Hi! I’m a research scientist at the UK AI Security Institute (AISI). I work on post-training, interpreting, and evaluating models for various forms of misalignment, such as reward hacking, eval awareness, and sandbagging. Here’s why:

I believe AGI can lead to extraordinary well-being if models are capable, efficient, and aligned (which is what I work on!). By understanding how AI models learn and how they think, we can make them safer, better-aligned, and more useful for the future of life. This informs my research focus.

Previously, I worked on RL for efficient multi-turn exploration at CHAI at UC Berkeley. I was also a scholar at MATS Research (twice), where I worked on deception, feature geometry, and mechanistic interpretability.

Before moving full-time to AI safety, I worked at Microsoft Research on language models. Prior to that, I was an associate research scientist at Wadhwani AI working on AI for Social Good and Healthcare.

Writing fiction and poetry along the way!