creators

Sam Bowman - alignment researcher at Anthropic

Sam Bowman leads alignment and evaluation research at Anthropic, on leave from NYU. He built the SNLI corpus and co-created the GLUE benchmark.

Sam Bowman works on technical AI safety at Anthropic, where he leads a research group on AI alignment and welfare with a particular focus on evaluation. He is on long-term leave from New York University - its Courant directory records him as on leave for 2025-2026 - where he remains officially an Associate Professor of Data Science and Computer Science, and where he led the Alignment Research Group from 2022 to 2024. He earned his PhD in 2016 at the Stanford NLP Group and Stanford Linguistics, supervised by Chris Potts and Chris Manning, for work on early neural network models for text. He joined the NYU faculty in 2016 and was granted tenure and promoted to Associate Professor effective the autumn of 2022, spending the 2022-23 academic year on sabbatical as a visiting researcher at Anthropic. Before his safety-focused work, Bowman was an influential figure in natural language processing benchmark design: he was first author of the Stanford Natural Language Inference (SNLI) corpus of 570,000 human-annotated sentence pairs, and he co-created GLUE, a widely used multi-task benchmark for evaluating language understanding models. His more recent research addresses how to evaluate and oversee AI systems that may exceed human capabilities, including scalable oversight via AI debate, for which he co-authored the ICML 2024 Best Paper Award-winning study on debate with more persuasive language models.

Specialization

AI alignment, model evaluation, scalable oversight, NLP benchmarks

Country

United States