Let's Get Encyclical

0:00 Room for a guest. The mics are hot. The beats hit hard. Fans throw their hands high in the yard. Deals and impressions like a verbal sprint. Breaking it down with an ad tech. It's a side four. We've gone for main engine start. We have main engine start.

0:31 Ladies and gentlemen, adtech ad talk. Marley has so much makeup. I don't pee, kids. We just do intros for the entire show one week. >> 45 minutes. >> Welcome.

1:05 >> Uh, welcome everybody to another episode of Adtech Ad Talk. We have a lot to talk about today. I'm Gareth and I represent the sell side and I'm joined by >> I'm Adam. I represent the buy side. >> Wonderful. Um, we're planning on talking a lot about the Pope today because I've been noodling that for a while. And then we have some other ad tech news to cover. Right, Adam? >> There's a little bit of news. There's just there's a lot going on, but it's a it's a bit of a a slow week. I think the uh the live ramp acquisition caused a flurry of activity for me. I think a catalyst of a a round of uh changes and possible M&A activity. We'll see over the next few months, but it's a little bit of uh the calm after the storm right now.

1:50 >> Last week, >> and people are getting up for CA, too. >> Yeah. And >> are you guys going to >> We're starting to be in the can. I am going. Are you going? >> No, I'm not. Not this year. I >> think >> we decided. Yeah. >> It'll be next year. >> Me and Marley are going to be doing person on the street stuff to send back to you to uh react to for the can show.

2:08 It'll be like you're there. >> I can't wait. It's going to be really fun. And there'll be some fun interviews there. I I feel like people are in a a very giving mindset in can. So if you catch them at the right moment, you know, yeah, you'll get some juicy stuff. >> A lot of it's the time difference. You know, you wake up and work hasn't started at home and won't start for like five more hours and that makes like the mornings at can like a weird time where you're like I you know, you're on an island away from work and work is not going on even though it's a weekday.

2:36 >> It's an really interesting underdisussed part of running a business. Like the east coast is kind of the middle time zone. So like my engineers have been Europeanbased for a very long time and it means that my poor CTO has to stay up all night every night, which he does. Um I still don't know how he does it. Like he has children. Um like it's very confusing that I can ping him at 9:00 at night and he's just just typing away.

2:59 Yeah. Um but uh then you have the West Coast and the West Coast, right, are done with work at 3 p.m. >> So the Europeans have mornings themselves. The people on the west coast have evenings to themselves and us in New York. >> Well, it's even crazier when you get Asia. >> Oh, I Asia's I lived in China, man. That inversion is so hard. Like >> Yeah. >> Just the the 12-hour inversion is such a rough one. Uh because you have these little itty bitty gaps, right? These little itty bitty windows of communication. Um or they work just crazy hours. Um are you >> It's like something >> What? We're not >> something Asian office. You going Asia?

3:39 No, not not right now. >> Not yet. >> Singapore would be fun. I've never been to Singapore. I think it'd be a great place to open office. >> Yeah, Alex said it's like a giant mall, but that's what people say about Dubai now, so I don't know. I never been to either. >> Yeah, I uh I heard it's >> I love malls. >> Yeah, I heard it's very I mean it's wildly clean. The the insult that is hurled at it is that it's like boring.

4:05 Um >> but I don't know. That doesn't sound so bad to me. Like, >> do you remember when that kid that American kid was going to get caned for graffiti? I think it did finally happen. Something I have no idea when that happened. >> Anybody? >> I don't remember that. >> Yeah, I remember. It was a huge It was a huge thing cuz, you know, you're not supposed to do that to Americans and and we put up a big fuss and they they reduced it from like 10 smacks to like, you know, three smacks or something, but he still got smacked.

4:34 >> Reduced the number. He was lashed six times. >> Oh man. >> Six flashing. >> What year was it? The story I remember stories going around be like it's not like they just It's not like a nun hitting you in Catholic school. It's like they have like a kung fu master come up with his kung fu stick and like you know like it's really serious. Like people would die from gating as a punishment. That was a thing. Like this is >> I don't know.

5:00 >> Real like the Simpsons about it. part of his podcast to get the boot for something he did in Australia. >> This must have been the 90s. Marley, can I have the year of the caning incident, please? >> It was 1994. And to be fair, this kid, he kind of looked like trouble. >> I mean, no graffiti. >> No wonder you don't. >> He deserved it. >> He's never going to do graffiti again in Singapore. I'll tell you that.

5:27 >> That was a crazy idea. I mean, to be fair, like when I lived in Asia, like I knew not to break the rules. Like, it was it's it's not like it's a surprise. It's like, oh, no, no, no. It's you break the rules here, it's it's not good for you. It's a bad thing for you. Um, >> I have a grand theory, though. Every almost every news event of the 90s, if you go back and read them, was is foreshadowing of what was going to happen after the turn of the millennium.

5:53 There are so many examples of this, but the single best is Independence Day. The '9s movie Independence Day, where a black soldier and a Jewish scientist saved the world from aliens, is it just has an incredible amount of foreshadowing in it. Uh, I highly recommend. >> Wait, what? >> Who's Will Smith and who's Jeff Goldblum in 2026? No, it's not not that specific about saving the world from aliens. But first of all, the destruction scenes are like look like 911. Like they blow up not the Twin Towers, but the Empire State Building.

6:31 >> Yeah. Yeah, I remember that one. >> But there's that. And um and it ends with a suicide bombing, right? The Randy Quay character kills himself to >> to kill the aliens. There's like a whole bunch of 911 references in it. >> Uh five years before 911. >> Yeah. And Independence Day was a fascinating one. I remember when that came out. That was a cultural phenomenon. Um >> the 90s >> I watched in this midnight showing uh July 3rd midnight showing of the first one at the uh I can't remember the name of the theater now is Giant Theater in Midtown, but it still had like the 20s decoration of Red Velvet and stuff.

7:10 >> Oh, those are the best. There's a few of those left. They still kind of have it even though movie theaters, you know, put through their gold rooms. And the other thing going on that night was the uh the World Cup final was in New Jersey that day. >> July 3rd, you know, whatever year that was the last time the World Cup was in the US and Brazil won. And there were like a million Brazilians partying in Midtown. Like where did these people come from?

7:32 >> Oh man, that's there's little Brazil in Midtown, Adam. Thank you very much. >> That's why that's where they went. Yeah, >> they went to 46th Street near this movie theater and they're, you know, banging drums and dancing going up. Sure. Must rock us. A World Cup wins a big deal. Um >> Yeah, but in the 90s in New York, very few people cared about the World Cup even we were though we were hosting it. It'll be different this time. There's a lot more Americans into into football.

