Open sourcing smol-podcaster

A big part of our Latent Space Podcast (opens in a new tab) 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 (opens in a new tab). At a high level:

• Uses whisper to transcribe the audio, as well as doing speaker recognition, labeling speakers and adding timestamps to each segment.
• Formats it in Markdown so that it's ready to copy into Substack.
• Uses Claude-100k to read through the transcript and create chapters with timestamps for all the topics mentioned in it.
• Uses both GPT-3.5-16k and Claude-100k to generate title ideas as well as tweet ideas.

It's been saving me a lot of time, and hopefully it helps other creators out there!