Reasoning
Enhancing language agents' reasoning capabilities to support complex decision-making processes and developing augmentation tools that enhance precision, creativity, and productivity.
Research
Research atlas.
We build language intelligence that remains reliable, efficient, and human-compatible.
Mission
Language agents for humanity.
We advance sustainable, trustworthy language agents. Our work spans responsible reasoning, efficient language modeling, scientific representations, and domain adaptation.
Findings of EMNLP 2024Zero-shot Commonsense Reasoning over Machine Imagination
Multimodality
Creating sophisticated language agents capable of integrating and understanding diverse modalities such as text, vision, speech, and numerical data to facilitate richer, more natural interactions.
Alignment
Ensuring language agents align with human intentions and values by developing robust alignment methodologies. Our research in alignment addresses safety, unintended behavior, and trustworthy interactions.
Efficiency
Developing computationally efficient methods for training and deploying language models, aiming to reduce environmental impact, minimize resource usage, and democratize access to advanced AI capabilities.
AI for Science
Adapting language agents to the languages of science. We develop domain-aware data construction, training, and knowledge transfer, accelerating scientific discovery.
Principles
How we work
Scientifically grounded
We connect model design with the structure and knowledge of each domain we study.
Human-centered
We evaluate language systems by how well they support people, not only by benchmark scores.
Open to alternatives
We look beyond false binaries and ask what better options a reasoning system can surface.
Projects
Research cooperation
Organizations and research communities represented across LILAB's cooperative work.
