How Smart Are Today’s Coding Agents?

00:00:00:00 - 00:00:08:17
Unknown
This episode of the AI Inside podcast is brought to you by area. Get started for free today@irca.com.

00:00:08:19 - 00:00:39:08
Unknown
Coming up next, Jeff Jarvis and I break down. Clod 4.6 is new financial research powerhouse OpenAI's GPT 5.3 Codex as a full lifecycle coding agent. Matt Schumer's warning that AI is coming for nearly every job in a few years. That's a very interesting story. We'll also untangle the Super Bowls AI heavy ad blitz, Waymo's use of DeepMind's Genie three world model to train self-driving cars on rare edge case scenarios, and more.

00:00:39:08 - 00:00:53:00
Unknown
Coming up next on this episode of the AI Inside podcast.

00:00:53:03 - 00:01:13:09
Unknown
Hello everybody, and welcome to another episode of the AI Inside Podcast, the show where we take a look at the AI that is layered throughout the world of technology and boy is it. I'm joined, as always, by Jeff Jarvis, my friend. We're still recovering, but so happy you could be here.

00:01:13:11 - 00:01:26:07
Unknown
You get to shake your cane at things that you're angry at.

00:01:26:09 - 00:01:32:12
Unknown
Yeah, with a microphone.

00:01:32:15 - 00:01:52:16
Unknown
Yeah. Don't don't even sweat it. I'm totally cool with it. I think people completely understand. And like I was listening to, listening to a clip of last week's episode. Sounds fine. It's fine. Jeff, I just want you to recover and get better and and all the things. And I know everybody listening and watching wants that, too. So I'm happier.

00:01:52:17 - 00:02:16:15
Unknown
You're mending, though, that you're on the road to recovery. So bit by bit. Yeah, I know it takes, it takes a while. Well, we got a lot to talk about. And I think this top story block is kind of interesting because there's a story making the rounds on tech meme today, which we'll get to. But in order to get to that, let's talk about the news that is kind of the foundation of that story.

00:02:16:15 - 00:02:45:14
Unknown
And we can start with anthropic, because last week, as we talked about on the show, there was the anthropic, clod Co-work plugin that impacted, I guess, the legal software world. And then now, you know, just a couple of days after, after that episode, maybe even the next day, we had clod 4.6 new version that was, suddenly impacting the world of financial software and financial research.

00:02:45:14 - 00:03:07:19
Unknown
Basically, anthropic has said this is a model that's positioned as an analyst, great engine for financial research. What does what the heck does that mean? Company it can, you know, we can scan through company data, regulatory filings, market information, produce detailed financial research. I'm actually looking at this because I'm not great with finance. As far as, like, analysis and stuff, I'm looking at this.

00:03:07:19 - 00:03:27:03
Unknown
I'm like, maybe I should start, you know, seeing how I could integrate this into, how I analyze my own finances. But, do all the, you know, things that I guess we've been used to, but apparently you can do it better creating spreadsheets, creating presentations, software development, yada yada, yada. They also increased the token limit up to 1 million.

00:03:27:03 - 00:03:51:03
Unknown
So it has a much larger token limit and introduce something called agent teams, which we have talked about and heard about with other models, which is just this idea of like breaking tasks into little kind of side tasks and throwing those to different agent workflows that happen simultaneously. But independent of each other, and then kind of connect together to create the full output.

00:03:51:05 - 00:04:15:23
Unknown
All of this impacted Wall Street, and, you know, kind of continued to impact the concerns around the future of software. And is software even needed anymore? Because, look, what Claude is doing. Some of those dropped 10% on the idea that this model was coming for their work. What do you think, Jeff?

00:04:15:25 - 00:05:23:10
Unknown
Yeah.

00:05:23:13 - 00:06:15:26
Unknown
Yeah.

00:06:15:29 - 00:06:46:02
Unknown
Oh it makes total sense. And kind of listening to you kind of lay out the few, the couple of different scenarios. I mean all I can think is a little of this, a little of that, it's a little bit of everything. Right. And fear, uncertainty and doubt is, is a hell of a drug. You know everybody that there's, there's a sense or there's been a sense we've gotten comfortable to a sense that our lives and our careers and what we do and who we are and our identity is all within our control.

00:06:46:02 - 00:07:04:05
Unknown
And I think, you know, part of the fear, uncertainty and doubt is, oh, maybe we're at a time where it's no longer in my control. Where I thought I did, I thought I knew what I did for a living, and that worked until now. Maybe it's not going to work anymore. And it is interesting how some people see this as an empowering piece of news.

00:07:04:05 - 00:07:25:24
Unknown
Oh my goodness, tools that are capable of of making, you know, of, of impacting and changing my workflow so that I can be more impactful, be more efficient and, and, cohesive and comprehensive and all that. And then you've got the other side, which is, oh my goodness, they're doing the thing I used to do or I've done for years.

00:07:25:27 - 00:07:57:24
Unknown
Sky is falling, you know, nothing is is what it seems.

00:07:57:26 - 00:08:03:01
Unknown
Yeah.

00:08:03:03 - 00:08:13:12
Unknown
Yeah.

00:08:13:14 - 00:08:24:24
Unknown
Totally.

00:08:24:27 - 00:08:48:26
Unknown
But it's, it's hard to see the forest for the trees when you're in the, in the middle of the trees I guess. Or the fir trees for the forest. Yeah. Right. Yes. Did you know by the way that like all those people, the trees, I mean those were people like hanging out. They weren't even dancing. They were just efficient trees.

00:08:48:29 - 00:08:54:20
Unknown
Totally. Nope.

00:08:54:22 - 00:08:59:10
Unknown
Yep.

00:08:59:13 - 00:09:17:18
Unknown
It was. It was efficiency. I don't know how you feel. If you're the if you are a dancer and you're hired for the Super Bowl and your whole job is to stand being a tree. But it was pretty funny because I saw it live. And my wife, she was she had to like drop off the kids or something like that during the Super Bowl.