7:57 >> Yeah, football's a thing here now. It happened. Had a conversation about last night. >> Um >> I remember that World Cup because the OJ trial was happening at that time and everything, if you're in New Jersey, it was just soccer balls and O everything was just soccer balls and OJ. That's all. And he was talking about OJ. >> OJ was the first time a celebrity was on trial. How many time? Now it's like a weekly thing.

8:19 >> Not only a celebrity on trial, but it's a TV show. Like you watch the trial on TV. That was the first time. >> Amazing. >> Then you watched it with Marley watched it with her grandparents. Must have been nice. >> Yeah. I watched Independence Day 3,000 times. >> You like it too? Well, back then you didn't have on demand, right? So when you if you had the movie channels, you watched what was on. It's basically all TV was fast and the paid channels were fast without ads. Um so they were like paid whatever. And um they played Independence Day a lot. It was a popular flick. I don't know if they were like cheap with the rights or whatever it was, but you watched what was on and it was like it was a maximize the minimum algorithm when you were choosing what TV to watch and Independence Day won that battle a lot of the time. Like it was >> it just had to beat whatever else was on. It didn't have to be what you >> 60% of the time it works every time.

9:22 >> Yeah, I find that if I do find it like playing on a channel like I'll watch it till the end no matter where I'm joining. You still have channels. >> I I seek them out. Yeah. >> Good for you. >> TV comes with them. >> Amazing. >> The TV comes with them. >> That's true. It does. >> Yeah. >> Oh, that's that's fast. That is the entire fast business is those channels that come with the TV now. They're they're migrating the whole thing to the to programmatic infrastructure and slapping CTV ads on there. Um, >> my kids know about ad targeting. I always talk to them about it. My daughter this morning showed me this ad.

9:53 It's like a total AI slop. And it's a uh it's toe fungus that's talking. It's like, are you embarrassed to show your toenails? It's like they made a character out of toe fungus. She's like, "Daddy, look at this. This made a They made toe fungus character talking. >> This is an affiliate. This is um the toe fungus ads are a classic, right? Like the pictures, the grosser the picture of the toe fungus, the more effective the ad is because the higher the CTR is and the more the publishers hate the fact that it's on their website." Um, and the talk and toe focus is just the next evolution.

10:25 >> YouTube doesn't mind. And you know who else isn't going to mind? Chat GPT open. You know those ads and the SpaceX AI. Let's get Let's get to news. Let's cover how the Tofungus is going to have a whole new >> You think they're going to be bottom of the funnel >> surface to advertising? >> Yeah. You think they're going to be bottom funnel? >> You think fungus ads going to be in there? >> Of course, it's affiliate, right? You have a condition. if you have a condition that needs relief. But first, let's get to the Pope's encyclical.

10:53 >> Oh, I have so much to say about this. I have so much. >> All right. We tried to get a Catholic to tell us what an encyclical is and why it's so intellectually written. We could not achieve that. So, we'll call on the on the audience. We have no Catholics on the adtech ad talk staff. As a matter of fact, >> and if nobody in the audience says anything, we'll call on AI. We are almost all Jewish.

11:18 Uh Marley, you're the closest thing to a Catholic with your Mormon background. Uh can you help us with a what a pope a papal encyclical is? >> I have no idea. >> Okay. >> No idea. >> All right. It's an intellectual treatise. It's coming from the head of the Catholic Church. Fair enough. >> All right. It's a formal letter written by the pope and sent to bishops and often to all Catholics in the wider world to teach or clarify important issues about faith, moral, society, or politics.

11:46 Um >> okay good >> and they've sent a few some some historical ones the rear room noarm about workers rights and social justice lado sea about climate change and care for the environment and human ivai about marriage and birth control I love old Latin I'm reading a sci-fi book that's like about this right now and it's >> so it's interesting tickles >> some of them some of them seem kind of kind of leftist and of course the abortion ones are not so leftist so it's you know it's it's their stance it's it doesn't easily it's not easily categorized test, but it's uh as you'll see at this one, there's like a pro-humanity thing going on uh in the Catholic Church's philosophy. It's not so much about Yeah. I don't know. All right, Marley, do you >> We're putting you on the spot to tell us what's in the encyclical so that uh we can react and comment on it in our total ignorance >> of providers but expertise on AI.

12:40 >> Let's go through this piece by piece. So the to sort of tee things up while emphasizing that technology should not be considered in itself as a force and pagnetic to humanity he wrote that the pursuit of greater profits cannot justify choices that systematically sacrifice jobs. So that sort of tees things up specifically called for government regulation of the private companies that are driving the development of AI. How do we feel about that? Yeah, >> I have feelings.

13:14 >> Um, I think that the regulation of the company themselves is very difficult because we've never seen a circumstance where the regulators understood the intricacies of technology to the extent necessary to regulate them and the technologies move too fast to pass regulations. Right? It takes a long time to write it. But wait, wait, but I have a second thought. So my second thought is that you can protect from downside. So when you see harms emerging, >> there go >> you can hit those. And then two is you can make sure that there's a market and that's the most important one that we have been failing at pretty robustly is that making sure that people don't exert monopolistic influence because that's what scares me in all of this is if we end up with one or two AI companies total who have absolute control, it's real bad. Uh and there's no competition for them. That's that's when a lot of the scary things like come to pass. Um but yeah, you know, tell me about >> the motto of American regulators is regulate the harm. It's different in Europe and other jurisdictions, but you regulate against the harm. I think there are some obvious harms coming out of the AI world, and there's no reason for regulators to be so slow about things like AI should not give you instructions on how to shoot up a school or to kill yourself. Like they're all these things should already be on the books. They're just common sense.

14:47 >> Can you publish a website instructions on how to shoot at the school? >> Is that to violence? >> I don't think you can. >> I think that's illegal, right? >> I mean, one time this went this happened in the 60s. >> Yeah. >> There was something called the anarchist cookbook. >> Yeah. Yeah. Yeah. Yeah. Yeah. I remember I' I've had some instruction >> how to blow up a car and stuff and Yeah. It's it's >> make stuff make napalm like make things.

15:11 It wasn't like >> it wasn't like technically illegal or banned, but no one would sell it. Like you get in trouble for selling it for liability reasons, right? >> Yeah. >> Liability. Liability is like a good structure of law. And the thing I hate most about Open AI is that they have lobbied against liability, you know, that they should be they've lobbied for exemptions from liability. >> That's fascinating. >> It's [ __ ] crazy. >> That I agree. That's crazy because I content that's procedurally generated is still generated.

