The AI Bubble Lives Another Day without Bursting - DTNS 5150

0:03This is the Daily Tech News for Thursday, November 20th, 2025. 5. We're going to tell you what you need to know, give you some important context and help you understand a bunch of stuff today. Lots of stuff. Today, did NVIDIA earnings actually mean that we avoided the AI bubble bursting? We'll talk about that. Also, I got a first look at the Nano Banana Pro. And so I'm going to peel it back for you. Yeah, this show is crazy. It's 5150, which is actually the episode number. I'm Tom Merritt. I'm Jason Howell. Let's start with what you need to know what the big story is there an ai bubble no one knows everyone wonders the entire tech industry held its breath wednesday to find out as nvidia's earnings were expected to point to the answer now spoiler they don't point to an answer any more than any one company's earnings reports can at least in my opinion but the good news is nvidia's earnings were good so the market didn't panic and the boom did not go bust for now.

1:06First, let's talk about the NVIDIA details themselves. The company beat expectations for sales and earnings. And I think more importantly for people who are bubble watching, promised better than expected projections for the next quarter. Net income rose 65 % compared to last year. NVIDIA doing fine, continues to be the the most valuable publicly traded company. For bubble watchers, though, the important numbers were for data center sales. If somehow NVIDIA had a great quarter, but it was all made up of selling GPUs to gamers, then the bubble talk would have got more serious. That was not the case.

1:41NVIDIA's data center sales rose 66 % on the year. Most of the sales were for its GPUs. About 20 % was networking products. You know, the things you put the GPUs in and connect to the GPUs. that is a growing part of nvidia's data center sales the company's best-selling chip is as expected the blackwell ultra the only one that disappointed in sales was the h20 which was the one designed to sell to china and they couldn't sell it to china so it's not big surprise that it didn't sell a lot uh if you're curious about that gaming segment it did well gaming sales up 30 on the year uh it makes up about 10 of nvidia's revenue now isn't that crazy jason wow the gaming the thing that made NVIDIA 10%.

2:26One of the newest segments is robotics. That was up 32 % on the year. Still makes up a very small part of the business, but growing. And like all other tech companies that reported earnings this month, NVIDIA is doubling its cloud spending commitment. I mean, they're not all doubling, but they're all increasing cloud spending. NVIDIA CEO Jensen Huang said on the earnings call, there's been a lot of talk about an AI bubble from our vantage point. we see something very different. Yeah, he kind of led with that. That was like at the top. It's like, hey, by the way, let's get in front of the AI bubble talk.

2:59Now, we could parse this a little too much, but he didn't deny. He said, there's been a lot of talk about it. We see something different. Right, that's true. I don't know. That's a good point. Or maybe it's a Taurus, not a bubble. Anyway, he was very optimistic. Maybe it's a balloon. Maybe. Let's remember, most of the reasonable bubble fears are that so-called hyperscalers, like OpenAI, Microsoft, et cetera, are spending too much money to build infrastructure without a clear path to profitability, right? So they're throwing all this money and commitments to like, we're going to give money to Oracle.

3:33We're going to give money to Microsoft. We're going to spend, spend, spend. And they're spending a lot of it on NVIDIA because NVIDIA is selling pickaxes to the gold miners. So NVIDIA should be making record money right now. It would be later when the hyperscalers might falter that you would see an impact on NVIDIA because they would stop buying those GPUs. NVIDIA's actual near-term concerns are about supply constraints. It is sold out on data center GPUs, it said. And so it's trying to figure out the logistics to make sure they can keep providing the products that people want to buy from them.

4:08They also have a problem that I think, Jason, you and I would like to have too much cash, too much money flowing in. What are they going to do with all that money? And they talk about that because using it to invest in startups has caused concerns about a circular investment where, you know, they loan money to somebody who's going to use it to buy stuff from NVIDIA, right? So what other signs can we look for for our bubble estimations? Server shipments from Taiwan to the U.S. are set to double this year compared to last. So it's not just NVIDIA. Foxconn, Quanta, and Wistron all ramping up production, according to Nikkei Asia.