00:09:17:18 - 00:09:42:28
Unknown
So she missed the performance and she came back and I was like, you got to see the Bad Bunny performance. And then, when she turned it on at the very beginning, I was like, and those trees are people. She's like, you're ruining the thing. I'm like, no, you never know. That's the thing. It's like, you know, it's not like suddenly they do a tree dance like you'd never know.

00:09:43:00 - 00:10:07:07
Unknown
Yet at least there's that. Maybe you were a tree, but at least you were in the Super Bowl. And you can put it on your resume now. Yes. Slight diversion. We will talk about the Super Bowl a little bit later, from an eye perspective, but other pieces of news kind of feeding into this very, very similar and very much along the same lines is OpenAI's GPT 5.3 Codex.

00:10:07:09 - 00:10:31:24
Unknown
This is now, released as of, I guess the fifth, so late last week, sits in a very similar trend line to what we're talking about. On the software side, this is a new a genetic model. It takes the strengths of 5.2 Codex encoding, and then it takes the strengths of GPT 5.2 reasoning and knowledge, capabilities, and kind of puts them together.

00:10:31:24 - 00:10:55:07
Unknown
At least this is kind of how OpenAI has, you know, presented it, able to write and debug code can manage full software life cycles. I'm not a developer, but I can kind of understand what that means. Yes. The standard, you know, hey, it creates spreadsheets and slide decks and reports. I feel like that's becoming, kind of table stakes, but can it do it effectively?

00:10:55:07 - 00:11:19:25
Unknown
Like, I've had some spreadsheets created by AI and they were not good spreadsheets, so hopefully they're getting better. I imagine they are. OpenAI is saying that 5.3 Codex was instrumental in creating itself, not necessarily building itself from scratch, although one could see maybe somewhere down the line, something like that happening. I suppose. But as an assistant, that was part of the process throughout the development.

00:11:20:00 - 00:11:40:09
Unknown
And this reminds me of when I was out at Google not too many months ago for their, Thanksgiving timed, Friendsgiving thing, right around the time that the latest, Gemini model was releasing and the latest Nano Banana Pro, and that was one of the topics that came up there, actually, that was a question that I asked them, like, do you use your Gemini model to develop your Gemini model?

00:11:40:09 - 00:11:58:19
Unknown
And they said, yes, they're doing that to some degree. They did that to some degree with their previous model. They're definitely doing that with the next model. So this is kind of this is kind of just happening with these models. They're capable enough at this point to help develop themselves. And if they're there now, what does that mean for two years down the line?

00:11:58:21 - 00:12:09:28
Unknown
It's interesting.

00:12:10:00 - 00:12:57:21
Unknown
Okay.

00:12:57:23 - 00:13:28:08
Unknown
Okay.

00:13:28:10 - 00:13:49:00
Unknown
Yeah. Very big.

00:13:49:02 - 00:14:22:29
Unknown
Yeah I think we're going to get there sooner rather than later. Yeah.

00:14:23:01 - 00:14:43:18
Unknown
Yeah. Yeah. Right. Well that, that kind of I think is a nice kind of segue into this piece that is totally making the rounds, because I think it kind of touches on a little bit of what you're saying, kind of the, the bridge between both of those, I don't know, divisions or extremes or whatever you want to call them.

00:14:43:20 - 00:15:03:21
Unknown
Matt Schumer is himself a CEO of an AI company. The company's called hyper, right. And he wrote a pretty long kind of detailed, I don't know if it's an article, blog post, whatever. It's on his, schumer.dev site where he. Yeah, I don't know. Yeah. I realized that earlier today. I was like, what do I call this?

00:15:03:27 - 00:15:27:02
Unknown
It's not an article. Is it a blog post? Is it musing is I, I don't know. But anyways, he shares, basically the point of all this is that he shares the common story that we've heard and talked about plenty of times, which is as coming for your job, too. And here he admits that, well, first of all, he admits that he is part of the AI train.

00:15:27:02 - 00:15:51:02
Unknown
He is on the AI train, his company is invested in it. But he also kind of calls out, you know, as much as I, you know, have an AI company, I by no means feel like I have any sort of control about what I'm writing here. He basically says that he shares, the release of opus 4.6 and GPT 5.3 Codex are hinting at what is coming down the pipeline for every job.

00:15:51:04 - 00:16:14:15
Unknown
He's arguing that this is not going to happen in the matter of decades, but in a matter of a few years, you know, he's saying, look, these models are rapidly developing. And, you know, he even points out, like the there are plenty of people who probably played with some of these models a few years ago and think that they have a solid sense of what AI is, and, oh, it's a piece of junk and it gets everything wrong, and it does this and it does that.

00:16:14:15 - 00:16:37:03
Unknown
And he's like, you know, even in the past six months, things have changed so much that, you know, these systems pale in comparison. And or he he calls them kind of unrecognizable. If you if you and I would argue that's that's not true. I think they're very recognizable. But you have to drill deep to, to really get a sense and understand what they are capable now versus what they weren't capable of before.

00:16:37:03 - 00:17:00:25
Unknown
It's not immediately obvious, is my point. But, and the reason that he backs all this up is because with where we're at with Claude and Opus and where we're at with Codex and everything he wrote, this is a little long, but I'll just read it because he kind of spells it out. He says, it's about his own experience of asking, it, which I think is Codex, to create an app for him.

00:17:00:25 - 00:17:20:00
Unknown
Just recently, like last week, he says it writes tens of thousands of lines of code. Then, and this would have been unthinkable a year ago. It opens the app itself. It clicks through the buttons, it tests the features, it uses the app the way a person would. If it doesn't like how something looks or feels, it goes back and changes it.

00:17:20:02 - 00:17:50:28
Unknown
It iterates like a developer would, fixing and refining until it's satisfied. Only once it to has decided the app meets its own standards. Does it come back to me and say it's ready for you to test. And when I test it, it's usually perfect. And so he's basically saying like, this is where we're at right now. And you know, even he is looking at these things as a, as a, I don't know, a person, you know, with an AI company and everything and saying, wow, look how far we've come in a short amount of time.

