Yeah I mean if an information source is so low signal-to-noise ratio I need to have it summarized, there are probably better sources. And if it's actually high quality, the summarization is lossy.
There's no point to doing this.
The point is figuring out if it is information dense or not. You don't know until you watch it. An LLM can process the entire transcript much faster than you can.
I can tell if a youtube video is worth watching in a few seconds of skipping around and listening to random parts.
Less time, lot less resource intensive than having a whole LLM infer text from it so I can think about it.