4:46That is propelling Taiwan to expect record export orders this year, according to Taiwan's economic ministry. That with U.S. tariffs. Imagine if there were no tariffs, that record export would be even bigger. Yeah. Again, that's all what you'd expect with hyperscalers building out. It's going to be, in my guess, 2027 before we might see the first cracks in this model, if there are going to be cracks. Yeah, no, you're absolutely right, Tom. The real bubble story, if the bubble exists, you know, because everybody has a different opinion. If that exists, it's yet to be played out. Everybody is so hyper-focused on that scale component right now.

5:27And as long as that's the case, at least related to NVIDIA, NVIDIA is in an excellent position because they're continuing to buy and to invest in NVIDIA and that circular financing that you're talking about. And really just like scrambling to keep up with the pace of that scale. NVIDIA stands to benefit a lot. If the dominoes begin to fall somewhere on the other side, and I'm not saying that's guaranteed to happen. I do think that there will be a correction, but that's when the companies themselves are going to start to really face the dead end, I suppose. If that scaling leads, it's a path to nowhere.

6:06I don't know. That certainly has a dramatic impact on the industry. I don't know what the – I mean, the impact would be dramatic on NVIDIA as well, but it's making its bag while it's canned right now. Yeah, it has a long way to fall before it would actually really suffer. So, you know, it's all relative when you're at that valuation. I am using the dot-com boom as my model right now. Totally. I think that fits this one pretty well. And the first thing you will see is probably some startups, some smaller startups going out of business. When that happens, then look to NVIDIA and the hardware makers to see if their projections start to lower as they realize, well, there's less demand for the services.

6:55And so the big guys are slowing down their build outs. I suspect that we will have a few Googles, Amazons, and Netflix, which everybody forgets. Those are all dot-com era companies. I feel like we'll have a few of those. Your candidates are anthropic and open AI and such. You'll have some Yahoo's, which will survive, but barely through this. And then you'll have your pets.coms that will just crash and burn, even though they reach great heights. Yeah. I, in my head, I think of anthropic and open AI at the top of that bubble. I feel like mistral and those are kind of like riding the surface and perplexity might be inside that bubble.

7:40I don't know. I think, I think so. Yeah. Adding perplexity in there makes sense too. And I think one thing we get caught up on this idea of a bubble and possibly it's not a bubble as much as it is just a correction, right? That doesn't mean that the world comes crumbling down it's more of a roller coaster right yeah exactly or it's more like waves in the ocean you know sometimes it's up sometimes it's down but it doesn't ever go away the way the ocean doesn't ever dry up entirely um and personally i think that's what we're probably looking at and along with that yes there there will be some adjustments on the lower end of the the kind of like the entry level uh players not being able to compete in in those waters if we want to continue the analogy.

8:23But yeah, I think that feels more real to me. We're headed for Mavericks is what you're saying. A bunch of people might wipe out, but somebody's going to ride that wave. Oh, ride the wave, bro. Yep. Interesting stuff. Well, DTNS is made possible by you, the listener. That's right. You want to give our big thanks to Brandon Brooks, Johnny Hernandez, High Tech Oki, and Stephen Howe. Thank you so much. Thanks for surfing the waves along with us. All right. There's more we need to know today. Let's get to the briefs. And apologies about my voice. I've got a kind of a throaty thing happening right now.

9:01So I'm doing my best. Oh, 40 years ago today, November 20th, 1985, little history lesson for you. Microsoft launched Windows 1.0. Happy birthday, Windows. You come a long way, baby. You're so cute. And you're only getting more features with age. That's a nice way to put it. That's like all of us. More features. Totally. More features, yes. Deep, deep etched features. For example, Qualcomm is releasing its Snapdragon control panel this week, which optimizes games you install on a system with Snapdragon X Elite, lets you install the latest Adreno GPU drivers to improve game performance. Along with that, Microsoft's Prism emulator supports x86 Advanced Vector Extensions, or AVX and AVX2 emulation.

9:49that will expand the number of games and vector graphics based apps you can run in windows on arm and fortnight will be able to run in windows on arm with kernel level anti-cheat support more anti-cheat support is on the way as well by the way and you can get fortnight through the xbox store which now lets you access arm 64 games as well yeah i mean uh i feel like if you're a gamer the criticism is, are we there yet? And the answer is like, no, but look, we're starting to see snow on the mountains. We're getting closer, right? We're getting there. Yes, exactly. You can do more. You can't just not think about it.

