00:15:16 · Jordi Hays, John Coogan, Tae Kim
[00:15:16] S1: build games on top of this? [00:15:18] S0: Yeah. I just think about, I don't know. I I mean, Dennis has a background in games and he was sort of alluding to the fact that he might go back into games at some point or or at least be able to like scratch that itch again. Famously, he wrote a programmatic code to generate vomit in a roller coaster simulator. Very fun story. But again, when I think about Roller Coaster Tycoon, was not, I think, I don't think he was actually working on that game. Was a similar theme park simulator, but we are moving back into the simulator world. But the mechanic is what is so enticing to gamers often. When I think about the games that I've spent a long time with, some of them have incredible graphics, AAA graphics. Some of them have two d graphics. But the mechanic is great. And so that is what gets me to feedback. [00:16:06] S1: The legend Bobby Chipman in the X Chat says, can't wait for smart glasses to fully replace my monitors. [00:16:12] S0: Yeah. Maybe you'll need augmented reality or something. Metaray Band display, certainly going that direction. The Orion. I've I've been surprised. Wasn't the first episode we ever did we were talking about Orion? And they still haven't shipped it. Right? I mean, they shipped their smaller version, the Metaray Band displays, which have [00:16:30] S3: Yeah.
00:16:30 · Tae Kim, John Coogan, Jordi Hays
[00:16:30] S3: Yeah. [00:16:30] S0: Sort of the Call of Duty HUD. It's not full augmented reality. Was expecting I was We we we demoed the Orion headset. And it and in it has a bit of a narrow field of view, but it really can put a screen right in front of you. And I I assumed that, everyone was saying it's really expensive, it's clunky, it's not ready for prime time. But look at how fast things are going. In a year, maybe two, we'll get it. And maybe that's coming at the next Metaconnect. Maybe this summer we'll see it. But hasn't been that many rumbles on it. Then obviously the massive pitch shift to AI CapEx might have taken a backseat. I don't know. I'm certainly hopeful. I'd like AR and VR. I think think we're we're overdue for a new fun product. I'm still waiting for the next Apple Vision Pro. Apple Vision Air, something just lighter. That's all I want. Cheaper maybe, but lighter and same screen. Screen was great. Anyway [00:17:24] S1: We know, John. [00:17:25] S0: Gemini Flash 3.5 looks pretty neat according to Tenebrous and extremely fast, but still largely the sort of incremental progress we've come to expect from Google, generally a pretty disappointing IO. Now, what's interesting is that Gemini three felt like a new base pre trained, felt like it had some of that big model smell, felt like it was really delightful to talk to. And I think a lot of people were expecting Gemini four here. We're still waiting for the next iteration here. And also Yeah. [00:17:55] S1: We're still waiting for Pro.
00:17:55 · Jordi Hays, John Coogan
[00:17:55] S1: We're still waiting for Pro. [00:17:56] S0: Yeah. Pro isn't out, but people are speculating that that 3.5 Pro won't necessarily be a new pre trained. And so, it seems like it's a there's a little bit of, you know, research being at odds with like the product cadence, like Google IO is scheduled probably like two years in advance. And whether or not the training run finishes on time is a little bit harder to package up and nail on a specific time. We see this with the independent labs or the OpenAI, Anthropic, the other labs, xAI, like they're launching models very much like when they're done and then they will like instantiate like something that looks like a conference around it or maybe a video or a blog post, a model card. But if you're grinding towards a specific date and the specific model isn't quite ready, you come out with something that looks a little bit more incremental. People were really, really honing in on the fact that the cutoff date was January 2025. Right? Was that the was that the date? That or was it December 2025? Either way January. Yeah. I I don't know. I don't know how much cutoff time cutoff dates matter because, you know, all these models, you know, can query the web and and get team [00:19:12] S1: Overall reactions from Yeah. Developers across the board were not good. Not good. [00:19:20] S0: Yeah. Pikash here. Unduewhelming. Cursor [00:19:24] S1: ranked it on CursorBench. It is below Composer two.
00:19:24 · Jordi Hays, John Coogan
[00:19:24] S1: ranked it on CursorBench. It is below Composer two. [00:19:29] S0: That is that a fair thing? I mean, I'd like to see you rank another another livestream on TBPN bench. It doesn't match up. Know? It's like [00:19:40] S1: No. I mean, they have they have all they the have all the other frontier [00:19:45] S0: And some of them are ahead of Cursor's own models on Cursor's [00:19:47] S1: It's data just point. But yeah. Yeah. But here's the other thing. Four times it costs four it underperforms Composer too even though it's roughly 4 times more expensive. Interesting. Interesting. [00:20:00] S0: Yeah. I I feel like for a long time Google's positioning was, you know, frontier or near frontier but best possible pricing. And this marks sort of a shift in the strategy perhaps. [00:20:11] S1: Yeah. Overall, starting to make more and more and more sense [00:20:16] S0: Mhmm. [00:20:16] S1: Why Google has put so much capital and resources behind Anthropic. Yeah. Right.
00:20:22 · John Coogan, Jordi Hays
[00:20:22] S0: Prakash says, seems to indicate that DeepMind is constrained by data rather than compute for what they intend to do, hence the TPU sales. Rest of Google now shipping their org chart. Ben Thompson talked about that a little bit. And there was some context on like, you know, we were asking the question like, will there be AI fatigue from stuffing AI in every product surface area? Allie Kay Miller shares one of the loudest applauses in the entire Google keynote. Nishtha put on the Gentle Monster plus Gemini glasses, tapped the side to summon Gemini, and all in one prompt said, take a photo and put a cartoon blimp in the sky that says Google IO twenty twenty six. And within seconds, the preview of the edited photo from Nana Banana appeared on her watch. I want to spend less time on screens. AI is really coming everywhere. And so much is driven by Voice AI as the interaction mode. Very cool demo, impressive technology, but Greg's gadget says these companies truly have no idea what regular people want because, yeah, that is a little bit of a niche use case. You need to be more creative with it for when you would actually use that because this is a very it's a perfect demo of the product and the functionality, but it lacks that like creative spark of like, yes, I did want a picture of that on my wrist at that key moment in time. Like if you're not doing a demo, I don't think his point is that regular people would not be excited about that particular feature. Right? Anyway, you want to move on to something? Where do you want [00:21:52] S1: go? What next?
00:21:52 · Jordi Hays, John Coogan, Marcus Milione
[00:21:52] S1: go? What next? [00:21:54] S0: IPOs? SpaceX IPO? We're getting more details by the day. [00:21:59] S1: The other the other thing The thing that is [00:22:01] S0: Yeah. [00:22:01] S1: Was going pretty viral last thing on IO was that the Google anti gravity team flashed a codex folder [00:22:10] S0: Oh, yeah. [00:22:10] S1: In their in their actual demo video. Gurgley says, I had to do a double take in the second minute of the launch video for Anti Gravity. You can see people use Codex on the Anti Gravity team. Did no one double check the launch video at the very least. Not a huge surprise. Obviously, anti gravity looks a lot of people were saying looks quite quite like Codex. So clearly [00:22:33] S4: More than I [00:22:35] S0: feel like it would be like a they would have just rebuilt Windsurf. Don't know. We'll have to see. [00:22:40] S1: No. So but but anyways, this isn't a huge surprise. Right? Mhmm. Google's been using a bunch of anthropic models. Mhmm. Clearly, they're using a ton of different models and products internally.