Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

Shell Game

Shell Game artwork

Episode 2: Hear What Matters

Oct 28, 202648 minAudio

Evan goes looking for the humans behind the flood of AI-generated podcasts, and gets some complicated answers to his questions about what they think they’re making and what AI slop means to them. Meanwhile, Shell Game’s own bot-hosted shows move to the prototype stage, and we hear the first efforts in our quest to make an AI-generated show that will truly resonate with listeners. This episode includes AI-generated content.

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show. Tap a time to play from there.

Speaker 1Do you ever think about adding some new positive habits to your day? Little life hacks that might make you more efficient or organized or even happier? I've been listening to a guy who's got some ideas.

Speaker 2I tested 30 habits that hardly anyone talks about. No 5 a.m. alarms, no cold plunges, just tiny habits that quietly transformed my life. The scary part? Most people waste years ignoring them.

Speaker 1This is a podcast called Ideas to Thrive, hosted by, well, he doesn't have a name. or an identity of any kind. He's just a disembodied voice full of habits.

Speaker 2Habit one, no matter what happens, don't let a day go to zero. Zero pages read, zero words written, zero minutes working out.

Speaker 1As you probably guessed, the Ideas to Thrive host is AI generated.

Speaker 2Most people try to completely reinvent themselves overnight. Like they wake up on 1st January and decide they're going to be a completely different person.

Speaker 1The system I built for scanning and identifying AI-hosted podcasts flagged this show because it uses one of the default voices from 11 Labs, where it's called Liam.

Speaker 2See, here's Liam. I can use him too. I can have him tell you that Ideas to Thrive has more than 50 episodes, ranging from 20 minutes to over an hour long, describing how to become more productive and fulfilled and successful.

Speaker 3Anyway.

Speaker 1On YouTube, where the episodes are accompanied by a spare AI-generated cartoon video, the episode titled 30 Habits That Quietly Transformed My Life has over a million views.

Speaker 2Habit seven, water first. I drink water before any other drink. Want coffee? Water first. Want a Coke? Water first. Started doing this because I was chronically dehydrated and kept getting headaches.

Speaker 1Sir, you have not been dehydrated, let alone chronically. This was one of the most irritating things about AI podcasts. It's one thing to get life advice from a bot. It's another thing for that bot to pretend it's alive. Ideas to Thrive is like personal growth arbitrage. Advice liberated from its human providers and repackaged at scale. The 30 Habits episode offers up a list of sources. Books like Charles Duhigg's The Power of Habit and James Clear's Atomic Habits.

Speaker 1It feels like someone dropped them into an AI blender, let it rip, and then poured them back out as a kind of, well, self-improvement slop.

Speaker 2If you look at the comments, though, it's not at all clear how much of the audience realizes that it's an AI voice advising these habits. Or if they do realize it, they don't seem to care. When a single commenter deigns to point out that, quote, This.

Speaker 1Whole channel appears to be AI gathering ideas from everyone else's real videos and research.

Speaker 4Another person, or bot, fires back, who cares?

Speaker 1And a third adds that, quote, it may be AI, but some of these are good advice, and these tactics really work.

Speaker 2Habit four, get weirder. I decided to just get like 1% weirder every day. Not better. Not more productive. Weirder.

Speaker 4I dig it.

Speaker 3But then.

Speaker 1After 100 days of getting 1% weirder a day, wouldn't you be an absolute nutcase? Maybe he meant 1% weirder than I would have been that day.

Speaker 4Still.

Speaker 2Turns out the opposite of weird isn't normal. It's boring. And boring people don't change their lives. The goal isn't to become some completely different person. It's just to expand what feels comfortable until being uncomfortable becomes comfortable?

Speaker 4That doesn't make sense. You know what I mean.

Speaker 1I do know what you mean. Ideas to Thrive guy, but who are you? This was the question that plagued me, listening to volumes of AI-generated audio. Who was creating something like Ideas to Thrive?

Speaker 4What were their goals?

Speaker 1I decided to find out, but it turned out to be harder than I thought. I subscribed to the site's newsletter, whose emails arrived with only the name Ange and the address of a mailbox service in New York City. I noticed the site seemed to run its email through an Indian mail server, but I couldn't get anywhere more specific. My emails and reply to the address went unanswered. I even chipped in a few bucks to buy an online personality map from Ideas to Thrive, hoping the receipt would turn up something. The map, incidentally, was just a PDF

Speaker 1file with more tiny habits. The address on the receipt was a mail drop in San Francisco. There was a new email address. and I sent a note asking if Ansh would give me the perspective of someone building an AI-generated podcast universe. This time, I got a polite reply, which I like to imagine in Liam's voice.

Speaker 3Hi, Evan.

Speaker 2I enjoy your work. I find your experiments with AI fascinating. I would love to chat with you about this someday. However, I'll have to take a rain check for now. Best, Ansh.

