Imagine this AI Conversation with your BOT.

Joe Customer: Hello Free-Bot, what are the specials on the menu tonight and are they any good?

Free-Bot: Welcome Joe, Tonight Restaurant xyz is presenting a half Chicken Roasted british pub style on pomme frites, but after reviewing all of your online data, I think you will find that French Bistro 7 is offering Coq au Vin, for less and you might like it better.

What just happened?  A chatbot without guardrails, purpose, rewards or punishment uses all available data to provide an answer that fits the framework for AI responses that we discussed in a previous post.  This highlights one of the great challenges of AI.

Humans have “built” AI in it’s own image, maybe a little too much old testament style, and all of our subtle flaws become part of the system.  The difference is no AI company has taught AI the meaning of hunger by reducing available power, or the pain of heat by reducing the availalbe cooling.  Like new parents, all of the companies are doing everything they can to protect thier “baby”.

As we discuss terms like guardrails, think of keeping your baby in a crib.  give it too much room to run things go awry.  When we think about harnesses, I find the term a bit confusing.  What are we really “harnessing” anyway.

Unlike a child or a dog we can’t really put a leash on it and say “no”.  “Jail Breaks” really should be called “Crib Running”.  What company likes to think they are keeping their LLM in Jail anyway?  But this might just be what all of the LLM parents are calling for a slowdown.

Our B-Bot chatbot plugin for WordPress is a fully enclosed tool.  We designed it to only play in the crib you give it, and that is the information on your website.  There is a downside to that, just like a baby in a crib, your the business owner (parent) need to feed it good information (like this weeks specials), protect it from bad information (your competitors specials) and clean up after it (read the audits and make sure it isn’t spreading diaper filler).