Marcus and Austin's post about the cost-effectiveness of AI safety YouTubers was very important. I come from a background in making videos for emergency skills training, where educational effectiveness is rigorously tested, so I’m delighted to see this issue tested here.

Aditya, one thought about the experiment design: viewing videos as part of a structured watch party, where participants expect to be in discussions afterwards, creates a very different context from typical passive YouTube viewing. Perhaps you could test solo viewers also? But I’m sure you’re taking this into account already. Good luck!