Introducing the Crosby Intelligence Fellows

Teddi Worledge is a final-year Computer Science Ph.D. candidate at Stanford University, where she is advised by Carlos Guestrin. Her research focuses on making AI systems more reliable, trustworthy, and useful by developing methods for understanding and verifying large language model outputs, including citations, data attribution, and explanations. She previously earned a B.S. in Computer Science from the University of California, Berkeley and worked as a member of technical staff at Crosby.

During her fellowship, Teddi will work with the Crosby team and Neel Guha of Columbia Law School to design and collect data for RedlineBench2, the next version of Crosby’s RedlineBench that uses preference data to measure how well AI systems redline and negotiate contracts. Alongside the benchmark, she will explore personalized verifiers that assist users as they collaborate with AI, and judgment profiles built from interpretable features, which capture how an individual attorney tends to make legal judgments in terms a person can read and check. The goal is to make it easier to understand and verify what AI systems produce in legal work.

Irena Saracay is a Computer Science Ph.D. candidate at Stanford University, where she is advised by Carlos Guestrin. Her research focuses on building AI systems that adapt to the goals and interaction needs of the human working with them. She previously earned an M.S. in Statistics and a B.S. in Computer Science from Stanford University.

During her fellowship, Irena will build review assistants that learn and adapt to the specific priorities of a human reviewer. People review AI outputs with different priorities: one attorney may prioritize legal protections while another prioritizes closing the deal, and one software engineer may scrutinize code architecture while another focuses on backward compatibility. Her assistants will learn each user's review priorities, as well as how each reviewer prefers to engage with an output, whether through summaries, excerpts, or design alternatives. The goal is to let reviewers work faster, catch more of the issues they care about, and carry less cognitive load during review.

Kylie Zhang is a Computer Science Ph.D. student at Princeton University, where she is advised by Peter Henderson. Her research focuses on how laws and legal doctrine can help coordinate multi-agent AI systems. She previously earned an M.S.E. in Computer Science from Princeton University and a B.S. in Computer Science from the University of Illinois Urbana-Champaign.

During her fellowship, Kylie will explore how AI makes strategic decisions in complex legal settings, with a focus on multi-issue negotiation and litigation. These settings require balancing competing priorities, working with incomplete information, and adapting as circumstances change. Using realistic legal scenarios and expert feedback, she will examine how AI systems can learn from the natural language decisions that shape a negotiation or case, such as when to seek information, make a tradeoff, or change course, and how those choices add up to longer-term objectives. The goal is more adaptable AI models that help lawyers evaluate the best strategies under incomplete information.