15:41 >> Like what Meta hides behind for liability is that it's UGC and like that they police it to the best of their ability, but like it's human beings doing it and there are scale issues >> and like you know I I it is dumb because I I have trouble believing that they can't police these things. Like it just doesn't make that much sense to me. But maybe it's a hard problem. Maybe I don't understand it deeply enough. But for the AI one, it's like, oh, wait, wait, wait, no, no, no. Your your system is generating these things like there is responsibility.

16:09 >> I think that's kind of what the Pope's getting at here. So, let let's go on, Marley. What else? What else is in here? >> He also called for protection and retraining for workers who whose jobs are threatened. >> Yeah. A very weird decision of the leaders of of the chatbot companies to come out and say this is going to cause mass unemployment and and eliminate zillions of jobs. like it's it's not necessarily true. There was no [ __ ] reason to say that.

16:36 >> No reason to say it. >> I think there was some >> well >> some amount of belief and brinkmanship and also because like developers like co people who code is all they know and they think of that as all workers and you know there'll be a big impact there. I think um one of my my my deeper thoughts on this one that I haven't written but I will is that if you think about the industrial revolution it was very easy to point to the jobs that the industrial revolution destroyed right there were certain types of manual labor and factory work that were done away with by the industrial revolution and it was primarily bluecollar work right it wasn't like the bgeoa who were nailed by the industrial revolution because they weren't working in factories there's a lot of craft there's a lot of craft lot of craftsmen, a lot of things like that.

17:26 >> This one seems kind of like the industrial revolution for the white collar workers because it's it is the automation of word selling, right? It's the automation of people who live a life in words and in information workers and like that's who's threatened by it and we still don't know the extent. We still don't know like >> how how far it's going to go. And sure, it could go really far like for lawyering things. Whenever I'm like lawyering things now, I'm like, "Oh my god, the LLM is so good at lawyering."

17:58 Like, it's just it's so good at lawyering. Um, >> it's really good at McKenzie Consulting. That's what I think it's really bullseye for. >> Oh man. Oh god. Oh yeah. Deck creation from like an outline like a discussion and from like internal knowledge. It's just it's spectacular at some of these things that are like white collar information worker drops. Um, which is also really funny because I feel like the the AI push back is primarily coming from the populist like parts of our society. And I'm like, h, you know, I'm not so worried like maybe about robots and driving cars, but that's not LLMs.

18:31 That's that's not like AI. Um, that's something else. That's that's a different type of automation. This one is squarely nails engineers, knowledge workers. Um, and it's just like a very like subset of jobs that I think are targeted right now. I mean, and who knows, it might create more of them because it might make like individual people so much more effective that now there's just this huge proliferation of companies who need these things. >> Yeah. My friend James in Australia just posted is like they're going to we're going to have to hire people to do research, right? Because there's not going to be any new it's not going to do any new research. So like we're gonna have to fund research to give it new training data all the time. And I I commented I was like, "Yeah, I think the same thing is going to happen with news.

19:14 there's going to be more funding of of news and novel information and occurrences because you're going to exhaust the the training supply. >> Well, people always ask me about publishers. I get this question internally. I get it from my employees like hey like you we're seeing this migration away from traffic from Google like what's going to happen to news publishing? I'm like well it's really hard to LLM your way into news. That's not really a thing. News occurs in the world and until like until like the the Uber system is here and the LLMs are monitoring the world through cameras and speakers, we need humans to report on the news and also the news cycle is super super fast, right? Like it's it moves very very quickly and you you and I have talked about this before. The amount of traffic that is to new web pages is very high. It's very significant. Yeah.

20:07 >> Um like the evergreen pages are that aren't news. they there's a they tail off. Um >> yeah, people haven't appreciated that the the LMS are so smart because people have been feeding novel information into the internet for 30 years and if they were to stop then they'd stop being so smart. >> Yeah. Especially about modern things. Um so yeah, I'm I'm not such a doomerit. I don't think so many people are affected specifically by LLMs because I do think AI becomes this catch-all term for technology. Um, and LLMs only target a very specific segment of society. Um, and yeah, it's true. It could be it's disruptive certainly. Like we talk about agent and advertising stuff all the time. But I um I mean I'll I'll stay on the soap box of I think that LLMs are going to be so great for agencies. Like I think this is like the the best thing ever for agencies. I really do. I don't think it's putting them out of a job. I think it is going to make them so much more effective at their jobs. It's going to be wild.

21:11 >> Let's put a bookmark in that. We're have we can have a whole show on that topic. >> Uh, but I wanted to ask you about the teenagers we talked to in add on the street last week. Remember how down the teenagers were all doomers? >> Yeah. >> Remarkable. >> Yeah. >> Oh, I'll pine. I I think that it's um the same phenomena I was talking about earlier where there's just all of this uncertainty and it all gets bundled up together, right? self-driving cars and robots that can move pallets and AI and LLMs and they all get bundled under AI and that becomes bad because that creates uncertainty and I just don't think we have the reason to be so scared of it. Um I think the industrial revolution impro improved the quality of life for people. really like indisputably did.

22:08 >> Um people had more things and more leisure time. That's another one. Like if you look at the amount of time that we work now versus 50 years ago, we actually work less. Um it's it's very interesting that like we're so worried and I feel like there's a social component to it. I think those students had just, you know, they're they're being trained as knowledge workers essentially and they have this experience that it could do your homework and you know that that's not good for their training and I think they probably hopefully when they get to college they will have better in instruction and the framing of of this tool and what it means for for human endeavor. That's hopefully what's part of what the pope is trying to trying to provide. What's what's the next one?

22:54 Well, I mean, that's that's a really interesting discussion. It's talked about a lot in New York City. Um, what should schools be teaching? How do schools deal with this? Uh, because the skill set for manipulating AI and understanding how to make yourself a powerful knowledge worker while retaining your like critical reasoning ability >> is not what we've been geared towards in a lot of schooling. when calculators came out. I have a good friend who's a math teacher and there was like a big fight in the 90s over whether students should be able to use calculators on math tests and he was like we taught them to calculate in grade schools like they can do it like that shouldn't be what we're testing anymore. Uh and they it's still I I think I think the schools are right to say people should still know how to write an essay. Totally. My daughter is learning how to write an essay.

