FollowFox blog

The AI Gold Rush — Or: It's okay to delay

Learn AI Blog
I picked up one or two pieces and examined them attentively; and having some general knowledge of minerals, I could not call to mind more than two which in any way resembled this, iron, very bright and brittle; and gold, bright, yet malleable. I then tried it between two rocks, and found that it could be beaten into a different shape, but not broken. I then collected four or five pieces and went up to Mr. Scott (who was working at the carpenter's bench making the mill wheel) with the pieces in my hand and said, "I have found it."

"What is it?" inquired Scott.

"Gold," I answered.

"Oh! no," replied Scott, "That can't be."

I said, "I know it to be nothing else."

— James W. Marshall

First, the cheap shot (and why it makes sense)

The year was 1848, and a carpenter named James W. Marshall had just made a startling discovery. As he stood by the American River in Coloma, California, he noticed something glinting in the sunlight. It was a nugget of gold. Little did he know that this simple discovery would set off one of the most extraordinary events in American history: the California Gold Rush.

Within a year, some 100,000 people from across the United States and abroad flocked to California in search of fortune, almost doubling the state’s population. They were called “forty-niners”, after the year they arrived; like a swarm of bees, prospectors descended upon the region, propelled by the hope of striking it rich. They arrived by land and by sea, braving hardships and dangers along the way. By early 1850s, about 300,000 people had arrived in total.

By 1852, a Wisconsin-based lawyer named Leland Stanford, having lost most of his property in a fire, joined the rush and set up a general store for miners in Placer County, California; by 1856, he was already among the top merchants in the state, and he went on to become Governor and Senator of California. By 1885, bereaved by his only son’s death from typhoid fever, Leland donates most of his estate for the creation of a new university on his honor – the Leland Stanford Junior University, or Stanford University for brevity.

Fast-forward to the 21st century, and the same spirit of innovation and determination that fueled the Gold Rush is once again alive and well. It turns out, gold was again found in California; the motherlode was struck in the Pioneer Building, 3180 18th Street, San Francisco – OpenAI’s head office –, led by Stanford dropout Sam Altman.

Yes, this is a cheap shot. Yes, it’s a corny way to start a text about AI. From Hollywood to Silicon Valley, countless people have made the cheap parallel with the Gold Rush. But something really feels different now, and it may be summarized by the graph below.

Figure 1. Estimated computation costs per decade.

The graph above lays bare the similarities between the AI Gold Rush and the California Gold Rush. It reframes the current story of AI as a story of disruptive innovation in the market of software engineering. Using large language models trained on huge repositories of code, it is now possible to directly translate natural language to code – from now on, non-technical people can launch functional apps in a matter of hours.

Now Imagine the scene: a ramshackle town filled with mud-caked miners, saloons overflowing with raucous laughter and whiskey, and merchants peddling their wares to eager prospectors. California in the 1850s was a world of grit, determination, and chance, where the only thing that seemed to stand between a man and his fortune was the distance he was willing to dig.

Substitute ‘gold’ for a ‘software use case’, and you may see what I’m talking about. It’s the same thing, minus, perhaps, the mud.

The name of the game: expertise

Situations like these push people into overdrive. When the barriers between you and unfettered success get this small, there’s a feeling of urgency that is both healthy and understandable. We just need to act now – dig now – ship something now.

Despite this, I’m going to argue against that and say that it’s fine to wait. Relax. Don’t worry. All will be fine; you will be fine.

It’s a difficult pill to swallow when we see the likes of Lensa AI. In early November 2022, Lensa was one of several apps launched by Prisma Labs, a little-known Silicon Valley startup that provided simple AI tools to enhance photos for users under a subscription model. While certainly not unsuccessful – its revenue run rate was estimated at about $5 million/year –, Lensa was not a breakout app by any measure.

Then, by late November, Lensa AI introduced its “Magic Avatars”: Lensa AI would receive 15-20 pictures of a user and, using Stable Diffusion and the Dreambooth fine-tuning algorithm, Lensa AI would reimagine their users’ image on a fantasy setting, providing them with hundreds of artistic drawings and renderings that were as appealing as they were faithful to the original images. (You can learn how to do the technique they used here in FollowFox.AI; if you're using Windows, we recommend you start by reading this post, followed by this one.)

It was an overnight global success. To give an order of magnitude: between December 06 and 10, market intelligence company SensorTower estimates that Lensa AI was the app with the largest daily revenues between all categories of apps in the US App Store. This means Lensa AI easily raked in well over $1 million per day during that time; since then, Lensa’s revenues tapered, but are likely well above the previous run rate. In light of this, how would I dare argue that it’s ok to wait?

Because, as weird as it may sound, stories like Lensa’s had little to do with the new technology, and a lot to do with who they already were. Prisma Labs was incredibly well-positioned to do what they did. They had a successful app, Lensa AI, that sold AI tools for users to enhance their photos; they had carved a niche and knew their customers well; they knew their infrastructure needs and how to organize and plan their workflow. When they absorbed Stable Diffusion + Dreambooth in their tech stack, they knew the scaling challenges they could face, and planned for them. In short, they were professionals in the field.

