Google I/O, Fertility Decline, Spotify Logo Drama | Diet TBPN

0:02John Coogan:Google I.O. starts today and the stock is ripping. I think people might have missed this if you haven't been watching closely, but Google is up 140 % in the last year. Absolute ripper. It's almost a$5 trillion company now,$4.68 trillion. I was really confused what chart you're reading because it's down 1.3 % today. Today? Oh, okay. No, it is up massively. We think in years, sometimes decades. Yes, yes. And yeah, they pulled in just shy of$110 billion in revenue last quarter. And they're in a great position for the next era of the AI story. So GCP is growing faster than AWS and Azure. Wall Street has basically fully repriced the company as a full stack AI winner.

0:49John Coogan:That's the new narrative across Google Cloud, Google Search, Gemini, the models, DeepMind, everything that they're doing. So long gone are the concerns about Google's search weakness because even core Google search is showing resiliency. Google search, the business continues to grow. Queries are at an all-time high. They're not reporting exact numbers of queries, but Sundar said that in the last call that it's at an all-time high. Certainly not going down. And search and other revenue, which is their bucket there, is up 19 % year over year. So holding up well. And Google I.O. generally offers consumers launches or previews of tons of new products.

1:29John Coogan:I'm getting called. Previews of tons of new products and features. The Verge was saying that there might be some like AI fatigue, which is maybe an overstatement, given that, you know, people are getting booed. Actually, the former CEO of Google. Yeah, understatement, given that the former CEO of Google, Eric Schmidt, was booed off stage at a commencement speech. That is a good point. But, you know, the people that watch Google I.O., the Google core consumers, they are fans of this stuff. I think they're generally pro AI, excited about new features. Some of the new features that we'll show are very, very cool.

2:02John Coogan:But there is this like goal of being ambient and useful instead of pushy and desperate. Many Google experiences now have duplicative Gemini panels. And I was writing this update in a Google Doc, and I noticed that I had two Gemini stars, basically. One Gemini star in my Google Doc. And then another in the Chrome browser that I'm using to load Google Docs. And it's a really hilarious outcome because I was writing this in sort of like a half window to the side of the screen. And if I open both Gemini panels, the Google Doc disappears entirely. And I'm just left with two chat boxes to interface with the Google Doc, which I don't really use AI in the actual Google Doc.

2:45John Coogan:I just kind of write it. But there's, you know, there's stuff it everywhere and then actually make it useful, make it ambient, make it delightful. And so that is, I think, what consumers are looking for more than just an AI button in a new place. But they're certainly showing that already. The new Gemini video model looks incredible. We'll play some videos of that. And there will be tons of delightful experiments that may turn out to be blockbuster products, or they may get shelved by year-end. That's kind of the beauty of Google's culture, is that they have plenty of opportunity for experimentation.

3:16John Coogan:We sort of, some people remember all the things that are in the Google graveyard, but most people just remember Gemini and whatnot. So yeah, we can play this video. It features eight cylinders, arranged in a V-shape, driving a single crankshaft. They take turns firing to deliver smooth, massive, That's pure mechanical genius at work. A V8 engine features eight cylinders. So I feel like this got rid... I mean, the video fidelity is incredibly high quality. There's no six fingers. It looks HD. The motion looks good. The lips are synced. And I feel like they got rid of that like hollow sound that you used to hear in AI video, where the audio was generated alongside.

3:55You can still clock it, but it's a lot more subtle.

3:57John Coogan:It's really subtle. It's crazy because you see these and you're like, like this is it like it's done like this is fully fully done and then there's just like ah we're at 99.9 now and i want to be at 99.999 also like this is kind of a nitpick but isn't that a v6 right oh is it i don't know is this good for video explainer channels on youtube bad for video explainer channels on youtube certainly commoditizing the production of video explainers i've seen a lot of these video explainers that will show you like inside of a rocket or inside of an RPG or an AK-47 or a Glock, and those get like tens of millions of views.

4:34John Coogan:They can be viewed in any language, but they're very intense from a CGI perspective. You have to go and model every little detail, every pin in the weapon or whatever the object is that's being visualized in this particular video explainer. Close to being on command, and then the question is where does the value sit? does if you prompt YouTube and you ask for a video explainer of, you know, a chair, break it down, explode it, show me the innards, will it just do it on demand for you? Will it just generate that? Or will this still sit below the creators? Yeah, I've always had the question, at what point do you go to YouTube?

5:10And there's just a series of videos waiting for you that were generated based on your interests, right? Sometimes, you know, you might be going to YouTube because your favorite sports team just played and you want some analysis on the game or, you know, your favorite fighter or something like that, where some news is happening. And it doesn't seem like we're that far from a future where you land on YouTube and YouTube is just, again, fully generated a video based on what it knows about your interests. That said, that would cause potentially a creator strike because it's YouTube starting to compete against their own content producers on the platform.

