Karan Goyal

Contextual entrainment in vision-language models reveals a duality that could fundamentally alter our understanding of multimodal interactions.

VLMs are often functionally blind, exploiting language priors instead of truly "seeing" visual data, and this problem paradoxically *worsens* as language models scale.

Existing citation recommendation benchmarks overestimate real-world performance because they fail to account for the temporal constraints of recommending citations for *new* papers.