The reading stack is piling up faster than I can process it for GonzoML, so I decided to experiment with NotebookLM and automated summaries for some of the material. It’s better than leaving things untouched.
For these summaries, I’ve launched a YouTube channel with AI-generated podcast-style reviews: https://www.youtube.com/@GonzoMLPodcasts/videos
The approach is fun but far from perfect. Sometimes, the summaries drift into broad discussions that miss the technical depth we love. Other times, they add “creative” details that weren’t in the original.
For example, I generated the review of the article "Gödel Agent: A Self-Referential Agent Framework for Recursive Self-Improvement" four times, trying to capture a focused breakdown with a slight nod to Schmidhuber’s original Gödel Machine paper. Each iteration veered off-track in its own way—sometimes wandering into unrelated topics like RNNs (however, it generated an interesting podcast on its own, maybe I will share it later), or speculating about probabilistic logic that wasn’t in the original paper. Eventually, I picked a version with the best balance of accuracy and imagination. Here it is:
You can read my handmade review of the paper here and compare:
Lesson learned: take these with a pinch of salt. They’re entertaining, and personally, I enjoy listening to them while out for a walk. The warm, in-depth hand-made reviews will still be posted on GonzoML. Meanwhile, the YouTube channel will feature automated podcasts with the same carefully chosen topics.
Looking forward to seeing where this experiment leads!