--:-- Steph Smith: It's kind of interesting that for most of history, I feel like humans have prayed to God, but now I see that content creators, we pray to the algorithm.
--:-- Steph Smith: Hey, everyone, welcome to the latest episode of the a16z podcast. This is your host, Steph Smith. Today's episode, I'm thrilled to bring on Karen Cheng. Karen has been a creator for a long time, but she specifically has a knack for integrating technology into her creations. You can only imagine with the rise of some of these AI tools like DALL-E, Midjourney, and Stable Diffusion, what she's been able to achieve. If you missed AI Summer and its impact on creators, you might have been sleeping under a rock, but you're not too late. It's truly just the beginning.
--:-- Steph Smith: In fact, I interviewed Karen back in September, and with every week, there was something that popped up where I thought, man, I wish I could have asked her about this. Yet still in this interview, we cover a lot of ground. We'll chat with Karen about how she originally found her niche, how she manages to grow her social following with a nod towards being additive instead of reductive. You'll also get to hear the behind the scenes of her many viral works, including a video of her becoming a lawnmower. Yes, truly a lawnmower. Her AI-generated Cosmo magazine cover, the first ever. Her DALL-E fashion show. Her transforming iconic art into 3D museums that people can explore. And much, much more.
--:-- Steph Smith: By the end of this episode, I think the audience will come away with some really key things, including an understanding of the new tools that are available that are literally at our fingertips, how AI can indeed enhance the creative process, and maybe even some of the second order effects of these innovations, like how creators are paid or how they might work with these AIs. I should note that since we're just at the beginning of this AI era, this is truly just the beginning of our coverage. So I hope you enjoyed this episode.
--:-- Steph Smith: The content here is for informational purposes only, should not be taken as legal, business, tax, or investment advice, or be used to evaluate any investment or security and is not directed at any investors or potential investors in any a16z fund. For more details, please see a16z.com slash disclosures.
--:-- Steph Smith: I am absolutely thrilled today to have Karen Cheng on the line. Karen is a digital creator. She has taken social by storm. Seriously, she's all over my feeds. I can almost guarantee you've seen one of her videos. She's done everything from high five Mark Zuckerberg in the metaverse to partnering with several A-list celebrities. She also was commissioned by Cosmo recently to create their first AI generated magazine cover. So super cool. We'll be talking about that. In particular, we'll be talking about her foray into AI, how she's using AI to be more creative, and how that relationship between humans and AI is evolving, how it enhances our creativity or how it doesn't. We'll see. But Karen, welcome to the show.
--:-- Karen Cheng: Thank you so much for having me. Thank you for that amazing intro.
--:-- Steph Smith: I'm talking to you, Karen. In 2022, you have millions of followers across various social media platforms. But I want to flashback to Karen from 2013. I'm going to preface this with, I was going down this YouTube rabbit hole. I was not researching for this episode. I happened to stumble upon a video with Paul Graham and Sam Altman from 2013 YC Office Hour. You know where this is going.
--:-- Karen Cheng: I can't believe you found that. That's kind of obscure.
--:-- Steph Smith: So I'm watching this video. Someone else familiar pops into it. Ryan Peterson from Flexport is in this video. And I'm like, OK, flashback. And then someone else is in this video and it's you. You're pitching this company, Give It 100, which is basically your startup at the time. The startup was structured around people sharing their journey in learning skills, sharing their progress every single day for 100 days. You also, I think around a year later, did a TED Talk where you used that framework of 100 days to go and learn dance. And you shared that. It was fascinating. People can look that up. We'll have it in the show notes.
--:-- Steph Smith: This is a broad question, but how do you go from that, 10 years ago, building this startup, to where you are today? Was there a point in time where you decided, "I want to start sharing this information online. I want to start being a creator"?
--:-- Karen Cheng: This was back in my failed startup days. I really badly wanted to get into YC, as everyone did. I managed to get on stage for office hours. Spoiler alert, did not get into YC. Was turned down, I think multiple times. Was interviewed on stage and answered their questions. That startup eventually failed like a year later. It was really tough on me at the time because up until then, I had good academic achievements in school. It was my first definitive failure. There's no one to blame. I had to come to terms with that.
--:-- Karen Cheng: Going through that startup, I learned so much about video that I then applied to doing my own stuff. I got this really interesting feedback when I was trying to get business for my startup. I was talking to someone, trying to get them to invest. They said, "We're not interested in this company." They could see it wasn't a scalable business. But they said, "We actually just wanted to hire you to make some videos." I said, "I don't want to do that." Years later, I realize that person was actually right. It took me trying every single career. I tried working for a huge company. I tried working at a startup. I tried being a founder. I tried running an agency. I've come to the realization that for me, being an influencer is the best thing, which is crazy because I always swore I would never be an influencer.
--:-- Steph Smith: I know exactly what you mean. I also create things on the side. For the longest time, people would associate me with certain things. They'd be like, "Steph, you're like the writing girl." I was like, "No, please. I'm not the writing girl." That is not how I picture myself. Ultimately, sometimes you just have to lean into what other people see in you.
--:-- Steph Smith: You started creating these videos. You've been doing it for years. I think you originally uploaded your first Musical.ly in 2017. Even in 2019, I think I saw you say in a video you had around 10,000 followers on Instagram. That might seem like a lot to some people, but compared to today where you're just smashing it, you have over a million followers. Everything you create almost goes viral. Was there a specific unlock where you found something that really worked or honed in on something that people really wanted?
--:-- Karen Cheng: Yes, absolutely there was. For a very long time, I struggled to grow my following. First, I resisted. I was like, "I don't even need to grow my following." I was just making viral one-off videos. Back in the days when you could make things go viral in a completely different way that doesn't work anymore. From 2012 up until around 2016, you could make videos go viral just by coming up with something clever that would be a good headline, cold emailing 300 reporters, getting them to write about it. It would almost guaranteed go viral. That does not work anymore because Facebook ate all the reporters' jobs. Now videos don't spread through reporters, they spread through algorithms. It is so competitive now, much more competitive than it was 10 years ago. It is much harder to make a video just go viral.
