AI Efficacy and Data Quality

While I was using the Wayback Machine to “bring back” my blog posts over the last 25 years, I backed in to a personal example of how quality data drives a higher degree of efficacy from AI. I’ll explain.

In this iteration of jimkirk.org, I’m opting for absolute transparency. To use this corner of the internet for my resume, my professional portfolio, and my blog posts (old and new). After spending a few evenings updating my resume, anonymizing aspects of my portfolio, and migrating old blog posts… I sat back and realized how much data this actually was.

… and then I had an idea.

What cool outputs could I get from AI using this optimized dataset about me? It didn’t take me long to find out.

I hopped over to Google’s NotebookLM, created a new notebook, and added my website as a data source. Within seconds, it opened a chat with the following heading:

Then I clicked the “Audio Overview” button to the right and it quickly generated an AI podcast featuring two hosts.

With zero prompt engineering, it spit out this 18-minute podcast which is effectively the story of my adult life. It captures the essence of my higher education journey, the story of my career, and even weaves in more personal aspects from my blog posts. It even contextualizes aspects of my career by pulling in details that aren’t discretely included in the dataset I gave it. An example of this is when it covers my transition to IT Security (3:55 mark) and makes reference to the fact that 2004 was the year of the computer worm. Indeed. Lived (and learned) through it.

I shared this with my folks as a working example of AI in action and my Father rightly commented that there wasn’t any reference to our children. Why? There’s not a lot of data. When the kids were young, I was too busy and tired to be writing blog posts and that’s a rather large hole in the dataset.

It’s really good. Not perfect, but if I was giving it a letter grade, it would be an A.

Standby for more AI adventures. I have another idea in the backlog.