10:29There are going to be some games that won't work. There are going to be some games that technically work, but maybe they don't work as well, but you're getting more and more of them. I think Fortnite showing up is a big one. Yeah, indeed. I agree. Austin's Paradromics, or Paradromics, received US FDA approval for ConnectOne, which is an early stage human trial of its Connexus brain computer interface. This is a brain implant. The test will assess long-term safety and how well Connexus can synthesize speech and text communication. So the implant is a metal disc smaller than a dime, like a very tiny coin, with 421 micro wire electrodes.

11:09It decodes brain signals and then sends them out to a model. So it's going to ask you as the patient to try to move your mouth. These are patients who can't move their mouth to speak, but try to move your mouth. And then it translates those impulses into machine generated speech or text. Trials will begin early next year in two people with impaired speech due to loss of voluntary motor control. That will go for six months at UC Davis in Sacramento, Massachusetts General Hospital in Boston. And I know they said only two people, but they listed three locations, University of Michigan in Ann Arbor.

11:45If that goes well, Pendromics hopes to expand the trial to more volunteers. Paradromics thinks it can reach about 60 words per minute with this system. That is a little less than half normal speaking speed, but it is a good amount of speed where you can carry on a conversation. Paradromics is one of a number of companies in varying levels and products for brain computer interfaces. Some are implants, some rest on the skull. And those include Cognition, Neuralink, Precision Neuroscience, and Synchron. What a time for technology right now. This stuff feels super sci-fi, a little nervousness around the let's go ahead and throw technology inside of your brain sort of thing.

12:28But honestly, stories like these where we're using modern technology, the advances that we have, be it, I know this doesn't directly mention anything about AI, but there's a lot of advancements happening in AI that apply in similar ways to health. It's a form of artificial intelligence that's interpreting those mouth movement signals and saying, oh, that must be this phoneme, et cetera, et cetera. Yeah. Yeah. I mean, these stories get me really excited about the capability and the potential of these technologies beyond like, write me a paragraph. You know what I mean? Like this is stuff that really has legitimate, deep impact for people who couldn't talk because they have a debilitating condition.

13:05That's just incredible. I'll try not to be snotty because I don't want to get Mel and people like Mel upset again. But this is an example of why we do need AI. When people are railing against AI, not all AI is the same, right? AI is a big old catch word that is a big old trash can full of a lot of different kinds of stuff. And this kind of stuff, which I believe is more machine learning than a large language model, although it does have LLMs. I read enough that it is using LLMs for some of the speech construction is great and useful and making the world a better place. Yes, indeed. Making things that didn't seem possible before possible.

13:46I think that's really incredible. California's Sunday Robotics plans to put its humanoid home robot called Memo. Hello, Memo. In people's homes starting next year, the bots have a cartoon face and a colored baseball cap. That looks pretty funny in my opinion. It looks a little bit like Homestar Runner if you take a look at it. They have two arms but move on a wheeled platform. They can slide up and down on a central column to get taller or shorter. I wonder if it can get six foot eight inches tall, you know, to match my height. The robot is trained on video from paid remote workers who use special gloves that gather data while they do all kinds of tasks.

14:28Sunday is creating this so that memo can carry a lot of those tasks out. That data is combined with training data from the robot's own sensors to make a model for its movements. It's capable of doing household tasks, things like operating an espresso machine, cleaning glasses, putting them in the dishwasher. You got to be really delicate, really touch lightly with those dishes and those glasses. Pretty impressive stuff. Complex tasks that require coordination, dexterity. And Sunday has been able to demonstrate all this. Sunday thinks the robots will be able to help with laundry and other tasks as well.

15:05Other robotics companies, of course, trying to achieve the same level of capability, including generalist, physical intelligence and skilled. And Sunday is planning to give memo to beta testers next year and then start rolling it out to the first set of users. So home robots, this is a topic that is bubbling up more and more, and it has long felt like sci-fi Jetson's future, but now it's kind of within the, oh, these things could be coming to someone's home, probably not my home, but someone's home next year, the year after. That's crazy. These can reach up to six foot 10. I looked it up while you were talking.

