Yewon Han
Titles as the sources gave them. In 1 entry, with 0 checked quotations.
In the Base
Every entry they appear in.
Safety for Whom? Benchmarking Understanding Beyond Refusal in Egocentric Multimodal Assistants
According to the description, a preprint introduces STOU, a benchmark of 2,902 first-person video queries for assistants that see what the user sees, grouped by whether harm would fall on the user, people nearby or society. Across 10 models, the authors report refusal varied with who would be harmed and models often recognized a harm yet still helped.