Recently a colleague shared a summary on an article in form of a podcast that had been generated by Google's NotebookLM. This reminded me of a growing trend among bloggers: turning your blog into podcasts. If you haven’t heard of Google's NotebookLM, it’s definitely worth checking out. This tool lets you generate podcast episodes based on your blog or website content free of charge. So let's jump in the bandwagon:
All the results impressed me. Google’s choice for the voices for the speakers felt spot-on: the deep masculine voice of Speaker A and the engaging tone of Speaker B made for a surprisingly enjoyable listening experience. Generating these episodes was straightforward—just add the pages for the AI to consider, click “generate,” and within moments, you have a podcast episode. While the first episode delved into my thoughts on leadership and how I approach the topic, the second episode took a more personal angle, focusing on my travels and adventures. Both offer a fascinating perspective, and here’s the second version:
The technology is here now and there is still room for improvement. As you can hear in the episodes there are quite a few jumps between topics here and there. Most of the topics aren't really connected with a fitting transition and the flow is too often interrupted with the back and forth between the speakers. I believe in the second episode they start at some point with one topic and then transition suddenly to a complete different one. I played the first version to my girlfriend and we both agreed as nice both speakers sound the way they speak and the handle the topics is putting us off. It sounds too superficial which may be the intention of the AI model. In both versions it is also dominant that the AI model believes I had broken my ankle. I had written about the drama my girlfriend and I experienced here. Let me know whether you think I the one with a broken ankle. Another aspect that comes to light and feels like I am able to view behind the curtain is when you generate several episodes. I believe I was able to identify certain building blocks which are often repeated and used by the AI model. This would explain to me how the model is able to generate the episode so quickly. Nevertheless what Google has created there is impressive. I will rest it there.
Reflecting on the technology I can't help but to think about the many times I use large language models (LLM) daily. As a developer I am now coding with co-pilot in my IDE, I am chatting with the LLM about code, ideas and technology. On that note I observe how social networks are now starting to use AI users to fill the conversations. Where is this going to lead, I ask myself. The DEAD INTERNET theory say's that at some point only autonomous accounts will create content in social platforms. Twitter has shown that not even verified accounts system prevail when it is taken over in the interest of a capitalist. If I can't trust another user to be human why stay and consume content on the platform? Furthermore big tech companies buy and restart old nuclear reactors to satisfy their energy needs. This reminds me of the recent CCC in Hamburg which had an interesting talk about the climate cost of the AI hype.
Let's end with my favorite catchphrase from the podcasts: "Keep diving deep"
Here is the talk: Multiple audio tracks are available.