Scott Belsky: Content Creators, Creativity, and Marketing in the AI Landscape (Encore)
31:41 Share

Bye.

Hey everyone, and welcome to Generative Now. I am Michael Mignano, and I am a partner at Lightspeed. This week, we're revisiting one of my favorite conversations with Scott Belsky. Scott is an author, founder, and all-around visionary at the intersection of creativity and technology. He was previously the co-founder of Behance, and he was previously the chief strategy officer at Adobe, but he recently announced that he has become a partner at A24, where

♪♪

Good to see you, Scott. Good to see you, Mike. Thanks for doing this. Yeah, thanks for having me. Thanks to everyone at Robinhood. So six months ago, we did a talk very similar to this one about the intersection of AI and creativity. And in AI, six months feels like an eternity. A lot has happened since then.

One of the main things that we talked about then was sort of the impact on creators and what impact AI is already having on creators. If you look back over the past, let's just say 2023, what impact has AI had on creativity? What are you seeing in the market right now? Yeah, well, it does feel like every day you need 30 minutes just to catch up on the research from the previous day that came out.

And in the space of creativity, I think one of the reasons why AI is so relevant is because in most areas of business and life, hallucination is a bug. Whereas in creativity, hallucination is a feature. And when creatives go in and prompt something, they actually want things that they didn't expect.

as a benefit of the product. And so you've seen creators kind of come in and start playing. I always feel like novelty precedes utility. So they come in thinking this is fun, and then they start to discover use cases. And a lot of those use cases are mood boarding, but then it becomes, oh, actually, I could use this image for some form of commercial use.

I think that my takeaway from the last six months or so is generally that the tensions that we're going to have to deal with for the next few years have surfaced. You know, this question of, is this technology democratizing creativity, in which case it's also maybe commoditizing it? Or is it raising the ceiling for what's possible?

When you talk to a creative professional and ask them how they come up with great ideas and solutions to problems, it's always a factor of time, like how much surface area possibility can they possibly explore in a given amount of time to find the best solutions for a client or an internal customer.

Of course, also it's lowering the floor in the sense that all of us, our peak creativity may have happened at like six years old when everything we did got put on a refrigerator by our parents. And then over time, we started to lose our creative confidence because we were worried about what critics would say or what other skills people had that we didn't have. Now we all have the ability to prompt and generate something, whatever's in our mind's eye. And so there is that democratization. So I think there's tensions like,

that outcome and where that will land, or the tension between skill versus taste, what's more important. There are a number of tensions, I think, that are now very relevant. This tension around democratization is really interesting. If generative AI is so good that any of us can be hyper-creative with typing just a couple of words into a text box, and you play that forward and you assume that it's going to get better and better and better, in a way, generative AI almost

seems like it could be existential for creativity, right? If you imagine the biggest companies in the world having access to this technology and deploying it to create content for customers, does the role of a creator actually become less relevant over time because of AI? Well, I think that the...

I mean, my firm answer is the role of the creator will be even more important. And the reason why is because not just that I'm biased, but every brand is going to start flooding the zone. And I think that this is a term I'm hearing a lot these days when I talk to folks in journalism or companies that are competing for SEO on Google.

and every brand has this opportunity to automatically generate thousands of original posts around any particular topic or objective and flood the zone literally with content.

But what ends up happening when anything floods our view is we start to look for things that are unique and different. Like that's just human behavior. And so when you start to see tons of generated images around you, generated by, let's face it, AI is a consensus machine. It's a recycle product that just kind of churns out variations and derivations of a consensus amount of material. You're going to start to be more interested in things that move you.

