Anders Søgaard

RAVENEA: A Benchmark for Multimodal Retrieval-Augmented Visual Culture Understanding

Lost at the Beginning of Reasoning

Mechanistic Interpretability on (multi-task) Irreducible Integer Identifiers

Towards a Progress Bar for Reasoning: Progress Prediction in Large Reasoning Models

Revisiting the Othello World Model Hypothesis

Toward a Sheaf-Theoretic Understanding of Compositionality in Large Language Models