By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.
Summaries, chapters, transcripts, and FAQs on this page are made by uListen, not YouTube.
CHAPTERS
- 0:00 – 2:49
Episode preview and Karen Cheng's background
Steph Smith introduces Karen Cheng, a creator who integrates technology into her work, and previews the topics: AI tools, viral projects, and second-order effects on creators. She then introduces Karen's viral work, including the Cosmo cover and her partnerships.
- •Karen is known for blending technology into her creations
- •Projects previewed include the lawnmower video, Cosmo's first AI-generated cover, and a DALL-E fashion show
- •Episode covers new tools, AI enhancing creativity, and effects on how creators are paid
- 2:49 – 7:01
From failed startup to full-time creator
Steph recalls a 2013 YC Office Hours video where Karen pitched her Give It 100 startup. Karen explains how the startup's failure and trying many careers led her to accept being an influencer.
- •Karen pitched Give It 100 at YC Office Hours and was not accepted
- •The startup failed about a year later and was her first definitive failure
- •She learned video skills and received early feedback that people wanted to hire her for videos
- •After trying big company, startup, founder and agency roles, she concluded being an influencer suits her best
- 7:01 – 9:04
The behind-the-scenes unlock and how virality changed
Karen describes how going viral worked in the reporter era versus the algorithm era, and why building a following became necessary. Her key unlock for growth was sharing behind-the-scenes footage.
- •Viral videos once spread through emailed reporters; now they spread through algorithms
- •Building a follower base raises your baseline audience
- •Showing the same shot with behind-the-scenes footage made it go viral
- •Steph notes a small adjustment can change the game
- 9:04 – 11:34
Praying to the algorithm: social media's changing landscape
Steph and Karen discuss how social moved from a social graph to an interest graph. Karen says creators now 'pray to the algorithm', a black box that feels like a cult or religion.
- •Social has shifted from friends and family to an interest graph
- •Social media is more competitive and past its honeymoon phase
- •Creators design for the algorithm instead of human taste
- •The black-box nature makes it feel like a mystical force
- 11:34 – 15:18
Running out of household objects: Karen's pivot to AI
Karen explains she felt stuck making low-budget camera tricks and began turning AI research papers into social-friendly videos. She found the novelty exciting, and notes that tools like DALL-E need no technical skill.
- •Most of her followers came from low-budget camera-trick videos like the phone-on-a-ceiling-fan Matrix effect
- •She ran out of ideas and sought new tools
- •Researchers weren't focused on storytelling, so she adapted white-paper research for social
- •She is novelty-seeking and enjoyed being first
- •She works with programmers; DALL-E requires no technical ability
- 15:18 – 18:28
The AI toolbox: DALL-E, DAIN, NeRF and EBSynth
Karen walks through the tools she uses, from image generators to DAIN's artificial slow motion for the lawnmower video, NeRF scene scanning, and EBSynth for getting DALL-E to work on video.
- •Image generators include DALL-E, Midjourney, Stable Diffusion and Disco Diffusion
- •DAIN smooths stop motion to create impossible movements like the lawnmower video
- •NeRF scans a scene with a phone camera and handles light, unlike photogrammetry
- •EBSynth, not an AI tool, helped her make DALL-E work for video
- 18:28 – 22:05
Combining tools and how AI products may differentiate
Karen says AI is best seen as specific tools that can be combined, such as using Facetune to fix faces from image generators. Steph and Karen speculate about specialization and consolidation among AI companies.
- •Karen sees AI as specific tools for specific purposes, not an all-powerful replacement
- •DALL-E images can go through CapCut for 3D and Facetune to fix faces
- •Steph wonders if tools will specialize, such as hyperrealistic photography versus illustration
- •Karen expects different strengths and acquisitions, like Meta buying WhatsApp and Instagram
- 22:05 – 27:18
Making Cosmo's first AI-generated magazine cover
Karen recounts how OpenAI asked her to generate Cosmo's first AI cover. She describes the woman astronaut direction, prompting tricks like 'walk with swagger' and a low wide-angle shot, and the inpainting needed to fix the result.