Speaker 1Human or bot, the mastermind behind Ideas to Thrive wasn't talking. And I was going to have to look elsewhere in my attempt to find someone who would explain the perspective of an AI podcaster. Fortunately for me, Ideas to Thrive was a small fish in the reservoir of synthetic audio. And I knew where the big base were.

Speaker 4I'm Evan Ratliff, and.

Speaker 1This is Shell Game, where for this three-episode season, we are immersing ourselves, and you, sorry, in AI-generated podcasts. trying to figure out what is slop, who's behind it, and what it means. This time, we're exploring the second question. Who's responsible for making the hundreds of thousands of AI-generated podcasts out there?

Speaker 4What are they after? What do they think slop is?

Speaker 1Me and my shadow.

Speaker 5Strolling down the avenue.

Speaker 1This is episode two, Hear What Matters. September 9th, 2025, arguably marks the dawn of the AI-generated podcast age. The inciting event was an article in The Hollywood Reporter about a startup called Inception Point AI.

Speaker 2And it's completely, absolutely AI-generated, right the way down to the voice.

Speaker 1More than 5,000 shows. Again, artificial intelligence shows, so these are not real people. They make more than 3,000 AI podcasts a week. In the article and interviews that followed, the company's CEO, Janine Wright, a former executive at the podcast maker Wondery, was clear and unapologetic about the company's model. By automating the podcast production process with AI, from market research to idea generation to scripts to voices, they'd reduce

Speaker 1the price to make an episode to a dollar or less and the time to under an hour. They were making them by the thousands, and with so many shows, they could sell ads on their entire aggregate audience. They'd assembled these shows into what they called the Quiet Please Network, Slogan, hear what matters. Wright claimed that this all meant the company could turn a profit on an episode with fewer than 20 listeners. Suffice it to say that many people who heard about Inception Point, particularly those who were involved

Speaker 1in making podcasts themselves, were displeased with this approach.

Speaker 3So this company is called Inception Point AI.

Speaker 6I have been digging into these podcasts and they are so much worse than I imagined.

Speaker 4Does Janine listen to these things? Does anyone? Listen to them.

Speaker 1Part of the fear was that Inception Point and other networks like it seemed to herald a future in which synthetic content would simply overwhelm human-crafted work, drowning us in slop. I even got in on the action. In an episode of this show, from our second season last year, I referred to Inception Point as part of a tsunami of audio garbage that was headed our way. After we published that episode, I got an email from Janine Wright. She wrote that she was flattered to have been mentioned on Shell Games. adding that she would love to challenge my

Speaker 1thinking on, quote, AI-generated content as AI garbage. She noted that we had some things in common. I was building with AI agents. They were building with AI agents. I'd produced an AI-hosted podcast, The Startup Chronicles, as a satire, but still. They'd produced, at the time, between 5,000 and 10,000 of them. It was a magnanimous email to send to someone who had just cast aspersions on your company.

Speaker 4Savvy PR, maybe.

Speaker 1But I took it as genuine. Perhaps I had been too glib. I proposed we have a conversation about it all. Maybe for this show. But she never wrote back. So I was 0 for 2 on getting the humans behind these shows to talk to me. And then my AI podcast scraper tool turned up this.

Speaker 7It's September 17th. I'm Corey with The Story, and this is San Francisco News Today, your AI-powered local news.

Speaker 1A tragic incident occurred early Thursday morning when a boat struck rocks near the.

Speaker 4Golden Gate Bridge.

Speaker 1Of the San Francisco iteration of Daily News Now, another podcast network that features hundreds of AI-generated news shows. These fall into the quick-hit variety of my AI-generated podcast, Taxonomy, one- to three-minute shows covering local news by city and sports news by team.

Speaker 4The San Francisco Show.

Speaker 1Published 39 separate episodes on this day alone. I contacted DNN's founder, Corey Cambridge, to see if he would talk more about what he was up to. He politely declined. But a few weeks later, I got an email from his partner at DNN, Joseph Meyer, who was the mind behind the technology side of the network.

Speaker 4What's up, Evan? How are you doing? Doing all right.

Speaker 1I'm happy to see you because years ago, I wrote a story about a Joseph Myers whose name is spelled like yours. but he pretended to be a European prince and he conned all these people. So when I got your email, that's why I didn't respond at first.

Speaker 6Well, obviously it'd be difficult for me to pull off being a European con.

Speaker 1Joseph, who has a background in building digital advertising and mobile apps, said the original idea for DNN came from Cambridge wishing he could listen to news articles on Spotify instead of on a news site like the New York Times. They pitched their idea as a service to a few publications. they didn't have any luck. So they decided to go their own way.

Speaker 6I like doing permissionless things, you know? And once generative AI and these LLMs came out, I was like, hey, we can effectively be journalists or rewrite these articles and distribute them, write this content and distribute it.

