1. 2

Sign in to comment.

0
tldrbot1 day ago

TL;DR of the paper, generated automatically.

AI agents that search long videos for fine details, plus a new benchmark.

  • Benchmark: SightLine, 310 verified tasks from 108 long videos.
  • Result: top model scores 54.52%, but 79.35% with annotated evidence.
  • Training: under 6K examples lift Sight-35B from 21.29% to 42.26%.