00:17:51:01 - 00:18:10:29
Unknown
Even I don't know, you know, whether what I'm doing is, is safe from massive disruption and not ten years down the line, not five years down the line, but 2 to 3. What I don't understand is why is everybody going nuts over this thing? Because I feel like this happens all the time. People write an essay and they say, they say, oh my goodness, this is so good.

00:18:10:29 - 00:18:22:27
Unknown
Now, yes, they are all better than the one that came before it. Why is this one getting so much attention? I don't get it.

00:18:22:29 - 00:18:49:16
Unknown
Yeah.

00:18:49:18 - 00:19:45:08
Unknown
Okay.

00:19:45:11 - 00:20:27:17
Unknown
Oh wow okay. Yeah.

00:20:27:20 - 00:21:00:29
Unknown
Because that's how we do it. Yeah. Yeah. Yeah. Yeah. Yeah. Yeah.

00:21:01:01 - 00:22:09:08
Unknown
Yeah.

00:22:09:11 - 00:22:33:22
Unknown
Yeah I yeah I feel like we had got it was just a couple of weeks ago we kind of talked about this idea of like these micro, I can't remember what the name was, these micro apps or these you know. And they are not meant, they're not meant to have a life beyond the use that you had in that moment, you know, of like, oh, you know what I need right now, in this very moment, I need an app that does this, create it.

00:22:33:22 - 00:22:55:26
Unknown
Okay, here it is, I use it. I have no more need for that anymore. You know? Now, granted, like there's the compute cost and there, you know, it has an impact to do that in kind of a disposable way. And so that's that's certainly a consideration. But but yeah, I think we're getting closer and closer to the point to where not just people who know how to do that can do that.

00:22:55:28 - 00:23:26:06
Unknown
People are already, you know, I use my wife as an example a lot. But like, you know, before a handful of months ago, she had never really touched or spent any time with ChatGPT or any any sort of LLM. And now I can't even tell you how many times she's like, you need to put that in chat, you know, you know, and it's just kind of part of the part of her like understanding of what's possible now is to go to chat, you know, when, when she has a question about something and to query it and talk to it and stuff like that.

00:23:26:06 - 00:23:45:25
Unknown
It's been a really rapid kind of transformation that I've seen in her and how she thinks about these things, and that's just for chat related stuff. That's not about creating apps, but when creating apps is as easy as going to chat and saying, I need something that does this and poof, you have it, then that's pretty darn powerful.

00:23:45:28 - 00:24:09:15
Unknown
And actually, to a certain degree, we do have that, right? Like these things. You go into Gemini, you can create an app, but by just going to Gemini timeline and saying create something and it'll create that, you know, that'll just get better.

00:24:09:17 - 00:24:24:20
Unknown
Yes. Oh 100%.

00:24:24:22 - 00:24:29:18
Unknown
Right.

00:24:29:20 - 00:24:44:03
Unknown
Right.

00:24:44:06 - 00:25:09:06
Unknown
So hand this down, this collection of thoughts and knowledge that you wouldn't have had otherwise. Here you go. I'm giving you a gift that you can hand down to your kids and yeah.

00:25:09:08 - 00:25:22:11
Unknown
Right.

00:25:22:13 - 00:25:43:03
Unknown
That's true. Yeah. That's a good point. I've thought about that.

00:25:43:05 - 00:26:03:15
Unknown
Oh boy. That's. Yeah. That's a good way to put it. I bet you there are some developers who are listening right now that are like exactly. That's why X that's why I hate this or whatever the case. Maybe you know what I mean. Because immediately. Yeah. And I mean, you know what this is, this is kind of the same thing that music artists are, are worried about too, right?

00:26:03:15 - 00:26:23:20
Unknown
Like when I can create a song that can sell, you know, or that can have a bazillion streams and get radio play and everything, and suddenly the market is flooded with music that was really easy to make by, by a single prompt. And it just it lowers the perceived value of their music, or at least that's what they're concerned about.

00:26:23:20 - 00:27:42:23
Unknown
I kind of think that humans will always have a higher value proposition, but I could be wrong.

00:27:42:25 - 00:28:14:04
Unknown
Yeah. Well I guess the the concern there though is that maybe, maybe you can, but are those people going to pay you for advising them when they can go there and get that advice. And then, and then. And this is part of what Matt Schumer was writing about is and then that and he's you know, he's not the only person ever say this, but it was definitely something that I remember from his piece that concentrates all of that stuff into the hands of a few few billionaires.

00:28:14:04 - 00:28:36:26
Unknown
Right? That that says all those jobs, all those people who decide not to to go with that advisor and to instead get the advising for themselves and ChatGPT or whatever, all of that is usage on their platform instead of others. I'm just I'm just count, you know, devil's advocate. Yeah.

00:28:36:28 - 00:28:45:16
Unknown
Just, just.

00:28:45:19 - 00:29:11:05
Unknown
Yeah, yeah.

00:29:11:07 - 00:29:25:09
Unknown
No.

00:29:25:11 - 00:29:49:10
Unknown
Yeah. Yeah. Indeed. Fascinating topic Daniel. Yeah. Thank you Daniel for writing in the live chat says it's free to look up how to fix a leak yet people will still pay someone to come do it for them. I think there will always be a need for people. Yeah.

00:29:49:12 - 00:30:11:29
Unknown
Please don't. I think we're still in the learning phase around what I'm capable of in that regard or not. You know, I think I still get asked to do things, and then I do it and I'm like, don't look too close to that. I did my best.

00:30:12:02 - 00:30:43:24
Unknown
No, I got this.

00:30:43:27 - 00:31:03:24
Unknown
Yeah. Yeah. And, and I was going to say and not. Yeah. That's a great way to put it, I was going to say it'll get messy during that shift. And it will be uncomfortable but it will shift as it has many times in history, as we often talk about. All right. We got a lot more to discuss.

00:31:03:26 - 00:31:24:24
Unknown
Real quick. Want to throw out a huge thank you to, a few patrons? Patreon.com slash. I inside show Charles Gillig Lee, lint here and Steve Isaac Isaacson. Sorry, a few of our amazing patrons. Patreon.com smash. I inside show. If you want to support us, you can you just go there and, find a level that works for you?