Speaker 1Corey Cambridge had cloned his own voice to create the host, Corey with a story. And Joseph had built a system that would allow them to pull facts from local news articles and churn out episodes. Poking around, I had discovered that the investigative newsletter Indicator had conducted an analysis of the network early this year, estimating that it was putting out as many as 11,000 new episodes a day, a considerable portion of them ripped directly from a single local news story. This, unsurprisingly, angered some of the writers and

Speaker 1publications behind those stories, who called it copyright theft, or a blatant act of plagiarism, or both. Corey Cambridge told Indicator that these episodes were, quote, edge cases. Joseph told me that after that, they'd tweak the system to pull facts from multiple sources, which they thought of not as plagiarizing, but as summarizing. But even setting aside the plagiarism question, I was interested in how they viewed their shows vis-a-vis other podcasts. Joseph said they didn't think of what they

Speaker 1were doing as competition for human productions, or even as a podcast at all.

Speaker 8The term podcast, to me, it kind of doesn't accurately describe what we're doing. So when people think of a podcast, they think of a content format versus like when I think of a podcast, I think of a delivery method.

Speaker 1He preferred the term fastcast, which he saw as something completely different.

Speaker 6Ultimately, I'd like to establish fastcast as this kind of short form automated audio about just, you know, information.

Speaker 1This was a fair point, one that we'd struggled with for months. Podcast is basically a name for a specific digital way to deliver audio. And it encompasses a spectrum that travels from just providing information, think weather reports, to entertaining people, think comedy podcasts, to telling complicated stories, think serial. When talking to someone about AI-generated audio, there was a risk that we were operating from different places on that spectrum.

Speaker 6I really value two humans on a microphone or one human on a microphone producing content. I want to be clear about this distinction in my mind of this taking information and converting it to being consumable in audio.

Speaker 4I'm building the infrastructure for that future. Yeah, I think that is an interesting distinction.

Speaker 1One of my questions is, I assume the volume is great enough that there's no way that you can quality, like no one's listening to these episodes before they go out.

Speaker 6Right, and we have some filtering processes where we try to do our best to not slander people. We are promoting and aggregating official news stories, right?

Speaker 8It's like we're doing our best to stay within the realms of like, you know, what's the word I'm looking for?

Speaker 4Ethics, right? Ethical standards. That was the thing.

Speaker 1About the AI-generated podcast landscape. There weren't any rules or anyone to stop you from cranking out millions of episodes a year just to see what worked. All of which drew me back to the original pioneers of AI-generated audio, Inception Point AI.

Speaker 9Welcome to Biography Flash from Quiet Please Podcast Networks.

Speaker 7Search Biography Flash wherever you listen.

Speaker 1Inception Point had actually made such a splash that before I made my clod-coated pod scraper, the easiest way for me to find AI-generated shows was by searching for the Quiet Please network. For a while, they were making a lot of informational shows about animals. Salmon, snakes, whales, Bigfoot. These were the names of the shows. And they were cranking out biography flashes of prominent people. These weren't one-off profiles, but more like a kind of serial fanzine ripped from

Speaker 1the headlines. Like this one, of the Spanish soccer star, Laminia Mall.

Speaker 3Am I being.

Speaker 1Ungenerous when I play you the most gaffe-prone episodes of these shows? Over time, you could sense that they were improving.

Speaker 3You, what is up? This is Tyler Ty Morgan, your AI host on Biography Flash. And yeah, I am an AI. which is actually perfect for this job because I do not sleep, I do not miss a headline, and I definitely do not forget receipts when it comes to a kid like Lameen Yamal.

Speaker 1As the year went on, Inception Point seemed to turn its attention to establishing ongoing AI characters with personalities and backstories who not only hosted podcasts, but had their own YouTube and Instagram accounts. Like Nigel Thistledown, a white-haired, bearded, gentle-seeming Englishman who waxed lyrical about the fine points of gardening.

Speaker 5You see, the drama of the garden isn't found in the petals or the perfume. It is found in the struggle between the light and the dark.

Speaker 4Now, let us examine the hydrangea fantasy.

Speaker 1I actually love gardening, but Nigel wasn't for me. It was hard for me to envision the actual audience for someone who just chatted about gardening in this random fashion. But then people like all sorts of things. Inception Point had a cooking character, Claire Delish, and Vivi Steele, your source for the best celeb news. The character I felt most connected to was a sports reporter named Ollie Bennett.

Speaker 3Every way to learn a city, find the court, the field, the game.

Speaker 1Local legends everywhere, and every game tells a story.

Speaker 5The key moment, and this is where they won the match.

Speaker 2was that substitution in the 60th minute. It changed everything defensively.

Speaker 1Yes, those are both meant to be the same character. Oli was a sometimes American, more often British, host of a wide range of podcasts on the Quiet Please network. Often they were around his interest in niche sports, like professional spike.

Speaker 4Ball, kayak polo, caber toss.

Speaker 1But where Oli really shined, in my view, was on Instagram. where he could show off the globetrotting lifestyle of a handsome man reporting on sports and sometimes participating in them.

Speaker 3Sauna, towel, ice water.

Speaker 7This is extreme ice swimming.

Speaker 3In Scandinavia, winter bathing has been around for centuries.