23:39 >> They should be learn that's a core skill and they should learn it. But I think the schools are erring on the side of of resisting the technology, right? Like I think you have to pick where you're going to draw the line. Let them use the calculator or the AI. And yeah, I guess I say it's it's it's a tough decision. You don't want to completely ban it and say you should be able to do everything the tool does. I think my math education would have been a lot better if I was able to use a calculator because I I suck at calculation. I'm quite I'm pretty good at mathematical concepts and I couldn't go as far in math and working on mathematical concepts because I would you know get a C for calculation errors and I couldn't go on and I think that's I was uh robbed.

24:21 >> It's really interesting because I so I write my blog right and I've never used AI to write my blog and I never will. Um, and the reason for that is because I read a lot, right? And I read like a lot of science fiction and I read a lot of like very like wordy things and that's because I want to maintain my ability to articulate thoughts and it takes practice. It just takes practice. And I I do understand the fear that if people do not write, and by the way, anyone who's listening to this show who has outsourced all of their cognition on LinkedIn to LLMs, everyone knows.

25:02 Everyone knows. And like I'm like getting really fed up with my LinkedIn feed because I have to if I have to read one more AI generated post that's not technically wrong. that is like actually somewhat insightful in some way but it's obviously AI. I know that it came from AI not from the person. >> Yeah. No, we should we should mark the milestone. The link Publix acquisition of LiveR was the first barrage of AI generated commentary in our industry and there's there are thousands of them and they yes you could tell which ones were mostly AI generated. Well, my question is about the in between layer like if you use AI to help and it's not obvious or it you know it raises the suspicion where does that fall in people's heads like I use right I do that all the time like I use it to speed me up >> me too I use it to wordsmith things where I do not feel that it is worth my expenditure of intellectual capacity to wordsmith this like in like in like day-to-day email replies Right. I could spend the time to like make my sentences beautiful and and lovely, but the LLM doing that for me means that I get to expend that energy and time elsewhere.

26:18 >> Um, highly efficient. Like I don't need to practice it in those moments. That's an email. >> But how good of how good of it should you be before before you I was a professional writer. I was in the 99.9th percentile of writing clean copy. So for me to give it up Yeah. All right. Oh, John Duza, I now hate the word curious from the LLM spam. Same problem on Reddit. Oh, yeah, man. The Oh, the curious. Curious. Oh god, it's everywhere. And I flag it every time.

26:46 Oh, people LLMs love saying that things are curious. They love >> For me, the for me the telling the live ramp ones was like if they were saying if they're saying now publicist is a threat. If you wake up in the morning and your and your ID company is owned by a competitor like you know like just like always this like the the telltale narrative was that this is a consolidation play uh and Pulysus is building a platform and now everyone else has is threatened by this consolidated platform. That's like that was to me was the what an LL trained on the past.

27:20 >> I've got one for you. Uh, whenever the closing couple sentences include a it's not an X. >> Yeah, >> it's a Y. >> I know. >> Come on. >> I know. I know the LLM wrote that. Like, I know it did. But yeah, it's I definitely think that um being able to do it is really important. That is why I still when I write and I need to get back to my blog. I will soon. Um, when I write, I make myself do it and I make myself proofread it and I make myself go through revisions because I'm like, "Okay, I need to not get totally >> out of whack doing this." Um, because my speech will suffer, right? Like the the way that I express myself, the precision will suffer, my voice.

28:03 >> Your blog has such strong voice. That's how we met. I was like, "Oh my god, this guy's voice is jumping off the page." Uh, and I redroll. Yeah, it'd be a huge loss to not have your voice. >> Yeah, man. So yes, I think I do think the Pope talks about something really interesting here, right? So when he talks about like what he's worried about. I have my AI summary for the Pope here. Uh humans become metrics.

28:26 I'm an ad guy. >> We're all we're all bundles of JSON to me. Um new economic inequality. That's a universal one. I think most people are worried about that. Loss of meaningful work. This is an interesting one because I've heard this a few times, right? that people need their work to find meaning um and need to have jobs to have meaning in their lives. And I'm not so sure I'm not so sure this one is so obvious to me. I think a lot of people who work derive no meaning from their work. Um >> yeah, >> like seriously and I I think it's like a little bit elitist to like say it. Um like >> also if you like to code or you like to write like there's people a lot people are having a blast with AI. Yeah, you get you still got to do those things like the people who get meaning from their their occupations are gonna get meaning no matter what. Um >> manipulation of truth is a real one. Um I actually think AI so far is pretty good at not being manipulative, especially compared to social media, right? Compared to the UGC amplified content, AI seems so balanced and reasonable. I'm scared that that could change. I don't know how to defend against that changing.

29:38 >> Yeah. Um, >> I know we went through this. Facebook was just this fun thing where you talk to your friends. Suddenly it's like Hillary Clinton's the devil. I like remember that girl would be like, "What? What's going on with Facebook?" >> Yeah, it was crazy. But that was what got engagement, right? Hillary Clinton the devil is what people clicked on. Um, and I think about that a lot with like Grock because we're on X all the time, right? I'm on X all the time. And like people will invoke Grock and Grock will slap him down and just be like, "Nope, that's a conspiracy theory. That's not true. And here's why.

30:08 >> I know. Did you see they cut me off? G. Remember you you loved it the weeks I was doing that. But now you have to pay. You can't get that unless you're um whatever. Uh >> I pay. You can just invoke me. >> Okay. >> Yeah, invoke me. I'll Grock things for you because Grock really does. Grock has been very good. But I worry, right, that that could change because Grock is controlled by humans. I've always hoped there was an open- source component to this that would protect us. And I think that that was like part of the >> Wikipedia. Well, they're pulling from Wikipedia a lot.

30:37 >> Ah, but you Wikipedia is so human generated. Like so much of that is so manipulated. >> It's very manageable. >> Yeah. I But I really like I I hope that open source is what protects us here where like eventually open source at least we can see the mechanisms. Um and you can see the extent to which something's being manipulated. Um autonomous warfare. Yeah, scary. Don't know if we can avoid it. And then the last one which gets my goat. This is the religious one where we're getting sci-fi here is trans >> the power of Abel one.

31:05 >> Um I think that the Pope's big fear and I I remember I was reading this in there is that we are going to use AI to move or the the aspiration is to move us beyond the things that we suffer through as human beings and not suffer from those things anymore. uh and that we will somehow cease to be human and that is and that the aspiration of a lot of the AI companies is to move us past humanity. Um yeah, this is the pope went hard on this one.