And I think that this is where creatives have an opportunity, right? They've always helped brands tell a story. They've always helped brands stand out and do something different. And that is going to be more in demand in this future of flooding the zone as opposed to less. Speaking of flooding the zone, it seems almost inevitable that the big platforms, TikTok, Meta, Instagram,

How soon will it be before we start to see those platforms flood the zone with content and start generating content that they then feed into their own algorithms, right? To serve the perfect ad to you at the perfect time. I think it's the right question, Mike. And I feel like the collapse, first of all, this collapse of creativity and marketing

is fascinating to me. These were two independent industries to a large extent. They didn't even use the same tools. They like to throw things over the wall to one another. And now it's collapsing. And I think these social media platforms and this desire to have real-time marketing that engages people is going to be one of those instances. In terms of the flooding of the zone, I think what we're going to find is that

this notion of core and periphery is going to become very important in both marketing and entertainment.

where the core is going to be original content creation, humans being, you know, channeling their emotions and their personal experiences and manifesting something unique and different. And then there's going to be tons of automatic variations that are made based on using AI for translation purposes and acculturation and variations that are towards different segments of customers and whatever else. And I think we're going to find that in entertainment.

And we're also going to find that in marketing. And so what that would mean is going to TikTok, having a really clever campaign that has meaning, that is something that is different and unique, that will catch people's eyes in the era of flooding the zone. And then TikTok will automatically generate tons of variations around that to optimize the performance. Yeah, you can almost see that being almost a new form of UGC, right? Where an original creator is making a piece of content and then using AI to

the fans of that content are making derivative works or, you know, remixes of content. Is that kind of what you're talking about? I think that's right. And I mean, the very practical example of that is literally just translation, right? Which every content creator was really limited to just their own English or whatever speaking audience. And now it's like, no, as a content creator, you can have your content automatically generated for every single language within minutes.

But then you can even go further. Why not make the tablecloth red in China and white in Italy and green? And you can really have a lot of fun in that way as well. Yeah, you can imagine the impact that that may have on advertising, right? Watching a YouTube video and seeing a specific product that you love just automatically dynamically inserted into the content.

100 percent. I think that's within years if not months. Right. You've actually talked a little bit about how customers, fans using AI tools may actually inspire product development through the generation of content.

Talk about this and what does it look like if we play that out a couple years into the future? Well, this is a fun thing I think about quite a bit because traditional R&D for any sort of product happens in a department within a company. And it's a small handful of people, right, whose job is to come up with the next version of the soap or the next beverage or whatever else or the next Lego set.

And it's a constrained group of people that are, of course, sometimes geographically bound and they have been intoxicated by the brand for so long that maybe they're somewhat myopic in their imagination.

And of course, ideas for brands, for companies, for products have always been everywhere. It's just we've never had an effective way as consumers, as random people around the world, of expressing what those ideas might look like. And then the rest of our community kind of curating based on whether they like it or not to the point where the brand even notices it and then can act upon it. Now, fast forward to today in the age of generative AI,

A great example, the other day I saw an Instagram post of a collaboration Lego set between Lego and Hermes. And they had all these Lego sets of these Hermes bags.

And of course that became viral and tons of people started to engage with it. And then before you know it, like the companies are like, you know, talking about, I don't know, should we do this? And that was an amazing example of like open source R&D in the age of AI. Now, I don't know, should Lego even have a department for partnership development if the whole community is empowered to do this and they can just look and see what gets traction, which is almost, you know, a more effective form of market validation than research groups. So I think that's going to

start happening across every segment of product. And I think companies need to adjust to accommodate that. This reminds me a bit of something you've written about called the personalization wave, where AI has the potential to really personalize brand and product experiences based on your tastes.

You talk about a world in which you show up to a website, it already knows the size of shoes that you like and maybe the styles that you like. What's it going to take to get the AI to the point where it can have that level of prediction to it? Yeah, well, I think that there's two threads of technology that have to really exist in order for this to happen.

I do, if we look all forward to 10 years from now, I think the idea that any digital experience or even physical experience in our lives treats us as a stranger will be like insane to us. Like how could Nike actually ask me to pick my gender and specify my shoe size

That's such a waste of clicks. Why can't it just, welcome, Scott. These are the shoes that are running size 9 1⁄2 for men. Why should I have to go through all this anonymous sort of experience, right? I still also think that way for restaurants, too. Why shouldn't restaurants know you're a vegetarian that doesn't like mushrooms or whatever the case might be? I think that we...