- •OpenAI called her about the opportunity and she worried about doing a good job
- •The woman astronaut concept avoided human faces and race statements
- •Prompts like 'walk with swagger' and 'wide angle shot from below' improved the image
- •Inpainting fixed the face and extended the boots, showing significant human work
- •Iterations vary from a few prompts to hundreds
- 27:18 – 34:54
Are artists fungible? Lowering the barrier to art
Steph asks how replaceable Karen was in the Cosmo process. Using the Girl with a Pearl Earring example, Karen argues different humans get very different results, that AI lowers the barrier to becoming an artist, and that standout artists will remain.
- •Good results depend on artistic eye, patience, and ability to describe what you imagine
- •August Kamp's Pearl Earring expansion moved Karen, showing artists aren't interchangeable
- •Karen likens image synthesizers to Peter Pan giving artistic skill to everyone
- •Steph compares skill displacement to taxi licenses in the Uber era
- •Karen says working with AI feels like a collaboration
- 34:54 – 41:03
Optimism versus pessimism: seeking respect, not attention
Karen explains why she avoids human-versus-AI clickbait and chooses wholesome, positive uses of AI even if each piece is less viral. She cites the quote 'Seek respect, not attention. It lasts longer.'
- •Steph contrasts 'human versus AI' with 'human plus AI' framings
- •Karen says her clickbait training pushed her toward human-versus-AI content, which she rejected
- •Letting go of follower growth freed her to make positive work
- •Positive content builds a long-term relationship with audiences
- •The quote about seeking respect changed how she thinks about content
- 41:03 – 46:26
Turning attention into business and the trouble with view counts
Steph and Karen discuss why attention must convert into business, and why viral low-effort content often doesn't lead to opportunities. Karen explains why likes and views are a flawed metric.
- •Followers are useless if they can't be monetized
- •Low-effort viral creators often aren't hired by major brands
- •Creators with 20,000 followers can earn well
- •Numbers feel objective because we were trained on grades
- •Focusing on what you're proud of goes against natural instincts
- 46:26 – 51:09
IP, ownership and downward pricing pressure for creators
Steph lays out where IP might sit in AI creation: training data, tool makers, or prompts. Karen has no answer on IP but wants humans paid for using AI as a tool, while acknowledging heavy downward price pressure.
- •IP could lie with training data, the AI tool builders, or the prompt
- •Karen says lawyers and Congress lack answers yet
- •She wants a standard that humans hired to use AI get paid
- •Steph notes democratization tends to push prices down
- •Karen expects well-known people to charge more while most face insane downward pressure
- 51:09 – 1:02:03
Is generative AI a gimmick? Adapting to change
Karen says that in 2022 companies use AI as a headline-grabbing gimmick but expects workflow adoption in 2023, noting backlash against early adopters. They discuss timelines, why to avoid betting on prompt engineering, and human adaptability.
- •Karen calls AI in 2022 a gimmick and headline grabber
- •A reporter was attacked for using Midjourney art instead of stock photos
- •Steph shares Sam Altman's framing that 5 or 50 years is short in human history
- •Karen advises an always-learning mindset rather than a specific skill like prompt engineering
- •Humans are adaptable but culture changes faster than evolution can keep up
- 1:02:03 – 1:12:10
Wish lists, AI ethics, labeling and cultural change
After discussing wish-list ideas like video re-saying and AI ASMR, Karen says her real wish is for everyone to take AI ethics seriously. They discuss labeling AI content like nutrition facts, hardware-level provenance, and culture changing faster than regulation, then Karen explains why she has no dream collaborator.
- •Karen sees technology as a neutral canvas we can influence
- •Her wish is cultural pressure to use AI for positive purposes
- •Deception is bad, so alterations should be disclosed, like nutrition labels
- •Provenance would likely need to be at the hardware level, starting in high-stakes cases
- •Culture can change faster than regulation, as the COVID handshake showed
- •Karen prefers making her best ideas over chasing celebrity partners