5:55So we'll see.

5:56John Coogan:Yeah, at least in the interim, it feels like the dawn of stock footage. YouTubers have been using these tools for a long time. They have been getting cheaper. Even the CGI world has become increasingly commoditized every year as you get more to templates and the tools become cheaper. Let's watch this other science explainer from the timeline. Gemini Omni explains science with video. Thanks a lot for this, says Chetoslua. Now every student will get a custom video for the topic of science and math. I'm so happy, like, while typing. I want to see all your reaction to this. This is about photosynthesis, I think.

6:37John Coogan:As this light enters our atmosphere, it crashes into molecules of nitrogen and oxygen. This triggers a phenomenon called Rayleigh scattering. Because gas molecules are tiny, they affect shorter wavelengths much more than longer ones. Blue light has a very short wavelength, so it's scattered in every direction, filling the sky with color. Meanwhile, longer red wavelengths pass through almost three... There has been a big push on YouTube for like, as people ask questions, like they would go to Google and say like, how do I fix this particular washing machine? You type in the number of the washing machine and it would take you to not just a single video about someone fixing that washing machine, but the actual section in the video with the solution to the exact problem you had and being able to read a manual and constitute a video on the fly of exactly that is pretty incredible and you can imagine satisfies that use case very, very quickly.

7:32John Coogan:And then of course there will just be entertainment and all sorts of different use cases. Logan Kilpatrick, friend of the show, says introducing Gemini Omni. Omni is our new model that can create anything from any input, starting with video. Think nano banana, but for video, okay. Leah, let's play this, because there's some amazing different styles here going on.

7:57John Coogan:I wonder if that motion graphic transition was created in Omni, because that's something that you'd normally bump out to After Effects for, or like the edit here, I wonder. I wonder if you'll be able to upload multiple clips and have it edited together to the beat of a song that you pick, or will it be able to AI generate a video and then match the footage to the beat of the video? So it says give it anything. So I think you could potentially give it a bunch of videos and it could edit it together into a vibe reel, something like that. Swap style, swap environment, swap angle. They've been having a lot of fun with this.

8:36John Coogan:The other news out of Google today is Gemini 3.5 Flash, our most powerful model to date. It pushes the frontier of intelligence speed and cost, putting 3.5 Flash in a class of its own. We spent the last six months making sure Flash is great for real world use cases. It's the strongest agentic coding model yet from Google. It delivers frontier level performance at 4x, the speed of comparable frontier models, often at less than half the cost. So dominating the Pareto frontier has been the goal for a long time. The speed is being heralded as a key feature. Google just showed a demo of Gemini Flash running between 600 and 1 ,400 tokens per second on TPU-8i.

9:21John Coogan:It peaked out around 1 ,480 tokens per second with an average of around 800 tokens per second. So very, very, very fast. The flip side is it's more expensive than previous flash models, but that's been the trend with smarter intelligence for a while. So investors are focused across three key areas, not so much the consumer story, more the next Gemini model. So where this fits in and then what adoption and diffusion looks like, how Google through Google Cloud will be getting this out into enterprises, into coding agents. Obviously, they have anti-gravity, but the Gemini CLI has not seen as much traction.

10:01John Coogan:And so a better model might pull that forward, might wind up seeing more traction there. Overall, I think token generation at Google is up 7x year over year, which seems great. It's unclear how much of that is because there's more reasoning happening. But given the fact that the Gemini models are sort of stuffed all over the product surfaces, I'm not surprised that there's massive growth. That makes a lot of sense. On the core Gemini model, everyone was wondering, are we getting four? 3.5 launched and there's a staged rollout with Flash going first. Andrew Curran had an interesting post here talking about the lack of vague posting.

10:34John Coogan:The DeepMind folks have not been vague posting about the new Gemini model. So he did some vague posting for them. At this point, everyone knows it's arriving tomorrow along with their personal agent named Spark. This reticence, of course, can be interpreted in many ways. I'm choosing to interpret it in accordance with my nature. I think they train the largest model they've ever successfully trained, possibly the largest one anyone ever has, and something unexpected emerged at scale. They had their mythos moment, but not in the same way Anthropic did. Gemini has always been a very different model from Claude.

11:05John Coogan:The benchmarks will go out tonight under embargo. They probably already are, but I don't think they will fully reflect what I'm talking about. I think they hit something even they weren't aiming for, something that surprised them. If I'm right, that surprise will be part of tomorrow's show. we shall find out together in the morning. I don't think tomorrow's show, because IO is a number of days, and there's a whole host of different announcements that could happen in the interim. There's a lot of other things going on. Yeah, has anyone been vague posting around, will there be a 3.5 Pro this week?

