3 ways to quickly level up your AI game
Delimiters, variables and thinking of AI like human counterparts can all bring a new dimension to your use of AI.
The first time I used AI, it was novel but far from a revelation. But I kept using it day after day, and I had some key epiphanies. These triggered some transformations to my workflow and, like a lot of life and business, 20% of these created 80% of what I’m finding is sizable value.
A fifth of a decade in, I looked back at the three shifts in my approach that delivered the most impact.
Key Takeaways:
Twenty percent of my AI learnings have delivered 80% of the value
Delimiters and variables make prompts more flexible and reusable while reducing ambiguity
Thinking of AI tasks analogous to human tasks makes it easier to identify leverage points
Epiphany 1: The power of delimiters
Delimiters are simple tools offering a straightforward benefit: structure your prompts to give more contextual details. These tools—which can be brackets, slashes, or any other set of characters—do more than just organize. They make it possible to offer much greater information without mucking up a prompt.
Why do we want to give more information? The more examples, direction, and data we can provide, the more pointed our responses get. Delimiters are, at their core, one of our best tools to reduce ambiguity and hallucinations.
Key Insight: Delimiters can also help modularize your prompts. Always remember this: The more you can modularize a prompt, the more it can be reused.
Example in Action: Generating Twitter account bio content
In the example below, I use two-sided delimiters for “About Me” and “Writing Samples”.
I need you to generate the content for my twitter account for me. I will provide you with details about myself and you should provide a list of all the content I will need for my profile. You must provide me this content, and please match to the writing style I've provided below.
-- About Me START --
* Dynamic Storytelling: As an author, I weave captivating narratives that resonate with readers through emotional depth and engaging plotlines.
* Versatile Creative: My writing spans various genres and formats, showcasing my adaptability and creativity across short stories, novels, and other forms of storytelling.
-- About Me END --
-- Writing samples START --
NYT bestselling author of Annihilation. Founder & President @FloridaRewild #VanderWild he/him
Breaker of reading slumps. Wayward Pines. Good Behavior. Recursion. Summer Frost. NOW PLAYING: Dark Matter on AppleTV+.
Official Twitter. Check out my Substack for writing advice, prompts and anecdotes from my career. New book out now! Newer book coming this fall.
-- Writing samples END --Actionable Tips: Include delimiters around longer pieces of content, like bulleted lists or copied data. Make delimiters a regular part of your prompts by:
Experimenting with them across different types of prompts (research, generation, data cleaning)
Naming your delimiters based on the content they denote (a more specific “email campaign data” is better than “data”)
Running side-by-side comparisons with and without delimiters
Exercise: Adding delimiters
Step 1: Choose a longer prompt (500+ words) where you’ve given AI longer pieces of content or data.
Step 2: If they aren’t already, separate out longer blocks of text (article content, list of emails, email copy, etc.). Put line breaks between sections.
Step 3: Choose your delimiter. You can use one-sided (placed before the content) or two-sided (place before and after). As you can see above, I use two-sided delimiters (“— Content START —”, “— Content END —”).
Step 4: Insert your delimiters before and after the relevant sections.
Epiphany 2: Thinking of AI as a partner, not a tool
And I don’t mean in a “please and thank you”, “how are you chatGPT?” kind of way. Thinking of AI as a partner might feel (and objectively be) weird, but it frames your strategy for AI. In other words, it gets easier to envision tangible use cases when I imagine it as a colleague.
Key Insight: This led me to a more organized approach in the tools I use. On the one hand, it gave me a way to accumulate related context and prompts within a single thread. On the other, it also reduced the emotional choas of having a thousand disparate chat threads.
Examples in Action:
Product Manager Assistant: Think of it like a junior assistant. Introduce dialogue to solo brainstorms. Categorize feedback. Draft product reqs.
Literary Agent Assistant: Get feedback on a manuscript pitch or query letter. Templatize the generation of agent queries or lit mag submissions with a little personalization.
Writing Assistant: Speak with your own characters. Have an expert on style and grammar. Break down story plots to ask what works and what doesn’t.
Audio Engineering Assistant: Discuss the pros and cons of recording techniques. Get personalized explanations of recording terms (some hallucinations included).
Legal Assistant: Get a high-level overview of legal tasks for a business. Understand connections between entry-level concepts and even help you work better with actual lawyers.
Actionable Tip: Name each chat thread for it’s specific role ("Marketing Assistant" or “Data Assistant.”
Epiphany 3: Using variables for prompt flexibility
Variables are the key to highly adaptable prompts. What does that mean in practice? It means you’re building templates instead of rewriting prompts from scratch every time.
It’s a time-saver for repetitive tasks. For example, let’s say you’re drafting a series of social media posts. By using variables for dates, specific product names, or calls to action, you can create one prompt that does the work of 30.
Key Insight: Variables give you a more efficient and systematic approach to recurring tasks like data categorization or bulk messaging.
Examples in Action: Researching an audio engineering term
You are a seasoned audio engineer. I need you to define the following term for me: {TERM}
Please tell me:
1. What this term typically refers to for an audio engineer?
2. How I know what my own {TERM} is?
TERM = “Comfortable Listening Volume”
Actionable Tips: Jump head first into variables! Prioritize longer prompts with more variables and you’ll see more value. Also, adopt a process to adapt your own prompts, including how you format each variable. Should they be in all capitals? Spaces or no spaces? Surrounded by curly brackets or square?
Exercise: Incorporating variables
Here is a three-step process to put variables into any prompt.
Step 1: List any details of the prompt that will change next time. They’re really the WHAT of your prompt inputs. Most often, they’ll be dates or numbers or names.
Step 2: Select your placeholder — I recommend “{ }”, or sticking with brackets but you can use most special characters. Replace all of the details with their corresponding variable. As an example, a prompt that had “Please create 5 social posts” would translate to “Please create {NUMBER_OF_POSTS} social posts.”
Step 3: Somewhere within your prompt, set your variable. The same prompt that has {NUMBER_OF_POSTS} in the direction should also have the following: ‘NUMBER_OF_POSTS = “[]”’. This should happen once for each variable you add.
Some pitfalls to avoid
Here’s how to sidestep a few common pitfalls for each of these techniques:
Not documenting your prompts: Recording prompts accurately ensures you can replicate or reuse them. It also helps you keep consistent with your choice of delimiter or variable characters.
Overusing delimiters: Don’t break down delimiters so much that you have 50 different delimiters. Keep them for bigger blocks of text.
Incorrectly naming your delimiters or variables: Making them too generic can be the same as not having any at all. Be detailed about marrying the name of your variable/delimiter to the content it contains.
Inconsistent formats across prompts: Pick a delimiter structure (such as “— content START —” and “— content END —”). Doing so reduces the micro-decisions you’ll have to make when prompting.
ONLY thinking of AI as a person: There is something to be said for NOT treating AI like a human. Looking at a problem from the Terminator’s perspective will yield different (and sometimes Terminator-like) results.