15:44Thank you. Finally, someone thinks of the tall people. Yeah, they can be taller than you even. I really like how these are trained. This is another example of a different spin on that big old catch all of AI, like providing LLMs along with computer vision data and capturing actual motions that are useful for the robot instead of just having the robot be teleoperated, I thought was an interesting way of doing this. and it seems to have paid off, I would like to try one out. I imagine these are going to be, you know, VIC-20 level. Yeah, totally. Some rough edges. You can do some stuff with them, but maybe they can't do everything.

16:30Yeah. But yeah, the demonstrations they showed on the Wired article were pretty impressive. So I think we're getting closer and closer, and it's these kinds of advances. This has got folks who used to work at Tesla and DeepMind and such going and saying, Let's apply what we've learned to a particular problem. I love that. I think people might actually be open to buying a home robot that can only do your laundry and do your dishes. If you could just do those two things, like that's enough for a lot of people. I spend a ton of time doing dishes. My wife spends a ton of time doing laundry. We kind of like split our tasks in the house.

17:05That would be amazing. Although I don't know how expensive it's going to be. And they're saying you might have to show it how to do some things. You might have to teach it how to do some things, especially, you know, if there's something unique about your layout, putting this in a home with pets and kids and stuff is going to be a whole lot different than a lab. And that's where the rough edges come in. But they're willing to try it out and beta test it. And I think they're getting close. Interesting. Which is good because we live in the real world, Jason. Years ago, Professor Nick Bostrom calculated there was a high probability we were living in a simulation.

17:37but there was always a chance this was all real and mathematicians at the university of british okanagan have published a paper proving that we probably are real they used goodell's incompleteness theorem to calculate that reality requires and i quote non-algorithmic understanding something that a computer can't replicate i'm going to oversimplify the explanation a lot but it is backed by their math. If you want to read their article, some truths cannot be reached by a sequence of logical operations alone. This is called a Goodellian truth. The classic example is something along the lines of this true statement is not provable.

18:21So if you prove the statement, then the statement is false, not true. Or if you can't prove the statement, well, then the statement is true and it's not provable and you couldn't prove it. the study was published in the journal of holography applications in physics okay so because living in this universe brings forth with it unprovable truths and and as it says an algorithm could never operate in that kind of world that means that the universe cannot be a simulation if we were sort of if we were living in a there's a lot more I'm sure to it in their paper, but yes. Yeah, I'm sure. I'm trying to simplify for the most part.

19:04If we were living in a simulation, the simulation would simulate logic that would tell us we are not. Tom. That's why you need to read their paper because they actually addressed that objection in there. Okay, fine. And show mathematically why that doesn't – they were able to prove this, right? Basically, what you and I just did was kind of a hypothesis that's like, if you can prove this was true, then that would prove that we're in reality. And what their paper does is prove that that's true. I trust them. I'm not going to do any more homework on this. I trust other physicists to check their math.

19:41Yes, exactly. And in a more down-to-earth story, real-life Earth story, the Connectivity Standards Alliance announced Thursday that Matter 1.5 will add smart cameras to the standard. That would be an awesome addition as that means you could buy any camera, set it up on any Matter-compliant smart home system, and then access the basic video. So no clips, no analysis, like facial recognition from any company's smart home app. So Samsung, Eve Home, Acara, Utec have already said they will support it with their cameras and apps. And, you know, Google, Amazon, Apple don't quite know yet, but I guess we'll find out.

20:25I have to imagine at some point they probably would want to hop on the standard as well. That would be awesome. And they're part of the standards alliance or at least a promoter. Samsung and Apple are promoters. Google and Amazon are part of the alliance. um so they all said things about how great this is and and they will continue to evolve their product lines to continue to support matter in all the possible ways but they didn't come right out and say like oh yeah no we're gonna have cameras that support this um i think they will and part of it is because they are involved in the standard so much but also part of it is they can reserve some of that like yeah but if you want to know if that was your dog that walked in front of the camera, you got to use the Google app.

