Understand this story
Google, together with institutions such as Stanford and Oxford, has proposed a video agent called StreamVibe, which can understand vague requests like "help me when I run into trouble" and decide on its own when to speak up. Its StreamVibe-VL model is based on Qwen3.5-9B, improving response timing and quality by 10.7% over comparable models, while its memory system further boosts proactive interaction by 12.3%. However, tests show that when the model processes an entire video continuously and needs to connect with its own previous answers, key content metrics drop by as much as 61.5%, and in videos longer than 5 minutes it struggles to pinpoint the right moment to intervene.
AI-generated from public reports
Developments
1 confirmed developments
- Source:The Decoder
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 1
The Decoder
Google, together with Stanford and Oxford, launches the StreamVibe video agent, teaching AI to judge when to proactively speak up and help
Google, together with Stanford, Oxford and other institutions, has launched the StreamVibe video agent, which can understand vague requests such as "help me when I run into trouble" and decide on its own when to speak. Its StreamVibe-VL model is based on Qwen3.5-9B, improving response timing and quality by 10.7% over comparable models, while its memory system further boosts proactive interaction by 12.3%. However, tests show that when the model processes an entire video continuously and needs to connect with its own previous answers, key content metrics drop by as much as 61.5%, and in videos longer than 5 minutes it is difficult to pinpoint the right moment to intervene.