Gaming Your Way to Better AI Prompts

A common sentiment that AI evangelists have expressed to me is their eagerness to outsource their day-to-day communications to it. I am not about to state that this is always a mistake. Certain missives are so rote and return so little for the invested effort that it’s hard to justify composing them from scratch. We’ve all sunk time into sending formulaic emails, which could have been better spent doing almost anything else.

But here’s the catch: if you want AI to do its best work, you have to stay sharp yourself. These tools are only as strong as the thinking and language you bring to them.

What I suspect yields the best long-term utility is to do some portion of your language composition completely unassisted. I don’t advance this out of some metaphysical concern for preserving the human soul (though it can double as that in a pinch), but as an appeal to the practical pursuit of conjuring the best AI output possible.

Just as Formula 1 drivers physically condition to better handle their finely tuned machines, practicing language skills without AI augmentation hones your ability to wield AI deftly. In this article, I offer three activities — really, two warm-ups and a high-intensity workout — that help extend what you can elicit from AI and cultivate your self-expression.

On the Write Track

To start, do some writing without AI. Most of the time, our thoughts crest and subside in unpredictable waves. This rhythm can surface profound realizations, but inconsistently so. Focused, linear thinking counterbalances the mind’s tendency to spontaneity with structure. By directing attention to one thought and playing it forward from there, you can pull at one of the innumerable threads lying all around us.

Writing is the documentation of this cognitive exercise — it’s like taking photographs as you follow your thoughts. The same sequential progression of reasoning at the heart of good writing is what organizes your instructions to the AI. As much as AI hype-chasers glorify what I just described as “prompt engineering”, it’s actually good, old-fashioned articulateness.

A Penny for Thoughts Is a Sound Investment

Once you’ve checked that box, start talking to people about things they understand but you don’t. Learning about something by discussing it with an expert is a kind of language puzzle, because you have to figure out how to figure out what to ask. You are prodding into the unknown. Without subject matter knowledge, you don’t know what rocks to look under.

But this is a conversation: you get the chance to iterate. If you hit a dead end, you change tack. When you find one breadcrumb, you’re closer to finding a loaf. Practice formulating the right diagnostic questions, and you get faster at discovering the tailored questions that unlock more arcane knowledge. Those opening questions are the trickiest, but once you get familiar with them, you will open doors you didn’t even realize were there.

Gaming the System

Now that you’ve gotten the hang of those, you’re ready to sit down and play an old-school tabletop role-playing game (RPG). It was only a matter of time until I worked this pastime of mine into one of my articles, but I can’t imagine a more fitting pretext.

For those unfamiliar with the genre, a tabletop RPG is a game in which one player, usually referred to as the “game master” or similar appellation, creates and administers an imaginary game world, and the other players each have a character that they use to interact with everything in that game world.

All players are free to attempt any activity they can think of, and if it is within the capacity of their characters, they usually carry out that activity. When the success of an activity is not guaranteed or there are irreconcilable conflicts between players (especially, and typically, with the game master), dice and character statistics are used to resolve the interaction.

The game interface is purely verbal: the game master describes the environment, the other players articulate what they want to do, and the game master pronounces the outcome of the players’ decisions.

Role-Playing as a Skill-Building Tool

The archetypal example of a tabletop RPG is Dungeons & Dragons (D&D). However, while entries in the genre are commonly likened to board games, the comparison is inapt. Board games rely on fixed visuals like boards, cards, and pieces, whereas tabletop RPGs need little more than character sheets and conversation. If visuals are needed, players can improvise with whatever helps everyone share a clear picture of the scene.

Moreover, while board game rules exhaustively prescribe the available (and unavailable) activities in the respective game, tabletop RPG rules are merely a framework for evaluating the success of player actions. Players may do anything, even what is not explicitly sanctioned by the rules — the rules just explain how to determine the result of their actions.

So what does it mean for an RPG to be “old-school” then? To condense what could easily be a PhD thesis into a brief primer, the main difference between modern tabletop RPGs and “old-school” RPGs is that while the former emphasizes character statistic optimization and rolling high, the latter focuses on player ingenuity and creative utilization of resources.

Whereas disarming a mechanical trap in an abandoned tomb in a modern game may simply be a matter of declaring the use of the character’s “Disarm Trap” skill (providing a numerical bonus) and rolling a high number, doing so in an old-school game would consist of describing your character’s technique for defeating the mechanism — provide enough detail, and outline a thorough procedure, and success may be assured without rolling dice.

What makes a stellar tabletop RPG player is a synthesis of the linguistic tools developed by the aforementioned warm-up routines. You start by “interrogating the fiction,” being curious enough not to take the scene at face value. Then you ask for more information about what’s around your character. You might learn even more by experimenting with the objects in your environment. Make no assumptions about what you think you know. That plain-looking bowl on the table could have magical properties.

Then, you engineer a path to your goals based on your observations. You are equipped to go further than merely stating the desired end goal — which can get your character killed when the stakes get moderately high — by expressing a procedural solution. If you outline a set of steps, including interventions for possible contingencies that can arise from the present game scenario, which would succeed without a doubt, then how could you fail?

Applying RPG Thinking to AI Prompts

You can employ the old-school RPG problem-solving technique with minimal modification when crafting large language model (LLM) prompts.

In both exercises, the surest route to success requires conceiving your goal in the minutest detail, establishing relevant roles (what do you think “role prompting” is?), listing all the steps to be taken, providing examples, and defining exigent circumstances and your prescribed reaction to them.

It should come as no surprise that technology pioneers like Steve Wozniak and Peter Norvig were avid D&D players — their respective corpus of work was largely before generative AI. Still, their accomplishments reflect mastery of the skills that D&D demands.

The order of the above exercises is deliberate. I sequenced them in order of fewest to most required players and lowest to highest difficulty of execution. There are fewer adept conversationalists than articulate writers, and even fewer stellar RPG players than engaging conversationalists. Each step builds the mental agility needed to frame better AI prompts and get more meaningful results.

With that in mind, especially if these are new to you, I encourage you to progress through these exercises in order. Not only can you practice each one without an internet connection to a billion-parameter model, but (even better) they’re fun.

In the age of AI, sharpening your communication skills is still the most powerful way to make the machines work for you.