21:06Yeah, right. Matter app is only going to let you look at the video. It's not going to do all the advanced features. Better than nothing. Yeah, yeah, no, it's a good step. And this could be the kind of thing that really propels Matter to wider acceptance. Yeah, for sure. Finally, some quick headlines. These are just good to know if you want to understand the news in the future. Samsung published a list of model numbers, which Android Authority interprets to mean you're going to get the S25, the S25 +, the S25 Edge and the S25 Ultra next year. Man, it's so confusing. I feel like that's flip-flopped all over the place.

21:40Along with a Z Flip 7 FE, Z Flip 7, Z Fold 7, and yes, a Trifold. These are not surprises, but yeah, okay. It's a bit confusing though. It's fun to see that stuff. A court in Spain has ruled that Meta must pay news companies in Spain a total of 481 million euros in damages for violating data protection laws to gain an unfair market advantage in ad sales. So they're not getting fined for violating the data protection. They're saying when you did that, you were anti-competitive in ad sales. And so you owe damages to these news companies. Meta, of course, will appeal. Speaking of Meta, we knew Meta's AI research head, Jan LeCun, was leaving at the end of the year.

22:29We had heard those reports, but Bloomberg reports that his startup is going to focus on systems that can understand the world, world models. That's what they call that. Understand the world around you. And that company will partner with Meta when that happens. That's the important part there at the end, right? Like, oh, he's leaving with their permission so that he can do a thing that they will benefit with. That's why it's all so friendly. Wall Street Journal has an article about the U.S. military testing a device that can measure your position by firing lasers at atoms to measure the Earth's magnetic field in real time.

23:04This is being pitched as a backup to GPS for times when, like, GPS is unavailable or being jammed. Oh, you know, lasers firing everywhere. What could possibly go wrong? waymo continues to expand announcing it will start testing and validating its autonomous cars with human drivers in minneapolis new orleans and tampa major league baseball added netflix as a tv partner more live sports coming to netflix a three-year deal for the streamer uh it will show the opening night baseball game there's an opening day and an opening night game uh netflix will also show the Home Run Derby and one of the special games every year.

23:43This year, it'll be the Field of Dreams game in Iowa. OpenAI has created a version of ChatGPT that gives teachers free unlimited access to GPT 5.1 through June 2027 with compliance for the U.S. Family Education Rights Act. Universal, Sony, and Warner Music Groups have all three signed a deal with a startup called Klay, K-L-A-Y, to license works to train a model that people can then use to remix songs. Also, Warner Music Group has reached an agreement with generative music platform Udio, and we're going to dig into that more with Andy Beach on Friday's show. Gotta hear that. Uber will start using robots from Starship Technologies to offer sidewalk robot deliveries in Europe, starting with the UK in December.

24:29All right. Those are the essentials for today. We're going to dive a little deeper. Google added a feature to Gemini that lets you upload an image to tell if it contains any trace of the synth ID watermark. So you can tell if it was machine made. That's good because Google is also contributing to the proliferation of machine generated images. Following the launch of Gemini 3 earlier this week, Google announced the launch of its image generator, which has an official name, but everybody calls Nano Banana Pro, including Google. It's available for free in the Gemini app with usage limits, as well as in writing assistant, notebook LM and developer enterprise and advertising products as well.

25:05But the cool news is that Jason got briefed on Google by this. So here you're going to tell us what's new in the banana. Yeah, it was fun. Yesterday was such an unexpected day for me, but I spent the most of yesterday at Google's San Francisco office. And part of the visit, I mean, it all had to do with Gemini. So they were very, very happy with the reception of Gemini 3, of course. Everybody freaked out about that on Tuesday, as we talked about on the show. And it was ahead of this morning's release of Nano Banana Pro. So big week for Google. There was a little bit of nervousness too. They're like, we just don't know how it's going to be received.

25:39I'm like, you know what? Viral success of Nano Banana. I think you're going to be all right, given it doesn't just go horribly wrong. And I think so far it looks pretty good. So I got to kind of test it slightly, but you'll kind of see the results of it. So I can give you a little bit of insight into what makes Nano Banana Pro so interesting. You know, it's just a lot of a lot of I mean, as all of these things are right, like the next improvement, it's bigger, better, faster. It does these things more, but it can do some really impressive things like text text representation is off the charts.