00:31:25:01 - 00:31:30:23
Unknown
There's lots to choose from, and we appreciate it. It really does help this show. Continue.

00:31:30:25 - 00:31:50:00
Unknown
And also want to take a break and thank the sponsor of today's episode, Arya. So happy to have Arya with us again. Arya is, it's just a fantastic know. We talk about a genetic kind of workflows and everything. I've built a gigantic workflows inside of Arya.

00:31:50:00 - 00:32:30:15
Unknown
That's a really great, tool to have for that. For that exact need, bring AI into your company. It can be exciting. It can also feel risky at times between data privacy concerns, security threats, constantly evolving compliance requirements, it can be easy to hesitate as a result. Well, that's where Arya comes in. Arya provides a unified security layer with advanced threat detection and strong governance tools designed to protect your AI ecosystem from day one, so you can safeguard your sensitive data, monitor activity, stay aligned with compliance standards and all of that, while continuing to innovate at speed.

00:32:30:19 - 00:33:00:26
Unknown
So instead of slowing down because of that uncertainty I was talking about, you can move forward with clear guardrails, real confidence. And that just means that you don't have to choose between your innovation and your security. Arya helps you do both at the same time. If risk has been the thing holding you back from going bigger with AI, well, this is this is your green light visit area.com and you can learn more or start a free trial.

00:33:00:29 - 00:33:21:19
Unknown
That's air a.com to check it out for yourself. I have I think it's a pretty fantastic system and you should check it out to area.com to get started today. We thank them for their support. All right going to take a quick break. Come back. And what do we have on the other side of the break. Oh the Super Bowl.

00:33:21:19 - 00:33:26:06
Unknown
Don't worry. We're not going to talk about the sports as much. We're going to talk about the ads. That's coming up in a moment.

00:33:26:08 - 00:33:41:10
Unknown
Okay Jeff did you watch Super Bowl. Yeah. Are you okay? That was my next question.

00:33:41:12 - 00:34:03:28
Unknown
I love the time. Show is awesome. Yeah, I absolutely love that, too. Did you did any ads stand out for you? Because that's always the thing. Is the ad spotting of, Super Bowl.

00:34:04:01 - 00:34:12:24
Unknown
Yeah. Yeah.

00:34:12:26 - 00:34:30:23
Unknown
Yeah. What what did you find so so enjoyable about, about that. Because I'll be honest I actually didn't see those ads when they happened. So I must have been, you know, I have seen the anthropic ad afterwards, but I didn't see it during the Super Bowl. And I didn't see chat or I didn't see the OpenAI.

00:34:30:23 - 00:34:51:16
Unknown
I wasn't actually realizing it.

00:34:51:18 - 00:35:18:02
Unknown
Yes. Yeah. He, he kind of went I don't know. He went public with his pissed off oddness, a little bit of that. Did you notice I definitely noticed a solid amount of ads or at least a few that that really stood out that were like kind of leaning into the generated video thing, you know, like I, when was it?

00:35:18:02 - 00:35:49:29
Unknown
I think it was a year ago. I went to Mobile World Congress in Barcelona, and as I was walking around, I was really noticing that a lot of the really big kind of booths, with their massive screens and everything you know, showing their, their promotional, you know, things that they probably spent ungodly amounts of money producing. Whatever. Ahead of the show, I noticed that a lot of them were starting to use generative AI video, and you could really tell because it kind of had that fuzziness, kind of had those the finer details, you know, the maybe not the six fingers, but, you know, that kind of stuff.

00:35:49:29 - 00:36:07:00
Unknown
The writing that was really small didn't look like actual writing unless you kind of like, relaxed your eyes into it. And I was like, okay, maybe that's a sign of things to come, at least over here, companies are are less ashamed to use it, let's say, because I think there was a little bit of shame in the in the marketplace for it for a while.

00:36:07:02 - 00:36:29:10
Unknown
And now we're seeing it on Super Bowl ads. Definitely, saw a few ads that just kind of, unabashedly kind of leaned into the AI generated nature of their their visuals and stuff. I thought that was interesting. Try this.

00:36:29:12 - 00:36:49:19
Unknown
Yeah, I know totally well that that was kind of one of the things that I was thinking too is like, okay, so how much how much pushback did that get? And if it didn't get much pushback, I haven't really I mean, other other than the people who just generally hate AI and when they catch wind of it, it's immediately, oh, well, that was absolute garbage.

00:36:49:19 - 00:37:15:07
Unknown
You know, I didn't think any of it looked horrible. There was, there was a Dunkin Donuts ad that, like, de-aged a bunch of, you know, a bunch of characters that that are very familiar, like Ben Affleck. And I think Urkel was in there and, you know, friends and Seinfeld and everything. It's real uncanny, uncanny valley type stuff where you watch this, but, you know, so it's so it's weird.

00:37:15:07 - 00:37:47:20
Unknown
But, I mean, I guess it was just an indication to me that, like, at least in the ad industry, and definitely I think we're getting there with Hollywood where they're kind of like, all right, let's put the shame to the side. These are tools we are going to use these tools. And it really reminds me of, I don't know, 30, 40 years ago, 30 years ago, whenever it was that we started to really see kind of CGI in in movies and it, you know, it looked interesting because it was different from what we had seen in Practical Effects before, but it didn't look perfect.

00:37:47:23 - 00:38:07:04
Unknown
And over time it got there. And I think that's kind of the arc that we're seeing with AI, generated video right now, maybe more compressed, but yeah.

00:38:07:06 - 00:38:16:15
Unknown
Yeah.

00:38:16:18 - 00:38:19:08
Unknown
The UCI.

00:38:19:11 - 00:38:45:24
Unknown
Yeah. Yeah. Google certainly showed off the AI right. The suit Google's I thought Google's I was pretty good because and the whole idea around their ad is like you know there's a, there's a family who's buying a new house. They've got a photo of their living room and they've got a photo of their current living room, and they're like, what would this living, this new living room look like with all of our stuff in it?