11:38John Coogan:Yeah, that's going to happen over the course of the next few days. They just started with Flash. Okay, starting with Flash. Cool. And then they also announced Spark, which is a personal agent that lives in anti-gravity. Oh, okay. That's my understanding. Oh, interesting. And so trying to make - When I hear personal agent, I think more like Gemini app, Google search, like Gmail, like the very, like the consumer product surfaces. I think, well, I guess I just think personal and I think consumer, but given how much people are using Codex Cloud Code for like personal-like things, like just because writing code creates a more dynamic agentic surface, OpenClaw, we saw all of this.

12:16John Coogan:It's helpful to have something running on a MacBook Pro that can go around and find different stuff. Yeah, just some additional context. 3.5 Pro is coming out next month. Next month. So not this week. A little bit of a delay there. I wonder what else is in the bag of like mythos like surprises because the cybersecurity one was like sort of predicted by the AI 2027. I feel like bio is next. Like it feels like, okay, we tested a bunch of stuff and we talked to a bunch of scientists and like this thing can come up with like super viruses and it's really scary. So we got to give it to all the pharmaceutical companies in advance and like Moderna gets it and creates like anti viruses or something like that.

12:55John Coogan:I don't know what else, but I'm sure there will be surprises. There always are in the AI era. So agent of commerce will also be top of mind for investors since messaging around the Google, the Gemini app has sort of strayed away from advertising as an immediate monetization engine. I think Demis said that at Davos. Google has a lot of capabilities when it comes to closing the consumer shopping loop. Like they have Google Shopping. They have a bunch of hooks into all sorts of different e-commerce services. People search for stuff on Google all the time to buy. E-commerce customer behavior seems to be lagging expectations here generally.

13:27John Coogan:There's been a lot of announcements from companies around agentic shopping protocols and the numbers. Whenever we dig into them, we're always like, is it gonna get to 1 % this year? Are we gonna see? And everyone's talking about the growth, which means we're growing from zero, obviously, because this didn't exist. But where is it going? Will Google have something to show here? Will they have some sort of demo of a new user experience, a new flow for agent of commerce that results in, you know, a faster takeoff of that adoption of that behavior? As I mentioned yesterday on the show, we had a lovely conversation with Joanna Stern from thenewthings.com.

14:01John Coogan:We had lots of fun takes about, like, the AI tools that I think most of us have interacted with. everyone's used agents, everyone's sort of felt what it's like to talk to a chatbot. But one place where she went deeper than I think most consumers and like AI fans have is in the wearables because she was wearing that recording device consistently. And she maintains that like humanoids are farther away, you need a lot more training data. The AI chat apps are here, we already know they're diffused, Waymo is now boring. But the next big wave she's sort of predicting is in the next few years wearables will have like a big moment and everyone will be sort of adopting these and contending with them.

14:39John Coogan:And it is interesting how, you know, we talk about a capability overhang in the enterprise with AI deployments, and that's why the big labs are partnering with consulting firms and private equity groups to get AI installed into large corporations. There's even more of a capability overhang in consumer hardware. Apple iterates extremely methodically. You know, they made a big story about Apple intelligence. Was that just one year ago? I guess that was one year ago because WWDC is in a few weeks. Feels like longer than that. I just remember they did a global billboard campaign for Apple intelligence.

15:16John Coogan:Yeah. But anyway, like the actually changing anything in hardware takes Apple a long time. They still haven't launched a folding phone. Like they take their time to deliver a great product at the right time. And then if you're a challenger and you just want to manufacture new devices at scale, that takes years to ramp up. And then you also have to distribute, sell. It's not one click away. It's go to the store or wait and wait for the mail. And Google's had some fun swings at these preview emerging hardware platforms. Google Glass, I mean, way ahead of its time. We're now there with the meta Ray-Ban displays.

15:52John Coogan:But even those are not selling by the millions and millions. They're very early stage. Google Cardboard, I don't know if you remember that one. This is, you put your phone in a cardboard box that they send you and then you can put it on your face and use it as a VR headset. Whoa, a little experiment of how can we strap someone's phone to their face? Basically. And serve the matter. And then they also did a Samsung Galaxy Gear. At point blank range. Which was, yeah, you'd slot it into like a piece of hardware but much cheaper than buying an Oculus at the time. Lisan Al-Gaib says, 3.5 Flash score is kind of low on coding index due to rough Terminal Bench hard scores.

16:33So I think the big question coming out of I.O. today is how do developers respond to the updates to anti-gravity to 3.5 Flash? The speed is amazing. We know how much people care about that in just like day-to-day coding. But the model has to be able to perform. So we'll see what people's reactions are, and we'll see if Google can really start to ramp revenue on the cogen side or still get exposure to that through Anthropic. It did come out yesterday that Demis is an angel in Anthropic himself. And not super surprising, although less pushback on that.

