My life changed dramatically a few days ago. Although I found myself without employment for the first time in over 30 years, that wasn’t the dramatic change. The change was that I fell in love again with AI. Like a smitten teenager, I understood in a flash that this technology that I had already fallen in and out of love with was, in fact, the one and only for me.

A week ago, I was an AI skeptic. Oh, I still used it regularly, but it was just not reliable enough and there were too many potential legal issues to use it for work. I likened it to a calculator that worked 100 times as fast but was only 90% accurate; to use it for anything important was out of the question. I got to the point where I wasn’t even sure of the quality of AI generated recipes it created for me!

On the first day after parting ways with my old company, I was having lunch with a good friend who came over to welcome me to my new life. She is an evangelist for the use of psychedelic mushrooms in therapy, having seen tremendous benefits of them herself. Like many influencers, for maximum reach she has content in many places: a book, various workbooks, YouTube videos, TikTok videos, posts on Instagram and X, as well as a website.

Her frustration is that she was always getting asked the same questions. She created an FAQ, which helped somewhat, but when she would respond that she had made a video about a topic, people would inevitably ask which video. Even if she remembered which one and where it was, they often would say they didn’t want to watch the entire video (often all of 5 minutes long!).

On the way back from lunch, I offered to create a product for her. I would use AI to create a chatbot that would gather all of her various creative content, answer questions about it and give links back to the source, including the sentence of the book or blog, or the video cued up  to the moment she speaks about the topic.

I figured this would be a good product to tinker with for a month or so while I decided what my next adventure would be. While I knew that the one place Generative AI was really succeeding was ChatBots, I was not so sure about the viability of my little project. But I knew I’d have fun trying!

Oh, I knew the basics of how it would work: write a Python program to get the data feeds from all of these sources, connect them to a Large Language Model (LLM) and use a Retrieval-Augmented Generation (RAG) which would parse the data, rank order examples according to the string it was looking for, return the top three examples, correlate it with the timestamp of the video and return a URL taking you directly there. Yes, I could clearly see how it’d work, but each piece is tricky.

My first problem was which LLM to choose. ChatGPT is the most well-known, but there are many, including open-source models. To start, I decided to go with what I know best, ChatGPT. I had previously played around with ChatGPT’s MyGPT creator with mixed results.  

Almost immediately I started hitting roadblocks. How do I get video information from YouTube? How do I make sure the GPT doesn’t go off the rails and start hallucinating or answering questions outside the topic?

I’ve been a software developer for 30 years (although, over the years less and less developer and more and more strategist and, well… bureaucrat…) and the answer is the thing that developers hate most– read the documentation! Go to each website, find the API documentation, and figure out where to look to answer our question.

Instead, I decided to ask ChatGPT to answer these questions. To my amazement, it did. And that’s when I fell in love again.

Don’t get me wrong, it didn’t answer every question correctly and I still had to go digging through some API documentation, but it did a yeoman’s job. As good… no… better.. than having another competent coder by your side helping you. Like any coder, it didn’t know all the answers and every piece of code had to be checked and rechecked for quality and security.

But within one day it was up and working! Up, working and super cool. All the wisdom from an influencer at my fingertips!

Of course, this is far from a production product. It’s version 0.0.1. But it works! and it works well! It still has a number of manual steps and it doesn’t work with all of the content sites she uses (yet!). A lot of tweaking had to be done to the GPT as well. But it works!

Most people have heard of the 80/20 rule; the idea that for some things the first 80% of the work is done in the first 20% of the time. Well, I’m pretty sure that the product I have is well under 80% done, meaning there’s a LOT of work ahead to make it a true product.

But I’ve gotten further in one day than I expected to in just a few days. And it’s all because AI was helping me build my AI! Well, plus a little, actually a LOT of help from my friends. The possibilities are endless!



Discover more from Lowry On Leadership

Subscribe to get the latest posts sent to your email.

Quote of the week

“AI will probably most likely lead to the end of the world, but in the meantime, there’ll be great companies.”

~ Sam Altman (apocryphal)

Designed with WordPress

Discover more from Lowry On Leadership

Subscribe now to keep reading and get access to the full archive.

Continue reading