paper web signal

First Streaming-Video Benchmark With 'Ask' Action Finds Best AI Scores Just 52%

By Published

Summary

Current streaming video agents default to answering or waiting when uncertain; AskStream measures a third option — proactively requesting missing information — and finds the strongest models barely hit half marks, exposing a key gap before real-world deployment.

Originally reported by paper

Read the original article →

Original headline: First Streaming-Video Benchmark With 'Ask' Action Finds Best AI Scores Just 52%