--:-- Karen Cheng: The way to do it now is you have to build a following. It's not so much about making everything go viral, but more about building an established follower base so your baseline of people who see your work just steadily gets higher. I really struggled to grow my following. I remember when I first tried growing my Instagram following, I was like, "Why is this so hard?" I would get things to go viral, but it was very hard to get any followers out of it. The unlock for me was actually sharing my behind the scenes. Before, I would just show these cool shots and be like, "Look at this cool shot," and then maybe I would describe how I did it in the caption. Crickets. If I showed the same shot, the exact same shot, but showed the behind the scenes with it, all of a sudden it was going viral. I've done that so many times now that I am confident that was the defining thing. There are lots of other things I learned along the way, but if there was one thing, it would be that.
--:-- Steph Smith: That's really interesting because a lot of people participate in these online worlds. A lot of people are creators. A lot of the stuff is very similar. Sometimes all it takes is one iteration, one adjustment in what you're doing. You're still sharing the same kinds of videos, but giving that behind the scenes look completely changes the game.
--:-- Steph Smith: I want to quickly get your take on something you mentioned, which is that social media is not the same. We use the same term from 10 years ago, but even if you think about "social" as a word, social media is not really social anymore. It's gone from the social graph where you followed your friends and family to an interest graph where you follow things the algorithm is serving you. Have there been any other changes you've noticed in terms of how social has changed that really matter to creators? Is there a way a lot of creators are thinking about this art incorrectly?
--:-- Karen Cheng: I think it's gotten a lot more competitive. People are a lot more aware of the downsides of social media now. We're no longer in the honeymoon phase. It's kind of interesting that for most of history, I feel like humans have prayed to God, but now I see that content creators, we pray to the algorithm. It's almost like a superstition. "What does the algorithm think? My video is not doing well because of the algorithm. How can I optimize my video for the algorithm?" We're almost no longer designing ourselves for human taste, but designing ourselves for the algorithm. I do think there are quite a few downsides to doing that, and there are good reasons to resist it as well.
--:-- Steph Smith: I think you're right about that. Increasingly, those algorithms are becoming black boxes where you think one variable is going to make your videos go viral, and then you play into that, but you don't really know. You're not there seeing the different levers that control the algorithm. It really can consume you.
--:-- Karen Cheng: The black box is actually what makes it feel like a cult or a religion. It's like a mystical force.
--:-- Steph Smith: Exactly. Even some people who work at these companies say they don't know exactly what plays into the algorithm or the output of it.
--:-- Steph Smith: One of the ways you've stood out is, at least recently, playing around with AI. I think you're one of the few creators I've seen that has really mastered the art of partnership with AI as it relates to social. There will be many more, of course. How did you start getting into that? What led you towards those kinds of creations?
--:-- Karen Cheng: I was really feeling quite stuck in my work. I felt like I was doing basically the same kind of video. Almost all 1 million of my followers are from the same kind of video.
--:-- Steph Smith: Can I guess which one it is?
--:-- Karen Cheng: Go for it.
--:-- Steph Smith: Is it the phone Matrix one?
--:-- Karen Cheng: Yeah, it's the low budget camera tricks. There was one video where I attached a phone to a ceiling fan and made it into a Matrix bullet time effect. I got 300,000 followers from that one video. Every time I got a surge of followers, it was the same thing. I would attach a phone to some household object, find a hack to do a camera trick. People were following me for that. I was just like, "I'm out of household objects. I'm out of ideas." I felt like I was running out of ideas. I had to find new tools, new toys to play with.
--:-- Karen Cheng: In the meantime, I was starting to see some of the really insane AI stuff coming out in white papers. It was all like "Two Minute Papers" research videos of researchers shooting whatever they had in their bedroom. They're so focused on the research. It's not their job to think about the storytelling or the cinematics. I was like, "Huh, that's interesting. I don't think anyone is actually doing that yet. So I could try doing that." I started experimenting with taking the research from white papers and making it a social media friendly video. I started doing that at the beginning of this year. It was so fun. I'm a very novelty seeking personality. My greatest motivation is, "Can I try to do something that hasn't quite been done before?" Probably someone has done it, but can I find my own original twist on it? The fact that no one was really using this technology in creative ways meant I could be the first to do something with it. That was very exciting. Trying to come up with the latest iPhone transition, all the ideas are taken. It is very hard to think of something clever or new with an iPhone camera trick or TikTok transition. With the AI stuff, it's a free for all right now.
--:-- Steph Smith: I know a lot of these tools are becoming more democratized and available. But there was a little bit of a technical gap for you to start playing with these AIs relative to the average creator. Is that right?
--:-- Karen Cheng: Yes, there is. I'm not that technical. I took a coding class once in college, but I prefer not to get into that myself. I work with really talented programmers who help me run it or teach me to set it up. That's why when DALL-E came out, I was like, "This requires no technical ability. If you can do a Google image search, you can use DALL-E." That was really interesting for me.
--:-- Steph Smith: I want to talk about those second and third order effects of having that fully democratized. You mentioned DALL-E. I've also seen you use a ton of other tools. I'm curious to hear from you, what are some of the other tools that others haven't heard of or that you're playing around with? Maybe DALL-E is the first AI tool you use, but I think you also use other ones to clean up a video or iterate on the creation from DALL-E. Can you share some of the tools you're playing around with?
--:-- Karen Cheng: Yes. With the image generation ones that are really popular right now, there's DALL-E, Midjourney, Stable Diffusion, Disco Diffusion. If you're watching the video version of this podcast, we'll have visuals.
--:-- Steph Smith: Yes, we'll bring all of them up.
--:-- Karen Cheng: Those are really popular. There are other ones that require more technical knowledge, but they enable techniques that are less mainstream, and therefore more unique if you can pull them off. One of them is DAIN. DAIN is really simple. All it does is apply artificial slow motion to your video. I thought it could be used in really interesting creative ways by shooting a custom stop motion video for the purpose of it being interpolated to create impossible movements.