26:12Good now. They also kind of told me that like, I don't know, some people might not like to hear this, But if you see a font and you want to do text in that font, you can load an image of something in that font and then say, now make that font say this, and it will understand how to kind of interpret it, things like that. improvements to aspect ratio adherence. They told me in the range of 90 % because I've had issues creating images and saying 16.9 and then it gives me something that's not quite 16.9 and everything. The model is now very, very good at that. Quality of things like shifting focus.

26:51So take, you know, one of the examples that they showed was it's an image of a girl kind of sitting in a flowery field and you can see the flowers in front of her and the focus is on the girl. So the flowers in front are blurred and taking that image and saying, all right, shift the focus to this subject or whatever, and having the image stay consistent, but do the camera move or do the kind of shift, keeping things consistent. I think consistency is a really big part of this model. Um, they also, you know, they, they took a picture of me, a selfie of me, uh, you know, by a white wall, and then they turned it into a Nana banana, um, holiday like picture.

27:34And through that conversation that I had with them and tell me what you think about this, cause I'm sure I'm guessing at some point you've ran your image through an image generator or video generator. And, you know, we see ourselves so specifically in this world. Like we look in a mirror. We are very familiar with our faces. So when we see ourselves represented in an AI representation, it's always just slightly off. And other people might not see those differences, but we definitely do because it's our image. And I mean, the consistency on the images that they were creating around me, I was like, oh my goodness, I'm having a real hard time seeing in that that that's not me.

28:12And so just the quality, obviously getting there. Another really interesting example was around creating infographic content, which my understanding is they didn't focus on this at all. This was just byproduct where it was like, oh, my goodness, this thing is amazing at creating infographic. And what was impressive about it is, you know, I can't remember the exact example, but it was like some genus, you know, some like infographic about, you know, some animal or something like that. And it had, you know, it looked like a totally legit infographic. All the text, I was scouring that thing, looking for spelling errors.

28:48There was nothing. There was no alien text. Everything was legible. The layout stylistically, it was super cohesive. It was interesting. All that to say, like Nano Banana Pro, as you can probably imagine, just better than the one that came before it. And the one that came before it was already pretty good. That's why people freaked out and it got its viral moment. I think this is going to be a really big deal. Great for marketing, much to the behest of people who want this to not be done with AI. I mean, these things are getting so good. You can load a logo and then you can load a bottle. I think you can load up to eight images and combine them into a single thing.

29:29I read 14, actually. Oh, okay. I could be wrong then. Yeah. In the article today, it said you could blend 14 and it would maintain consistency on five characters. Yes. And OK, there you go. Consistency of the characters. OK, maybe that's what I'm thinking. Just the examples that they show were just really impressive. Now, granted, you go to something like that, they're going to show you the examples that rise to the top. Right. They're going to show you the cream of the crop. Oh, man, I wish I had written down the the name. They had a terminology for it to explain the kind of the percentage of the pickable images in that scenario.

30:06And I thought the terminology was funny. I'll have to dig it up. Can't remember what it was. But they said that it used to be that if you created 10 images, two of them might be pickable. And now it's kind of flopped. Now it's like eight of them are pickable and two of them might not be. So just all in all, things are just improving at a tremendous rate. And yeah, pretty interesting stuff. Did you get a sense of what use the infographic, like how the infographic thing would work? Because I saw the examples they posted. I'm like, yeah, those look really good. But like, how do I take my resume and turn it into an infographic?

30:40What's the prompt for that? Well, yeah, I mean, that's that's a really good question. I couldn't tell you the exact prompt. Obviously, with everything, you know, you kind of got to I imagine because I haven't I haven't like sat down with it and tested because I definitely want to test that out. But, you know, you probably want to play around with it a little bit. And one thing that they did mention that maybe ties into that is that this model is really good at giving it something as a as a inspiration point. So feeding it an image and saying, all right, this is kind of what I'm going for, but here's what I want it to actually be about or here's what I want it to represent.

31:18You know, and I guess that style transfer is probably what it is more than anything, but it's become really good at that. So I imagine if you wanted to, I don't know, this is how they did the infographic thing, but I'm sure you could show it an example of an infographic that you like or that you're going for and then say, all right, now starting for square one, this is the thing that I want it to represent. Yeah. And because it maintains consistency, you could probably, you know, again, they used resume as an example. So I'll use that. Say, turn my resume into an infographic. It does it. And you're like, actually, now that I see it, I want you to focus more on my experience.