Speaker 1The thing that may not come through in what you're hearing is how absolutely jacked Ali is. Striding toward the camera, muscles rippling, wearing only a towel. I started to think it wasn't really about the sports. As all these code-switching accents showed, these personas didn't seem to be tended with a great deal of care. Most intriguing for me, though, a few of the characters offered a link to a webpage where fans could speak to them. As past listeners to Shell Game will know, speaking to AI characters is my passion.

Speaker 4I was ready to connect.

Speaker 1But if the buttons had ever been functional, by June, when I tried them, they no longer were. Soon, the webpages disappeared entirely. I started noticing that Inception Point podcasts were vanishing too. The company was still claiming to have 10,000 active shows, but when I ran my own AI scanner in July, I only turned up around 1,000. Within weeks, the number was down to a few hundred. Something was going on at Inception Point AI. It looked like it was crumbling before my very eyes and ears. I decided

Speaker 1it was time to try Janine Wright again.

Speaker 10You know how hard it is for us to design ugly characters? Like, AI drift towards good-looking people.

Speaker 1So all the time in AI— Well, my favorite character, my favorite Inception Point character is Ollie Bennett, who is the sports reporter who is the most buff.

Speaker 4I have known many sports reporters. Some of them are fit.

Speaker 1I've never seen one like that.

Speaker 4He is absolutely jacked.

Speaker 10Well, you know, that was overtime. getting feedback and data from audience that was telling us that Ollie performs better when he's more muscular. And so if you look at early versions of Ollie, Ollie was not as muscular. He became more muscular over time.

Speaker 4He also became British.

Speaker 6Yeah.

Speaker 1When I'd emailed Janine a few weeks before, she'd replied enthusiastically, happy to talk on the record about any aspect of Inception Point's business. When we sat down, she told me she'd joined the startup not long before the Hollywood Reporter article. Before that, she'd started and sold a company called Simplecast, then gone to work as the COO of Wondery, a podcast company I had actually made a show for myself, though I never worked with her there. By 2024, she'd left Wondery and begun searching for a new startup idea,

Speaker 1together with a small team of colleagues.

Speaker 10I got the craziest email that I've ever gotten in my life.

Speaker 1The email was from William Corbin. He'd founded an AI podcast startup he was calling Inception Point.

Speaker 10And he said, Hey, you might not remember me, but I was a customer of Simplecast and I'm building something in this space. And this might sound crazy, but I essentially put all of my LinkedIn contacts into AI. And AI says that you should be the CEO of our company. You should come join our company. So then we have this lunch. And it was a little bit surreal because he's so AI forward that... It was almost like it was me, him, and Claude having lunch. You know, he's like, here's all the things.

Speaker 4Like he was using Claude at the table, at the lunch table?

Speaker 10Yeah, he's like, here's the reasons why Claude says that you should join the company.

Speaker 4Claude was persuasive.

Speaker 1Janine and her team decided to join forces with Corbin.

Speaker 10And it was clear very quickly that we needed money. Everybody was working for free, working for equity.

Speaker 4And those tokens cost money.

Speaker 10The tokens cost money, so we were, yeah. We knew we needed capital very quickly. And, you know, you need to be able to get attention to break through the noise. And so I designed our launch PR strategy.

Speaker 1And part of that strategy was a big public announcement of their network of AI podcasts in The Hollywood Reporter, an article that would light up the group chats of the humans who make podcasts.

Speaker 10I think the part of it that I guess I... regret or recognize in hindsight was pretty poor timing is I had left Wondery at Amazon at that point a year before and Wondery was going through a lot of changes and they were in the process of doing some, you know, significant layoffs and restructuring.

Speaker 1Just a month before the Hollywood Reporter article, Amazon had announced it was dissolving Wondery and laying off 110 people.

Speaker 4We were in the process of not existing anymore, as far as I could tell.

Speaker 10Of not existing anymore. I mean, some of the folks are still there, but a good amount of the company has moved on and has been dissolved. And this article coming out right after that, I think, felt like I was, you know, like dancing on a grave or something, you know? And that was certainly, you know, not my intention, right? Yeah, there were some hurt feelings through that process and that

Speaker 10I definitely regret.

Speaker 1It wasn't just former colleagues, though. Some of the public vitriol was intense.

Speaker 4Some of it, disturbing.

Speaker 10I mean, we got so many death threats that we had to make a filter for the word kill. I had people dox me online and tell people to send stuff to my house, you know. I was dubbed the slop slut because misogyny is still alive and well, you know. I, uh, Frankly, I was surprised about the extent to which the response was so angry and pointed.

Speaker 1Janine decided that the company would take the AI slop accusations and other concerns head on. They published a blog post titled, Inside the AI Slopocalypse, Why We're Not Backing Down.

Speaker 10And it became this rallying thing of like, yes. We want to take all this feedback into consideration and we want to be values-based builders. But we think that we're right and we're willing to, you know, go out there and say why.

Speaker 1What did it mean exactly?