We all want to be treated special. We just, the technology doesn't exist. And also the ability to augment these digital experiences in real time to make us feel like a personalized experience is happening also is not widespread. So a couple of things have to happen, right? Of course, the stack that provides these digital experiences for all of us needs to leverage AI to be able to

radically redesign it based on who we are without any sort of work happening. And then also our preferences, our own interests, you know, also need to be stored and relayed to the brand somehow. Now, my answer to that might be the idea of agents. And we'll get to that, I know, a little later. But this notion that, you know, if we're really well known by our devices or by some application that we use, maybe our experiences become personalized, you know, using that.

Now, the argument against what I just said would be that, do we all want a hyper-personalized experience in every part of our life? I mean, we certainly want to know how we're known. We don't want to feel like it's creepy, right? So we need to have some control and some degree of privacy around this. I mean, most of us actually do want to be known in instances where we want to achieve something. We just want to know how we're known. And then the second point is around entertainment.

You know, I've had this debate with some people in Hollywood around personalized entertainment. You know, why shouldn't a TV show that you watch have a slightly different ending than what I watch based on our own, I don't know, where we live or our own situation, our own preferences?

But unfortunately, if it's personalized, if we all have a different version of succession the last season, we can't connect over it. We have no shared experience. And without a shared experience, we kind of miss part of the sensation of entertainment. And what's the cost of this from a data privacy standpoint?

point? How does this not end up just being, you know, internet advertising on steroids to the point of privacy invasion? Right. Well, you know, my view on ad tech generally is that it was sort of like the dinosaur era, you know, of life on earth, you know, that

just went in a completely crazy direction. You know, instead of always inferring what we might like and sniffing our activity and retargeting us and creating all these databases of information about us without our permission. Like that was the era of, you know, that went on for decades, if you think about it.

I feel like the antithesis of that would be having our own preferences dictate our experiences. Like instead of me guessing everything about you and secretly trying to make your experience better, I don't know, why don't I just ask you? And then why don't I give you the option of having maybe Michael's agent inform my experience I'm about to provide for you?

Super interesting. Let's talk about investing in AI. Everyone wants in on the AI action. Obviously, there's a ton of venture capital flowing into AI. You've had some contrarian views, I think, about the opportunity for AI and investing. Where do you think is the best stage to invest in AI right now?

Is it at the venture level? Is it at the Magnificent Seven, the biggest companies in the world? Or is it maybe something in the middle, the Spotify's of the world that maybe can't hire hundreds of thousands of content moderators now using AI as a resource? Like where do you think the opportunity is from an investing standpoint? - Yeah, well, I mean, you're the highly sought VC in New York DC is in the product area. So maybe you should be telling me, but I definitely do believe that, you know,

if you look at the stack of what's enabling all of this,

AI to happen at scale. I mean, first of all, it starts with energy. You're going to need a lot of energy for very efficient AI into the future. And I do believe that there's questions around where this energy will come from and geopolitical, blah, blah, blah, right? Then there's the actual housing of these chips and servers. There's this company called Vertiv that's been on fire lately. There's a few companies that outfit the centers to be able to cool and optimize for machine learning at scale. And

There's just a shortage of places to do this stuff. Then there is the chips, which everyone knows that story. And then there is, of course, the models that are being built on these chips. And I think that there's a few kind of big LLM companies that are going to ultimately win. And then I think that there are some folks like us building very specialized models for particular areas, like you just saw Firefly in action there.

I didn't pay for that cameo, but I appreciated it. And you're going to have really deeply specialized vertical models for all the different areas of creativity. And so then you have the question is, okay, now what products do people use on top of everything that I just described? And

This is where startups are taking a lot of these capabilities as APIs and building nice veneers or interfaces on top of them. And I think that some companies who have established reach to customers already are just inserting those API endpoints wherever a customer needs a workflow to be achieved with superpowers.

and being able to do that really quickly, which makes this very different from previous platform shifts where startups ran circles around incumbents. I think now incumbents are like, wait, I can just get these APIs and cert them and I can have tomorrow a lot of these capabilities where my customers already are and I can even charge them more for it.