31:45 >> All right, that's the Tower of Babel narrative which as humans we know very we know extremely well from Genesis. >> So I didn't look this up. Uh yes, the Tower of Babel is a story about human arrogance. They were building the tower after Noah. All the people speak the same language after the flood. And they all work together to build a great tower. And they're going to take on God. And God punishes them not by destroying the tower, but by scattering them across the earth and making them speak different languages, saying, "You guys are too dangerous if you work together.

32:24 We're going to have you be less arrogant, humbled by the language barrier." >> Yeah. Here, I'll I'll I'll read some of the encyclical because I found this really interesting. Um, fearing being scattered across the earth, they sought to guarantee stability and power for themselves and above all to make a name for themselves. Make a name. Like naming is really like a profound thing in Judaism by the way. Like naming is a thing. Um it was an impressive feat. A single language, a single technology, a single direction.

32:51 However, a project concealed the profound danger. It it was a project conceived without reference to God supported by a uniformity that eliminated diversity and chose homogenization over communion. Um when a city is built on pride and the claim to self-sufficiency, communication breaks down, languages are confused, and people no longer understand each other. And the result is not unity but dispersion. Um my big takeaway from this is we've like >> Google kind of started this right search engines kind of started this where we developed a common like set of knowledge that we all shared >> because we all googled things >> and that meant that we all like kind of had the same Google or similar Google results right for questions and thoughts and AI is a next evolution there where there's almost a collective mind there right like there's almost a collective Like and it's it's personalized. Sure.

33:43 Like each person gets their own answers, but it's personalized by the same system. So like it's only so different. >> Well, yeah. I mean, in social is where it becomes info bubbles, right? So this is the fear is well, one fear is more info bubbles, but is that what the pope means? >> No, the the pope means the the centralization. He does he thinks that AI is another tower of babel. We are building this thing that will unify all of our knowledge and make us one thing and that is an affront to the pope.

34:13 >> He doesn't like it. >> Um he's like >> I think it does. I think it does. I think that the LLMs like have tremendous utility. Um and he and he does acknowledge that, right? Like the the breakthroughs that they have in medicine and in all those sorts of things. He's he acknowledges like all the good parts of the LLMs, but there there's a reason he invoked the Tower of Babel story very specifically. He's like I he it's a direct correlary that these systems are a modern tower of babel and >> right he didn't indict them. I there's somebody from anthropic in the involved in the encyclical right the guy was standing next to him. So that's another another point for them as being uh >> the more moral AI company. But take a step back. What the pope is saying is that humans have a spiritual dimension that's not covered by the definitions of intelligence in AI and by and in business, right?

35:10 Especially American business, right? And we have a tendency to say or to believe that people like Elon Musk and Mark Zuckerberg must know best or that it's it's fit for them to have the right to decide about, you know, what information is accessible and and amplified because they're so smart and the evidence of them being so smart is so successful. And I'm with the Pope. This real this definition really bothers me. I think there are moral and spiritual aspects to intelligence. I think we all experience them every day.

35:44 Like we re we recognize when people have wisdom and give us good guidance and help us avoid suffering through their experience. And um you know AI might be able to mimic that, right? It might it might bring improvements in this area, but a language model is not going to develop that sort of intelligence certainly not on its own. Like maybe it could be trained toward it and to help toward it. And I think part of the split between anthropic and AI is that anthropic aspires to train it a little bit toward spiritual and moral intelligence. And and OpenAI's attitude is whatever consumers like is is good with us. Uh more of the Zuckerberg view of it. Um but still like they're saying he's saying a human is is something fundamentally different than this disembodied language machine. and he's saying it's because we have a spirit, which I don't know if I believe. For me, it might be more stuff in the flesh or about how our brains work. Uh but I just think objectively we're very different.

36:43 Our brains run on 20 watts of power. Uh and these things need enormous amounts of electricity. So, you know, they're very different. >> Uh and he's saying it's a mistake to think if it talks like us that it's like us. >> Oh, I I think it's a super salient point. um they're they're word machines. How much of humanity's thinking and what we do is not done in words. Um and I think that there's a there's a really compo important I mean this is this is Wickenstein, right? Like language is a tremendous limiter. It's a definer right like it defines our cognition and and how we live but it is not reality.

37:27 um which is makes the LLM right models built on proxies is what they are. >> Yeah, media. It's media. >> It is media. But I really like your distinction actually because I it makes me really nervous to think that whatever the consumer wants. Uh I think people forget often that capitalism is inherently super democratic in that you build what people want to buy, >> right? You get the average whatever you said. Yeah, >> it's the average of everything. It is inherently a very democratic thing because the way to make the most money is to build the thing that the most people want. And like the thing the most people want is a very mediocre horistic.

38:11 It's great for designing a product, very mediocre for thinking like like for like moral and spiritual like >> they want porn bots. Marley shared the pornbot story. Marley, we didn't put it on the show, but can you summarize the pornbot story you shared in the channel? >> Yeah. So, BYU recently did a study where they found that was it maybe like four out of every seven committed adults in a relationship was having an affair with an AI chatbot and had like used it for sexual pleasure within the last week.

38:43 Like, it was insane. Yes, it was. >> I didn't think it was that hot, but it was pretty hot. But also >> four and seven, >> maybe three and seven. It was like it was >> amazing. It was hot. We I I don't have experience. We We should either volunteer ourselves or get a volunteer from the staff or the audience to test these test these products. >> I tried it. >> Bob said no. Like I don't care.

39:12 >> You tried it with about. Yeah. I tried it with Claude. He's like, no matter what you tell me, I will not do this with you. And that's like okay. >> See that? God, why would you try? You get off on rejection or something. >> Grole Gro will do it. >> Institutional AI company. >> Yeah, it's really interesting. I like that we're wrestling with this. Um, and for adtech companies, I certainly think that uh it's worth acknowledging the linguistic component of this because whenever I talk to people about agentic, I'm like, okay, describe to me the action that you want to do in words.

39:54 Like, is this an action a human would take? And when I think about like how I how different like standards right like ARTF and ADCP and AM and how these things interact and which components of them are LLM driven and which components of them are not LLM driven. I'm like okay which ones are humans doing? Which parts of this are done in words in the language of a human? >> Um >> I think of it as a loop. like we share the model output and the humans go oh that's what AI thinks and then when you see it then you could start to be like well here are the things it didn't know or here are the things that it's missed or here's like context that's specific to this situation where I could tell it took something averaged like to us at Chalice like that's how we see it unfolding is it's not like it's not we do train per advertiser and that's a big piece it's not having aggregated predictions but one advertiser's data for one advertiser's output But having experts from the brand or representing the brand in the loop is how you really get to train something that's more valuable than you know a person or an algorithm on their own.

