A colleague of mine was telling us about Google’s “Notebook LLM” which enables you to upload or point to as specific body of content and then get a summary, chat about the content, or even have the LLM generate a semi-plausible sounding synthetic “podcast” in which two perky morning-show type hosts discuss what you’ve shoveled into them. (They had used it to digest three weeks of missed meetings into a 30-minute “show” one could listen to while doing the dishes.)
My first experiment was to point to the first page of this blog. This produced a semi-hilarious podcast about a blog author who is losing his job, finding community, releasing an album thematically related to collaboration and relying on others, gets a new job, and also blogs about random experiences and thoughts.
So I tried something a little more complicated and gave it all the main content from the website I’ve been readying for the record. Then I had it spin up a new podcast about the record and… it’s not terrible? I mean, some of it is objectively terrible or filled with cringe, but… it also kind of gets at some interested ideas that I’ve definitely attempted to convey about this work.
Before I had to get back to my day job, I experimented with a beta feature where you can interrupt the podcast to ask questions. I introduced myself as the person they were talking about and offered to answer any questions they had.
The personas immediately pivoted and started asking me about my process, what the band name “Layers of Meta” means, and so on. It was not without its hitches but it was a goofy kind of fun, and I could even imagine generating some good material this way just by virtue of creating a context that prompts me to respond and elaborate on things.
We’re not too far from having artifacts derived by such methods from sources themselves derived by such methods, layers of meta all the way down.
Leave a Reply