Scholar
Liam Carroll
Google Scholar ID: aCsN5VoAAAAJ
Independent Researcher
Citations & Impact
Citations
54
H-index
5
i10-index
1
Publications
11
Co-authors
0
Publications
Resume (English only)
Academic Achievements
- Co-authored 'The Developmental Landscape of In-Context Learning'; will soon be publishing other projects studying essential dynamics of linear regression transformer trajectories; turned his Master's thesis into a sequence on LessWrong.
Research Experience
- Currently working as a Researcher in AI Safety, jointly funded by Timaeus and the Gradient Institute, focusing on building tools for AI safety assurance using Singular Learning Theory and Developmental Interpretability (DevInterp); previously independently funded by Lightspeed Grants to do research on DevInterp, co-authoring 'The Developmental Landscape of In-Context Learning'; worked as a Research Assistant with the ARC Centre of Excellence for Automated Decision-Making and Society (ADMS), producing a simulation of a music recommender environment using ChatGPT to model user behavior, called recagent-music.
Education
- Master’s of Mathematics from The University of Melbourne, with a thesis on 'Phase Transitions in Neural Networks' supervised by Dr. Daniel Murfet; also completed a Diploma of Music in Clarinet performance.
Background
- Mathematician, musician, and hiker. These interests feel deeply interconnected to him, bringing joy through a combination of beauty, love, and wildness.
Miscellany
- Lead hiking guide on the 6-day Overland Track with Tasmanian Walking Company in 2022/23, traversing the route over 16 times; spent over 120 days in wilderness in 2022/23; passionate about making Australian classical music more relevant and engaging for its audience, transforming what is played and how it is played, in service of a more connected community and sense of identity.
Co-authors