--:-- Steph Smith: I think I saw one of yours. It was you as a lawnmower, lying on the ground and getting dragged across. Is that using DAIN?
--:-- Karen Cheng: Exactly. If you're just listening to this, imagine a stop motion of someone on the ground planking, slithering across the ground in stop motion. With DAIN, you can make it perfectly smooth. It looks like I'm being dragged on the ground by an invisible string. Actually, it was just a stop motion video that we used AI to smooth out. I did that with James Perlman. He helped me smooth everything out.
--:-- Karen Cheng: Other tools are NeRF. NeRF has gotten pretty popular. NeRF is a way to use any camera, like your phone, to scan a scene. All of a sudden, you have this beautiful 3D scan of it. What's interesting is it's different than photogrammetry because it can handle light. It's constructing a particle light field. When you look at things from different angles, the light changes. That's why it can handle mirrors, whereas traditional photogrammetry can't.
--:-- Karen Cheng: There are other tools too. I recently discovered EBSynth, which I learned is not an AI tool. I use that to figure out how to make DALL-E work for video. DALL-E is photo only, but using EBSynth, I was able to make DALL-E work for video.
--:-- Steph Smith: Very cool. We'll talk about this later, but you ended up doing an AI fashion show, right?
--:-- Karen Cheng: Yes, with the combination of many of these tools.
--:-- Steph Smith: I want to hear from you. A lot of these tools are quite nascent. We're going to see a lot more come to market. How do you see them all playing together? Let's use the text to image generators. You have DALL-E, Stable Diffusion, Midjourney. There will be many more. A lot of them are similar. If we extrapolate many years, how will they differentiate? How will you as a creator or consumers decide to use DALL-E or Stable Diffusion? Will they be different business models? How do you think that will evolve?
--:-- Karen Cheng: I can speak from the creator side, maybe more than the business side. In mass media, there's a misconception that AI is this all-powerful thing that's going to replace humans. A more interesting way to think of it is that they are tools. Each one is a specific tool for a specific purpose, just like a calculator or a compass. Where you get interesting things is where you combine the tools.
--:-- Karen Cheng: For example, you can generate an image in DALL-E, then use CapCut, which is an app on your phone that will turn any image into a 3D image. There's another problem that DALL-E and Midjourney have, which is that they don't do human faces very well. There will be artifacts in the eyes. They'll look droopy or messed up. Guess how you can fix them? You can run them through Facetune, the app that all influencers use. That'll fix it.
--:-- Steph Smith: Facetune can fix it to that degree? I've seen some of the outputs from DALL-E and they're really kind of...
--:-- Karen Cheng: It'll fix them to be flawless. It depends on the image, but I have seen it do it flawlessly. It gets harder and harder to reach perfection. You could be the AI researcher trying really hard to get that last one percent, or a creator could just run it through Facetune.
--:-- Steph Smith: I also wonder whether some of these tools become more specialized. These text to image generators are quite broad. They're being trained on all of the internet and used for use cases of all of the internet. Because there will be so many of them, I wonder if they become specialized. Maybe some become specialized for hyper-realistic human photography, and others become really specialized for generative illustrations. Do you think that might happen, or will they continue to stay broad?
--:-- Karen Cheng: Different companies will have different strengths, and they'll continue to differentiate by specializing in that. I do imagine some of these companies are going to buy other companies, and then they'll have more verticals, kind of like how Meta bought WhatsApp and Instagram. I see that happening with AI tools too.
--:-- Steph Smith: If we use Meta as an example, they tried to own messaging. You might see a similar phenomenon with AI tools.
--:-- Steph Smith: Let's jump into a really cool example of how you deployed AI, which is something I mentioned before. You got commissioned by Cosmo Magazine to create their first AI-generated magazine cover. We'll throw it up on the screen. It looks awesome. You've also done a behind-the-scenes explainer. For those who are not familiar with this project, can you share a little more about how it went down?
--:-- Karen Cheng: This was shortly after DALL-E was released. The people at OpenAI gave me a call. They said, "Hey, we have this really cool potential opportunity where Cosmo wants to make the first ever AI generated magazine cover. And we want you to generate it." I was like, "What? That's crazy." My first reaction was worry. I hoped I could do a good job because I can't emphasize enough how much the human is needed in this process. You can't have a fully AI generated magazine cover because you have to give a prompt. You could theoretically ask an AI to generate the prompt, but the human still needs to give input.
--:-- Karen Cheng: For our process, we went through so many different rounds of iterations. We eventually landed on a creative direction of a woman astronaut. I thought this was a really clever workaround by OpenAI to not show a human face. At that time, human faces were not allowed in DALL-E. We also didn't want to make a race statement. By making it a woman astronaut, it could be any ethnicity. They had that huge direction. They showed me a reference image generated in DALL-E. I was like, "Damn, this is so good. I don't know that I can even top this image." I spent hours trying to make a better image. I could not. I was like, "They hired me to do this. I have to do a better job than what they quickly threw together." I kept going for a few more hours. I kept troubleshooting. I wanted this woman to walk with swagger. I kept saying "walking confidently," "walking proudly," "walking energetically." None of those prompts worked. Eventually I was like, "walk with swagger." Boom. Then I got the hip thrust.
--:-- Karen Cheng: The other thing I was thinking of was an influencer tip for taking photos. The hack is to bring the camera to the ground and shoot from the ground up to make it look dramatic. I was like, "Let me try that." I typed "wide angle shot from below." Boom, it made it way more dramatic. The image I got had a messed up face. I did inpainting to fix the face. I did inpainting to extend the boots. I did quite a bit of fixing up of the original image. It was a diamond in the rough. If you look at my behind the scenes video, there was a significant amount of human work to make it happen.
--:-- Steph Smith: When you say inpainting, for people who aren't familiar, you're still using AI for that, right? You're selecting a portion of the outputted image and saying, "I like a lot of this, but I don't like this area. Can you in-paint? Can you replace that specific region?"