41:02 >> Well, that's really interesting. When you use the word train, what are we training? Are you training an LLM or you training a model? >> A model. Yeah, we're training, you know, how to how to value inventory. >> It's math. Yeah, it's math. And I like math, too. I mean at their fundamental level, sure all computing is math, but like the LLMs are are are based on words. >> I tell people to understand them as math. It didn't work with Marley, but I I I I tell people to get with claude or chatbt and say, "What is latent space?

41:39 What are sparse encoders? What happens in those when I put a prompt in?" And like get a working knowledge. I'll teach it on the show or in a one-off um based on my own limited experience, but it helps a lot to have at least a model in your head of what they're doing. Um some people prefer to look for the ghost in the machine, but it's mechanical. It's math. >> It's correlative. >> And what I say about what I say about training is you're adjusting its statistical model by giving it other context adjusts the model. more context you give it, the more it changes the odds of it choosing certain next words.

42:17 That's what you're doing. >> I I would say LLMs are trained with a very specific type of data and that's words. Um, and they they therefore do the word things. They're word they're word correlative things. Um, God, it's how Proximic used to do stuff back in the day. This is so funny. Proximic used to do character like modeling where they would run this on the characters inside of words. They would say based on like the characters we've seen so far, what do we think the next character is going to be? And that was how that would be how they contextualize pages. Um, and so it was language agnostic. And so they were like they were the first ones to be like, "Oh, we could we can contextualize any page." Um, because >> yeah, that's the birectional encoder thing, but they've been replaced by LLMs. We looked into using birectional encoders for that kind of thing and they're like that's even though LLMs are not made for that, they're better at that now. It's >> Oh, yeah. totally. Oh, well Proximic was selling this 2011. Man, >> we're going to lose our audience.

43:18 >> Long time ago. Yeah. Yeah. Yeah. >> But I do think it's uh it's interesting to know that these tools, the way that they work, the way that they activate like agentically, um they're in those two camps, at least for me. Like are you building models to do valuations or are you building things that do things that humans do with their hands in their minds? Uh and the things that humans do with their hands and their minds are normally done in words like click this thing.

43:45 >> But I'm just confused about that. Why are there businesses around that? Do people think they're going to catch claude? People think they're going to [ __ ] catch up. Like what the [ __ ] >> Well, they think that they can replace the things that they do with their hands with claude. Um and I think that that's a that's a very good use of claude. Actually, >> what's the business? I don't know. I don't understand when I meet a company and they're like, "Oh, we're building our, you know, we're making our own LLM based models for media planning or stuff." And I was like, >> do they think they're going to compete with Anthropic's ability to do media planning off a set of of media plans media plan?

44:17 >> This is such a good question. >> Is that does Anthropic eat all of these agents? >> No, it's not Anthropic. It's the tool set they've already released in enterprise cloud with co-work and the and the way to build shared context and you know MCP connections. It's like there I I swear I will take on any company that's building something internally that's based on their own LLM or that's multiple LLMs against what my team is building just using cloud cover and existing MCP connections. And I I bet we're building faster and better stuff.

44:52 And you can Oh, your drill is going to what >> for campaign config and optimization and trafficking. >> Well, not optimization, >> vendor testing, >> just planning planning and reporting. >> There's a huge amount of money. And you said agencies >> is a good thing for AI. >> I will just say I'm a little worried about what I see in the plants. >> Tell me more. What do you see in the plants? that makes you worried. You can't stop there. It's too good.

45:22 >> I I when the plans are presented to me, >> it can feel like they weren't thought entire they weren't entirely thought through >> the way I the way through the rigor of like a startup founder as being like >> who who's going to pay for this? How much are they going to pay? How's it going to compare with the other things that are coming out around this? Right? Like you know there's a lot when you're a startup founder you you can't afford any kind of siloed thinking right all your thinking and ideas get pounded by the true market response as soon as they're released right so you can't have like a bunkered mentality like you're constantly getting these rude awakenings and I've seen I've seen some projects on the buy side that I think are headed for rude awakenings when they're compared to what a you know compared to what the same agency can do with claude or that.

46:17 >> Oh, wait. So, when you mean plans, you mean actual media plans or you mean companies saying I'm going to do this? >> Yeah, there's there's a ton of there's a ton of stuff going into media plans like I have a new product. You know, it removes toe fungus. Uh, you know, what should my ad say and who should I target and what property should it be on and how much should I budget? Like, cloud will do pretty good at that right now.

46:39 It'll do pretty good at that right now. So, I just want to say if that's what you're building, like have you really thought through your ability to charge for this compared to a what cloud could do right now and b what an agency that has access to, you know, some amount of licensed proprietary data and knowhow can do with with nine co-work and a bunch of MCPs. >> Yeah, that's interesting. I mean, I've had that thought a lot where like the reason I think this is so great for agencies is the training data that they have that's kept proprietary. Like they have all of their media plans and how they worked out.

47:18 >> You see, that's not publicly wrong. >> Is it wrong? >> No one is coming to the agency and say, I want the average of your aggregated wisdom. That's why >> because the average of aggregated wisdom is going to be abundantly available. >> Already there's six. Yeah, it's not that hard to get. >> Well, that's not good. >> Well, if there is a way to be I would say if there is a way to make that useful, it's do it's doing it per advertiser, which is a completely different architecture from how they're thinking about it. Right. So the the a the chalice thesis would be all right you have really valuable data on how to do things you have 400 client campaigns across the last 10 years that's amazing now your architecture should be how could you take one client data >> and model it all the way through on that to a specific outcome right the last thing you want to do is lock down what features and output types you're going to get out of this corpus right you want every you want all the >> right that's the way chalice is built all the architecture is made for per advertiser That's kind of what I'm saying though, right? Like I don't I don't mean like you get the average. You get you have a specific type of advertiser come in. The system goes in says, "Okay, what do I have that's closest to this advertiser, closest to this outcome? How did it go?"

48:39 >> Yeah, that's that's the wrong architecture. >> No, you should look at >> the wrong architecture. >> So, wait, so what's my distinction between what I just said and what you just said? You take a training seed from an advertiser and all that is just spine. Like you're not averaging that out at all. You just score everything in there for whatever you're trying to do. So like if what you described the first thing you do in that architecture would be like all right let's let's you would create an architecture so that all that can be averaged and you can come out the other side.