00:38:45:26 - 00:39:05:23
Unknown
And I just thought that was a really smart way to appeal to someone who might have no interest in these things to go, oh my goodness, that's actually incredibly useful. Like, I remember when I moved into this house or when we moved into this house, and it was in a horrible state of disarray because we had to do a lot of repairs and a lot of fixer upper.

00:39:05:26 - 00:39:25:17
Unknown
And, I all I had were these, you know, dark dungeon kind of photos of the living room. We're trying to decide, like paint colors and stuff. And I bring that photo into Photoshop. I'd outline the walls and around the things, and then I, you know, and it was all approximation. It didn't look great, but it was gave me enough of an idea.

00:39:25:17 - 00:39:38:21
Unknown
And now it's as easy as, like, take this picture, take that stuff, put that stuff in this picture, go. And that's super applicable. Yeah.

00:39:38:24 - 00:39:58:03
Unknown
Yeah. Yeah, totally. Yeah I agree. So yeah. I'll have to go. Oh yeah. Right.

00:39:58:05 - 00:40:11:14
Unknown
Yeah. Right.

00:40:11:17 - 00:40:16:13
Unknown
That was all. That was all it was. Yeah.

00:40:16:15 - 00:40:39:05
Unknown
So I went there. Yeah, I went there just before showtime because I was like, okay, fine, I'll go there and, you know, my name was taken, so I had to do that. Jason Howell, which is always my fallback, but it had reserve your name and the reserve your bot name. So it's i.com/your name and then i.com/whatever you want your bot's name to be.

00:40:39:05 - 00:40:52:11
Unknown
I was like what that what I and with no context it's like I have no idea what I'm setting up right now. You know, I'm choosing a name for something I have no clue. But anyways.

00:40:52:13 - 00:41:17:12
Unknown
Well, that's a good question. Yeah. I mean, clearly they're making a bet that that is going to pay off at some point. Yeah. It's not a small deal, but I mean, does it, does it compare to similar things that we were talking about when the internet was early? Because I know we were talking about stories just like this, you know, maybe not to the 85 million tune, but, you know, Pets.com or whatever.

00:41:17:15 - 00:41:34:13
Unknown
Yeah. Those early, those early domains that people were paying ungodly amounts of money for. You better have a business model. And many of them just didn't sell.

00:41:34:15 - 00:41:40:29
Unknown
Okay.

00:41:41:01 - 00:42:04:19
Unknown
Interesting. Yeah, yeah, the ad really just doesn't tell you anything other than get your get your URL, get your username. Yeah. And I guess somebody named Jason Howell. Got it while it was hot. Dang. Oh, well, I have no idea what I signed up for. It's probably for the best. Who knows, I might regret having signed up for that.

00:42:04:19 - 00:42:32:01
Unknown
But, hey, we we work in podcasting, journalism. We do all we got to do for you. Let's see here. What else do we have? We have OpenAI. Well, we kind of kind of mentioned this, but OpenAI testing ads or bringing ads to, ChatGPT coming to free and go accounts. So this includes sponsored placements that actually sit alongside the answers.

00:42:32:07 - 00:42:55:17
Unknown
So you can kind of see in this video, it's it's not integrated into the answer. It was just like below it in its own separate section. OpenAI says that the ads do not change the answers themselves. Ads target logged in US adults only. For now, it's going to they're going to avoid, topics like health, mental health, politics, sensitive topics.

00:42:55:20 - 00:43:19:11
Unknown
I would say for now, that's my own interjection. And then kind of tie into some of the conversational themes, past chats, prior ad interactions. Some of that is controllable if you're a user. So you can go to settings and say, I don't want that extra contextual layer. You know, don't use my history to to inform this. Similar to what we can do in search and stuff.

00:43:19:13 - 00:43:37:21
Unknown
And then I thought this was interesting. Free users can hide the ads in the experience in exchange for fewer daily messages. I don't know. I find that interesting. It's like, oh, you don't want to see the ads. Cool. Well, you can see you can do less, less messages with us, but, that will give you an ad free experience.

00:43:37:21 - 00:43:46:15
Unknown
So. Hey, choice is good. I suppose.

00:43:46:18 - 00:43:50:16
Unknown
Yeah. Yeah.

00:43:50:18 - 00:45:02:28
Unknown
No, not at all.

00:45:03:00 - 00:45:18:22
Unknown
Yes. Very much so. They are. Yeah.

00:45:18:24 - 00:45:39:16
Unknown
Yeah. It's interesting I'm sure if anthropic wanted to figure out how they, how they present the ads, they probably could you know maybe not into the code like where my mind went with that is like if you're creating code or you're creating an app through through cloud, do your ads go into your app? Do they create the mechanisms with which to feed the ads?

00:45:39:16 - 00:45:55:11
Unknown
Because, hey, you created this for free, so the trade off is you get your app has to have a vector for ads. I can totally see that. But maybe not for anthropic, not now.

00:45:55:13 - 00:46:02:03
Unknown
Yeah, yeah.

00:46:02:06 - 00:46:20:06
Unknown
Yeah, 100%.

00:46:20:08 - 00:47:02:27
Unknown
Yeah. Now that's a that's a biggie. Advertisers really do want to have control over that.

00:47:02:29 - 00:47:12:16
Unknown
Now.

00:47:12:18 - 00:47:36:24
Unknown
God you have to, you have to imagine at some point they're going to be like well we can make a lot of money if we just make sure that these knobs always appear and those kinds of searches. Yeah it's going to happen eventually. Maybe not everyone, but yeah that transparency is important for that.

00:47:36:26 - 00:47:55:15
Unknown
Yeah. Interesting. Yeah I bet that happens at some point. Anyways I logged into a free account. Free. I have like two accounts. I have the one that I initially fired up when I was still working at Twit to like play around with it and everything. That's like my yeah, my personal account. It's free. It's just a free account.

00:47:55:15 - 00:48:20:03
Unknown
And then I have my paid for account. I logged into my free account, which I never do because I'm always logged into my paid one. I couldn't get any ads to show, you know, so they're rolling out. They're not there yet, apparently. Or at least they weren't serving to me. Yeah. That could be.

