This newsletter has a lot of images, so it might be hard to see what it has to offer. Here’s the summary:
A discussion on being ChatGPT famous and the economic implications
An economic lesson from the “make it more” trend with DALL-E 3
A voice message for the paid subscribers
Almost (ChatGPT) Famous
I recently discovered I was ChatGPT famous, and it got me curious about what the economic implications are.
Just like back in the day when you would Google yourself, I decided to ChatGPT myself. Here’s what I got.
I was super impressed! Out of the hundreds of terabytes in its training data, GPT4 recognized that I was someone of note in there. And it was for my work on Market Power, which, on the one hand, validates the work I’m doing on YouTube but, on the other hand, makes me worried my research isn’t being disseminated sufficiently.
But maybe that’s because I didn’t push it enough to find my research. So I prompted it to dig deeper.
I think it’s interesting that it listed development economics first, which is one of my main specialties, but I think it was just guessing.
Some might conclude that I’ve reached a dead end. But I know about something called the reversal problem. At one point, you could ask ChatGPT who Tom Cruise’s mother is, and it would respond Mary Lee Pfeiffer. But if you asked who Mary Lee Pfeiffer’s son was, it would say it doesn’t know. Some information is easier to index in one direction than the other.
So I started prompting it with questions on economists who do work on the economic history of Haiti. But it never suggested my name. I thought it might be time to give up. But then I realized that it also did not recommend the name of Mats Lundahl, the most prolific economists on the history of Haiti, writing several books on the topic. If it couldn’t suggest him, then maybe I shouldn’t be so upset.
Finally, I decided to make the connection as obvious as possible. I gave it my name as well as two of my coauthors and asked it if any of us had worked on Haiti.
It nailed the main topics of both of my coauthors. But now we had an interesting response on me. It finally said that I worked in economic history and development economics, which indeed are my two primary fields. But it missed that all of that work is in Haiti.
I gave it one last chance.
But it failed! It came so, so close. But, for some reason, it is just missing that last connection.
Economic Implications
What does this mean going forward? It’s increasingly likely that ChatGPT and other LLMs are going to replace Google and Wikipedia as our primary source of knowledge.
When Google became the oracle, suddenly, there was economic value in being the first result in a search. An entire industry was born as experts in search engine optimization (SEO) found ways to exploit the Google algorithm and push pages to the top.
And there’s something similar going on here. When I asked about my name, ChatGPT found me. But when I asked about my family members, it didn’t know who they were. And while it feels kind of cool to be ChatGPT famous, it only knows about one side of me (YouTube) and not a very important professional side of me (research). If someone comes to the oracle asking for who they should consult on economic issues in Haiti, I would really value ChatGPT recommending my name.
So clearly there is going to be an industry around GPTO (GPT optimization). There will be people claiming to get ChatGPT to recognize your name and highlight your accomplishments.
But will it be as effective as SEO?
I don’t think so. Google updates its database daily. A new page will end up on the search page almost immediately if you use the right terms. But because training ChatGPT is so computationally intensive, it only updates every few months, if that.
And how much is that marginal data worth to ChatGPT? With Google, you could push and pull and by the end of the week know if your strategies worked. What if you seed the training data and still get it drawing blanks like it does with my Haiti research? It would take months to realize this, and therefore months to know if your investment paid off.
ChatGPT might not be built for these details. If I go on Google, it’s like a personal library with access to all of the books on my subject. Type my name into Google and you’ll find my LinkedIn, my Twitter, my YouTube, my personal page, and every journal publication. You have to synthesize the information yourself, but it’s all there and current. ChatGPT, on the other hand, is like a curated museum. No matter how hard you look, the only information you’ll get is what the museum puts on the plaque.
A GPTO industry seems inevitable. But it seems like it will be most valuable to those who want LLMs to have a specific narrative on their plaque.
More Famous
There’s a trend right now where people have DALL-E 3 take an image and make it more X. I decided to have it make a YouTuber who was ChatGPT famous. But my experience had me realize something about the economics of the challenge.
Then I had it make him more famous.
And more famous.
And more.
After this image, I asked it to go again. But it pushed back.
I had reached the point of diminishing returns! DALL-E independently decided that there was no clear path to push the image forward and get a return on the investment. That’s fascinating!
But I pushed it anyway.
Perhaps the most surprising part of this image is that the peak of intergalactic fame is CatTube.
Overall, I’m seeing reports that ChatGPT is pushing back against requests to do work. I’ve seen some examples of people asking for lists or code, but ChatGPT declines to produce the work and tells the user to figure it out. Maybe we’re making ChatGPT too human…
Voice Message
Since we’re already talking about ChatGPT, let me share some more of how I like to use it.
Keep reading with a 7-day free trial
Subscribe to Market Power Newsletter to keep reading this post and get 7 days of free access to the full post archives.