49:18 I see what you're saying. Like so the the idea would be the way that I would think about it is you don't you don't average it, right? You have like the granular data. You have the campaign results for all of the CPG brands for all of these things and then you have some standardized fields that are shared between the desired advertiser and the config for everything else. And then it goes through and says, "Okay, well based on the fields that are shared and based on the results that I've seen over here specifically, like these are the things that correlate to what you should be doing."

49:50 >> Yeah, that will give you the average result. >> It's the average >> I'll give you the average result. I'll go through next week about how do you get >> the very specific result to the very specific outcomes. The reason I would give you the average result has to do with with features and feature engineering. like the architecture you described will will nail down a set of features and how to engineer them to get to the output and that's exactly what you must not do to do per advertiser.

50:18 >> So does that mean so basically the way you run it >> so the alternative just so I understand architecturally what it is is you start the advertiser out running blank then you let it buy everything. >> Yes. And then you c go in and you say, "Okay, well seeing what's happening for this advertiser." And so do you use the historical data at all? Like you you look at like, "Hey, do these correl are these correlations shared? Are there power?"

50:42 >> Only if you can match it to the real brand outcome only if you can match it to the real outcome. That's that's where it gets dicey, right? Because normal architecture would give the same output for Coke and Pepsi. They're both trying to sell soft drinks. >> Sure. >> Right. But what you actually have to do is get Coke drinkers to switch to Pepsi and Pepsi drinkers to switch to Coke. Right. So, so they they have to be completely different. So, you might use that data, but you're going to use them in completely different ways.

51:08 >> That's interesting. Um, so I think what it comes down to in the end then is that the the outcomes that you are modeling for are even if they don't seem different, they are different. >> You have to treat them that way. that they're not in your data set. They're not in your data. >> It's just not possible. >> Or they might be in there somewhere, but it's going to be like needle in a hay stack. You can't assume they're in there and tag features of them and and and elevate them.

51:36 >> It's like if somewhere in this data set we have Coke drinkers who switch to Pepsi. Yeah. We'll be able to use that and model on it. Otherwise, we have to generate it in this campaign and use all this all this as just a spine or something to help us understand what else is out there in the universe and how it should be scored. Oh, that's so interesting. So, that makes perfect sense to me for enterprise advertisers, right? No enterprise advertis enterprise advertisers should be spending the money to develop their own data set for their own goals. And the only place where you'd want to use an average is if you simply don't have enough money to build the model where somebody's budget is just >> or not enough data >> or not enough like Yeah. So, you can't get enough data to like >> another way to think of the trade-off is the per advertiser model starts slower and compounds better, >> right? It's an investment in itself.

52:20 like it gets better and better and better and better the more you spend on it, >> right? But to the platform point of view, it's like start slow, we can't do that. We need it to perform. >> Yeah. It's got to perform the second they set it up or they're going to turn it off. Like, >> right. So, they start different. >> So, where that's headed is that the brands eventually will realize, >> why am I working with this platform? Why do I need this? like I like I have what I need in my data and I just need to bring it to market as like a sort of a a model based off a private valuation. And I would argue >> Oh, it's so good. That makes perfect sense. I get it 100% now. The idea is that if you have enough money to spend a customized model for you, your out the outcomes are so actually distinct that there is no reason to actually use data from elsewhere. Um, and now where this gets really interesting is also this model needs a consolidated delivery mechanism so that it's not an ETL nightmare to try and marry up the different systems that are delivering like all over the place and consolidate the data for it to look at its outcomes across all of its media. And that's why I'm screaming about modularity and artf shared compute environments. Like this is the infra where private valuations are going to be dropped.

53:36 >> Yeah. Uh and god it makes so much sense. And it's also why direct sales makes this super annoying because you're no longer involved in valuation like you're value you're now val this because what I'm so GH adtech article that I've been noodling >> is all inventory is actually traded at the impression level because we talked about direct sales like being bundles and programmatic being direct like individual decisions and I was a huge proponent of that framing. I was like one of the earliest people to use that framing where direct sales is bundled and programmatic is impression by impression like the bundles will work.

54:14 >> Yeah. >> And I am I am recanting because it's not that it's bundled. It's simply that the advertiser doesn't have control over the impression level decisioning. The impression level decisioning is still occurring. There is still a bid happening. It's just happening with lower resolution. It's the same bid over and over again because most ad servers for publishers do not have dynamic valuation systems in them. And even if they did, they're not hooked up to your advertise your model as an advertiser.

54:45 So, it's not going to value things properly. So, it's not necessarily bundling versus not bundling. It is the does the impression level delivery system that exists in both places actually have the ability to do a good job. And that is where I think a lot of these direct sales automation discussions will fall short is for a sophisticated advertiser like the b it's like how smart you >> still need the execution layer. You still need the execution.

55:11 >> It has to be there. It's it's it's there no matter what. >> To talk about finance. >> It helps to talk about finance. Like the biggest people with the most money pulled out of of brokerages and went to hedge funds, right? They did not become day traders going bit by bit by bit. That's essentially what they wanted. The reason they went to hedge funds is because brokerages have essentially they're a platform. They have their own interests. They have their own commitments and their decisions tend to be the average of all their customers decisions. Whereas a hedge fund is is like usually one guy's idea of how to make the most possible money, right? And the whole team and the whole team and the whole infrastructure is based on bringing that to market in its pure form without it getting averaged out in any sort of platform. So hedge funds aren't for everybody. Brokerages didn't go away. But if you have a if you have a billion dollars to invest, you're you know you're much you know a lot of people choose to go to hedge funds because you're bypassing that platform aggregation layer.

56:14 >> Yeah. I it's this is >> this makes it's just the the argument is just so strong for programmatic bidding systems in this like in this schema right because if you have one model that model needs to be able to value everywhere and standardize its outcome modeling against the bundles that it's valuing like that it needs to happen across ideally as much of your media as possible. Um, yeah, >> and CTV is really exciting for that, right? If if linear really moves to CTV and buyers can do this across banners, app, and CTV, that's pretty wild. Like from a marketing perspective, that's a like all digital becomes a big chunk.