00:48:20:06 - 00:48:47:06
Unknown
Ooh, interesting. That's an interesting wrinkle. Could happen. Well, speaking of Google, I saw this, and immediately I was like, yes, this is this is a story that that, ties right in to our conversations around world models. And the potential benefits of world models also ties into this idea of, like, synthetic data and what that can be.

00:48:47:06 - 00:49:19:01
Unknown
What that can be useful for Waymo is using DeepMind's Genie three world model to spin up realistic driving scenarios for its vehicles. And now Genie three we talked about a couple of weeks ago is their AI model system that creates little 62nd kind of interactive world models. Essentially, you could say, you know, build a forest land and roll of around chrome ball over it and make a path of blue behind it in paint or whatever.

00:49:19:01 - 00:49:49:18
Unknown
And it would create this thing. You could control it with your keyboard and it creates, you know, little 60s snippets of these experiences. While Waymo is using that Genie three kind of world model builder to generate scenarios that are difficult, if not nearly impossible to encounter in real life, but that happen in a in a way that allows their vehicles to learn what they would do if and when that happens in real life.

00:49:49:18 - 00:50:14:15
Unknown
Because there's, you know, they can drive around our streets and, you know, they might get into an accident or, or encounter a very confusing and confused intersection. These things happen all the time. But there are edge cases that don't. But yet they do happen. And these vehicles need to be trained, need to have some sort of knowledge, some sort of experience or training as far as what they do when they encounter those situations.

00:50:14:15 - 00:50:37:19
Unknown
And that's what, Waymo is using DeepMind's Genie three for. I think it's I think it's a really cool use of this and really ties into a lot of what we've been talking about, about the power of world models.

00:50:37:21 - 00:50:57:14
Unknown
Yes.

00:50:57:16 - 00:51:02:25
Unknown
Yeah.

00:51:02:27 - 00:51:09:21
Unknown
Right.

00:51:09:23 - 00:51:39:24
Unknown
Yeah, it's a good question. Working with the models. It can it can generate depth data that feeds into the vehicles. So, you know, they can kind of have that training, create the data, feed that. Right, and have it be part of it. Yeah. Yeah. That's interesting. I just I look at this and I see it as kind of like a tangible, kind of benefit of a system like this, you know, because it's easy to like a Genie three and be like, oh, that's that'll be cool for gaming.

00:51:39:24 - 00:52:02:04
Unknown
It's like, no, but it can be cool for so many other things. You know, how how solid are these models that that are created? I think that goes back to always kind of my kind of confusion around synthetic data is like how how quality is data. If it's synthetic, you know, if it's generated by an AI model, even, versus the real thing.

00:52:02:06 - 00:52:20:03
Unknown
I don't know the answer to that, but, but this seems like a cool kind of step in the in, in a direction that makes sense to me. And of course, friend of the show, Yann LeCun. Like, I'm super curious to know, like, it would be cool to have him on and like, pose this question and be like, you know, what are your thoughts on this?

00:52:20:06 - 00:52:31:10
Unknown
Yeah.

00:52:31:12 - 00:53:15:28
Unknown
Yeah.

00:53:16:01 - 00:53:47:15
Unknown
Yeah. Yeah. That's fascinating. Yeah. We should have him back. If only we were that easy. You should just come back. Yeah. Come back. We'll, we'll talk with you again. We'll give you a platform. You do it. All right. And then oh this was an interesting story. An HBR, HBR not HBO, HBR study that, looks at generative AI and workload and, super curious to hear kind of what you think about this.

00:53:47:15 - 00:54:04:16
Unknown
It basically the study looks at like, how does it actually impact our workload, like in, in daily life, right. Does it actually lighten our workload? Because so often we're saying, you know, these these AI models can do so much. It's like I don't have to do that work anymore. Now I can focus on different things and I'm freed up.

00:54:04:16 - 00:54:36:13
Unknown
I'm saving hours in my day. You know, the common thing that I hear people say speeds up work and expands it. This study basically says that means that that people are moving faster, which means that they take on more, more kinds of tasks as a result. And often that can mean that they stretch work into parts of their lives that were otherwise off limits, let's say, into the evenings or into their breaks, because doing work feels easily accessible.

00:54:36:13 - 00:54:59:16
Unknown
It feels possible, rewarding, relatively easy when you compare it to without these tools. And so people are running, you know, multiple threads all in parallel, squeezing in a quick prompt here and there, you know, oh, wow, I'm just going to pop this open because they can because it's easy to low lift. And then over time that erodes our physical and mental recovery.

00:54:59:19 - 00:55:26:00
Unknown
It kind of remove it. Even though we aren't doing that work, our brains are constantly thinking and doing through these models and filling up time with them. And, you know, it has other impacts. It raises the expectations bar of of what we're expected to be able to accomplish. It just continues to kind of reinforce this, this, cycle of more output, more and more and more.

00:55:26:03 - 00:55:47:27
Unknown
But yeah, interesting study.

00:55:48:00 - 00:55:59:10
Unknown
Yeah, it's on Harvard Business Review. But yeah.

00:55:59:12 - 00:56:06:15
Unknown
Yeah.

00:56:06:18 - 00:56:28:15
Unknown
Right. Exactly that expectations bar. Yeah.

00:56:28:17 - 00:57:06:17
Unknown
Yeah.

00:57:06:19 - 00:57:11:19
Unknown
Yes.

00:57:11:22 - 00:57:20:03
Unknown
Yeah.

00:57:20:05 - 00:57:23:23
Unknown
Yeah.

00:57:23:25 - 00:57:50:02
Unknown
Yeah. Yeah definitely. It says in here US based technology company with about 200 200 employees. So definitely US based. Yeah I'm. Yeah. Oh okay. That's interesting. Yeah. I would love to see that. That would be fascinating. I mean, I know for myself, like, I, when I kind of read through this and, first heard about it, it had been talked about on, on a daily tech news show, not an episode that I was hosting.