So the startups that I would be most interested in now are actually bringing AI to spaces where there actually isn't really great technology yet. I mean, you look at the legal area, legal fields, you look at government and government efficiency.

You know, if you look at government and municipalities, the types of technology that they are using is bulky, cumbersome, and tons of human sort of in the loop type stuff happening constantly. Whereas AI could just kind of truncate a lot of that quickly. And then maybe on the interface side, if you can come up with a interface that's 10x better than what an incumbent can do, but that means you have to have like exceptional designers. And that's why I actually think designers are sort of the unsung heroes of the era of design.

Everyone's talking about the opportunity for AI in the enterprise, and obviously there are a lot of highly capitalized enterprise AI startups. But if you look back on the two biggest success cases of AI startups over the past year, it's really been in consumer.

talking about Midjourney and ChatGPT, which I think is reported to be doing over a billion and a half in revenue, which is mostly in their consumer product. What is holding AI back from the enterprise? Why haven't we seen any truly breakout enterprise AI companies yet? Well, I think that, first of all, I think we definitely will. I believe that every enterprise right now is having an internal conversation. I know we are at Adobe around AI.

how AI refactors the way we work. And if you are a CFO or if you're the general counsel, you're going to your teams and you're saying, "Hey, in this area, no more headcount. I want to find a way that you're going to refactor how you work before you add more resources to your work."

And that's going to then stimulate the demand for some of these startups that you're describing. But I also think there's a lot of kind of nervousness in the enterprise around using AI, not knowing how it was trained. And I think that's actually one of the big kind of delays in adoption.

I want to talk about that. That's a big topic that we'll talk about in a minute. But before we jump to that, talk a little bit about SMBs. Obviously, startups are leveraging AI, but you've talked a lot about how the potential for AI really may be with SMBs, small companies of a dozen people or less leveraging this to get much more leverage than they could from hiring huge teams. Well, I'm so excited about this. And if

if any enterprises are going to scale down some of their departments and stuff as a result of AI, it's going to come back in terms of opportunity for employment with more and more and more SMBs scaling like we couldn't believe. And the reason is because if you had a small business or you have a small business

you know to have an hr department to have a finance department to have a marketing department to hire that data analyst and that marketing person and you know that becomes a big business right and so uh so smbs have been inherently constrained by their ability to have these functions operating at scale enter ai where suddenly you can have a financial analyst

AI tool that proactively suggests to you as the business owner where you should and shouldn't be spending. You have a little marketing analysis agent that you can talk to in natural language and you can say, "Hello, agent. Tell me which Facebook ads perform better." And it actually walks you through it and it suggests things to you in new campaigns and new regions.

And I think that you're seeing across the stack of what makes an SMB operate at a greater scale, you're seeing all kinds of cool AI tools emerge. And so if we fast forward, maybe we'll have 100x or 1,000x more small businesses that actually have some of the superpowers that were previously only accessible by the big enterprise. And that's like super exciting. Do you think we might see, you know...

five-person team worth a billion dollars, much in the way-- - 100%. And I actually think that I would say like, you know, within five years, I would not be surprised if there is a five-person team making north of 100 million in revenue using AI tools to cross, you know, have cross-functional capabilities.

Let's get into the IP stuff and the legal risk stuff that you just talked about a second ago. Starting with the training data side, so these large language models, they're trained on, in some cases, the entirety of the internet, every word that's ever been written, including stuff that is copyright protected and owned from an IP standpoint. It's the total Wild West out there. Everyone is doing this. There doesn't really seem to be law preventing it.

How is this likely to play out if you fast forward? Yeah. Well, this is like the very big, I would call it the billion dollar question right now, right? Is how are all these regulatory policies and guidelines and laws and also in many different regions? I think when we think about this, we typically think about the US and copyright law. But remember, like if you're a global company or even one of the SMBs I just described that wants to work globally,

you are constrained by the lowest common denominator of law anywhere you operate. So if the EU comes out and says, actually, in order to use a certain type of model, it needs to be trained on licensed content, then everyone has to adhere to that level for commercially viable content.