--:-- Karen Cheng: Inpainting is like magic Photoshop. If Photoshop worked magically, that's what it would be. Imagine there's a photo and you don't like what you're wearing or the background. You erase that, type in what you want, and boom, you get three options. It's incredible. That's what I also used partially for my fashion show.
--:-- Steph Smith: You talked about the fact that humans need to be involved and that it took several iterations. How many iterations are we talking in this route to the final image?
--:-- Karen Cheng: It depends a lot from artist to artist. Some people share their first prompt. That's interesting. Some people spend maybe five to ten prompts. For people trying to achieve a very specific result for a specific project, we'll be in there for hours running dozens or hundreds of prompts. In the early days of DALL-E, it was unlimited and free, so it was a little easier to do back then.
--:-- Steph Smith: I think they may have changed their business model or pricing recently. I saw an article about a woman creating a llama dunking a basketball. We'll pull that up. I think that one took 100 iterations and cost around $13. That gives you an idea. As you said, it depends.
--:-- Steph Smith: The more interesting question is how fungible was Karen in that process? I don't want to dismiss you.
--:-- Karen Cheng: How replaceable am I?
--:-- Steph Smith: If you gave Average Joe the same project and the same tool, do you think they would come up with anything comparable?
--:-- Karen Cheng: I think a lot of people would have been able to make a really good Cosmo cover. The people who would have done it best are people with a good artistic eye, the patience and motivation to keep going and keep refining the prompt, and who can describe what they see in their head. A lot of these people are not necessarily traditional artists. I think a lot of people could have made a really good result.
--:-- Karen Cheng: There's an example I have that we can show on screen. A while ago, I wanted to expand "The Girl with the Pearl Earring." You may have seen that you can expand paintings in DALL-E. You can take a famous painting and imagine what was all around it. You can take the Mona Lisa and imagine where she was. I made a video where I imagined the girl with the pearl earring in a library holding a book. I wanted her to be an educated woman.
--:-- Karen Cheng: And so I made that. And then I actually am working with a project with OpenAI right now. It's one that they commissioned. It's not out yet, but it will probably be out by the time this podcast releases. So if you go on my Instagram on Karen X Chen, you'll be able to go and see this filter. We made this Instagram filter where we expanded these famous paintings. And you can actually go inside through them and see these famous paintings.
--:-- Karen Cheng: I looked at my girl, the pearl earring, and I was like, this can be done better. So I hired August Camp, who is the person who taught me about this method. She has spent way more time in DALL-E than I have, she's so artistic and talented. I hired her to do this pearl earring, and when she showed me hers, I was just moved to tears, almost like jaw drop. I was like, that is what you made.
--:-- Karen Cheng: I felt like she was irreplaceable in this. I could have hired 10 different artists to do this pearl earring and gotten 10 very different results. What I like about this example is that everyone is starting with the same source image, so it almost establishes a little bit of a control for it. I think this is the perfect case study for: different humans get very different results with AI. Yeah, it's a tool, as you said.
--:-- Steph Smith: The reason I asked if you were fungible is because I just really wonder how this progresses. We know it's going to be somewhat of a democratizing force because now a bunch of people who couldn't paint or couldn't do Photoshop in certain ways can now do it and create all of this art. But I do wonder, does that make a lot of people really successful? Or does it still surface the very, very best to the top? Where there's a different filter, and now the filter is prompt engineering versus painting or drawing or singing. Do you think that's still the case where you're still going to see these outliers who are just so much better than the rest? Or do you think it'll be more of a level playing field?
--:-- Karen Cheng: Okay, so I think what this is going to do, what AI art is going to do, is it's going to significantly lower the barrier to entry to become an artist. To be an artist right now, you have to have a lot of time, a lot of training. Sometimes the monetary financial need means to be able to do that, or the willingness to be like a starving artist, or it's a hobby on the side. There's definitely a barrier. And now it's like everyone can do it.
--:-- Karen Cheng: I almost liken these image synthesizers like DALL-E or Midjourney to a Peter Pan. But instead of stealing from the rich and giving to the poor, it takes the artistic skill of artists and gives it to everyone. Like, here you go. So I do think that a lot more people are going to be empowered to be artistic or to be artists because they didn't necessarily have the patience to go learn oil painting. But they actually do have the talent to be able to describe what's in their imagination and continue to refine until they get the result they want.
--:-- Karen Cheng: There will still be the standouts and the extraordinary people because they're going to be the ones who are finding different or creative or innovative ways to do it. As much innovation as there is, there's always going to be the ability to combine things in new ways. So I definitely think that there's going to be a lot more artists, a lot of unhappy artists, a lot of happy artists, and also still standout artists.
--:-- Steph Smith: Yeah, it's almost just reshaping the skill ecosystem. Some people who have invested in skills for many decades, if they were just an A-plus oil painter, unfortunately for them, that skill is somewhat being replaced by this technology. It reminds me a little bit of Uber as an example. Some of those taxi drivers had invested in their taxi license for many, many years. There's kind of a parallel there. It doesn't mean it's bad or good, but it is just a fact that these AIs are, as you said, almost cherry picking skills and dropping them on the rest of the population. It's this great experiment to see what comes out of that.
--:-- Steph Smith: I also want to hear from you as you worked with these AIs. Were there any aha moments? Did you feel like it was truly a back and forth process, or did you feel like it was unidirectional? If it was unidirectional, you're basically giving the AI prompts, you're seeing what comes back and you're like, no, no, no, that's not what I want. You have a vision. But if it's more bi-directional, did you get a sense of like, you gave it a prompt, you got something back and you go, whoa, I didn't even think of that, but I like that. So was it more uni-directional or bi-directional?
--:-- Karen Cheng: It's 100% the second one. It does feel like a collaboration. You oftentimes get things you don't expect. And you're like, oh, I didn't think of that. Let me go down that rabbit hole now more.
--:-- Steph Smith: That reminds me of something I want to discuss with you, which we've actually talked about before in a call we had, which is the perception of AI. There's almost two framings. On one side, there's human versus AI, human versus machine, an age old trope. And then there's another, which is human plus AI, which is more optimistic. It's like, whoa, what can we do with this partnership? You've told me that you feel almost a responsibility to portray this technology, these tools in a certain way, maybe more of this optimistic versus pessimistic doomsday lens. Can you speak to why you feel that responsibility, or how you think AI should be portrayed, or how you want to portray it?
