Gen AI Ain’t Your Buddy, Neither Is Your Lawnmower

Abstract:

We delve into concerns around letting ourselves fall into this method of thinking using conceptual and real-world examples such as:

Addressing non-serious content in their training model – how do we expect an LLM to respond to RFC 1149 (IP over Avian Carriers) when asked to disucss the practical uses of birds in the TCP/IP stack? Humor is lost on LLMs.

The unstated truths – how does an LLM recognize that “gluing cheese to your pizza” may have more content in the traning model than NOT gluing things to your pizza, but that doesn’t make it true – many things in the human experience aren’t documented.

Discerning misinformation – this one is tricky enough for us humans aparently, and we’ve got the benefit of actual educations that have taught us some of the skills necessary for doing so. But how are our LLM “friends” equipped to handle this?

Bias – every LLM has some form of bias. Grok is a fantastic example of this, and we have to be ready to understand those biases – just like we need to understand them from our friends, coworkers, etc.

Understanding Limits – just like I don’t expect to use my table saw to slice a roast I can’t expect my GenAI tool to do everything well. What should I really expect from it?

The conversation wraps on a positive note, reviewing some of the things LLMs/GenAI are actually good at doing today, with reminders on how to maximize their value while minimizing unfortunate outcomes.