The fact that they stumbled upon digital gold was only possible because that they knew how to mine that gold beforehand.

Social Media and AI: a modern-day parable

I’m absolutely not arguing against shipping products fast. I’m arguing for the fact that we must become experts first. Whatever it is that you’re building with AI now – take your time to learn about your target market; delay launching your main idea until you do. Find a nice little niche to work with; get to know it very well; grow from there. See where that growth leads you.

Also, since AI is a blue ocean environment where almost everything is new, this also means I’m arguing against the urge of being the first one building something. Don’t be the proverbial pioneer; these get the arrows. You want to be the settlers, since these get the land.

To illustrate this, a modern-day parable — the advent of social media — comes to mind.

Social media as a business model generally aims to connect users to advertisers, who foot the bill on behalf of users. Facebook started with just Harvard students; its first product, launched in February 2004, was designed to get all of Mark Zuckerberg’s classmates signed up, which was hardly enough of a market to ensure the platform could cover its costs. History would prove this a spectacularly successful model – as Peter Thiel would say in his book Zero to One, “This is why successful network businesses rarely get started by MBA types: the initial markets are so small that they often don’t even appear to be business opportunities at all”.

In contrast with Facebook, Friendster, Myspace and Orkut – all of which started before Facebook – believed that, in the newfangled world of social media, the winner would likely take all. While all three were lauded as a success at the time of their respective launches, their launch strategies eventually planted the seeds of their own demise (see table below and texts below).

Figure 2. Rough comparison between Friendster, Myspace, Orkut and Facebook.

Pioneer #1: Friendster

Friendster, the first true social network, was launched in March 2003, originally intended to be a better dating website than the offers available at the time. There was no defined go-to-market strategy – the idea was to test the platform by launching a first version, and to grant access to anyone over 18 years old with an email account. Friendster quickly exploded in popularity, reaching 1 million users only four months after the beta release.

It soon became clear that few users used the site to find dates; instead, they were using the platform to invite their friends to keep up with them – who, in turn, invited their own friends. The uncontrolled viral growth that came out of this led to Friendster growing a large user base in Asia, particularly in the Philippines and Indonesia, markets which the company was completely unprepared to serve; in addition, the exponential growth soon overwhelmed the platform, which was simply not prepared to scale (these were days way before Cloud-as-a-Service, after all). To make matters worse, the lack of registration control led to a few users setting up fake profiles, which hindered meaningful social interactions between real users. Friendster eventually started removing such profiles, however there were little controls in place for identifying the profiles and impeding new ones to show up. With all these difficulties piling up, Friendster would never be able to present an attractive value proposition for advertisers, eventually experimenting with several different business models until it decided to quit social media entirely in 2011.

Pioneer #2: Myspace

In key ways, Myspace was Friendster put on steroids.

Launched in August 2003, it had been better engineered to scale, and aggressively targeted disgruntled Friendster users as part of its go-to-market strategy. Myspace gave users full control over their pages’ look-and-feel and openly embraced fake profiles – in fact, the company actively recruited some of Friendster’s most well-known ‘fakesters’. Fundamentally, Myspace was better than Friendster in understanding who their users were, and was quick in introducing new features in response to customer feedback (such as, for example, tools to better connect bands and fans). By early 2005, Myspace had overtaken Friendster as the dominant social media platform, which led to it being acquired by NewsCorp for over half a billion dollars.

Despite the successful exit, Myspace would also suffer from the mistakes in its go-to-market strategy. Lack of content curation, coupled with the acceptance of underage users, led to a proliferation of controversial content; real-world cases of sexual assaults connected to Myspace severely hurt its reputation. As a result, users abandoned it, and advertisers refrained from associating their brands with Myspace. Ultimately, NewsCorp’s new management could not never turn it into a profitable venture after its purchase, and it was sequentially sold to different groups for a fraction of its original value. The remnants of Myspace continue to exist in a weird form in

Pioneer #3: Orkut

Anyone remembers Orkut? Brazilians do; at its heyday, almost 70% of Brazilians were present in it. Quietly launched by Google in January 2004 as a side project of Turkish engineer Orkut Büyükkökten, it went on to become the leading social media company in Brazil well into 2012, when Facebook overtook it. Two years later, in 2014, Google deactivated it. Talk about a fall from grace.

At the time of its launch, it was said that Orkut was a social experiment that could prove the Six Degrees of Kevin Bacon law – that is, any person would be at most six degrees of separation from any other person on the planet. Accordingly, Orkut’s launch strategy involved bringing people to the network through invitations from its users, preventing new users outside of any friend networks. The result was extremely successful from the point of view of user growth – trading access to the network created a sense of scarcity and selectivity that attracted users into the network.

However, this strategy brought unintended consequences. First of all, by outsourcing to users the work of disseminating access to the network, Orkut had no control on the geographical growth of the network; its main user base eventually became Brazil and India. Similar infrastructure issues faced by Friendster would then become standard in the network as well.

