Sequoia, OpenAI, and Citadel Host Pokerbot Tournament at Harvard

Students from Harvard, MIT, and other Boston-area schools competed in a pokerbot building competition held Monday at Harvard’s David Rubenstein Treehouse.

The event, hosted by OpenAI, venture capital firm Sequoia Capital, and trading firm Citadel Securities, was the Boston regional of No Limit, a national poker bot-building tournament.

The top five teams at each of three regionals, including the Boston event, will advance to a Nov. 15 final in San Francisco to compete for a $100,000 grand prize

Abhishek A. Malani ’22, the Sequoia partner who proposed the tournament, acknowledged that a poker competition is an unusual project for a venture capital firm.

“It’s not like we’re recruiting,” Malani said. “This is truly about helping people and reaching this generation of founders.”

The event reflects a shared concern between the three hosts, Malani said that artificial intelligence models can write most code, what separates top performers at companies like OpenAI and Citadel is creativity and discernment about which problems to work on.

“Creativity and research direction are the core bottlenecks to the top one percent performers at a lot of these organizations,” Malani said, “Is there a way that we can create a competition that’ll be really fun for students to compete in that helps foster this?”

Poker has long been a recruiting tool for quantitative trading firms, which value skills like estimating probabilities and making decisions with limited information. Firms like Jane Street, Citadel, and Hudson River Trading sponsor Harvard’s student poker club.

In the No Limit tournament, students had 60 minutes to build a bot that could play a version of poker none of them had seen before. After the build period, the bots played 10,000 simulated games against one another.

Malani said the format was meant to test what students choose to build in a world where AI tools can handle much of the coding. The tournament drew more than 10,000 applications, he said, with more than 330 students attending.

“We have shifted from ‘can I build this’ to ‘what should I build,’” he said.

Malani said the inspiration for No Limit was class tournaments in COMPSCI 136: “Economics and Computation,” a course on game theory and computation taught by School of Engineering and Applied Sciences Dean David Parkes. The course’s approach to strategic game design, Malani said, is a “really interesting way to express learning, especially in the post-ChatGPT era.”

Juliana Li ’29, a Harvard student who competed in the Boston event, said she and fellow competitors had shared interests in startups, quantitative finance, and AI research. To ideate the best strategy before the event, Li directed AI agents to research recent articles, blog posts, and social media posts on building poker bots. She said the top teams she spoke with had done most of their research ahead of time and studied why their bots lost.

But Li said that since the format allows competitors to score well without much knowledge of poker, strategy becomes more important.

“The problem with AI and AI coding in general these days is you don’t really have to understand what you’re doing. You can do pretty well with no idea what the poker variant even is or how to play poker,” Li said.

The pace of AI development means that no one has extensive experience building with the newest models, Malani said, advantaging students who are not tied to old assumptions about what technology can do.

“It’s a core belief of Sequoia that the world has accelerating change, and so we need to constantly be adapting to it,” Malani said.