Charlie George

Charlie George

Machine Learning Engineer at Elicit formerly studying maths at Warwick. I am interested in AI safety, particularly in ideas around scalable oversight for smarter-than-human systems.

Contact: [email protected]

Blog

Training Strong Models to Verify IMO Solutions Using Unstructured Internet Data with Tinker

October 2025

Training models to accurately grade International Mathematical Olympiad problems using only unstructured internet data and reinforcement learning.

Publications

Factored Verification: Detecting and Reducing Hallucination in Summaries of Academic Papers (2023)

Charlie George, Andreas Stuhlmüller

Published at the Second Workshop on Information Extraction from Scientific Publications (WIESP) at IJCNLP-AACL 2023

Iterated Decomposition: Improving Science Q&A by Supervising Reasoning Processes (2023)

Justin Reppert, Ben Rachbach, Charlie George, Luke Stebbing, Jungwon Byun, Maggie Appleton, Andreas Stuhlmüller

Ought Technical Report. arXiv:2301.01751 [cs.CL]