Easiest way to prove it can't handle the full context in reality is to upload a one hour documentary movie with audio and ask it to write timestamps of chapters/critical moments. It can't handle this beyond 10 minutes even remotely reliably.