Callum McDougall
Principal Investigator
Callum McDougall is a research scientist in Google DeepMind's interpretability team. Previously, he worked at Anthropic doing interpretability research.
Callum founded the ARENA program and ran the first three iterations, acting as head TA and curriculum designer. He created materials for people studying the program virtually, including a monthly series of mechanistic interpretability challenges that have helped train hundreds of researchers.
He was accepted into Neel Nanda's Mechanistic Interpretability SERI MATS stream with Arthur Conmy, studying negative attention heads in GPT2-Small. His research found a GPT-2 attention head that "copy suppresses"—responding to naive copying from earlier layers and suppressing it.
At Google DeepMind, he led the release of Gemma Scope 2: SAEs and transcoders on every layer of every Gemma 3 model (270M-27B, base and chat), enabling deep dives into complex model behavior for open source safety and interpretability work.
Callum was a fourth-year Cambridge maths undergraduate who interned at SIG and Jane Street before transitioning to AI safety research. He is also a Guinness World Record holder for the largest number of clothes pegs held in one hand.
Begin Your Journey
The application takes 10 minutes and is reviewed on a rolling basis. We look for strong technical signal—projects, coursework, or competition results—and a genuine curiosity to do real research.
If admitted, you will join a structured pipeline with direct mentorship to take your work from ideation to top conference submission at venues like NeurIPS, ACL, and EMNLP.