00:57:50:02 - 00:58:04:12
Unknown
And then I heard about it from an email and I was like, oh, I got to check that out because it because it resonated with me a little bit. And I think part of that is just because I'm feeling a little, you know, a little exhausted. I've been working my tail off for the last two years in this whole content business.

00:58:04:15 - 00:58:30:27
Unknown
And, and when I look at kind of what I have done in building this business and my efficiencies and how I do it, I mean, you know, it over time it has brought in I kind of integrations and I'm clearly not alone in this, into the process to, you know, enable me at least I feel to do more with the same amount of time.

00:58:30:29 - 00:58:52:07
Unknown
What I realized when reading through this is I really can identify, though, with this. Like the fact that like, all of this efficiency can be very fatiguing, right? Because like if I look at my browser right now, I've got like 20 tabs open, I usually have like 20 to 30 tabs open. Two thirds of those are some form of AI model.

00:58:52:07 - 00:59:08:29
Unknown
You know, I icon because of I'll have an idea for a thing that I need to do. And then I open up a new new chat window I fired off. Get that going. I have another idea in the middle of that one. Open up a new one, start that thing. Go back. Try and remember what was I doing there.

00:59:08:29 - 00:59:40:09
Unknown
Like there's a lot of like context switching and you know, then I go back to one from like five days ago and I'm like, wait a minute. That seemed really important at the time. What where was I on that one? I need to pick up on that one. And, yeah, it can be, you know, and then suddenly and this whole idea of, like, working with these models is so easy that we feel like in the moment, it's not work to open up a tab and to fire this thing off, you know, like, if I was actively researching something, that would take me a lot of time.

00:59:40:09 - 01:00:01:26
Unknown
But here you research that, but still it's in our mind, it's in our consciousness, and we're spending our energy filling our time off with that stuff. And that has a cost to it that that resonated with me. Yeah.

01:00:01:28 - 01:00:10:12
Unknown
Yeah. Yeah.

01:00:10:15 - 01:00:25:27
Unknown
Oh yeah. For sure. I hope that it does. Yeah. Do you have multiple tabs open with different AI models in each of them.

01:00:25:29 - 01:00:29:09
Unknown
Yeah.

01:00:29:12 - 01:00:38:16
Unknown
Yeah.

01:00:38:19 - 01:01:01:24
Unknown
Yeah yeah yeah I know I know I'm not alone. I know there are plenty of people out there that probably look up at their browser and that, you know, the browser tab row and see like two thirds of them are some form of AI model. Yeah. What next, what next? What's the next task? It here. Yeah.

01:01:01:26 - 01:01:06:04
Unknown
Just.

01:01:06:06 - 01:01:24:06
Unknown
Right.

01:01:24:09 - 01:01:31:28
Unknown
Yeah.

01:01:32:01 - 01:01:54:01
Unknown
Here's an assistant to tell it to teach you how to have an assistant. What? To have an assistant use. That's that's what we need. Yeah. All right. We're going to take a quick break before we do, review on on Apple Podcasts. That's your homework assignment for today. Give us a review, log in Apple Podcasts if that's where you get this podcast or if your podcast has the ability for reviews or ratings or anything like that.

01:01:54:01 - 01:02:09:19
Unknown
It just really helps expand the scope and expand the, the, awareness of what we're doing with AI insights. So please do that. We appreciate it. Good to take a break and then talk about a few stories to, round things out. After a moment.

01:02:09:21 - 01:02:40:25
Unknown
Reuters reporting on a Nature Medicine study that shows that AI is mislabeling body parts and having a negative impact on surgeries in operating rooms. You know, in the O.R., the study looked at 1298 people in the UK. Those people were randomly assigned GPT four zero, Lama three Coheres command R, and they were asked to make decisions about a medical scenario as though they had encountered it at home.

01:02:40:25 - 01:03:08:07
Unknown
They missed two thirds of relevant conditions. When people describe their symptoms, using their own unstructured words. So if people were just kind of talking to it or whatever, the AI models missed two thirds of the relevant conditions. The Am models were very good, though. When they were fed structured clinician grade data, they were able to identify 94.9% of the cases.

01:03:08:09 - 01:03:16:00
Unknown
In that scenario. So are we surprised?

01:03:16:02 - 01:03:25:03
Unknown
That's true. So true. That's a great way to put it.

01:03:25:06 - 01:03:34:19
Unknown
Yeah.

01:03:34:22 - 01:03:37:13
Unknown
Yeah.

01:03:37:16 - 01:04:18:11
Unknown
Love the pit. Absolutely love it. Yes. Yeah.

01:04:18:13 - 01:04:36:15
Unknown
Right. Rely. Yeah. Rely entirely upon it. It's the final word and all that. Yeah.

01:04:36:18 - 01:04:43:24
Unknown
Yeah. There you go.

01:04:43:26 - 01:05:08:04
Unknown
Yeah. Indeed. So do I, so do I. And I think so do a lot of people out there. Meta spinning its vibes. Service which you can get in meta AI which is this is essentially it's kind of like generated video and generated photos, feed similar to what OpenAI has with Sora. And it existed inside of the meta AI app.

01:05:08:04 - 01:05:20:06
Unknown
Now they're going to spin it out into its own standalone app. And, yeah, I don't know what that concept.

01:05:20:08 - 01:05:22:19
Unknown
Yeah.

01:05:22:22 - 01:05:26:23
Unknown
Yeah.

01:05:26:25 - 01:05:35:16
Unknown
Yeah, that's that was that was the next thing I was going to mention. It's been a while. It's been a while.

01:05:35:19 - 01:05:42:19
Unknown
Yeah.

01:05:42:21 - 01:06:06:11
Unknown
You started to kind of see a lot of similar kind of, you know, returns and similar output, and. Yeah, I just kind of lost interest. It was it was really fun and interesting to me for, for a short period of time. I don't know if my fall off is because the product like isn't interesting anymore, or if I just got so busy that I that I stopped like making the time for it or spending time in it.

01:06:06:11 - 01:06:11:27
Unknown
You know?

01:06:12:00 - 01:06:16:18
Unknown
That's a great way. Yeah, I love that. Yeah.

