⊕ An atlas of ideas · Sheet 01
I'm Vansh Gupta, a PhD student at MIT Sloan. I study how new ideas get made, especially the ones nobody pays for directly, like the open-source code inside almost every app and the technical standards the internet runs on.
That question runs through all of my work. With coauthors, I studied what happened when GitHub started letting people pay open-source developers directly. The developers who began receiving money went on to do less of the community work their projects depend on. In a paper of my own, I use committees that pick their members by lottery to show that serving on one makes experts less likely to volunteer again, even though past members look more committed. In Brazil, my coauthors and I are reading more than 16 million paragraphs of local government records to measure what local governments actually do each day. I'm also part of a large study that tested AI models against human scientists at building theories.
What these projects share is an interest in public goods, the things everyone can use but nobody has much reason to pay for. Open-source software and internet standards are good examples, and much of both is made by volunteers. As AI makes it cheaper to build things, ideas and tools that anyone can use matter more, and I don't think we are good at producing enough of them yet. I want to work out how to get more of them made.
I got here the long way. I did two bachelor's degrees at Cornell, in statistics and in applied economics, and while there I checked other researchers' results for the American Economic Association. After that I spent two years at Charles River Associates working on antitrust cases about healthcare mergers. I often end up building my own tools when the data I need doesn't exist yet, like an R package for causal inference or a program that uses language models to read Brazilian government gazettes.
Outside work I like cities and their transit systems, draw maps for fun, and am still trying to learn to ride a bike with no hands.
When open-source developers on GitHub started receiving sponsorship money, they did less of the unpaid community work their projects rely on. I want to understand when paying for ideas helps and when it backfires. Revise and resubmit, Management Science
Most new inventions come out of a handful of cities. I'm interested in how much of that is down to the people who live there and how much to their being close to each other.
Some of the difference in how inventive people are may start in the brain, and I'm using UK Biobank data to look at it. I'm also a coauthor on a study that compared AI models with human scientists at building theories. The models wrote more elaborate theories, and the people got more out of simpler ones. Under review
With coauthors, I measure what Brazil's local governments do each day and how floods disrupt that work. On my own, I study experts who serve on committees for little or no pay, and find that serving once makes them less likely to serve again.
We followed about 100,000 developers around the launch of GitHub Sponsors. Those who started receiving money went on to do less, and the drop lasted. It came from the community and service work that open-source projects depend on rather than from writing code.
Internet standards bodies and university hiring committees depend on experts who serve for little or no pay, and people who have served tend to come back. Using three institutions that choose members by public lottery, I show that this is selection. The people who come back were more willing to serve from the start, and serving itself makes them less likely to return.
Every Brazilian city publishes an official gazette recording what its government did that day. We are reading nearly all of them, more than 16 million paragraphs so far, with the help of language models, to build the first daily measure of what local governments actually do.