William Rudman

vision-language model (2) embedding analysis (2) cosine similarity (2) visual reasoning (1) vector space (1) geometric reasoning (1) hidden state (1) multimodal large language model (1) mechanistic interpretability (1) attention head (1) chain-of-thought prompting (1) multimodal reasoning (1) cross-attention mechanism (1) transformer decoder (1) knowledge prior (1) word embedding (1) activation steering (1) model prediction (1) surprisal analysis (1) syntactic ambiguity (1)