Anya Sims

Hi! I’m a final year PhD student at Oxford, supervised by Jakob Foerster as part of FLAIR, and Yee Whye Teh as part of OXCSML.

My research focuses on reinforcement learning, LLMs, and the intersection of the two. I am particularly excited about ‘closing the loop’ to enable recursive self-improvement in LLMs, and I am currently exploring adversarial approaches to this. More broadly, my interests include meta-learning, open-endedness, and improving understanding of what transformers learn and why, through the lens of optimization dynamics.

Before coming to Oxford, I completed an integrated Masters in Computer and Information Engineering at the University of Cambridge, where I did my dissertation on language models for automatic speech recognition supervised by Phil Woodland, and also worked with José Miguel Hernández-Lobato in the Cambridge Machine Learning Group (MLG) on causal representation learning.