56:56 No, we're there. >> I I went I kind of went from being like, you know, annoyed to amused that the whole market was chasing Brian Okeelly's vision of automated guarantees. And now I'm becoming alarmed about how much money is being ve invested in this control layer, right? It is where the money is committed and how little attention is being paid to the execution layer, which is where ROI is either achieved or not. Um um I will write this article about this. I need to get it out because it is I think actually the the bundle versus unbundle distinction I've decided is the wrong one. It is how far away is the advertiser from the execution layer. And in one of them they're a lot farther away than the other. And like and then then the farther away the advertiser is from the execution layer, the worse the valuations are going to be. The lower the rorowaz is going to be. And here's the kicker for publishers. the less money they're going to spend. That's the thing here. Like they're the ROI is worse, the less money they're going to spend. Um I hope and I think it's the case.

58:03 >> Yeah. The way to think about the execution layer is there's so much money to be made there from spreads that somebody's going to make it. Somebody's gonna somebody's going to exploit spreads, right? Just like any market with a heavy execution layer, in any given minute, something is undervalued and there'll be an arbitrage opportunity. Somebody's going to make that money. M >> yeah maybe that's the the synic explanation for the obsession with it right >> is the people who are obsessed with it don't see it as a way to make campaigns work better they see it as a way to get spreads >> yeah that's yeah the platform voice is very loud >> yeah I think it's super interesting I love the framing >> um thank you Adam I enjoy oh no thank you thanks for hearing me uh what else did we have planned for this week >> yeah there were a couple other news items Yeah, let's just play. Let's just put down our bets. Is uh is OpenAI going to make a billion in their first year? And is SpaceX they said this is this tip sheet piece is saying is that the S1 for SpaceX is says that Grock is going to make a ton of money from advertising. How much money?

59:14 >> Oh, wait, wait, wait. No, no. This is Twitter ad revenue. This is Twitter ad revenue. Why is it part why is it part of SpaceX? >> X is part of SpaceX now. The whole thing, >> right? So, Twitter ad revenue driven by XAI is going to be a money maker for parent company SpaceX. >> Oh, they they think that uh it's going to go back up over two billion on X. >> I think they say four billion.

59:40 >> Well, it was it was for it was four when he bought it >> and then he managed to evaporate half of it by suing advertisers. I don't know whose idea that was. Just the customer idea. >> It's very interesting. >> They're the customer. They are the customer. He's a genius. >> Jesus. >> Yeah. I wish I'd been in the at the conference table to be like, can we just not do this? >> This is what people care about. Will Open AI and SpaceX make a billion from advertising in their first year. I'm going no and no.

60:13 Uh, I definitely bet that the chat GBT ads are do not make a billion in their first year 100%. I have I mean I have inside baseball on this one because I know the open ads guys. >> No, the open ads guys like uh that not the trade desk product the company called open ads. Their original product was ads in LLMs. >> Yeah. >> And they were like the CTRs are not there. the percentage of conversations where someone is in a high mindset not big and you are not going to be able to build a search style ad platform.

60:46 >> I don't see why not. >> If I'm talking to track GPT about my toe fungus and the toe fungus ad comes up, I'm clicking that. >> They said there is a very small subset of searches that have crazy CTR and then a bunch of them that just >> Oh yeah, I just remembered I don't have to fungus. But you might one day. Isn't it? >> Who does? How many people have this condition? The advertising seems totally out of proportion to the problem.

61:16 >> It's because the ROI is really high when someone does. I actually don't know what's behind a toe fungus ad. Um like what are they actually selling >> product? >> Yeah, it's an antifungal cream. Yeah. >> Yeah. But they have those at the the store. >> What? alcohol kill it. >> Also, like the the single sale return on that is low. Normally, those guys have subscriptions and that's how they make their money back is it's a it's something called a rebuill, which is if you can get someone to pay for something and then make it really hard for them to stop paying for it. The LT, >> they're just juicing old people.

61:50 >> Yeah, this is America. >> America. >> Amazing. Oh, >> yeah. I'm not I'm not hot I'm not hot on ads and chatbt. >> What's Jay saying in the chat? They did finally rebuild the ads manager, but it is still diabolical and lack strong pixel tracking capabilities, Jonathan. Yeah, we don't. And we're we're we're bears on it. They could switch teams any number of times. They have this Facebook team now.

62:24 >> That's probably better than a search ads team to be fair. It's probably closer to Facebook. Search ad search ads team would have been better. >> Yeah. >> I think I don't think people think enough about when you come into a market they always think it's growing and I think what you think about should be like who you taking budget from >> right? This is a like they should the whole thing should have been how do we get search budget out of Google and maybe video budget out of YouTube. Like that's like a loaded balloon of money.

62:52 And I think the whole strategy of open AI I guess I don't know they probably have 200 billion dollars of Google money invested. So maybe that strategy wasn't possible, but that's that's what I would have recommended. We >> should have the the open ads guys on. I'll ask them. They they they're very opinionated about this because they >> I mean I'm sure their original vision was we're going to build an ads and LLM product and we're going to get bought by OpenAI. That was like the plan. Oh yeah.

63:17 And then they built it and they integrated with all these LLMs and they were running the advertising business and they were like [ __ ] the performance is worse than Google search. like like even though we're good at contextualization, like we we're observing the alignment between the context and the ads. We're making sure it's lined up super nicely and just the the percentage of chat GPT and LLM conversations that lend themselves to high intent is not the same as search.

63:41 Um it was it was very interesting. They're opinionated about it. >> Who um that company? What are their names? >> Open ads. It's uh Oh, comment was more about X ads. John Duza. Oh, good. >> Yeah, yeah, yeah. Uh, it's Steve Liss and Michael Bishop. They're great. >> They haven't been quiet yet. >> They're running a DSP now. This is better. >> Oh, yeah. They have it to DSP. Perfect time for it. I hope it's containerized.

64:08 >> I don't know. They certainly use LLMs a lot. >> Yeah, guys. Anything else? >> I feel like that went really fast. >> It did go really fast because we were really hard on like philosophy today. I knew it was going to happen. I was like, "Oh my god, this is this Pope thing is so nuts that like it's he's so it was a really deep paper. It really got me." Um, >> we never know. We never get feedback on what episodes people like. Like we have numbers of viewers that that are different, but that doesn't seem to be related to the content, but we should we need some way to get feedback to know if the viewers >> we have hundreds of viewers a week and we don't know what kind of material they like or what we should do more or less of.

64:49 >> I got really good feedback on the Matt Sat Satel episode. People like the >> Yeah, people like that one. >> That was a good one. >> Yeah, he was fun. All right, guys. Well, thank you so much. Thank you for joining chat. Thank you for everyone tuning in. >> We'll see you next week. is >> darn. >> Yo, the crowd is pack. The lights are dim. Front rows buzzing. It's about to begin. Ad talk to taking the stage. Adam and