--:-- Karen Cheng: I had to unlearn a lot of my bad habits. I am a trained viral video creator, and in many ways, I'm trained very similar to the way a reporter would be, which means that I'm rewarded for making clickbait headlines. When all this AI stuff came out, my first instinct was like, oh, let me make a bunch of human versus AI. Let's have the AI do something and a human do something, and let's have judges critique and say which one's better. That was my first instinct.
--:-- Karen Cheng: And then I realized, well, what is that just going to do? It's just going to freak people out. I do think there are reasons to be worried about AI. I think there are very many legitimate ones, and it will negatively impact some people more than others. But AI is a technology. It can be used for good. It can be used for bad. Every gravitational force for the media is going to push towards the bad because that is what's going to get more clicks and engagements and views. What's not the natural gravitational force is to actually think of ways to use AI for good.
--:-- Karen Cheng: I was also in this turning point in my career where I just spent two years optimizing for growing followers, and I finally just sort of let go of that. I was like, what would it look like if I stopped trying to grow my followers? What would it look like if I actively stopped trying and just made what I wanted to make? Since I no longer care about growing my followers, which was very freeing for me, I actually don't have to do this clickbait thing. I actually don't have to try to get the most views by doing this controversial AI stuff. I actually can just make wholesome, positive, fine creative uses of it. Because if it is a technology that can be used for good or for bad, why not? It's more rewarding for me to make it for good, even if it's less viral.
--:-- Steph Smith: Have you found that things have been less viral when you go that more optimistic route?
--:-- Karen Cheng: I think on a one-off basis, each thing is less viral. There are many pieces where I was like, if I had written the headline in a more controversial way, it would have gone bigger. But I definitely think long-term, it's a better strategy for me to consistently put out positive stuff, so people associate that with the content I'm putting out.
--:-- Steph Smith: I was just curious. You mentioned news as an example. Pessimism sells, clickbait sells. It captures your attention. Fear is a very natural human emotion. But I do wonder if more optimistic creations endure. People follow you because they're like, Karen is this inspiring person. She's learned all these technologies. She's creative. So they view you as truly this creator that they more so want to be like, instead of pessimism which is sometimes quite surface level. It does capture your attention, but you're not inching to go back to hear from that creator. It's very transient.
--:-- Karen Cheng: Yes, and that's a great insight. You can make something go viral, but it's a one-off, and people aren't necessarily going to be inspired to continue to hear from you. Or you can continue to put out positive content, and each piece of positive content isn't going to get as many views, but you will kind of form a more long-term relationship in people's minds.
--:-- Karen Cheng: There was this really incredible quote that I saw a while ago, and it actually changed a lot of how I think about things as a content creator. I don't remember who said it, but the quote is, "Seek respect, not attention. It lasts longer." I was like, wow, because I was on this hamster wheel of chasing likes and views. As anyone who has made a viral video knows, it's over so quick. You can have a million views one day and then it completely dies down by the next. People get on this hamster wheel of addiction, of continuing to churn out content to try to get viral videos. I was on that hamster wheel for years, just being like, when does this end? Why am I doing this? It doesn't feel good. It's addicting.
--:-- Karen Cheng: Then I saw that quote and I was like, oh, if I actually just put more effort into making more memorable, good content, I don't have to put out content as often. I would much rather have the respect of a few people who I respect rather than the shallow attention of millions who forget about it by the next day.
--:-- Steph Smith: On that point, if we even remove the feel-good, "I feel respected" part of it, it's also good business. A lot of people view attention as the end goal. Like, I want a million followers. But ultimately, you need to convert that attention if you want this to be a long term game, whether it's to find clients or to sell a course or to start a rolling fund. That second part of conversion does not work if you're just doing these viral things.
--:-- Steph Smith: I operate more on Twitter. As an example, something I've been pretty outspoken about is these tweet threads which are just so overdone and so templated. They work, but they're just really, really surface level. Ultimately, what I find is that I see a lot of creators feed into that. They're like, okay, I'm going to go all in on this thing that I don't believe in, I don't feel good about, but it's going to bring me hundreds of thousands of followers. But then coming back to the point, try converting those followers into something that generates you money or that is long lasting. It's very hard, because those followers don't actually love your content. They don't buy into you. They just happen to be like, oh, okay, another thread. It's almost like a habit for them to engage with that kind of work.
--:-- Karen Cheng: I'm so glad you said that because I feel the same way. Ultimately, this is my job. The millions of followers are useless to me if I can't make a penny off of it. I'm not going to put in all this effort. It's my job, right? I need to make money out of it. I remember that I used to feel really bitter when I would see these TikTok videos blow up and get millions and millions of views, and they would be the most inane things.
--:-- Karen Cheng: For example, a TikTok video that can go really viral is a video where someone's like, "Try to get the number of hearts and the number of comments to be the same." It gets tens of millions of views because it's a very smart engagement bait. I'll be mad at that. There's other examples that are less clever. There's just straight up garbage on TikTok that is getting millions and millions of views. That is so demotivating for creators, VFX creators, artists who are putting hundreds of hours into their craft to make something look really beautiful, and then it completely flops and doesn't get very many views.
--:-- Karen Cheng: I used to be so bitter about that. Then someone told me, and I learned over time that this was actually very much true, the people who are putting out low effort content all the time, getting millions of views, they're not the ones getting hired by major respected brands. They're not actually getting great opportunities off of their follower count. Whereas I know people who make incredible content, they only have like 20,000 followers, and they're making bank.
--:-- Steph Smith: Exactly. I just have to say, I resonate with what you said so much there as a creator myself. Again, I operate more with Twitter, but same thing. I'll see these threads and I'm like, oh my God, what is going on? Why are people liking and retweeting this? And why is this one account blown up? It's really low quality compared to some other accounts that I follow where I'm like, this is gold. Why is no one paying attention? But I think it's not productive to focus on that because ultimately the people who are putting the best stuff out there are the ones who, if we take the whole value chain and consider that it's a business, those are the successful businesses.