17:17John Coogan:Yeah, when did they meet? I wonder what the story is there, how early he got in. He might be sitting on a bag. Well, who else is going to Anthropic? Andre Carpathi has gone from OpenAI to Tesla to Anthropic. I think he went back to OpenAI at one point in there. And Andre, a different account, is pointing out this KMT general who defected and subsequently betrayed five different countries in Asia, ending in Japan, jumping around. He's seen it all. Certainly the world tour of AI labs. I guess Andre Carpathia was never inside of XAI because he was sort of the precursor at Tesla. Yeah, he was poached by Elon in the early days.

18:05John Coogan:I feel like he might have been in Google before OpenAI. I don't know. So he interned there. He interned there, so he's been everywhere. He's got the Thanos rings. Huge pickup and excited to see what they do together. He's apparently, according to Alex Heath, going to be working on basically RSI. RSI, yeah. Yeah, I think he's continuing on his auto research project. Oh, yeah. He's been doing RSI basically in the open source world. Auto research is open source, right? Yes. Okay. Yeah, I think you can read into this that it was effectively an acquirer of the company he was working on. Oh, interesting.

18:41Yeah, I'm assuming.

18:41John Coogan:He said he was going to get back to the education project that he was working on. I thought he had raised for it. I don't think he did. Maybe not. I don't know. That's always helpful. But that was a cool idea. And I wonder how that fits in. It was always interesting to think about, like, LLMs are really good at education. I mean, we're seeing that today with Gemini Omni. Like, it can generate a video for you. Now, we haven't really pushed it to the limit. Like, I wonder, if you give it a PhD-level problem, is it going to teach you as well as, you know, a great professor who has thought about all the different responses.

19:14John Coogan:is like maybe it's not fully there, but education certainly seems on like the core path of the models. Going to a computer and asking, teach me something, felt like something most of the AI models would get very, very good at because there's a lot of training data. There's a lot of open source educational materials. All the textbooks have been scanned. Wikipedia is in the models. There's so much information that's readily available. It isn't tightly held secrets that are hard to bring to bear in the pre-training data. One more thing out of I.O. that we forgot to cover, Google's new Synth ID framework that 11 Labs, OpenAI, and NVIDIA are joining forces.

19:52This is to help identify AI-generated content, basically creating a standard for across platforms so that when you generate an asset, 11 Labs, OpenAI, Gemini Omni, it should be auto-detected by the different platforms.

20:09John Coogan:Yeah, I've seen that on X recently. there's been a little tag that says like made with AI and but I feel like you can get around that if you screenshot it and so well so I think the ones on x are just in the metadata in the you can actually change it like fairly easily I don't think it's actually using like like on on nano banana images on gbt2 that's like watermarks yeah you've seen these like weird patterns people posted yeah I think subtle changes to the to the saturation yeah I think they've just been it's just been metadata so far yeah but yeah the trick with all those is that like it's in theory pretty easy to like rip that out if you're running like an advanced AI, you know, slop avoidance detection system or something.

20:48John Coogan:But just to know, okay, for the average, for the average poster, if this is an AI image, that's certainly helpful. But as you start bringing different assets and you bring in some stock footage, you bring in some AI footage, you blend them together, you're doing a lot of different things, you'll probably lose a little of that AI detection ability, but hopefully people aren't too annoyed by it. If it's used tastefully, I guess it shouldn't matter at the end of the day. Anyway, do you think Spotify used AI to create their new disco ball icon? This was burning up the timeline this weekend. I was shocked at all the negative reactions to this icon.

21:25Me too, for a bunch of reasons. What's wrong with you? What's wrong with you? Seriously, if you don't like this, seek help. I will say at first it threw me off. I was like, where did my Spotify app go? Because it's too dark.

21:37John Coogan:Genius. I think it was genius. I opened up my phone and I was drawn to it immediately. My eyes jumped because I was like, something's wrong with my phone. Something's wrong with my home screen. Things don't look the way they normally look. It drew my eye. I saw, oh, Spotify. Okay, look a little bit deeper. The icon looks a little bit different. The color's a little bit deeper. Oh, there's something else going on there. Peel back the onion. You see that there's a disco ball. And then, of course, that there is a meaning behind it. They didn't just, there's a whole reason why they did this. It's the 20th anniversary of the company.

22:08John Coogan:And so lots of people complained, but party, your party of the year. It's so funny because I don't know, prior to this, were people sitting around being like, wow, I really hope they never changed the Spotify logo, even for a few weeks. I just love it so much. Yeah. Right? I think it's fun. I think it's a nice change from, you know, the flat minimalist logos that we've all grown accustomed to. Keep it. Dylan said, I thought this was fun. I'm sure the complainers thought so, too. But when tapping an icon is second nature after being used to it. I told my wife to cancel our subscription. Oh, no.