Sean Welleck

Sean Welleck is an Assistant Professor at Carnegie Mellon University's School of Computer Science, specifically within the Language Technologies Institute (LTI). He leads the L3 Lab and serves as an advisor for the AI for Math Fund. His academic journey includes a PhD from New York University under Kyunghyun Cho and postdoctoral positions at the Allen Institute for Artificial Intelligence and the University of Washington with Yejin Choi.

Dr. Welleck's educational background shows a strong foundation in computer science. He earned his PhD in Computer Science from New York University, where he worked under the mentorship of Kyunghyun Cho and Zheng Zhang. Prior to this, he completed his MSE and BSE in Computer Science from the University of Pennsylvania, demonstrating a long-standing commitment to the field.

Dr. Welleck's research focuses on bridging informal and formal reasoning with AI, with particular emphasis on developing learning, inference, and evaluation algorithms for large language models. His work spans multiple cutting-edge areas including mathematical reasoning, code generation, inference algorithms, and AI reasoning agents. A significant portion of his recent work involves combining AI with formal methods for mathematics, where he has developed frameworks like Llemma (an open-source language model for mathematical reasoning) and meta-generation (for inference-time algorithms). His research is characterized by a strong theoretical foundation coupled with practical applications that push the boundaries of what AI systems can achieve in formal reasoning domains.

Analysis of Dr. Welleck's recent publications reveals a clear research trajectory focused on enhancing language models' capabilities in formal reasoning and mathematical problem-solving. His work demonstrates an evolution from foundational research in neural text generation to increasingly sophisticated approaches that integrate formal methods with deep learning. Key trends include the development of inference-time algorithms that improve model performance without additional training, frameworks for mathematical reasoning that connect informal and formal proofs, and novel evaluation methodologies for language models. His publications consistently appear in top-tier conferences including NeurIPS, ICLR, ICML, and ACL, reflecting the high impact of his contributions to the field.

Dr. Welleck's scientific achievements have been recognized with several prestigious awards:

• NAACL 2025 Best Paper Award
• ICLR 2025 Oral Presentation (Top 2%)
• ICLR 2025 Spotlight Presentation (Top 5%)
• NeurIPS 2021 Outstanding Paper Award (Top 0.1%) for MAUVE
• NVIDIA AI Labs Pioneering Research Award (2017 and 2018)

As an educator and mentor, Dr. Welleck actively guides the next generation of AI researchers. He currently advises multiple PhD students including Pranjal Aggarwal, Weihua Du, Andre He, and Seungone Kim (some co-advised with other faculty), along with MS students Riyaz Ahuja, Jiewen Hu, Qinyue Tan, and Thomas Zhu, and undergraduate Tate Rowney. At CMU, he teaches advanced courses such as Neural Code Generation and Advanced NLP, and has previously taught at New York University and the University of Washington. His commitment to education extends to creating resources like the Thesis Review Podcast and developing tutorials on neural theorem proving that have been presented at major conferences.

Dr. Welleck leads the L3 Lab at CMU, which focuses on the intersection of language, learning, and logic. The lab brings together students and researchers to tackle challenging problems in AI reasoning, with particular emphasis on mathematical reasoning and code generation. Recent initiatives include the development of Llemma, an open-source language model specialized for mathematical reasoning, and work on inference-time algorithms that enable language models to improve their performance through additional computation during inference rather than through additional training.