[togume]

Playing last weekend I extracted the audio using yt-dlp, ran that through Whisper (found the quality of that better than YT subs/transcript).

However, then I ran into the 2048 token limit for longer videos. Because it doesn’t hold the full context, it wasn’t good enough at summarizing or providing insights.

The solution is to do smaller summaries of 2048 chunks recursively until you have a single one.

This felt and worked… meh.

We’re you able to get around this in some other clever way?