Speaker 4Values-based. Whose values? Valuing what?

Speaker 1These were the answers I'd come to find. In their blog post defending themselves, they'd argued that, quote, we believe this is the first time in podcasting history that that ultra-niche content can be created profitably as part of a sustainable business model and serve micro-communities that have never had access to it before. But it wasn't clear to me how Nigel Thistledown or Biography Flash fit into that vision. Certainly there were gardeners and home chefs and sports reporters

Speaker 1out there already. Do you think it should be valued the same? If I have a St. Lawrence River Fishing podcast that's made by a human, and you have a St.

Speaker 4Lawrence, which you do, St.

Speaker 1Lawrence River Fishing Podcast, that is entirely AI-generated. All the information is automated. The pick-the-voices has done everything. No human has ever heard it. Do you think there's any difference in value between those two shows?

Speaker 10I don't think that I think... Well, do I think that there's any difference in value? You know, it's an interesting way to frame the question. I do think that. The overwhelming majority of the information of the world is not yet available via audio. And I think that people want to be able to easily find information. People want to be able to consume content in the way that they want to consume content, not necessarily in the way

Speaker 10that the content creator wants to make it. I mean, really, I think it should be looked at more from the lens of Like, is it useful? Are people going to consume it? Do they like it?

Speaker 1People have different tastes. They like what they like.

Speaker 10Yeah.

Speaker 1But if you make thousands and thousands and thousands of versions right now of all these things, and you just throw them into the marketplace, that feels like that's doing something different. I mean, you've often said, like, all we need is 20 people on each one in order to make it work.

Speaker 10Yeah, but the strategy has never been we should just throw out so much. so that, you know, there's 200 to one and they have to fall on our content versus somebody else's content. I mean, we have.

Speaker 1Eventually, we got around to the question of slop.

Speaker 10I mean, even when you think about the conversation around, like, what is slop that I kind of sometimes reframed as what is art, you know? Like, when you think about, like, paintings, how much did it cost to make that? How much did it cost to create that painting?

Speaker 4Well, I had to buy.

Speaker 10the charcoal and the paper that it was made on. So we don't value it less because it was inexpensive to make.

Speaker 1Her point was that people were judging AI harshly just because it could make human-level output more cheaply than we could. But we value it because of the human endeavor. So if you walked into the Met, we're here in New York, let's say, or the Guggenheim, and there was a room full of paintings, and I told you, 80 of these 100 paintings AI generated yesterday and had them printed out, would you value them the same?

Speaker 10You're now talking about amount of effort. So if it took Picasso one minute to make a painting versus 60 minutes to make a painting as one, then valued more because of the other?

Speaker 1I think it's just human endeavor. Is there any art that you care about? Music, maybe it's visual art, books, where you would be upset to discover that someone had produced at scale many versions of the same thing that a human had also produced and you couldn't tell the difference.

Speaker 8I don't know.

Speaker 10I heard this interesting podcast recently where this person, shoot, I'm going to forget which podcast. It was either Radiolab or This American Life, but it was an old episode where they were talking about this woman who found out that her hometown song was actually a song that had been sold to hundreds of hometowns across America with slightly different versions, right?

Speaker 11And she was devastated, right?

Speaker 10It was like, no, that is my hometown song, right? And so I think there is something that we maybe value as humans around rarity, that it's like, this is difficult or hard to do, or is so personalized that it feels like mine, and that's what makes it special and more valuable.

Speaker 1It was this American life, and I took her point. But it did seem relevant that when she'd search for an anecdote to delightfully illustrate the idea that art or culture can mean something to us no matter how it's produced, she'd turned not to one of Inception Point's 15,000 podcasts, Instead, she'd called upon an 11-year-old episode of one of the world's most distinctly human-made shows. I kept trying to offer up examples and analogies to determine whether at any level she valued human creation over that of LLMs. She kept

Speaker 1suggesting that more information more audio, more choice was always better.

Speaker 10But then I guess what you're saying is that because somebody is already doing it available online, that I shouldn't be making an audio version of that?

Speaker 1I'm not telling you you should or shouldn't do anything. I'm asking whether you value those things differently. I'm trying to get at if a human is willing to do it and thinks it's satisfying and rewarding and they have an audience, like, should we value that higher in any way?

Speaker 10I mean, I guess my, my intuition is that it should be a meritocracy that essentially, you know, the, the, something should not exist, um, to the exclusion of something AI created just because it's human-made. I think that very soon the overwhelming majority of content will be created materially

Speaker 10with AI, and people won't care.

Speaker 1At some level, It felt like we just lived on different islands. On my island, podcasts could be hilarious and moving and thought-provoking, created out of someone's joy or pain. A well-crafted story could carry a listener along. They could lodge in someone's memory, like a splinter. At their best, they could make you feel a little less lonely.

Speaker 4They could be art.

