Cornell University Team Wins $50K in AI Puzzle-Solving Challenge

By Daniel January 27, 2025
Alarm 2 Min Read
Copied
Featured

A team from Cornell University has clinched first place in the 2024 ARC Prize competition, winning $50,000 for their innovative AI models. The team, led by Kevin Ellis, Assistant Professor of Computer Science at the Cornell Ann S. Bowers College of Computing and Information Science, submitted a paper titled “Combining Induction and Transduction for Abstract Reasoning.” Their AI models were able to solve 56% of problems, bringing them close to the average human performance of 60%.

The winning team, including lead authors Keya Hu, a visiting undergraduate, and Wen-Ding Li, a doctoral student, worked on the ARC-AGI set, which involves solving complex puzzles presented in coloured grids. The challenge requires participants to detect patterns in examples and apply them to solve the puzzles. The Cornell team’s model combines two approaches—one based on slow, logical thinking, and the other on fast, intuitive thinking—resulting in enhanced accuracy.

Li described the victory as a “milestone,” particularly as the competition has endured for five years, with numerous high-performing AI models presented. Hu added that combining the two distinct thinking approaches proved crucial to their success, as it significantly boosted performance.

The research was supported by several other members of the team, including Cornell undergraduates Carter Larsen, Yuqing Wu, Caleb Woo, and Spencer Dunn, along with doctoral students Simon Alford and Hao Tang.

Cornell’s achievement highlights the university’s growing prominence in AI research and its commitment to pushing the boundaries of machine learning. The $50,000 prize will be shared among the lead authors, marking a significant milestone in their academic and research careers.

Source: SHIKSHA

More Articles