Liam Dugan
PhD Candidate
Philadelphia, Pennsylvania, United States
Summary
👤
Senior🎓
Top SchoolLiam Dugan is a PhD candidate in Computer Science at the University of Pennsylvania with a decade of experience at the intersection of large language models and human–machine interaction. His research probes how people detect machine-generated text and how those insights can improve automated detection systems, with practical applications for educational technology. He contributed to the high-profile BIG-bench benchmark by developing and refining the "real_or_fake_text" task, splitting it into subtasks and creating easier examples to improve evaluation fidelity. Trained in engineering, robotics, and computer engineering with strong academics and multilingual study in Japanese, he blends rigorous experimental methods with hands-on ML engineering. Based in Philadelphia, Liam combines academic rigor with open-source collaboration, aiming to make LLM behaviors more interpretable and practically useful.
10 years of coding experience
High School, 4.00, High School, 4.00 at St. Joseph's PrepBachelor of Science in Engineering, Computer Engineering & Japanese Studies, 3.63, Bachelor of Science in Engineering, Computer Engineering & Japanese Studies, 3.63 at University of Pennsylvania
Japanese, English, Korean
Github Skills (6)
json10
jupyter-notebook10
python10
machine-learning9
nlp8
tensorflow4
Programming languages (9)
Github contributions (5)
google/BIG-bench
Mar 2021 - Mar 2021
Beyond the Imitation Game collaborative benchmark for measuring and extrapolating the capabilities of language models
Role in this project:
ML Engineer
ML EngineerContributions:13 commits, 2 PRs, 4 comments in 23 days
Contributions summary:Liam primarily focused on developing and refining the "real_or_fake_text" (RoFT) task within the BIG-bench benchmark. Their work involved modifying a Jupyter Notebook to generate JSON data for the task, including adjusting how scores are assigned to model generations. Furthermore, the user contributed to task definition by splitting the task into subtasks and added examples for an easier version of the task based on article-switching. These changes suggest an effort to improve the benchmark's functionality and its evaluation of language models.
bertmachine-learningbenchmarkmeasuringbenchmarks
liamdugan/raid
May 2024 - Mar 2025
RAID is the largest and most challenging benchmark for AI-generated text detection. (ACL 2024)
Contributions:10 releases, 32 PRs, 56 pushes in 10 months
acl2024ai-generatedai-generated-textai-generated-text-detectionbenchmark
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.