31:51And then it changes it. That's pretty interesting. They did point out that quality degradation happens the longer you push the same conversation. That's something that they're actively working on. So the deeper you get into a conversation and iterating and everything like that, at some point things start to fall off and get a little fuzzy. One thing I thought was really interesting, it was a conversation that kept coming up throughout the afternoon and evening, is this idea of Nano Banana Pro being its own checkpoint inside of Gemini products instead of just being a part of Gemini. And what I thought was interesting about that is just kind of the calculus that they make on the inside from the perspective of, do we put all our time into making a better model that we release a point release somewhere down the line or a whole number release down the line?

32:40Or do we spend that time merging it into Gemini so that you don't have to switch modes that you can just say, create this image and it just does it integrated in all the Gemini products. And right now they're very firmly in the, let's keep iterating, let's keep making it better. And maybe eventually these things all collapse into a single product. It's just a matter of when, right? Yeah, I would imagine so. I also, you were telling me that it can do some really interesting localization of stuff. Yeah. Well, yes. So you could, again, kind of talking about like ad campaign or something like that, you could create something and tell it about, if you've got a campaign and then you want to target a different country, you could potentially just say that.

33:25I don't know how you word that, by the way, but just the concept of it. Not only would it change the text to match the localization, but it would also shift along a bunch of other details along with that. So would it be clothing styles of where that place would be or the streetscapes, the look and feel of the place? And yeah, it's just crazy how far these things have developed in such a short amount a rainy Paris cafe, you know? Yeah, yeah, totally. Right. And it knows all the, all this, the little details that make that convincing and unbelievable. Well, cool. I'm glad you got the inside scoop on this.

34:02It's much more interesting and much more substantive to, to get that, that kind of look at it. So I think it was, it was interesting for me because I, it was an opportunity for me to get close to the people who are designing these systems and understand kind of what their mentality is going into them, we see the product on the other side and we're like, oh, that's a neat product. But to hear them talk about the process of how they iterated, you know, one of the things that really stood out, and I know we're running short on time already, but I asked about Gemini 3 and I said, did you use Gemini 3 to test itself?

34:37Like they were talking about all this information they were getting from all these different places of people that were using it and sharing notes and everything like that. And in my head, I was like, well, you know, Gemini 3 would probably be pretty good at like synthesizing that. And they said, absolutely they do. So we're getting to this point to where the models recursively are helping to develop themselves. And that's pretty crazy. That is crazy. Well, folks, if you would like to thank Jason for bringing all this great information, buy him a gift from the Daily Technique Show store. We've got some great stuff in there.

35:08We've got a shirt now, thanks to Charles' request with the Len Peralta TechSanta logo on it, but you can get TechSanta on a greeting card. You can get them on a hoodie. There's a DTNS pint glass in there now that R.W. Nash and Wen Tway Dao are both testing out. Go take a look. There's lots of great stuff at dailytechnewshow.com slash store.

35:33We end every episode of DTNS with some shared perspectives. Today, Anand Jr. reacts to Cloudflare's explanation of the bug that caused its big outage earlier this week. Yeah, Adnan Jr. noted. So here we are in the tail end of 2025 and memory allocation errors are still causing problems. Have we learned nothing? Looks at how static the Open Web Application Security Project top 10 list has been. Nevermind. Womp womp. I know, but sometimes you need to allocate memory and you don't think it's going to go more than the amount that you allocated And that's, you know, going to be a never ending story.

36:14I think. Thanks for that. Yeah. Thank you for that. What are you thinking about? If you have insight into any of the stories we talked about today or ever share it with us, feedback at dailytechnewsshow.com. Big thanks to Anand Jr. for contributing to today's show. You can contribute to the show too. You can send us an email. You can support us on Patreon. Our Patreon is patreon.com slash DTNS. You get value from today's show. Just give a little value back. Patreon.com slash DTNS.

36:48The DTNS Family Podcasts. Helping each other understand. Diamond Club hopes you have enjoyed this program. My finger twitched.