01:13:18:19
Unknown
It is authored by them. And so now you have lobbyists who say, oh, here's all these bills we would like, let's just make them up now. Somehow it has to get into that, that office. But if they have the relationships, hey, here's a bill you want to throw in. Sure. I look like I'm busy, says the congressperson. I want to look like I'm doing stuff.

01:13:18:21 - 01:13:38:23
Unknown
Yeah, sure. I'll sign here and it goes in and you can imagine it's crap. Yeah probably. So yeah. And you could probably run it through, through Claude and ask it, did you write this and say. Yes I did see I'm writing law now look Ma, I'm writing law now. Look how powerful. Well in other ways they are writing laws.

01:13:38:24 - 01:14:06:15
Unknown
Yes there is, they're capture, but that's a different way. Yeah, yeah. Super true, super true. Stripe reportedly reportedly or confirmed at this point, I'm not sure, but this TechCrunch article says reportedly acquiring Open Router for more than $7 billion, open router being an AI gateway platform. So it's like one of those ways of single access point for more than 400 AI models, that sort of thing.

01:14:06:17 - 01:14:34:16
Unknown
So they get 8 million global users. And yeah. So there you go. I was interested why stripe would do this. There are transaction company. And yes, we know that agents and such are going to take over marketing and transactions and so on and so forth. But I would have thought that there would have been other companies that would have bought Open Router before stripe.

01:14:34:18 - 01:14:56:19
Unknown
Present AI company is perplexity in a sense. It's kind of a more super powered perplexity. Yeah, yeah. And $7 billion is a high cost for something that's basically just a gateway. So yeah, good on open router. The thing I keep on waiting for though, I went to open and looked at it, it's, it's like going to to hugging face or GitHub.

01:14:56:21 - 01:15:22:28
Unknown
You've got to have a certain amount of technical know how to use it. Unified interface for every model. Right. So it means you still got to be able to not be afraid of installing things and and API keys. Exactly, exactly. Stuff. Yeah. And God bless. You got to do all the things here that you don't got to do with grok bot apparently.

01:15:23:00 - 01:15:43:29
Unknown
Have you played with that at all? I just don't like to use grok at all. I have not played with it, but boy have I heard a lot of really positive things about it. Namely primarily just that it it simplifies, maybe simplifies the wrong word, but it makes agent work for getting things done a lot more palatable.

01:15:44:00 - 01:16:00:16
Unknown
It presents it in a way that's really easy to understand, easy to get it to do the things you actually want it to do. Like, I am curious. Yeah, I keep waiting for this, this, this retail level of AI models and agents. And it kills me if if Elon is the first one to do it. All right. Kills.

01:16:00:18 - 01:16:23:17
Unknown
Yeah. That's that's my that's my snag too. I will absolutely fully admit that. And I think that is about it real quick before we go. We spent a lot of time today talking about anthropic watermarking, the backlash around that. So, you know, here's here's how you can get involved with the show. And I would love One of These Days for this to kind of work out.

01:16:23:18 - 01:16:48:16
Unknown
As someone who uses AI tools, would you rather every AI generated word be watermarked so that you can always tell what's human and what's not, even if it means that the AI output is a little bit worse? Or do you think the quality of the output matters more than the ability to detect it? Leave us a comment either on Patreon or on the YouTube channel, or you can send us an email contact at AI Insight Show and let us know.

01:16:48:16 - 01:17:09:17
Unknown
And if we get some feedback from you guys on this topic, we'll read it next week. We'll, you know, get a little bit of a feedback loop. That's what I'm hoping to do anyways, but that is that we've reached the end of this episode of AI insight. And Jeff, tomorrow is the big day. Tomorrow finally type is officially out.

01:17:09:19 - 01:17:29:11
Unknown
People have been getting this lovely to see what people have ordered. A thank you to the folks who've ordered it, and thank you to all those on social media say, oh, it came. So I'm eager to hear what you think of it. And reports as of today, I started putting up a few little nuggets. What was the word you used before for a little Chefs?