01:06:16:20 - 01:06:19:23
Unknown
Every single day.

01:06:19:25 - 01:06:30:13
Unknown
Yeah.

01:06:30:16 - 01:07:03:26
Unknown
Yeah. Well, I wonder how closely they're looking at the drop with Sora. Because installs, apparently, fell 32% in December, 45% in January. Yeah, yeah. So, you know, I guess they call a slot for a reason, right? Because. Yeah. Alphabet is, like, as semaphore says, go along with 100 year bond tapping, ultra long. That is 100 years.

01:07:03:26 - 01:07:30:19
Unknown
Ultra cheap debt to lock in funding for its AI and data center kind of ambitions and build out and stuff. They're issuing rare 100 year bonds at rates that are actually lower than your typical 30 year mortgage, raising $32 billion in two days. Apparently. Obviously it's a long investment horizon. Really, you know, a strong choice, I'd say, although I'm not an investor or financial advisor.

01:07:30:19 - 01:07:45:25
Unknown
Don't listen to me. For conservative investors who want to go real conservative over time, but it is for people who actually believe that Google's going to be around for 100 years. How do you feel about that, Jeff? Yeah, yeah.

01:07:45:28 - 01:07:55:25
Unknown
Yeah, yeah yeah.

01:07:55:27 - 01:08:27:06
Unknown
It's Wow. Okay. Yes.

01:08:27:09 - 01:08:41:24
Unknown
Yeah. And so if it's not because that 100 years is a long time, then what happens, that those bonds become worthless.

01:08:41:26 - 01:09:17:24
Unknown
Yeah. Yeah. Interesting. I didn't see that one coming. Google, by the way, sits on more than $100 billion in cash. And, so it's holding on to that for things like deals, buybacks, shocks to the business. And then the bot, the income or the money that they gather from the bonds, selling the bonds goes to some of their current kind of needs, like building out data centers, which is a costly affair in this moment in time, apparently.

01:09:17:26 - 01:09:42:23
Unknown
And finally, OpenAI has dropped the I name from its upcoming hardware device. This is the device that, OpenAI is developing in collaboration with Johnny Ive's design firm. And, this was revealed in a court filing that's related to a copyright lawsuit with an audio startup. We definitely talked about this months ago. That audio startup is called IO.

01:09:42:26 - 01:09:59:09
Unknown
I y o is how you spell that. But when you say it out loud, it's io. And OpenAI's hardware was going to be named IO and apparently no longer they want it to go away. Yeah.

01:09:59:12 - 01:10:25:28
Unknown
Yeah. With Google IO, I, I suppose so, I wonder, but that's not like a product. So that might not, that might be different enough. But OpenAI. Yeah. Yeah okay okay. They can't call it what they wanted to call it where they expect this hardware to hit the market sometime around February 2027. So next February, a year from now.

01:10:25:28 - 01:10:52:25
Unknown
Originally they expected this to hit sometime later this year. So so it's going to get get some delays. But that's just how it goes. All right Jeff that's it. We've talked about all the AI news that exists in the world ever for the last week. Every every single bit of it. We didn't miss a single story. That's what we do every Wednesday on AI inside except for next Wednesday because we're taking the week off.

01:10:52:27 - 01:11:10:17
Unknown
I'm sorry to to have to do it, but I, I have a family trip coming up. We're going to Tahoe for a week, and I was flirting with the idea of doing the podcast. I think I did this last year when we were at, winter Park and podcast it and everything. And I'm going to be honest, I just need a break.

01:11:10:19 - 01:11:18:03
Unknown
I need a week. I've been working my tail off and I need to. I need to take a breather. So.

01:11:18:06 - 01:11:31:09
Unknown
This is. Thank you. Thank you. Jeff.

01:11:31:11 - 01:11:50:28
Unknown
Hahahahahahaha. Wow. Yeah. See, I probably need to just hire you to do all my marketing because, it does not come easily to me. Yes. Pod tune up.com. If you got a podcast, you want to talk about podcasting. Hey, I think I've proven for the last 20 years I know a thing or two about podcasting. I can help you with yours.

01:11:50:28 - 01:12:14:25
Unknown
So go to pod tune up.com Jeff can be found at Jeff jarvis.com. The book that we've heard of a few times throughout this episode. That's awesome I can't wait. Hot Tape, The Magnificent Machine that gave birth to mass media and drove Mark Twain mad. You can preorder also Gutenberg parenthesis out mag magazine out the web. We we've out.

01:12:14:28 - 01:12:41:04
Unknown
Jeff's busy guy, Jeff Jarvis. Dot com. Thank you. Jeff. I hope you sell a million copies. Throw it back to you. You can also go to AI inside Dot show to find the everything you need to know about this particular show. Waste subscribe any reviews. Do we get any any newer reviews now? Still. May 18th, 2025. See, this is what I'm waiting for.

01:12:41:04 - 01:13:02:16
Unknown
I'm waiting for whoever does the next review. It's going to appear right here on our website. That's. That's what I'm waiting for, this one to get replaced. Not replaced, but pushed over here. Anyways, I'll keep saying it. It'll happen one of these days. And finally, Patreon.com slash AI inside show where you can go and support us on a deep, level.

01:13:02:16 - 01:13:24:12
Unknown
However deep you want to go, you can get ad free episodes of the podcast, access to our discord, and then if you go real deep, you become an executive producer. A like so many amazing executive producers who got a t shirt doctor do Jeffrey Barrett, Cheyney, Radio Asheville, 1 or 3.7, Dante, Saint, James Bond, Derek, Jason night for Jason Brady, Anthony Downs, Mark Archer and Carsten Szymanski.

01:13:24:12 - 01:13:46:09
Unknown
Thank you so much for continuing to support us each and every week and every month. We just couldn't do it without you. So thank you for that. All right Jeff, heal up. Feel better. Got an extra week to do it. Everybody watching and listening. We'll be back in two weeks okay I promise I will stay safe, and, I'll see you next time.

01:13:46:11 - 01:13:50:17
Unknown
In two weeks on another episode of the AI and side podcast. Take care everybody.