Video-Index 1.1 arrives with claims of higher accuracy using far less context
A post argues that letting the model decide what to watch beats feeding it 512 fixed frames.
TLDR
A post celebrating Video-Index 1.1 claims agentic video understanding reaches higher accuracy with far less context. Its author favors letting the model decide what to watch over feeding it 512 fixed frames and asks which vision tasks are still hard for agents.
Combined views
303
2 Sources, first seen ago