Right now, the laws are very ambiguous around, can you scrape the entire internet and create an image model? Can you scrape the entire internet and create a large language model? And also, if you can, well, then what if you conjure up copyrighted material about Spider-Man or Mickey Mouse or whoever else and use that? Obviously, you can't do that. And so this is what the jury's still out now. But where our bet is, is a few things. Number one, what we've heard from our customers at Adobe is,

is that they will only use models that are trained on licensed content to create commercially useful assets. So they will not put something in a marketing campaign unless it was trained on a model that is clean, number one. I think number two is what we've also realized is that when you pack these models with more and more scraped content, the quality actually does not necessarily go up.

It may know more copyrighted terms than a clean model, but it won't give better outputs.

And, you know, when we launched Firefly, I remember a few people hit it right away and started comparing to Midjourney. And, you know, some people obviously put in Spider-Man and said, oh my gosh, like I got a spider that looks like a man. Like what the heck, Firefly? And, you know, of course we had to remind them that also is a feature, not a bug. And that's how you can know you can trust this model. So I think that's kind of how this is playing out. How

But how is it actually going to work from a practical standpoint? I mean, how can this even be enforced? Also, if you think about it, so much of what is written or what is created from an imagery standpoint is already inspired by existing work. So how are you going to account for this and make sure that every piece of content is tracked?

So I think that when it comes to ingestion of content to train a model, we had Adobe Stock with 300 million assets. Each one has a building release or a model release. And each one, we know the origin of that piece, and we have a license to use it for training.

I think that, but if I were to forecast a few years into the future, I think every single piece of content generated with any mainstream AI tool will have kind of credentials in that asset that says what model was used to train it. And it will be relatively common knowledge which models are commercially viable and safe and which ones are more for like fun, but not commercially useful.

I also think that the big platforms like the Metas of the world will display counter-credentials on every asset you see. And so you're going to determine whether you can trust or use that asset based on the credentials that are exposed. And when you see assets with no credentials, you will actually ask yourself, is this real by default? And I think that this whole notion of trust but verify, well, the new version of that is called verify then trust.

We're going to look at an asset. We're going to say, does it have credentials? If so, I might peek and see where it came from and who actually made it.

And then I will determine in my mind whether I can trust this or not. Right. And then, you know, if we think about the output side, right, where a model is spitting out content, obviously, you know, IP infringement is much clearer there. You talked about the Spider-Man example. If a model spits out Spider-Man, clearly that's infringing. Is there actually an opportunity here for certain brands or properties, right, where maybe back to our point earlier about fans leaning in and creating derivative works, right?

What's the opportunity here for letting people actually play with IP? Well, it's funny you ask. Right now, we're actually doing some experiments with some of our customers, like big media brands, where they're making what we call custom-tuned models. So they're taking a version of our Firefly model, which is the basic image generation model, and then they're adding all of their IP into it. Like imagine SpongeBob SquarePants or some other character. They're putting all the assets they've ever generated

They're adding it to that version of the model for their own purposes only. And it's pretty amazing, the outputs. I mean, they can make marketing graphics, they can do character exploration, and they can use it because it's all commercially viable. It's all stuff that they own or that we have license to.

And when I see this, I'm like, okay, that's cool for your brand, but why not let your viewers play with this? Why not let customers? Of course, right now, a lot of them are very scared of that proposition. Who knows what pose someone will prompt SpongeBob to be in, right? This could be a little scary, but I do think we have to kind of lean in to your point and allow people to engage and embrace some of this, see what comes up from it. Yeah.

You talked about agents earlier, and everyone's talking about agents. This next wave of AI is all going to be about agents. We're all going to have agents running around doing work for us. It seems kind of easy to generalize that as an outcome. But maybe what are some of the practical ways you expect people to leverage agents in the near term?

