Michael Hahn

Automata Learning and Identification of the Support of Language Models

Understanding the Emergence of Seemingly Useless Features in Next-Token Predictors

Softmax Transformers are Turing-Complete

Benefits and Limitations of Communication in Multi-Agent Reasoning

Decomposing Representation Space into Interpretable Subspaces with Unsupervised Learning

The Bayesian Origin of the Probability Weighting Function in Human Representation of Probabilities

SafeReview: Building a Robust Deep Review Assistant Against Prompt Injection

Contextualize-then-Aggregate: Circuits for In-Context Learning in Gemma-2 2B

Born a Transformer -- Always a Transformer? On the Effect of Pretraining on Architectural Abilities

A Formal Framework for Understanding Length Generalization in Transformers

Lower Bounds for Chain-of-Thought Reasoning in Hard-Attention Transformers