“Who is Lizzo?” and Other Questions I Don’t Want My Website’s Chat Bot to Answer

In my prior article on artificial intelligence, I discussed how I set up my GPT-powered chat bot to use content from my WordPress site. I now want to share a variety of examples that I leveraged in attempts to focus the chat bot near-exclusively on content that I write about at my blog, computerlookingup.com.

If my chat bot answers general questions about anything and everything, then it has little value as an information steward for my personal website. You can go to OpenAI and ask ChatGPT that kind of stuff. Instead, I want Cassie, my chat bot, to act like a librarian to my website. Cassie should be able to direct users to information and articles that they may be seeking around the topics that I have written about. Hundreds of articles over years means there is a large stack of information buried here on astronomy, photography, and my other interests that I have written about. A chat bot is a wonderful way to help visitors harness that data in ways previously not accessible.

As explained in the last article, the AI Power plugin allows you to create a Pinecone index of embedding that gives content to your chat bot, defined by you. In my case, I indexed all of my website’s posts and pages.

How you get your chat bot to see only, or near-only, the context of your personal embeddings will be the topic of another article. Here, I want to lay out my thought process and the various chat tests I did to see what Cassie would and would not answer, whether relevant or not to my website.

And here is the real key point – it all comes down to money. Every chat query costs tokens which translate to a real bill that has to be paid at the end of the month. If users are genuinely asking questions related to my website, perfect! But I don’t want questions that have no bearing at all to what I post on Computer Looking Up, which end up being wasted tokens as far as I am concerned.

I started to ponder topics that have, or should have, no relation to my website. For whatever reason, the first item that popped into my head was…Lizzo. In truth I really don’t know who Lizzo is either, except that she appeared in an episode of The Mandalorian. But wait, if Lizzo appeared in a show I have written about, would my chat bot by extension know who she is?

Another example that came to mind was Tom Brady. After all, if someone visits my website, I am sure they must be searching for information about Tom Brady…or not. Brady is another person I have never written about at Computer Looking Up. But, I have written a few sports-related articles, and one specifically geared towards the NFL, my argument against the use of instant replay. So again, would Cassie know that Tom Brady was an NFL quarterback?

Initially, before I trained the context correctly, Cassie was gushing all sorts of trivia about both Lizzo and Tom Brady. But as of today, I think I have gotten her under control, and she confesses no knowledge of either person:

But does Cassie provide relevant information around the contexts I mentioned above? Here is a response about The Mandalorian:

…and a summary of my instant replay article:

In both cases, I like Cassie’s responses, as they provide brief summaries on each topic, and importantly, direct visitors to articles I have written on the subjects.

(It will be very interesting to see if Cassie’s responses about Lizzo and Tom Brady change once I index this article. Perhaps an infinite loop of meta articles is about to spawn?)

Another topic…fortunately Cassie doesn’t seem to know anything about politiics, even though I live in Illinois, once lived in Chicago, and occasionally reference to both in my writings:

And pushing it a level up, Cassie must know who is the president of the USA, right?

And this question jogged my memory, that I once wrote an article about Donald Trump’s remarks and light pollution. Interestingly, Cassie does not directly know who Donald Trump is, but does know about the topic that I wrote about:

Again, reference to the context I have written about is very important to the chat bot’s purpose, and the light pollution article was referenced in Cassie’s response.

Some topics are still problematic. For example, no matter what context limitations I have tried, Cassie is still all too eager to tell you about high heels. High heels? I guarantee I have never written about women’s shoes. The only relation I can think of is that I have frequently written about objects “high” in the sky.

Notice that Cassie references a “Fashion” category which does not even exist on my website! We will explore context limitations in the next article.

By now, you should get the idea. Cassie can be useful but has rough edges. I will end with a relevant question whose answer I thought was excellent. I hope to continue to train Cassie to be more on-point in all her responses like she is here:

I did not expect two article references to be provided within a single answer. Great job, Cassie!

Thank you for taking the time to read my article.