Yesterday was the day an AI floored me. I used Google’s NotebookLM to review the manuscript of our upcoming book,
The Open Executive
NotebookLM did a great summary and gave excellent advice on improving it, better than ChatGPT. But that’s not what this is about. As an aside, a party trick, NotebookLM will also generate a “NPR-style” radio segment reviewing the document. It is astonishing. Two faultlessly realistic voices have a discussion about the doc.
That first cut was great but missed the key point of the book. It talked – brilliantly – all about Open Leadership as philosophy and missed that the book is primarily about the personal journey of the leader. This is very realistic . Our experience with interviews and reporters is that they often pick up on points 2 and 3 and miss point 1 entirely. So I had a little chat with the LLM and pointed out its error, which it said was good point. Then I regenerated the “radio” and the result was even better.
It mangled the Vietnamese names and got the gender of one of our clients wrong – again, very realistic. But otherwise the errors are minor. It’s awesome. Most of all, it didn’t just regurgitate the words of the book. The two announcers use their own words, their own metaphors, and their own opinions. It’s unbelievably real. And the two radio segments are quite different: different structure, approach, ideas, and reasoning. It’s not working to a formula or template. It created these.
Here’s another one