Liam Carroll

Liam Carroll is a Researcher in AI Safety based in Melbourne, Australia, jointly affiliated with Timaeus and the Gradient Institute. He holds a Master of Mathematics from the University of Melbourne (2021), where his thesis on Phase Transitions in Neural Networks was supervised by Dr. Daniel Murfet and Dr. Thomas Quella, and a Bachelor of Science in Applied Mathematics alongside a Diploma of Music in Clarinet Performance. Prior to his current role, he was an independent researcher in Developmental Interpretability (2023-2024), funded by a Lightspeed Grant, during which he co-authored papers on the stagewise development of neural networks using Singular Learning Theory. He received a Long-Term Future Fund grant to write a LessWrong sequence called Distilling SLT, which translates the technical foundations of Watanabe's Singular Learning Theory into accessible form for the AI safety community. Carroll also organized the inaugural Australian AI Safety Forum in Sydney in November 2024, and brings a distinctive background as a wilderness hiking guide in Tasmania and a music educator.

No endorsements yet.

Personal Website

https://www.liamcarroll.au/

LinkedIn

LinkedIn

Twitter / X

Twitter

LessWrong

liam-carroll

EA Forum

-