معرفی
Graham Neubig is an Associate Professor at the Language Technologies Institute (LTI) within Carnegie Mellon University (CMU). His research focuses on advancing artificial intelligence, particularly in natural language processing (NLP), multimodal reasoning, and large language models (LLMs). He explores topics such as AI safety, generative AI, and human-AI interaction, with an emphasis on practical applications like machine translation and web-agent systems. His work often involves developing frameworks for evaluating AI systems, such as OpenAgentSafety and BehaviorBox, which assess real-world agent performance and model behavior. Neubig's research also delves into improving LLM capabilities through reasoning analysis, hallucination detection (e.g., ZINA), and culturally aware systems (e.g., CAIRe). He has contributed to open-source projects like Pangea (a multilingual LLM) and frameworks such as Cmulab for model deployment. His recent work addresses challenges in agentic tasks, self-improving agents (Skillweaver), and benchmarking across domains like visual reasoning (VisualPuzzles) and software engineering.
Notable achievements include advancing evaluation methodologies for LLMs, developing tools for ethical AI, and creating benchmark suites that test systems under realistic conditions. His lab collaborates on projects like the BrowserGym ecosystem and OpenHands platform, which aim to standardize web-agent research and AI-driven software development. Neubig's contributions span theoretical advancements and practical implementations, bridging the gap between cutting-edge research and real-world applications.
He advises students such as Apurva Gandhi and actively publishes in top venues, addressing topics from instruction-following improvements to the societal impacts of AI. His work frequently emphasizes the importance of transparency, controllability, and cultural awareness in AI systems.
