I'm interested in developing
human-centered AI systems.
My primary goal is to understanding and harnessing the capabilities and risks of LLMs, and better use them for the benefit of human, society, and experts.
My group's current research is focused on building human-centric NLP systems by learning from human perception, societal values, and interaction with experts:
-
Cognitive Scaffolding LLMs: Collect different types of human cognition signals from explanations, eye-tracking, feedback, or instructions, and
develop cognitively-inspired AI models that are capable of planning and complex reasoning.
-
Pushing toward the expert-level AI: Develop interactive systems
to facilitate productive and creative collaboration with human experts like scientists, lawyers, and teachers, and
create complex, compositional, and domain-specific expert-level benchmarks
-
Societal Alignment with Pluralistic People Values: Develop inclusive NLP systems that align with diverse and subjective perspectives, and develop pluralistic
representation, alignment models, and evaluation of AI systems.
Writing
(AI or/and Human) Writing, Discourse, Planning
Alignment
Societal Alignment, Diversity Modeling, Personalization
Interaction
Interactive and Collaborative AI systems
Cognition
Perception, Interpretability, Styles
Others