Edison Marrese-Taylor

Performance gaps in multilingual medical evaluations reveal that proprietary models outperform open-source ones, but translation quality can swing results dramatically.

Evaluations reveal that current models struggle with long-range narrative integration and cultural reasoning, highlighting a critical gap in video understanding capabilities.