About Me

I am a research scientist and deputy head of the Institute for Human-Centered AI at Helmholtz Munich.
My research employs computational models to uncover the fundamental principles behind human cognition. I believe that for a more complete understanding of human cognition, we must consider the human mind as a whole. My current goal is therefore to establish foundation models of human cognition – models that cannot only simulate, predict, and explain human behavior in a single domain but those that offer a unified take on our mind. To accomplish this, I use tools such as neural networks, Bayesian inference, meta-learning, information theory, and large language models.

News

• Check out the Centaur webpage.
• Evaluating alignment between humans and neural network representations in image-based learning tasks to be presented at NeurIPS.

• New preprint: Centaur: a foundation model of human cognition.
• How should the advent of large language models affect the practice of science? provisionally accepted at PNAS.
• New preprint: Sparse Autoencoders Reveal Temporal Difference Learning in Large Language Models.

• The commentaries and our response for our BBS paper are now available.

• A tutorial on open-source large language models for behavioral science now accepted at Behavior Research Methods.

• We are organizing a workshop on In-context learning in natural and artificial intelligence at CogSci 2024 (Rotterdam).

• Three papers accepted at ICML [1, 2, 3].
• I will be at ICLR in Vienna presenting our work on Turning large language models into cognitive models.

• New preprint: CogBench: a large language model walks into a psychology lab.

• Two new preprints: Ecologically rational meta-learned inference explains human category learning and In-context learning agents are asymmetric belief updaters.

• Turning large language models into cognitive models got accepted at ICLR with ratings of 8, 8, 8, 8.

• I moved to Munich.
• Two new preprints: How should the advent of large language models affect the practice of science? and A tutorial on open-source large language models for behavioral science.
• Two papers presented at NeurIPS: Reinforcement Learning with Simple Sequence Priors and Meta-in-context learning in large language models.

• Meta-Learned Models of Cognition got accepted at BBS.

Key Publications