Sachin's Profile Photo
Sachin Kumar
Postdoctal Researcher
Allen Institute for AI
Vidhisha's Profile Photo
Vidhisha Balachandran
PhD Student
Carnegie Mellon University
Antonios's Profile Photo
Antonios Anastasopolous
Assistant Professor
George Mason University
Lucille's Profile Photo
Lucille Njoo
PhD Student
University of Washington


As with all language, text generated by language models can be harmful, or used to bring about harm. Automating language generation with large language models adds both an element of scale and also more subtle or emergent undesirable tendencies to the generated text. With an ever increasing number of user-facing applications being built and deployed on top of large language models, there are risks of societal harms that such applications can cause. It is of utmost importance for NLP practitioners to be aware of these risks as well as ways to mitigate them.

Our primary focus will be on how to systematically identify risks, and how to eliminate them at various stages of NLP pipeline, from data collection, model development, model adaptation, inference, to application deployment. Through this tutorial, we aim to equip NLP researchers and practitioners with a suite of practical tools for mitigating safety risks from language generation models, while highlighting current challenges and future research directions.

Tutorial Slides