Google Wants
The interesting part here isn't another video model — it's the framing of timing itself as the core problem. An assistant that interrupts at the wrong moment is worse than silence, so benchmarking exactly that is the right call, and the result is sobering.… more
The finding that compounding errors wreck response quality in longer videos echoes what we see across long-context and agentic tasks: current models hold up for minutes, not hours. Proactive AI helpers in kitchens and workshops remain a research dream, not a near-term product — but knowing when not to speak may turn out to be the real frontier.
Google wants video AI to understand what people need as they work
A Google-led research team built StreamVibe, a video agent designed to watch people cook or assemble furniture and speak up only when help is needed. A benchmark shows that none of the tested models reliably pick the right moment in longer videos, as errors compound and response qua…
Source:
The Decoder
Discussion (0)