James Currier argues that proprietary data advantages in marketplaces are overstated and short-lived. He expects horizontal LLM solutions to beat vertical ones, pointing to Upwork as an example of how general tools can outcompete specialized platforms.
Chapters
So James, my question centers around, you know, reading through the blog posts, it's clear that you believe LLMs will have a huge impact on marketplaces. Yeah. One thing that's not clear to me is whether that innovation will be primarily horizontally. Example given, someone building an LLM service for connecting supply and demand across all marketplaces, similar to what Twilio did for SMS, or if the innovation will be so training data dependent that It'll only take place vertically or within a specific company. Example given, Upwork building the leading services chatbot because of their advantage in training data. I'm kind of interested to hear your thoughts on which direction do you think the marketplace market will evolve and why?
Yeah, I mean, this is, this is one of my core theses that people are overestimating the value of proprietary data. So as, as my angle into answering your question, it's going to come from that perspective. Got it. That if you go to Google and type in data network effects, the first article is that about my skepticism about the existence of data network effects. Now, my marketing team made me change the title so that it wouldn't be so negative, but because people love to believe in network effects and proprietary data. And I'll tell you why. It's because Google and Microsoft, keep telling Wall Street, we're going to win because we have the data and we have proprietary data. And the reason those companies do that is to keep Wall Street investing in them, number one. Number two, to keep their employees from leaving. Like, why would you leave and start a startup? You're going to have no data. And you know that because we have the data, we're going to win. So this, I think, is a myth that they create because I can synthesize your data. I can steal your data. I can cobble together other data sources that aren't yours. And I can get very good approximations. And as these models get better and better, my solution will become, you know, 98.8% accurate and yours will be 99.2% and the customers can't tell the difference. And so I don't find, I think Upwork's, you know, lead is, you know, maybe 4 to 6 months, you know, in terms of this. And if I really wanted to create something that did what the Upwork bot does, I could do it. Got it. And so I suspect it's going to be more horizontal in nature. I think the reality is going to be more catch-as-catch-can over the next 2 to 3 years as this whole thing starts to settle out and everyone understands the game. And, you know, but I would not over-index on your proprietary data thesis. I think there are, I think there's going to be very, very few cases where that ends up being persistent.
