A big part of our Latent Space Podcast audience is people who actually don't listen to the podcast, but only read the writeup + transcript. We've made it a focus to always have high quality transcription for each episode, as well as "chapters" to skim to based on timestamps.
A lot of that work has been automated using large language models. I've open sourced a lot of it at fanahova/smol-podcaster. At a high level:
It's been saving me a lot of time, and hopefully it helps other creators out there!