Google's NotebookLM uses RAG to fact-check prompts on limited data sources and can generate a podcast with two AI hosts on the resulting summary. The example audio in this post is honestly crazy - I can't easily discern that it's AI apart from the low fidelity and some of the intonation is a bit weird.

https://blog.google/technology/ai/notebooklm-audio-overviews/

I uploaded my dissertation from 2016 "Is personal data still relevant after death?" to NotebookLM.

The model clearly has been trained on more recent texts as it references deepfakes, which I didn’t in the paper. So it can draw some relevant topics into the response/conversation.

I find it fascinating that the conclusion for the “listeners” was actually the subtext of the essay.
I danced around it but never actually landed on that takeaway.