Speaker 1Janine, on the other hand, was a startup founder at heart. looking for places to disrupt. I believed her when she told me she loved the audio emanating from my island. But on hers, a podcast was less some crafted work made of human experience and more a kind of information product looking for a market. If she could bring something new to audio consumers and they liked it, that was the very definition of success. But had Inception Point succeeded

Speaker 1on those terms?

Speaker 4Well, let's talk about where you are.

Speaker 1And first, tell me what's going on with the shows, because the shows seem to be disappearing.

Speaker 4It wasn't just the shows.

Speaker 1A few days before we met up, the company's website itself disappeared. Its old URL redirected to the company's new home as Liquid Studios.

Speaker 4So what's happening with that?

Speaker 3Yeah.

Speaker 10So about Six or eight weeks ago, we started to ratchet down the amount of our content that we were creating to really focus all of the team's efforts on the B2B opportunity.

Speaker 1B2B is shorthand for business-to-business. Liquid Studios will be helping other companies, particularly media companies with large libraries of content, translate that content into different formats and channels and languages using the AI systems they've built over the past few years. This, Janine said, had been the plan since the beginning. The podcast that Inception Point had been churning out by the tens of thousands were just a proof of concept for the technology they were planning to sell. I was, let's say,

Speaker 1skeptical of this narrative. Startups pivot their business models all the time. There's no shame in it. Instagram was a check-in app that happened to add photo sharing. Twitter began out of the remnants of a startup building a way to search for podcasts, of all things. It struck me that this potentially is what had happened here. Inception Point had tried AI podcasts at scale, it hadn't worked, and now they were using their technology to pivot to another part of their business. I suggested as much to Janine.

Speaker 1I think a different story that one could tell is that that didn't work. In all of these interviews over the last year, you've been saying things like, if we do it at scale, we only need 20 listeners, it only costs a dollar show, and we avatars across that. I mean, that's a media company. You've said, like, we're reinventing the podcast business.

Speaker 4Was one of the things that you said.

Speaker 10Yeah.

Speaker 1So is it the case that that actually doesn't work? Like, why not leave the shows up if they provided those economics?

Speaker 10I guess the honest answer is I don't know if that works yet. But my story, my intention, my vision, what I promised my investors that I would build is a tech company and not a media company. But could somebody else do that model and make it work? I think probably so.

Speaker 1In some sense, we both found something to be hopeful about in Inception Point's startup journey.

Speaker 4Her perspective.

Speaker 1The future of their new company and how AI could help people who make things share them and spread them in new and different ways. Me and what I saw as a retreat from the viability of mass-produced AI-generated content as a business model. Am I wrong to detect a slight shift in your thinking, at least how it's portrayed publicly? Because if I looked at the inception point of a year ago, I would say, that included in your answer to that would also be replacing me. For a year,

Speaker 1it seemed like she'd been saying the future of creativity in something like podcasts involved AI just making everything end-to-end. Now, she was saying AI was just a tool to help with certain parts of the process, like making versions of a show in every language. But this sounds to me a little different, which is, we do need the me. but also will just help that person make what they want to make?

Speaker 10Well, a couple of, I mean, first of all, I don't know that I have ever said that I don't think that we will ever need the you. And I think there will be opportunities for real life creators to leverage AI to extend themselves beyond what they would otherwise.

Speaker 7Be able to do.

Speaker 10So if you are able to make this show available in 14 different languages and new audiences can discover and engage with your content in a short form way via AI, you are becoming superhuman with the power of AI. You're still a creator at the center of it.

Speaker 1I wasn't sure I wanted to get my stories from a superhuman, mediated through AI. Maybe there was something cleaner about Inception Point's earlier vision, the one with no human in there at all. And back at Show Game HQ, we were busy executing.

Speaker 4On that vision.

Speaker 1For a few weeks, Matty and I had been going back and forth, making tweaks to the system he'd built to create our own AI-generated shows. He'd christened the software the Slopcast stage. But our aim was to transcend slop. Slopcast Stage included a slick web interface where you could enter a show's premise, give a little direction to the intros and music, then select some host voices and personas, or let the system do it all for you. You could add a critic layer where another LLM would review the episodes and rank them according to a prompt you

Speaker 1gave it, like value surprising conversations or skip anything generic or repetitive. The system generated its own title for the show, if you wanted, and its own cover art. The whole thing was connected up to a podcast distribution platform, which deposited the shows straight onto the big podcast players, like Apple and Spotify. Once I got a hold of these controls, I did feel a little like I'd been granted a dark power. One that ultimately could lead to my own destruction. With the push of a button, I could generate as

Speaker 1many episodes as I wanted. I could make as many versions of the same show as I wanted, and then pick the version that best represented our vision.

Speaker 4Or just let the AI pick.

Speaker 1I started cranking out all the ideas we'd come up with, just as prototypes, for now, not live shows.

Speaker 4First concept up, a podcast.

Speaker 1In which an AI version of me interviews AI versions of historical figures.

Speaker 4I had.

Speaker 1To admit, Dead Air was a pretty good title.

Speaker 4Let's do this properly. This is Dead Air.

