Amir Zamir

Modus achieves competitive performance across diverse benchmarks by treating all modalities symmetrically, eliminating the need for modality-specific heads or pipelines.

Achieving competitive performance with multimodal models trained solely on test environment data challenges the necessity of large-scale internet datasets for effective learning.

Synthesized videos reveal systematic differences in how the brain's visual pathways respond to dynamic stimuli, surpassing traditional methods.