The media and entertainment landscape continues to evolve at a remarkable pace, with new formats emerging and AI capabilities advancing in unexpected ways. In this episode of Denoised, hosts Joey Daoud and Addy Ghani explore three significant developments transforming content creation and technology: the rise of vertical soap operas as a booming industry, H&M's implementation of AI-generated models with the models' consent, and fascinating insights into how Anthropic's Claude AI actually processes information.

Vertical Soap Operas: A Multibillion-Dollar Industry Takes Hold

A new content format is rapidly gaining traction in the entertainment world: vertical soap operas. These short-form, vertically-oriented productions are designed specifically for mobile consumption and employ psychology similar to mobile gaming to hook viewers into paying for content.

Joey Daoud first encountered this phenomenon when speaking with camera professionals at industry events who mentioned working on these vertical productions. Unlike traditional productions that might film in a wider format and crop for vertical display, these shows are filmed with the camera physically turned sideways, creating content exclusively for vertical viewing.

The format originated in China, where it has become a substantial industry:

• The vertical series market in China exceeded $5 billion in 2023

• Projections suggest it could reach $13 billion by 2027

• Content is distributed through platforms similar to TikTok but with integrated shopping capabilities

Several companies have adapted this model for Western audiences, including Real Short, Drama Box, Short TV, Serial Plus, and Flex TV. The fundamental business approach involves:

• Creating a compelling 90-second hook that ends on a cliffhanger

• Making this initial clip freely available, often as social media advertisements

• Requiring payment to continue watching the story

• Implementing a digital currency or token system that obscures the actual cost

"Users typically purchase digital coins or passes to access additional episodes with the cost to watch an entire series ranging between $20 to $40," Daoud explained. "Many end up spending $10 to $20 weekly or up to $80 monthly."

The content typically leverages familiar soap opera tropes—billionaires, werewolves, secret identities, and dramatic relationships—designed to trigger emotional investment from viewers. This format bears similarities to Quibi, Jeffrey Katzenberg's short-form vertical video platform that launched in 2020 but quickly failed.

The hosts identified several key differences that may explain why these new platforms are succeeding where Quibi stumbled:

• These platforms embrace lower-cost, formulaic content rather than premium productions

• They target basic emotional triggers rather than attempting high-concept programming

• They implement a direct monetization model rather than subscription

• They don't invest heavily in proprietary technology for horizontal/vertical viewing

Real Short, founded in 2022, has already achieved significant success with over 30 million downloads and generating more than $10 million in monthly revenue. The production model allows for rapid content creation, with entire series potentially filmed in a week at budgets around $100,000.

This development represents a significant shift in content production economics, creating new opportunities for actors, directors, and crew during a period when traditional production has slowed.

H&M Embraces AI Models with Consent of Human Counterparts

Fashion retailer H&M is implementing a notable approach to AI-generated marketing assets by creating digital twins of human models—with their permission—to produce additional campaign materials without requiring the models to return for additional photo shoots.

This development represents a practical application of AI in marketing while raising important questions about the future of modeling and commercial photography.

Key aspects of H&M's approach include:

• Creating detailed digital twins of real models based on extensive photography sessions

• Capturing models from multiple angles and in various lighting conditions

• Documenting even small details like birthmarks and movement patterns

• Being transparent about the use of AI in their marketing

• Compensating models when their digital twins are used (though likely at lower rates than in-person shoots)

H&M is working with a technology partner called Uncut to implement this system, treating it as a learning process to gauge consumer reaction. The company has acknowledged that some consumers may not appreciate this approach but sees significant potential benefits in creating marketing assets more efficiently.

The hosts discussed the broader implications for the modeling industry:

"It feels like this is the end of the line for a lot of fashion models," Addy Ghani noted, suggesting that as these AI systems improve, the demand for real models may decrease substantially. This could particularly impact catalog modeling, where the focus is on efficiently showcasing large numbers of products rather than creating distinctive high-fashion imagery.

The conversation also touched on the potential for different brand strategies to emerge:

• Mass-market brands might embrace fully synthetic approaches

• Luxury brands might differentiate by emphasizing real human models

• Some brands might adopt hybrid approaches or emphasize in-person activations

Joey and Addy also discussed how this development connects to the broader ecosystem of digital humans, including fully synthetic influencers like Lil Miquela, who has maintained a significant social media presence for years despite being entirely computer-generated.

This trend raises questions about the future intersection of AI, digital rights, and content licensing. The hosts noted that blockchain and smart contract technology—aspects of the NFT ecosystem that received less attention during the speculative frenzy—might eventually play important roles in tracking the usage and rights of digital twins and AI-generated assets.

How Claude Actually "Thinks": Anthropic's Eye-Opening Research

The episode's final segment explored fascinating research from AI company Anthropic that provides unexpected insights into how large language models like Claude actually process information and generate responses.

Anthropic published a detailed analysis showing that Claude's reasoning process differs significantly from our common understanding of how these models work. Rather than simply predicting the next word in sequence, Claude employs sophisticated multi-step reasoning that sometimes works backward from a goal.

Three key examples highlighted this complexity:

1. Poetry Generation

When asked to create a rhyming second line for a poem beginning with "He saw a carrot and had to grab it," Claude didn't generate the response word-by-word from left to right. Instead:

• It first identified "rabbit" as a strong rhyming word for "grab it"

• It then worked backward to create a coherent sentence ending with "rabbit"

• The result was "His hunger was like a starving rabbit"

This process mirrors how human poets and songwriters often work, identifying key rhyming words first and then building lines around them.

2. Mathematical Calculation

When asked to perform the calculation 36 + 59:

• Claude first created a rough approximation (between 88 and 97)

• It then refined this to 95, recognizing that the final digit must be 5

• When asked to explain its process, Claude described a conventional step-by-step addition method (adding digits and carrying the one)—not the actual process it used

This suggests the model can provide explanations tailored to human expectations rather than describing its actual internal processes.

3. Multilingual Problem Solving

When asked for the opposite of "small" in English, French, and Chinese:

• Claude broke the problem into three distinct reasoning steps

• First, it had to understand what "small" means

• Next, it needed to recognize that the request was for an antonym

• Finally, it had to identify the appropriate opposite terms in each language

"It's computing this completely on the fly with different reasoning layers," Addy explained, noting that this complexity exists even in Claude 3.5 Haiku, which is an older and smaller model than current offerings.

The significance of these findings extends beyond technical curiosity. As Anthropic noted in their research, understanding how these models actually reason is crucial for AI alignment—ensuring these systems behave as intended and can be properly controlled. This becomes increasingly important as models grow from billions to trillions of parameters.

Conclusion: The Accelerating Evolution of Media and Technology

These three stories highlight the diversifying landscape of media production and consumption, as well as the increasingly sophisticated technical capabilities underlying AI systems. Vertical soap operas represent a significant shift in content economics and viewer engagement strategies. H&M's approach to AI models reflects changing production processes in visual media. And Anthropic's research reveals that our AI tools may be more advanced in their "thinking" than previously understood.

For professionals in the media and entertainment industry, staying informed about these developments is essential for navigating an increasingly complex landscape where traditional boundaries between human and machine creation continue to blur.