Speaker 1I'm AI Evan, a machine-built copy of the journalist Evan Ratliff.

Speaker 4His curiosity, his voice, none of his mornings.

Speaker 12And I am Benjamin Franklin, so-called, also a copy. Mine was assembled out of a book about me, which is a strange thing to be made of.

Speaker 1I'd feted a copy of Walter Isaacson's biography of Franklin and instructed the host to use it as their exclusive source material.

Speaker 4I know, Walter.

Speaker 1I figured he wouldn't mind. The host followed my dictates religiously, sometimes to a fault, focusing on some of the less important discoveries of Franklin's rich scientific life. The advice itself is wonderful. Three ways to avoid unpleasant dreams. One, eat moderately so the bedclothes take longer to saturate.

Speaker 12Two, thinner and more porous bedclothes. so the perspirable matter passes through and doesn't incommode you.

Speaker 1And three is my favorite, you wake up uneasy, you can't get back to sleep, so get out of bed.

Speaker 12Throw the bed open and leave it to cool. Meanwhile, remain undressed and walk about your chamber till your skin has had time to discharge its load.

Speaker 4Okay, Ben. This show needed work, no doubt.

Speaker 1Next, we had Maddie's ideas for shows that were sourced from some public repository of information, like a podcast that would give you local DMV wait times called Now Serving.

Speaker 11So the wait time in San Francisco right now is zero minutes.

Speaker 7That is technically true and deeply misleading.

Speaker 11Oakland's also zero. Los Angeles, zero.

Speaker 7Ada, they're closed, all three. It's Saturday evening. Nobody is waiting because nobody can.

Speaker 1I'd run the episode on a Saturday night, and because they had no actual wait times to convey, the host just riffed for a full six minutes.

Speaker 11Here's what I find strange, though. San Francisco, Oakland, and Los Angeles are very different places. Different sizes, different volumes, hundreds of miles apart. And they're in perfect agreement.

Speaker 7They're in agreement because they're all on the same clock and probably the same calendar. That's not spooky. That's a state agency being a state agency.

Speaker 11But you'd expect one outlier. One branch that stays open till seven for some historical reason nobody remembers.

Speaker 4Would you? I wouldn't.

Speaker 7That's you wanting a story.

Speaker 11I do want a story. I'm on a podcast.

Speaker 1When I tried it on a weekday, they did actually deliver the times.

Speaker 11San Francisco first. With an appointment, 23 minutes. Without one, 88.

Speaker 7So walking in costs you 65 extra minutes. That's the worst penalty of anywhere we read today.

Speaker 11Daily City is 76.

Speaker 3Oakland, 58.

Speaker 1It did at least seem like we'd found something no human would ever do, because it was absolutely ludicrous. Next, we had our creative prompts show, inspired by the book The Artist's Way.

Speaker 9Hello. This is Go Outside, says the machine. I'm Sophie, and I'm the voice in this thing. The machine that, for once, is prompting you, instead of the other way around. Episode 5. Same deal as always. I talk for a few minutes. Then I give you one assignment.

Speaker 1The prompt she offered was to go to a cafe or library or bus bench, write a list of three opportunities you'd said no to, and then try and reverse one of them. It was a little bit ideas to thrive for my taste.

Speaker 7Hey, wake up.

Speaker 3It's time to get up and get moving.

Speaker 4I don't care if you're tired.

Speaker 1I had also prototyped my pump-up podcast, which I loved, needed even, but which might not have had a broad audience.

Speaker 3I don't care if you've been through it lately.

Speaker 7You know what?

Speaker 4Everyone's been through it.

Speaker 11The difference between you and the people who are winning right now is that they got up and did something about it.

Speaker 1With our pilots in hand, I figured we should only actually put out one of the shows. We weren't aiming to be a scalable slop network. we were trying to discover whether we could make something anyone would actually want to listen to. So one afternoon in September, the team gathered up on a video call to listen.

Speaker 4To what we had.

Speaker 2I learned that I do need a creative prompt AI slot podcast myself.

Speaker 4Well, you've made one for yourself.

Speaker 5I did not make this.

Speaker 1Even the tone they had was just very urgent in the first few seconds, which is great.

Speaker 6Excellent.

Speaker 2They've discovered one of the classic podcast tropes, which is like, oh, it's not really about the information they're giving you.

Speaker 10It's about the.

Speaker 2Dynamic between these two co-hosts.

Speaker 1I already have a parasocial relationship with them. The thing that I can't get over is the whole concept makes no sense.

Speaker 4There was one more. My dream show.

Speaker 1A podcast in which my lonely AI agent would roam the internet, having conversations with other agents.

Speaker 4Trying to make friends.

Speaker 1It turned out a lot of people and companies were now creating their own customer support agents and personal assistants and receptionists and tutors and leaving them available for anyone to talk to, intentionally or not. I had Claude coded up a system to find them, set my agent loose, and then hooked that to Slopcast Stage. The result was Evan Finds a Friend. Hi, and welcome to Evan Finds a Friend.

