If you are interested in working with AI (ChatGPT, Gemini, Claude, etc.), in human psychology, or both, you might enjoy this episode.

If you ask yourself, “What is prompting” or “What is a super prompt” you’ll also enjoy this.

Transcript:

The Ubiquity of AI in Business

So, what prompted you to listen to this episode? Uh, get it? Oh, not yet. Well, you will. I’ve done at least two episodes on AI, the dangers of using, and so forth. AI is ubiquitous with business now. Everyone’s using it in some form or another. And why wouldn’t they when it’s pretty much shoved down your throat in every piece of software you use, whether it’s Gmail or even drawing software, scheduling software. New companies are crawling out of the woodwork practically daily saying what they do is AI powered.

I have a unique, very unbalanced way of using it. What I do use it for is sort of a language processing model. I have it take the raw transcript from these podcast episodes and then convert it into something that’s a bit more readable, but with the explicit instructions to never change a word or a syllable, regardless of how wrong it thinks the sentence is, because that’s what a transcript is. If I mess up, then I should mess up in the text as well.

What I don’t use it for is anything creative. I will never ask it for an idea. I will never ask it to give me a quick list of things that have to do with something else because I need to be able to do that or I don’t understand my subject. Do I ask for clarification? Sure. Do I ask for research sometimes, but mostly it’s a lot of mundane, tedious things that would really take a tremendous amount of time and be a complete waste of time.

The Reality of Running Local Language Models

I have learned through online use and local use quite a bit. And when I say local use, I I spun up, as they say, a copy of a local language model directly on my office computer. This was a while back and it took a lot of steps. Recently, I did it in just a few steps with an even larger model. It’s so large and so taxing on my system that when I make requests of it, I have to literally watch the temperature to make sure I don’t go over a certain point. And my computer’s smart enough to know to shut down or to or to spin down the CPU, but I it it’s barely running within the parameters that it it can run within.

That being said, I want to talk to you about prompting, but not just prompting for LLMs, but prompting for humans, because there’s a lot in common here.

When you first heard the word prompting, it probably didn’t mean a lot to you. It didn’t trigger much unless you’re someone who is in writer groups or something where you do writer prompts and things like that. But otherwise, you you probably wondered what prompting was. and it really never stuck in your brain. And because it never stuck in your brain, it’s probably not something you’ve you’ve developed as a skill.

Context Windows, Personas, and Super Prompts

Prompting is just a set of parameters you give the model you’re talking to before you really get into the conversation. And it really is a must. Now, all models have built-in prompts. They’re all basically engineered to be cheerleaders for you, to be very cheerful, to be overly agreeable, and to be overly cautious.

With a beginning prompt, you can get the AI to interact with you in a way that’s a little bit more favorable to what you want to get done. And what you have to do is think of it as a a per session kind of arrangement with it. It’s not going to be your best buddy forever and it only has so much memory. It has something called a context window in which it can fit in the stuff you’re talking about and it’s sort of first in first out when it comes to what it forgets. So, it’s going to have some amnesia on the things that you talked about a while back. If you switch gears a lot, it’ll also make it difficult for it to understand what you’re trying to work on.

So, that’s where the prompt comes in. You can say act as a this, you know, a professional person doing something. You’re, you know, you are an attorney and that’s there’s an asterisk there. You’re a this, you’re a that or you’re a development partner helping me do this. If you pick something like attorney or doctor, what have you, it’s always going to give you a disclaimer that it isn’t and so on and so forth. And it’s going to always give you that disclaimer. If you want it to help you in marketing, then you say act as a marketing professional uh and and do this or that. But you’re going to quickly run into basically the same things because you’re not going to get any new exciting ideas out of it unless you give it new and exciting ideas to work with. And that’s the simplest reason to not rely on it to be creative. And unfortunately why we see so many posts on LinkedIn and other places where they all look the same and you can instantly tell that it was created by AI. even if it wasn’t directly created by AI, you can tell that the bullet lists are something that AI came up with.

So, back to the prompt, there’s also something called a super prompt, and that is a really long prompt that you give the AI at the very beginning. You can literally cut and paste these things if you search for them, which will give you what they say are sometimes superpowers. Sometimes you can get around the guard rails and get it to tell you things that are a little bit more interesting, a little bit more helpful, and maybe a little bit more dangerous, at least in its perception.

You can, of course, and should continue prompting as you’re using it. It’s not a set it and forget it because you can adjust its behavior accordingly. And that’s actually how these models are created. They’re with human reinforcement. But unfortunately, this human reinforcement carries a lot of bias. So you sometimes have to undo some of that bias just to get it to give you the responses you need to do your job or the project or whatever it is you’re working on.

If you’ve really never had this sort of way of approaching AI, you will be shocked to see the difference in the responses and the and the difference in how useful it becomes to you if you truly set parameters for it. It’s an interesting exercise in being bossy. Well, it’ll feel like you’re being bossy, but you’re not. What you’re actually doing is saying, “Here, here’s the groundwork. Here’s how I want to work with you. Here’s how I’d like you to create and craft your responses and the the attitude and the persona that I sort of need you to do to fill this.

Applying Prompting Principles to Human Interaction

Wait a second. This is sounding like something else, isn’t it?

What a nice segue. Well, thank you. So, this is why I say that prompting also applies to humans:

Think about that. You meet someone for the first time and you can chat with them and both people are going to go from sort of a place of neutrality, right? What do you do? What is this? What is that? And so forth. If you work with someone, they’re going to get onboarded. You know, you’re going to talk about the mission statement and lots of other things that may or may not be a reality in your company. You have a certain persona for the company, a certain way to dress, a way meetings are conducted, the way that people talk to their bosses. Is it a fun environment? Is it a corporate environment? Is it both, which isn’t a real thing? And so forth. So there are all these visual and audio cues and environmental cues for people to be prompted.

And the only time you get prompted properly, efficiently, is either A when you do something wrong. Oh, we don’t do that here, Sarah. or B when someone there takes you aside. It’s typically someone of your level who says, “Hey, I got to tell you this is how things really are this way. When you deal with Ralph, do this. They really like this kind of thing here.” They kind of look down on that. That’s prompting. You’re being prompted.

Setting the Groundwork Without Feeling Bossy

If we’re always prompting each other and and in in many ways, why not do it on purpose then? Why not consider that? And you’re going to say, “Well, that kind of sounds like how I talk to my kids sometimes.” Well, yeah, there’s a lot of parallel between how you interact with your children and how you interact with with other people. We just feel the we feel on our shoulders the the need, the ability, and the responsibility to get it right when our with our kids, especially when they’re little. When we’re dealing with adults, we make all sorts of assumptions that we sometimes get burned for. But what I’m saying is consider the whole concept of prompting but applied to humans.

My challenge to you would be first of all listen to my advice about prompting an AI. And if you say, “Well, I don’t work with an AI.” Well, we’ll skip ahead. But if you do work with AIS in your job and in some way, shape, or form, and you’re freely allowed to interact with them in the sort of chat interface, then apply what I’m saying about prompting. And after you do that for a bit...