Neel Nanda
Senior Research Scientist at Google DeepMind
London, England, United Kingdom
Summary
🤩
Rockstar🎓
Top SchoolNeel Nanda is a Senior Research Scientist at Google DeepMind in London who focuses on mechanistic interpretability and the safety of transformer-based language models. His recent work includes contributions to sparse autoencoders (Gemma Scope and Gated SAEs, NeurIPS 2024) and an ICLR Spotlight paper on progress measures for grokking, reflecting a blend of practical tooling and theoretical insight. He is the creator of TransformerLens, an open-source library used for mechanistic interpretability, and has a track record of mentoring and publishing work that helps onboard others to the field. Formerly at Anthropic and affiliated with FHI and CHAI, he brings seven years of research and engineering experience grounded in a BA in Mathematics from Cambridge. Outside of research he’s active in the Effective Altruism community and has committed to donating 10% of his income to high-impact charities.
8 years of coding experience
3 years of employment as a software developer
BA, Mathematics, BA, Mathematics at University of CambridgeThe Latymer School
Github Skills (6)
transformers10
transformer-models10
pytorch10
machine-learning10
python10
natural-language-processing10
Programming languages (6)
Github contributions (5)
TransformerLensOrg/TransformerLens
Aug 2022 - Jan 2023
A library for mechanistic interpretability of GPT-style language models
Role in this project:
ML Engineer & Data Scientist
ML Engineer & Data ScientistContributions:5 releases, 89 reviews, 285 commits in 5 months
Contributions summary:Neel appears to be adding to and modifying a library designed for mechanistic interpretability of GPT-style language models. Their contributions include implementing and refining the model architecture (Embed, Unembed, PosEmbed, LayerNorm, Attention, MLP, TransformerBlock) and adding support for different models. A basic demo code for the model is added.
neelnanda-io/neelutils
Oct 2022 - Mar 2024
Contributions:13 pushes in 1 year 5 months
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.