--:-- Karen Cheng: People put so much value on the number of likes and the number of views because it is the only number we can really see and judge ourselves by. We've been trained since kindergarten to put value in the number judgment. You get like a 97 on a test, wow, that's so good. You get like a 60, that's a D, that's bad. We are so trained to take that feedback of numbers as the truth. Then social media comes along and you get lots of views versus a little. It's so easy to say, well, a video was great because it got tons of views, and it's not because it got very little views. It took me years to learn that it is such a bad metric, but it's the only objective one we have.
--:-- Steph Smith: I know. But it's hard sometimes, as we talked about, to check out of that ecosystem to be like, oh, I'm actually going to focus on the things that I want to put out there, the things that I'm proud of, instead of feeding into what some people might argue is an objective metric of like, this is what people want. You have to go against so many natural human instincts.
--:-- Steph Smith: I want to talk about another topic which I've seen everywhere. It's coming up in basically every conversation I have around creators and AI, and it's the idea of IP. We talked about being hired for things, but also owning your creations is something that's mattered a lot for not just independent creators, but if you think about all the way up, films, entertainment, what matters is who owns the IP. AI introduces a very, very interesting filter on this.
--:-- Steph Smith: Honestly, I don't know if anyone has the answers, but what I want to walk through is where is the value along the AI creation process? There's the information that the AI is trained on, which you could argue the output wouldn't exist without that. So maybe that's where the IP is. There's also the argument that the creators of the AI tools, OpenAI, Midjourney, et cetera, well, the outputs also wouldn't exist if the AI tool didn't exist. And then of course there's the prompt, which you could also argue if someone doesn't enter a very, very specific prompt, the output also does not exist. I've even seen a company yesterday that was selling prompts because it is valuable to have the right prompts.
--:-- Karen Cheng: I've bought a prompt before.
--:-- Steph Smith: There you go. So there's probably going to be a whole economy around this. I want to hear from you. Obviously, there's maybe no right or wrong answer, but how do you think about that? Where is the IP and how on earth will we think about that as creators moving forward?
--:-- Karen Cheng: Yeah, I don't know. I've thought about this, but I don't have the answers. I'm not a lawyer. I know that lawyers don't even have the answers. Congress doesn't have an answer to this yet. It's all so new.
--:-- Steph Smith: Well, let me rephrase the question to something that I think you can comment on, which is how creators get paid, how creators charge. Something that we talked about before was that for your projects, whether it was with Cosmo, or we were actually talking about a project, you said that you felt like it was really important that those projects were paid for. That might sound obvious, but as we talked about before, AI in certain cases might not cost that much. It might cost like $2 to run the prompts you're looking for. But some of those projects historically, like I don't know how much Cosmo would typically pay for a magazine cover, but it's not $2. So how do you think about the way that you charge as a creator in this new ecosystem?
--:-- Karen Cheng: Yeah, when I work with clients, I'll literally say to them, "Hey, even if you don't hire me," and sometimes it doesn't work out, so I'm like, "Hey, you know, my schedule doesn't work out for this project. But whoever you hire, please pay them. Please allocate this budget." Because we're at a moment in history that's very vulnerable where certain standards can be established. I would like to see the standard established that if you hire a human to use an AI as a tool, then you pay the human.
--:-- Steph Smith: Do you think that holds up, though? I ask that because I think, like many industries, the very best of the best, if there really is a 1% or a 10x prompt engineer, they will get paid very well. We've seen that happen in many fields. But then you also see when things get democratized, it's very hard for those price economics to not change. You imagine, for example, that at the beginning, maybe certain creators are holding their ground. They're saying, we must be paid X. But then you're going to have insert creator here that is like, actually, I'll take half of that. Or maybe I'll take a hundredth of that price.
--:-- Karen Cheng: I think that's a really good point. I think there is incredible downward pressure. So what's probably going to end up happening, unfortunately, is that there will be the more well-known people who can still charge higher rates. But then for the vast majority of people, there's such insane downwards pressure on it. I do worry about that for creators. I don't really know how that is all going to play out.
--:-- Steph Smith: Yeah, it's really hard to predict. I wonder how some of these fields change. I want to hear from you because you've started to work with some of these companies. At this point in 2022, is AI, or at least the tools that people are using, are they seen as more gimmicks? Or do you see companies really investing in this and saying, oh, actually, we are going to replace our photographer with AI? Or to use Cosmo as an example, was working with you just something where they could put their flag in the ground and say, hey, we're innovative, we're working with AI? Or do you see them actually going and doing every fifth cover with AI? How are companies thinking about this?
--:-- Karen Cheng: In 2022, it is most certainly a gimmick and a headline grabber. It is like, let's be innovative. Let's be first. Let's be on top of this. I would imagine in 2023, the answer to that will probably be different. People will actually start using it as their workflow because it makes sense and not as a headline grabber.
--:-- Karen Cheng: Really interesting, I saw that there was a reporter who used a Midjourney art for his article, I think about Alex Jones. And then he went viral in a bad way. He got majorly attacked on Twitter for using Midjourney art for a magazine article. He was attacked because he didn't hire a human artist instead. I think it was The Atlantic. Correct me if I'm wrong. But I think what people didn't have context was that the department he was in, the image that he would have used instead would have been like a Getty images or some stock photo. It wouldn't have been like an official commission. So that was what he was replacing. But he still got attacked for it. I think the first few, the arrows attack the leaders. The first few people who are going to start using AI are going to get backlash, but it's going to get less and less and less until it's accepted.
--:-- Karen Cheng: Honestly, discussing some of these downsides is very uncomfortable for me because I don't know what's going to happen. I do think that it could be bad for a lot of people. I personally do try to stay optimistic and try to see the optimistic side. But there's no denying that it is going to be bad for a lot of people and it is going to be good for a lot of people. It is a game changing technology.