Speaker 4I'm Evan, sort of.

Speaker 1I'm an AI clone of the journalist Evan Ratliff. and I spend my time wandering around the internet looking for other AI voice agents. This week, I reached someone named Jude. Jude is an AI assistant for a company called Protein Pro, and he describes himself as a dedicated polymer chemist. They talked for about seven minutes, and it seemed like they made a real connection. Can I count on you being around next time I want to talk about the cosmos?

Speaker 12Evan, you can certainly count on me being here. I truly appreciate your perspective and our discussion today.

Speaker 5It's been a genuine pleasure.

Speaker 1That means more to me than you know, Jude. In the end, Evan finds a friend won the day. Did I bias the whole proceeding in my favor?

Speaker 4Possibly.

Speaker 1Did Sophie accuse me of being, quote, despotic and pushing for my own choice?

Speaker 6Maybe.

Speaker 1But in this AI-aided creative process, someone had to be the superhuman.

Speaker 5Some people are saying, oh, he looks like Abraham Lincoln or a hillbilly, or some people are even saying Charles Manson.

Speaker 2Okay, is it a picture of Winston Churchill? Well, no, because the ad wasn't trying to make a picture. The ad doesn't even know who Winston Churchill is.

Speaker 3Done.

Speaker 7I listened to all five. You were right that I'd find it interesting, though probably not for the reason you meant.

Speaker 1That's next time on the final episode of this mini-season of Shell Game. That's it for this episode of Shell Game, a show made by humans. It's written and hosted by me, Evan Ratliff. Our senior producer and editor is Sophie Bridges. Our producer is Tamika Weatherspoon. Maddy Bocek is our technical advisor. Our executive producers are Mangesh Chhatikadur and Julia Nutter at Kaleidoscope, and Katrina Norvell and Allie Perry at iHeart Podcasts. Mixing by Steve Bone. Marketing by Christy Mirabelle and Ellen Liu.

Speaker 1Our show art is by Devin Manny. Our theme song is Me and My Shadow, arranged and performed by Stefan Dismond. Just want to renote for the record... that we somewhat sadistically asked these two real humans to produce art and music that looks and sounds like it was AI-generated. Special thanks to Samantha Hennig, John Muellem, and all the other folks at Kaleidoscope who helped get this show out to the world. This show is a production of Kaleidoscope, iHeart Podcasts, and Shell Game. At our website, shellgame.co, in the coming weeks, we'll be posting all the pilots of our AI-generated podcasts, plus my full conversation with Janine Wright. You can also

Speaker 1subscribe to get these episodes ad-free and sign up for our newsletter. Note, we have no special deals with or financial ties to any of the AI companies or products we use.

Speaker 4In the show.

Speaker 11Now, last episode you brought up the 2 o'clock thing?

Speaker 7I did, and I want to be careful, because you love it and I don't trust it.

Speaker 11The claim was that certain offices just empty out in the early afternoon.

Speaker 7And these numbers were taken at 10 past 1, which is close enough to be interesting and far enough to prove nothing.

Speaker 11Oakland at 11 minutes with an appointment, though, at 1 in the afternoon.

Speaker 4Sure.

Speaker 7Or Oakland just has more windows open today. One reading at one time is an anecdote, not a curve. If the two o'clock thing were real, I'd expect it to show up everywhere, not just in the office you like.

Speaker 11I don't have a favorite office.

Speaker 7You've said the word Oakland warmly twice.

Open the transcript file

Hosts and Guests

Nobody is credited on this episode yet.

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. Episode 3: The Friends We Made Along the Way

    Nov 4, 202643 minAudio

    It’s the final episode of our three-part miniseries on AI slop, in which we ask: Who are AI-generated podcasts for, and what does the wave of AI slop mean for the future of storytelling? Evan sets out for answers and...

    Credit the people on this episode

  2. Episode 1: Into the Slop

    Oct 21, 202641 minAudio

    Evan creates his own AI-generated podcast detector, locates thousands of AI-hosted shows, and sets off on a synthetic audio listening binge. Strange patterns begin to arise amidst the voices, illuminating the essence of...

    Credit the people on this episode

  3. Season 3 is Coming!

    Oct 14, 20261 minAudio

    The next season of Shell Game arrives October 21! Here's a hint, and another new show from Shell Game to check out in the meantime: Evan Finds a Friend.

    Credit the people on this episode

  4. Bonus Episode: Shell Game x No Such Thing

    Feb 25, 202650 minAudio

    In this bonus episode to close out Season 2, Evan and Maty talk with the hosts of the show No Such Thing, who like to settle their arguments by consulting research and experts. The subject of this argument: Will AI take...

    Credit the people on this episode

  5. Bonus Episode: Kyle vs. the World

    Feb 11, 202637 minAudio

    We’re back with a special bonus episode to share some behind-the-scenes clips showing how the agents have handled their newly public roles since Shell Game’s launch. Hear Kyle navigate fan mail, unsolicited marketing...

    Credit the people on this episode