Well, I'm imagining a day where you get a new version of your mobile OS operating system and it sort of says to you, "Hi, do you want me to immediately make

an agent to help you make great decisions that lives locally on your phone, so there's no privacy concern. It's not going into the cloud somewhere. And you're like, sure. And then it says, great, do you want me to mount all possible data sources that live on your device so that we can really personalize this experience for you? And again, you're like, well, if it's staying on my device and there's no real risk because the stuff is already on my device, why not?

So you say yes. And then suddenly, every email, every text, everything you ever look at on the internet, everything you ever buy, all of that stuff digested locally onto a model that's super intelligent that then starts to advise you on things. So say you're on a shopping website and you're about to buy something because you were brainwashed by the brand marketing that you saw unknowingly on the billboard.

and might say, "Michael, you're sure you want to buy those gloves? I found a 30% cheaper version that is ranked on Consumer Reports to be far better. And by the way, seven of your 30 friends have these, and so you should probably buy those gloves." And you're going to be like, "Oh, okay."

And then you're going to get a call and it's going to sound like your grandmother asking for money. And it's going to say, 99.999% chance this is not your grandmother. Like, do not send her money. And you're going to be like, thank you, agent. And then this is going to go on and on and on. And you're going to start to really trust

and value this local agent that accompanies you through life. And eventually, you're going to start deploying this agent to talk to other agents on your behalf. So you're going to go to Delta, and Delta is going to have an agent saying, how may I help you? And you're going to say, agent, go talk to Delta's agent. You know me. Figure it out. I got to get there and get back. And then it's going to know everything about you.

So in the world I just described, what are the implications? Number one, first of all, brands have less power over our decisions because the agent is going to be trusted. Number two, everyone's so concerned about AI killing us. Well, maybe AI will save us because this AI will actually always tell us when AI is trying to trick us, right? So I think this sort of has to happen. And I do think it's kind of the next big thing. Now, whether it's two years from now or five years from now, like, I don't know.

Yeah, I mean, if you play that out and we all just have agents running around for us, that sounds like some, almost like some utopian society where we assign all of our work to our agents and then we're free to just, I don't know, improve ourselves or entertain ourselves. Like, is that where this goes ultimately? I have that image of WALL-E with like the really big people being like carted around by wheelchairs. Sounds terrible. I know, drinking milkshakes. Yeah.

You know, I am an optimist. Like, I believe that whenever we reduce, like, lower order work for us as humans, we end up taking on higher order work. And so my hope is that the more human ingenuity we unlock per person, the more humans we are going to want to employ on things. And by the way, the proof point of this is engineers. Over the last 20 years, engineers have become more productive every single year, like materially.

And yet, companies keep hiring more engineers. Why? Because they want to create more features, more products. They want to optimize. They want to make them better. Unless you're a penny-pinching, private equity-owned company, you want to... Are there any? I'm sorry. Wrong audience.

But you're going to want to hire more of these engineers to do more, right? And so I think that the insight here, whether it's creators or engineers, I actually think that in the future, we will have more people on the floor at the Nike store helping us.

Because they will have unlocked some of the people that had to work behind the scenes and they'll have more capital to invest in like the interpersonal relationship building stuff that actually truly makes us make great, maybe irrational decisions of where we buy. It's like the impact of humans and relationships, you know, is very material in our purchase decisions. That seems like a perfectly optimistic place to end. Scott, thank you so much. Thanks, everyone. Thanks, Mike. Thank you.

Thanks for listening to Generative Now. If you liked what you heard, please do us a favor and rate and review the podcast. It really helps. And of course, subscribe to the show on Spotify, Apple Podcasts, YouTube, or wherever you get your podcasts. And if you'd like to learn more, you can follow Lightspeed at LightspeedVP on YouTube, Twitter, LinkedIn, Instagram, or anywhere else. Generative Now is produced by Lightspeed in partnership with Pod People. I am Michael Mignano, and we will be back next week. Thank you so much.