00:33:32 · Eric Seufert
[00:33:32] S2: Yeah. It's like, what's your bid? Well, your bid should be the value that you get back. I mean, that's like, the whole point of a second price auction is that you should build your bid your true value because you're you're gonna make money if you win. Mhmm. But, like, the thing is, like, they've built like, lot of their initiatives. Right? So JEM, foundation model for ranking. Andromeda is is a is a whole system for doing retrieval. And then Lattice, which is transfer learning. But, like, the whole point of Andromeda was they were reacting to a lot more creative being deployed. Now the creative being deployed, though, was created with third party tools. Right? This creative being deployed is being built without the benefit of the actual performance data. If you actually want this to work, you need to train it on the ROAS. You care about the end result. You care about what the person does when they go to your website or your app, And that no one has that except for Meta at that volume because they get it passed back to the Capi and the Pixel. Right? And so their ability to build this foundation model, I think, really unlocks a lot of value. And, you know, I think you'll probably see that show up in, you know, in the in the revenue growth. Now do they hit 40% to satisfy these, these needs of these hedge fund people? I don't know. But, like, my sense is if you can point to the output, you can say, look. This addresses the core bottleneck, which is we need a lot of creative, but it needs to be built with the knowledge of what actually drives the outcomes and not just a bunch of variations that clog up the system.
00:34:43 · John Coogan, Eric Seufert, Sean Frank
[00:34:43] S1: Mhmm. I wanna push back on that idea that no other lab can do anything like this. What about DeepMind? What about YouTube, Nano Banana, v o three video generation? It does feel a little bit farther off. And, also, it feels like the hedge fund analysts that you're talking to aren't asking the same questions of Google's investments in AI because they have such a incredible business with Google Cloud Platform, and they're able to strike these massive, compute deals. So they have some off take there. But, how is how is the situation different in Google? [00:35:21] S2: Because they don't have a history of tilting at windmills. They don't have the they they don't have an albatross around their neck, which is metaverse. Mhmm. That was a misadventure. It cost a lot of money. It never resulted in anything meaningful. Mhmm. Right? Now I would actually make the point that that whole rebrand was just a distraction. It was a smokescreen. They had to get away from the whole Facebook Files thing. Sure. And that took everyone's eyes away from, you know, that that scandal. [00:35:43] S6: Mhmm. [00:35:43] S2: And was it worth it? I don't know if you achieved that, which they kind of did. They kind of did that with the meta rebrand. Yeah. Maybe it was worth it. Right? You know, not
00:35:43 · Eric Seufert, Jordi Hays, John Coogan, Bernt Øivind Børnich
[00:35:43] S2: And was it worth it? I don't know if you achieved that, which they kind of did. They kind of did that with the meta rebrand. Yeah. Maybe it was worth it. Right? You know, not [00:35:50] S0: necessarily also the name even if you ignore, like, the actual investments they made in in metaverse and, like, Horizons and things like that, like, meta platforms is is the metaverse. Like, it is the place that people exist online. Mhmm. So, like, the name makes sense even if you ignore that and it makes sense for the strategic reason, like you said, to get away from the Facebook files. So, like, they could have just rebranded and never done the metaverse and they'd be in a much better position. Right? They'd have that [00:36:16] S1: Yeah. [00:36:16] S0: Sort of [00:36:17] S1: But you can't really do the rebrand unless you tell the story or else everyone accuses you of what you just described. [00:36:23] S2: Right. Yeah. Yeah. So you gotta [00:36:24] S5: I mean, you gotta put [00:36:25] S2: your money where your mouth is.
00:36:25 · Eric Seufert, John Coogan
[00:36:25] S2: your money where your mouth is. [00:36:26] S1: Yeah. But on on video generation in particular, do do you feel like we are further out just further away from that? I was I was demoing the latest VO model and it's good, but it's still clockable as AI generated. There's still I had some cars spinning around and, you know, it's three cars and then it's four cars and then it's two cars and they're sort of melding into one another. It's an incredible video model, definitely state of the art, but I don't know that that that's ready to be deployed in YouTube across a ton of video ad impressions. So do you have a timeline for that, or do you have some thoughts on, when that will actually be important to the business? Because you have to imagine that Instagram video ads perform better than just image ads, and so they'll try and do this. But what what how are you seeing the AI video advertising model evolve?
00:37:20 · Eric Seufert
[00:37:20] S2: Here, you gotta look at ByteDance. What ByteDance is doing is incredible on this front. Right? So they put out this paper a month ago. Mhmm. I did a summary on Twitter and and LinkedIn, but, what they're doing with TikTok shop is these real like, you know, basically, photorealistic three d avatars that are selling stuff. Right? Like, so infomercials. And they've, you know, they've built custom models to build those ads. And those are all ads. A lot like, I mean, not all of it, but, like, if you go on TikTok and you're looking at the TikTok shop stuff, a lot of it is AI generated. Mhmm. And so what they they in this paper that I summarized, like, they invested in, you know, in in, essentially, like, fine tuning this model to make sure that there was no collision with the hands. What they were finding is that, like so when you get in that uncanny valley Mhmm. Situation where people can tell it's AI, then they turn off. Like, there's lot of research that's been done in this. If people know that it's AI, they penalize the ad. But when they don't know it's AI, the AI ads outperform the human creative ads. It's it's really fascinating. But so what they found was, like, when you saw the collision between someone holding something in their hand and the object, then people the the the click to rates dropped, the conversion rates dropped. But so what they did was they fixed that. So they built this whole, like, visual interpretability model that just focused on that with it with with an expert, like, in the model. And so it just addressed the hands. Mhmm. And so, like, if you but that's use case specific. You needed a general purpose model that's gonna build photorealistic video. We're probably pretty far off of that for, like, all ads of all types. But I think with something like, you know, okay. Well, we need human photorealistic kind of, like, infomercial style. I think you could get to that point now. And maybe we were there now, and maybe ByteDance is there now. But I think it it takes a lot of investment. Right? I mean, they had something that if I remember correctly from the paper, it's been a while since I saw, like, twelve thousand hours of live human product interactions. I mean, it takes a lot of data to do that. Right? And so, you know, it's just it's whatever you wanna invest in.
00:39:03 · Eric Seufert, Thibault Sottiaux, John Coogan, Jordi Hays
I think if YouTube wants to do that for a general purpose photorealistic video ad tool, we're probably pretty far [00:39:08] S4: off. [00:39:08] S1: Mhmm. [00:39:09] S0: How do you think that the market will react if Muse 1.1 is like a very much like a base hit on the API, where they they get some you know, big companies move over some workloads, but it's not, you know, this runaway hit. And and I say that because so many models that have been good, not great, have a little demand just because there's a lot of demand for AI, but but they don't sort of, like, have these sort of breakout, revenue charts.