When a new technology breaks into the mainstream, it seems like a sudden and transformative shift. Naturally, a gold rush follows, as corporations, venture capitalists and enterprising entrepreneurs alike all scramble to invent the ideal use case, or at least the use case that can most quickly get to market and find traction.
How Entrepreneurs Should Use ChatGPT
Entrepreneurs should use natural language generation products such as ChatGPT and Bard as tools, not as technology to build products on top of.
You might remember that not too long ago, an iced tea company tried to create the ideal use case for blockchain.
The rush is on with natural language generation and the recent commercialization breakthrough heralded with ChatGPT. Believe me, everyone I know is trying to figure out how to weave the NLG wave into their own business model, including me.
But I’m a special case.
As a writer, I’ve got a bone to pick with ChatGPT. As an entrepreneur, I hold a patent for the first commercial platform to use natural language generation in a corporate environment. So I’m forced to play both sides.
A Brief History of NLG Commercialization
Back in 2010, sports data entrepreneur Robbie Allen and I, along with a handful of young developers, built a platform and an engine to create human-sounding content out of large sets of data, mostly sports data at that time. We didn’t call what we were doing NLG because we weren’t aware of the term yet. It hadn’t even been coined. We were just trying to do better sports data. We built a company around that platform called Automated Insights, which was acquired by private equity in 2015.
I wrote algorithms to analyze the data and determine what could be said and when it should be said, paying special attention to patterns, deltas, trends, and milestones. In the closed universe of sports and eventually industries like finance, marketing and even healthcare, my algorithms could produce some pretty amazing and magical content. Thanks to Robbie and the coding team, we could spit out those articles at the rate of tens of thousands per second, each one unique, each one reading like it was written by an industry professional.
In 2011 and 2012, we pivoted from writing about sports to become industry agnostic and instead focus on creating content where humans couldn’t. So for Yahoo Fantasy Football, for example, we produced millions of fantasy matchup recaps every Tuesday morning. For the Associated Press, we wrote hundreds of quarterly earnings report news articles, some of which the AP added a human touch to, some they just let go to press. These and many other examples are still in use today.
Were We Replacing Writers?
Automated Insights was VC-backed and grew quickly, and as we launched more and more examples of our ever-improving tech, we got more attention and more press. Without fail, every single interview either of us did ended with the journalist going off the record to ask if we were coming after all their jobs.
The honest and obvious-to-us answer was no, not a single one. We weren’t automating journalists, we were automating data science.
The question became so pervasive that eventually, in an attempt to answer before it was asked, we added a ticker to our website:
Number of Journalists Automated Insights Has Replaced: 0
And the joke was that the ticker never moved off of zero.
I constantly fought the easy perception that what Automated Insights was doing was writing. And in that same sense, ChatGPT is not writing. Yet. It’s aggregating and analyzing data and producing results in a narrative format, albeit a narrative format that “writes” exceptionally well.
But what looks magical and scary now is not a lot different than what looked magical and scary to everyone back in 2010.
So what are the entrepreneurial implications of this magic? Where’s the gold mine?
Well…
Just Because You Can Doesn’t Mean You Should
One lesson we learned pretty quickly at Automated Insights was that you shouldn’t automate everything.
Once we had a handle on our tech, we decided we needed to show it off. And to do this, we hit upon a pretty great idea. We spun up more than 800 websites, one for every pro and college football, basketball and baseball team in the United States. Then we populated those websites with content, up to five times a day, for game recaps, previews, players of the week, your common sports content.
Great tech. Lousy use case. I’ve come to realize that those five words kill most young, promising startups.
Again, people thought it was amazing and it showed great potential. One problem: Every pro team and almost every college team had at least one human covering that team. Regardless of how awesome our tech was, the humans were better. They could get quotes, add gut speculation, source strategy, stuff we could never do.
Great tech. Lousy use case. As I’ve gained more experience as an entrepreneur, I’ve come to realize that those five words kill most young, promising startups.
But then something interesting happened.
Finding the Optimum Use Case
We started noticing that smaller college conferences started promoting our content on social media, even blasting out press releases when we selected one of their players as that conference’s player of the week.
Light bulb. No one was writing about these teams, or any teams at these smaller schools. We realized we were unearthing a new audience and fulfilling an unmet need.
Same tech. Optimum use case. This led to our project for Yahoo Fantasy Football, and even before it was released, the potential was so overwhelming (based on actual revenue from ads they were able sell against that content), that we decided to go all in producing content where it couldn’t be produced — large sets of financial data, marketing data, traffic data, population data. Insights that would take a human weeks to produce, if a human was even working with that data, would take us seconds.
There’s no screaming unmet need that ChatGPT, as magical as the tech is, would instantly fill.
What does that mean for ChatGPT? Well, to boomerang back to the title of this post, I don’t see another obvious optimum use case on the horizon. In other words, there’s no screaming unmet need that ChatGPT, as incredible and magical as the tech is, would instantly fill.
Although there are indeed a lot of possibilities. And so, I’ve been swatting down these possibilities for months now as plucky entrepreneurs excitedly drop them at my doorstep. No, ChatGPT shouldn’t write your blog posts for you. No, ChatGPT shouldn’t bug your co-workers to get that report to you by the end of the day. No, ChatGPT is not a friend for hire.
All my “no” begs the question: Where do I see that optimum use case rising to the top?
The Web is Broken
Along with the idea that we could produce written narrative from data, at Automated Insights we shared and were driven by the idea that the internet itself was buckling under the weight of the data being generated and aggregated into it. And in 2023, I hold the belief that the web is irreparably broken.
The Internet was introduced to the mainstream as an information superhighway, and like most superhighways, it has evolved into eight lanes of backed up traffic.
Search engines are optimized to produce results that are paid for. Social media is full of fake facts and sometimes proactively placed disinformation. Professional media is mostly a hot mess of clickbait, advertorials and dwindling revenues. The younger among us are turning to Reddit for their web searches.
Let me rephrase that for emphasis. Reddit is becoming the single source of truth.
Can AI and NLG Save Us?
Maybe. Perhaps. Possibly. And I believe unclogging the information glut is the optimum use case for ChatGPT at scale. Although we’re seeing less than stellar results in the early days (looking at you, Bing and Google).
I think we need another catalyst.
Automated Insights could only fill the need for machine-written data-rich narratives once the promise of cloud-based on-demand processing power could be harnessed and directed at our optimum use case. In other words, the only cost-effective way to produce content at scale was to spin up hundreds of servers for a couple hours and then spin them back down when we were done. We broke AWS a lot, and we worked with them to bridge our gaps.
ChatGPT is going to need something like that. Right now it’s loaded with possibility, but until another catalyst comes along to allow this new flavor of NLG to fill a dire business need at scale, it’s best used by entrepreneurs as a tool, not a technology to build a product on top of.