Secondly, and perhaps more importantly, the user-invitation system led to more sociable and communicative users dominating in the network. Just like social networks outside the virtual world, initial interactions between individuals served as a basis for future conduct norms – think of a first day of a school where all students were popular ones. Disorder reigned on the network; Orkut was ridden with user communities with funny names, like “Mom I love you – give me money”, “I love unicorns”, “As Godzilla used to say”, “Dishes: they wash me”, or “No trans fat? No good”. Interactions within Orkut were rarely genuine, and user engagement could never be truly converted into advertising interest.

Settler: Facebook

Facebook’s launch strategy could not have been more different. According to Dan Olsen, Friendster’s former Senior Director of Product, the most masterful thing Facebook did while growing the site was having an effective go-to-market strategy. "They started at one US college campus, then another, then another. When you control your roll out like that, you can control scale." Facebook overtook Myspace in Monthly Active Users (MAU) by early 2009, following an unparalleled growth story that led to one of the largest IPOs in US history in 2012.

Having the advantage of witnessing Friendster, Orkut, and Myspace develop, Facebook focused exclusively on students from top universities in the United States for its launch. Starting at Harvard in February 2004 and launching at Stanford, Columbia, and Yale the following month, the social network's "first day of school" couldn't have been more contrasting than Orkut’s. The values of elite students were carried over to the social network, where they found their peers/people they wanted to impress, which led to positive behavior patterns; these patterns were ultimately copied by other users when access to the network became unrestricted. Advertisers clearly saw the value in accessing such a focused and well-off customer segment, and lined up to participate. Eventually, Facebook would connect most of the top US schools before finally opening the platform to the general public in September 2006.

With its slow, deliberate growth, from connecting college students within institutions, to connecting colleges with each other, and finally to connecting everyone else, Facebook discovered an optimal strategy to create an environment of trust and positive engagement at that time. By relentlessly iterating and improving upon its platform to its original users, Facebook would distinguish itself from its competitors with a clean, simple interface and an unwavering focus on user experience, such as by creating the “wall” experience which would infinitely scroll down and show what your friends were up to. Whether deliberately or not, Facebook would discover that success hinged not on the strength of one's technology or growth speed, but mostly on content curation and the ability to create a seamless, enjoyable experience for users. Facebook was able to win over users in droves, eventually surpassing its rivals and emerging as the undisputed king of social media.

Figure 3. Google Trends on search of social networks (Facebook in Dec 2008 == 100).

There is naturally some truth with regards to speed as key to win. Surely, strategic moats get wider as network effects intensify, assuming that the competitive landscape stays roughly the same – an expanding shopping mall would surely disincentivize new competing shopping malls to arise in its vicinity, just as Microsoft’s pre-GPT Bing has a difficult time standing up to Google as a viable search engine.

However, it is the emergence of new business models that learn from the pioneers’ mistakes that seems to define success – which is to say, it’s Amazon, not another competing shopping mall, that disrupts retail businesses. In other words, as Peter Thiel puts it, in Zero to One, chapter 5:

You’ve probably heard about “first mover advantage”: if you’re the first entrant into a market, you can capture significant market share while competitors scramble to get started. But moving first is a tactic, not a goal. What really matters is generating cash flows in the future, so being the first mover doesn’t do you any good if someone else comes along and unseats you. It’s much better to be the last mover—that is, to make the last great development in a specific market and enjoy years or even decades of monopoly profits. The way to do that is to dominate a small niche and scale up from there, toward your ambitious long-term vision. In this one particular at least, business is like chess. Grandmaster José Raúl Capablanca put it well: to succeed, “you must study the endgame before everything else”.

In conclusion

So Facebook’s story connects with Lensa AI: it carved itself a niche (Ivy League students); committed time to learn from its niche, and learned tons from it; and then rolled out a product that grew at breakneck speed. It found its gold exactly as it acquired the means to mine it.

The thing about this story is that, while it may seem obvious in hindsight, it was very hard to see this at that point. There were no written rules about how to build a successful social media company. Few people would see that the launch strategy of a social network could decisively affect its social mores and the content its users would ultimately share in them; fewer people would argue against scaling fast and in favor of creating scarcity. As a consequence, almost no one in 2007 would believe that Facebook would dethrone Myspace within a year.

The Gold Rush was an era of ingenuity and resourcefulness. Miners devised creative methods and products – stamping mills, hydraulic mining and denim jeans would be invented then –, while opportunistic entrepreneurs made their own fortunes by providing services to the prospectors. However, the harsh reality of the Gold Rush was that, for every person who found gold, there were thousands who didn't.

As we reflect on the past and look to the future, we can learn valuable lessons from these historical events. We should celebrate other peoples’ achievements in AI, as there’s plenty of fish in the sea and we can only learn from their successes and mistakes. We should keep our heads down, learning and building – even as we keep our ears up, taking the lay of the land and being up to date with what’s latest around us. First comes expertise, then comes success.

And so we should relax. Remember: Leland Stanford was not a forty-niner; he came in 1852. Facebook was not the incumbent social media. If the titans of the Gold Rush and the Social Media era could wait, so can you. Delaying is fine.

Don’t worry. Relax!
Made on