Video (360p re-encode)
Annotated transcript —
Each case is one egocentric cooking video in which the cook deliberately makes mistakes
(wrong quantity, wrong tool, skipped step, ...). Expert annotations specify what an ideal streaming
assistant should say and exactly when: step Instructions, Success confirmations,
corrective Feedback at the moment a mistake occurs, plus clarifications and user turns.
The benchmark scores whether a video LLM intervenes at the right time with the right feedback.
Fields per utterance: index, text_type, text_timestamp (seconds), text.
0:00
| # | Time | Speaker | Type | Utterance |
|---|