REPRESENTATIONS / MECHANISMS / EVIDENCE
Understanding how models represent information.
Investigating how category structure is represented across model layers.
This work asks how category structure changes through a language model, and how those patterns can be compared across layers.
Paper recordStudying state and motion in video-model representations.
This work examines state and motion in video representations, using causal interventions to investigate what a model encodes.
Paper recordExploring how document structure appears inside OCR models.
This work examines the representation of document structure in OCR models, with attention to how that information is organized.
Paper record