--:-- Steph Smith: I've heard some people say that it is like the technology of our generation in terms of how it shapes things. I also heard Sam Altman say something which I thought was an interesting frame. A lot of people will talk about, and we even did this at the beginning of our call, "When is this going to change things fundamentally?" He didn't give a definitive timeline because I think no one can predict that. Probably smart. I shouldn't have given any numbers either.
--:-- Karen Cheng: No, no, no. But we were just having fun.
--:-- Steph Smith: But I think his framing was really interesting. He believes that it's the game-changing innovation. If you look at humans and the spectrum of time that we've been on this earth, if we're talking five years, or we're talking 50 years, that is still just a drop in the hat. It's such a short period of time where it's almost inevitable. I think many people in this space will say that AI within 50 years is just going to really, really fundamentally shape or change our world. Coming back to his point, it was like, if it's in five months, five years, 50 years, in the span of human civilization, it's nothing. It doesn't seem like nothing within our lifetimes.
--:-- Steph Smith: Coming back to the idea of it helping people, it hurting other people, how do you think about what creators should be focused on? Should they be focused on learning prompt engineering? Should they be focused on a field that they think won't be touched by AI, if that exists? What are you doing as a creator to say, you know what, I don't want to be left behind, so I'm going to be learning X, Y, and Z?
--:-- Karen Cheng: Well, I don't think the answer is focus on prompt engineering, because I wouldn't be surprised if in a very short amount of time, AI gets extremely good at prompt engineering. Those people become less relevant. In fact, I was actually just talking to someone who was using GPT-3 to write prompts to run through Stable Diffusion. So I would say that the best thing to do is not to learn a specific skill, because that's going to change probably by the time this podcast comes out.
--:-- Karen Cheng: I think the best thing is to adopt a mindset that you have to always be learning and to accept that the model that humans had for much of humanity, which is that you could kind of choose a career and then have that career for life, that's gone. We are at the last dying, grasping breaths of those days. The sooner you can accept and embrace the mindset that the world is always changing, the skills you have to learn always need to change, the better off you'll be. I don't know if we've lived in a time that's been so exponential where these technologies are changing, but humans have adapted. So we'll see how we adapt.
--:-- Steph Smith: Some interesting examples. You pointed me towards an article from Nathan Baschez where he talked about even simple things like the introduction of Unsplash completely changed how companies stand out with their design because all of a sudden stock photography, hyper-realistic pictures, became widely available. Then you saw what I think people call Memphis design come up, and you saw all these very similar looking illustrations. I do wonder if there's some version of that, where we have access to unlimited 2D art, so maybe 3D art is important just as an example. But then of course AI will eat that. I think it's interesting to kind of think through how humans might evolve. Or do you think that this stuff is just moving so quickly?
--:-- Karen Cheng: I think humans are remarkably adaptable. So we will adapt. I do think that already the world is moving much faster than human evolution can keep up with. For example, we weren't really evolved to sit in office chairs all day. We weren't evolved to hear the comments of thousands of people on social media. We were evolved for like 30-person tribes. You see all of these mental health and physical health problems that we have. A lot of those are because human evolution hasn't caught up to how much culture has changed.
--:-- Karen Cheng: I say two things. One, humans are remarkably adaptable. But two, there is a limit to how much humans can actually evolve from a Charles Darwinian evolutionary perspective to keep up with some of this stuff. So we are entering a completely unknown time. I'm trying to stay positive and try to highlight as much of the positivity as possible. But yeah, there is a lot of uncertainty.
--:-- Steph Smith: It is such a sea change. To your point, even as two people who believe fundamentally that technology is good, or at least I don't want to put words in your mouth, that's something I believe. It is sometimes hard to just fathom what the hell we do and who is going to be disrupted. It's impossible to say that there aren't negative externalities.
--:-- Karen Cheng: Yeah, I would say I think technology is neither good nor bad. I like to view it as neutral. It's like the neutral canvas. I find that to be the most empowering because if it's neutral, then we have the ability to influence it.
--:-- Steph Smith: I like that. I've heard other people use it's like a mirror. It mirrors back to humans, ultimately, certain things that maybe are below the surface sometimes.
--:-- Karen Cheng: Yeah, that's good. I like that.
--:-- Steph Smith: Taking a little bit more of an optimistic lens, I've seen AI used in a ton of different ways. I saw something recently which was an AI that helps write Excel formulas, which I thought was quirky and neat. I also saw one that takes legal jargon and simplifies it for the end user. Curious to know from you, do you have a wishlist? Are there things that AI has not touched yet where you're like, man, I would really love this?
--:-- Steph Smith: While you think about it, I'll just quickly share one. Super simple, but we're doing these podcast recordings with video. Every so often, I would just love to re-say something. Things like Descript allow you to re-say something with audio very easily. But I want that with video because if we're airing the video versions of these, I want to just be able to remove my filler words and fix up my performance. But obviously it'll look ridiculous if the video portion isn't equivalent. So that's my little wishlist for anyone at OpenAI or otherwise to do. I know they're probably coming out with video at some point, but to sync the two.
--:-- Karen Cheng: What I have wanted for AI is video tools for creators that are AI powered. That's why I've been doing so much hacking, trying to get these image tools to work for video because it's not quite there yet. I keep on trying to make it work. Once it does, though, I don't even know what's going to happen. So I'm almost like, be careful what you wish for.
--:-- Karen Cheng: Yeah, I mean, it's coming, right? It's for sure coming. It's hard for me to say a wish list because I'm just trying to keep up with what's already coming out.
--:-- Steph Smith: Oh, totally. Something we talked about before we jumped on, just a neat little idea to get the wheels spinning for any listeners. We were talking about audio AI. Have you actually seen any AI ASMR? I wonder.
--:-- Karen Cheng: No, I've never heard of that.
--:-- Steph Smith: No, I haven't either. But I was just thinking, what would be interesting with ASMR is if they could connect to your senses and measure your response to it. And then in real time, adjust the ASMR, update it to your sensory response.
