Ethan Perez
Research Scientist at Anthropic
United States
Summary
👤
Senior🎓
Top SchoolEthan Perez is a research scientist with 11 years of experience focused on reducing catastrophic risks from advanced machine learning, currently conducting research at Anthropic and advising NYU’s Alignment Research Group. His PhD work at NYU investigated aligning language models with human preferences, and he has published and interned at top labs including DeepMind and Facebook AI on topics from red-teaming LLMs to unsupervised question decomposition and multimodal reasoning. He contributes to major open-source projects like Hugging Face Transformers, improving training and model-loading robustness for architectures such as Longformer and BART. Ethan blends deep theoretical grounding with practical engineering—having shipped reproducible code, fixed subtle evaluation bugs, and adapted training scripts across model families. Notably, his background spans both foundational ML research and hands-on systems work, giving him a rare fluency across model development, safety, and deployment.
11 years of coding experience
1 year of employment as a software developer
Doctor of Philosophy - PhD Computer Science, Doctor of Philosophy - PhD Computer Science at New York UniversityBachelor's Degree Computer Science, Bachelor's Degree Computer Science at Rice UniversityHigh School, High School at Cistercian
English, Spanish, Tamil
Github Skills (13)
transformers10
pytorch10
machine-learning10
nlp10
python10
natural-language-processing10
pre-trained-model10
hub9
bart9
xnet9
roberta9
bert8
tensorflow8
Programming languages (6)
Github contributions (5)
huggingface/transformers
Aug 2019 - Dec 2020
🤗 Transformers: State-of-the-art Machine Learning for Pytorch, TensorFlow, and JAX.
Role in this project:
ML Engineer
ML EngineerContributions:6 commits, 6 PRs, 24 comments in 1 year 4 months
Contributions summary:Ethan primarily contributed to fixing bugs and improving the functionality of machine learning training scripts and model loading within the Hugging Face Transformers repository. They addressed issues related to unused arguments, incorrect samplers during evaluation, and improper model loading in prediction scenarios. Additionally, they made adjustments to handle specific model types (Longformer, BART) by modifying training scripts, specifically addressing the use of token_type_ids. These changes focused on adapting existing functionalities to new model architectures and ensuring the correct behavior of the models in different training and evaluation scenarios.
pythonbertspeech-recognitionstate-of-the-artflax
inverse-scaling/prize
Jun 2022 - Nov 2022
A prize for finding tasks that cause large language models to show inverse scaling
Contributions:6 commits, 1 PR, 8 pushes in 5 months
nlpprizelarge-language-modelsfindingcause
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.