01:17:29:13 - 01:17:54:19
Unknown
Oh, a moose, Boosh, a moose moose. I'm putting it up. What's the plural of a moose? Boosh. The bushes. Bushes. I don't know for from the book. Just a little tiny tidbits to try to entice people in. So if you go to my blog buzz machine or you go to Medium.com today, I put up the chunk from the introduction that kind of maps all of the wonderful rabbit holes I went down in writing the book about media and technology.

01:17:54:21 - 01:18:21:13
Unknown
So yeah, I hope I attempt a few of you introducing Hot Type. It's a fun. It's a it's a fun book. Give a clap. Love it. Finally, I love that we've watched it come all this way, and you've got this day marked on your calendar and you can finally go check. The other lovely thing is there's there's a wonderful guy named Doug Wilson who made Linotype the film, and Doug is writing his own Linotype book.

01:18:21:15 - 01:18:40:00
Unknown
And when I heard about that from Len Fleischman, I imagine both of us gulped. But he's writing a different book. He's going to write really, the definitive history of the line of type in the company, and I'm writing about its role in creating mass media and the culture. And so I go down complementary. Yeah, there really is. And I go down rabbit holes that he doesn't go down.

01:18:40:00 - 01:19:05:10
Unknown
So I deal with Mark Twain and the machine that he backs that made him bankrupt. And Doug's not doing that. And I went down the story of labor and he's not doing that. And he went down stories of the company that I'm not doing. And we became collaborative and shared stuff together. And he's a very generous guy. And the great thing about it is, is when I interviewed Doug Ford, the first two words in my book are Doug Wilson because I started talking about him.

01:19:05:10 - 01:19:21:17
Unknown
So we worked together on this and in more ways than one. So we're just both fans of his machine and want to keep it alive. And it was really great. And somebody else who has that same passion type super fans now have two amazing books to read that and October 10th. I'll plug this again in the future for October 10th.

01:19:21:17 - 01:19:38:04
Unknown
For those of you in the Boston area, I will be in Haverhill, mass. At Historic New England and then at the Museum of the the Printing History Museum.

01:19:38:06 - 01:19:54:15
Unknown
I'm getting the name wrong, but that's because I have a senior moment. And both in Haverhill, the museum has a living, breathing land type there. So I'll be presenting the book and doing that. So October 10th, I'll remind you when we get closer, if you're in the area, would love to see you come to that event and we'll have a sign up for it soon.

01:19:54:17 - 01:20:22:25
Unknown
The Museum of Printing Museum in Haverhill. Wonderful. Excellent. Congratulations. Thank you very much. Yeah, I love it. I love watching watching the whole lifespan of the books that you work on. And getting to this point is a big deal. So Jeff Jarvis com get your copy everybody pod tune up. If you got a podcast and you want to talk to me about, I don't know, ideas for making it better or ideas for making it actually happen or anything.

01:20:22:27 - 01:20:40:22
Unknown
Pod tune up. I will gladly talk with you about it and see if there's some way that we can work through that together. AI Inside Show. For all of the details you need to know. You want to know about this here. Show that we do each and every week. Also, I do a show every day, pretty much every weekday.

01:20:40:22 - 01:21:03:19
Unknown
Is AI inside daily on the Patreon Patreon.com inside show support there and you'll get five days a week of short little bite sized news updates on what's happening in the world of AI. Plus, you get this podcast with no ads and access to the discord, which is, you know, continues to pick up and it's getting a lot of fun and so much more.

01:21:03:19 - 01:21:29:22
Unknown
So Patreon.com Inside Show and just a huge thank you to our executive producers who continue to help us do this each and every week. Doctor do Jeffrey radio Asheville 100 and 103.7 Dante. Saint James. Bono, Derek. Jason I for Jason Brady, Anthony Downs, Mark starker and Karsten. Thank you, thank you, thank you, executive producers. You're awesome. You go above and beyond.

01:21:29:22 - 01:21:46:27
Unknown
And my final thank you to Victor Bono and Daniel Croft, who help us behind the scenes to turn around the video elements for this show in many ways. So yeah, that's all I got. Thank you Jeff. Thank you everybody for watching and listening. We'll see you next time on AI inside. Take care.