My research interests bridge cognitive science and artificial intelligence, often with a focus on learning, generalization, and representation. My work on these topics spans various subdomains including language and vision models, reasoning and explanation, agents, and memory.

Some representative lines of work include:

• Analyzing the broad spectrum of contextual capabilities in language models, including understanding how in-context learning emerges from simple data distributional properties, and how language models generalize better from information learned in context than via finetuning, and why this may mean that episodic retrieval helps to reuse learning experiences more flexibly.
• Examining how training data, architectures, and objectives shape model representations, and how these representations are biased by irrelevant properties, posing challenges for interpretability and neuroscience.
• Reviewing the interdisciplinary literatures on representational alignment, and exploring how aligning vision model representations with human semantic knowledge can make them both more robust for ML applications, and better cognitive models.
• Studying how explanations can support learning in RL agents and language models, and how explanations and interventional structure in internet data could allow language models to learn about causality from passive data.
• Comparing language model capabilties to those of humans, including in processing of recursive grammar structures (with lessons on how to perform careful comparisons), and how language models, like humans, show content effects on logical reasoning.
• Exploring how richer environments can improve compositional generalization of grounded language agents, and attempting to scale grounded language agents across many virtual environments.
• I also consider broader issues such as how to build more generalizable cognitive models and theories, how we should think about symbols in AI and how publishing fast and slow could improve the generalizability of research.

I am a Member of Technical Staff at Anthropic. Previously, I was a Staff Research Scientist at DeepMind. Before that, I completed my PhD in Cognitive Psychology at Stanford University. Prior to that, my background is in mathematics, physics, and machine learning. In my spare time, I enjoy climbing.

For more information check out my publications, media & talks, and CV!