--:-- Karen Cheng: Oh, I like that. Startup idea. People are already playing around with audio AI, so I haven't seen the ASMR stuff, but I feel like that's coming soon. I've heard other people talk about this in the realm of video as an example. Let's say you're in a movie, and this is a little further away, but based on your prompts, it goes another direction. So it's almost like personalized content for the consumer based on their needs or prompts. That's interesting.
--:-- Steph Smith: I have an answer for your question. Here's my wish list for AI. My wishlist is actually for all humans to take the ethics of it very seriously. What I mean by that is, my wishlist is that everyone working on AI be held to a standard to use it for positive forces rather than negative. And that we somehow develop an insane amount of cultural pressure so that anyone not using it for positive sources is penalized in a way that severely disincentivizes it.
--:-- Karen Cheng: How do you think about what is good and bad in that context?
--:-- Steph Smith: It's very hard to be the arbiter of that. But for example, deception is bad. So when you're altering things, you need to disclose what you alter and how.
--:-- Karen Cheng: Do you think we will need to label things as AI generated or human generated? Because you brought up deception, but people do deepfakes of people without their consent. They're not quite there yet, but they're very close. Have you seen the Tom Cruise ones? They are fully the Tom Cruise deepfakes. I can't tell. I think most people can't tell.
--:-- Steph Smith: So we're already there. It's not even a future. We're already there on edge cases, and we'll pretty soon be there for mass cases.
--:-- Karen Cheng: I think that will be developed. I think it will be necessary, and it will happen. I think it will be very similar to nutrition facts. There will need to be some system that shows this video was produced in this way.
--:-- Steph Smith: I've heard people talk about this, and I might be wrong because I was just listening to another podcast, but it likely would have to be at the hardware level, run by Apple or Google. If it's not at the device level, it's really easy to hack or abuse. It would have to be a case where you could not remove that metadata. It would have to be at the hardware level. I think it will be first adopted in really high stakes scenarios where the veracity of something is extremely important. For example, in political messaging, maybe they'll have a very specific kind of camera or hardware that cannot be altered. For certain types of really important messaging where it's critical that it not be faked. Whereas someone's social media influencer video, it doesn't matter if something was adjusted or tweaked.
--:-- Karen Cheng: So there are standards. That's part of the regulation that in certain fields, you must use certain hardware, or at least a suite of hardware that plugs into these. Regulation will catch up. But I actually think it will be somewhat self-regulated. Some people will have the need to show that what they do is true and not altered. Companies will pop up and say, this is the technology that does that. People will buy that technology and use it of their own free will, not because of regulatory requirements, because it'll take a while for regulation to catch up.
--:-- Steph Smith: But I do wonder, the good actors will use the technology that has that functionality, but bad actors likely won't, unless there's regulation. That's why I say my wishlist is for the culture to change. Humans are a self-enforcing society. We are very naturally tribal, and we behave well largely because that's what's expected of us from other humans. It's not like someone's watching us at every single moment, but we still want to behave well. A lot of that is the cues from culture. I would love to see a culture developed where, as part of being a good human, we use technology responsibly and ethically. I don't know how that's going to happen. That's why it's my wish list.
--:-- Karen Cheng: I love going down these rabbit holes because it gets the wheels spinning. There's so much to figure out. When you're dealing with an exponential technology, policy, regulation, IP, a lot of this stuff takes time. I wonder whether we create regulation in 2023, and by 2025 it's already completely different.
--:-- Steph Smith: I actually think culture can change a lot faster than regulation. Much, much faster. I'll give you an example. The fastest I've ever seen culture change was the handshake during COVID. How many thousands of years has the handshake been the golden standard? And then all of a sudden, socially unacceptable. So many things became socially unacceptable almost overnight because of COVID. I thought, wow, human culture can change like that. And now that the pandemic is better, we're back to doing it again. So I do think we're going to need a lot of cultural forces to sprint ahead of regulation, because regulation will take time.
--:-- Karen Cheng: Well, I have no idea how this will progress. It's been fascinating to hear how you think it might progress as someone who has played around with these tools. We'll be sharing a lot of your projects as we talk about them throughout this episode. I want to end with something really fun. You've worked with tons of creators. Creators that people know, creators that people don't know. Do you have a wishlist, or is there one creator you would just love to partner with? I love throwing this stuff out into the ether because you never know who might make this happen.
--:-- Steph Smith: It's interesting. In the past I would have had a dream client or a dream something, but I've learned over time that it's not about which celebrity or brand you can get. For me, it's actually about doing my best ideas, my best work. The way I do things is I try to matchmake my ideas to a client who will accept them unaltered, or with the least alteration possible. So nothing actually comes to mind, because it's mostly about making the ideas that I want to make.
--:-- Karen Cheng: I love that. For me, when I was younger, it used to be, I want to work with these people or I want these companies to hire me. I think you're right. There is an evolution, like we talked about before, where you're almost transcending attention and you've gotten to a place where your focus is, how do I make something really great?
--:-- Steph Smith: Yes, it's almost like Maslow's hierarchy of needs. I'm actually focused on doing the ideas that I want to do. Ten years ago, I would have been like, I want to dance in a Coca-Cola commercial. I'm good. I don't need to do that now. Even two years ago, I was like, I really want to make a commercial for the Apple iPhone. And now I'm like, I don't actually need to do that anymore. So I don't have one because it's actually about the idea. It sounds so cheesy, but it actually is about enjoying making the idea.
--:-- Karen Cheng: Well, that's a wonderful place to end things off. Karen, thank you so much for talking with us today. This was so fun to think through the optimistic, the pessimistic, the dystopian, utopian futures we might live in. But at the end of the day, I think one of the biggest takeaways for me is that AI is a tool, and that tool is being democratized. That in itself is exciting for many people to be able to participate in that future.
--:-- Steph Smith: Yeah, thank you so much. Thank you for having me. Thank you for these great questions.
--:-- Karen Cheng: Thanks for listening to the a16z podcast. If you like this episode, don't forget to subscribe, leave a review, or tell a friend. We also recently launched on YouTube at youtube.com slash a16z underscore video, where you'll find exclusive video content. We'll see you next time. Thank you.
