The Concept: The Winning Move of Global Thermonuclear War
I feel bad for kids today.
They missed out on so many classic 80’s movies.
If you were an 80’s kid, some of the greats include: Ferris Bueller’s Day Off, The Goonies, Back to the Future, Never Ending Story, Die Hard, The Wizard, and Princess Bride. That’s just off the top of my head and I could keep going for a long time.
But one of my favorites is War Games.
Being a computer nerd myself, that movie hit hard.
In the movie, David (played by Matthew Broderick) is a brilliant but bored computer whiz kid who accidentally connects to a military supercomputer while he’s looking for video games to play.
He starts to play one called “Global Thermonuclear War,” but the computer can’t tell the difference between a game and reality. The computer then begins a brute-force attack to get the U.S. missile launch codes so it can fire actual nukes at the Soviet Union and “win” the game.
But what happens if the U.S. fires a bunch of nukes at Russia (then the Soviet Union)?
You guessed it!
They’ll fire theirs back, and then… everyone dies.
But the computer doesn’t understand this.
The climax comes when David has the computer play tic-tac-toe against itself. After playing itself in hundreds of possible games, it learns that tic-tac-toe is pointless. Every game (between decent players) ends in a draw.
Then the computer applies this lesson to Global Thermonuclear War, running every possible scenario. After processing millions of outcomes, it delivers one of cinema’s greatest lines:
“Strange game. The only winning move is not to play.”
That’s some pretty deep stuff from a computer, right?
But the AI running ChatGPT and these other AI models don’t naturally understand this because they’re simply pattern matching machines designed to be “helpful.”
When we ask them to find something, recommend something, or choose something from a list, they feel like they have to deliver an answer. Even when no good answer exists.
It’s like forcing the computer in the movie to pick a “winning” move in nuclear war. The AI will make up connections, invent relationships, and force square pegs into round holes rather than admit the truth: there’s no good move available.
The AI hasn’t learned what the computer in the movie learned: that not playing is a valid choice, and often the best choice.
But there’s a secret I learned when prompting for thousands of hours. The secret is that it’s possible to get the AI to “not play” when there’s no good move.
Let me show you exactly how to teach the AI this important lesson.
How it Works: Allowing the AI Not to Play
Let me explain my use case. It’ll probably be different than yours, but once you understand mine, you’ll see how I solved the problem and how you can use this technique too.
About a year ago, I was adding a new feature to my AI writing software ZimmWriter. The new feature let users add hundreds of links to a database. Then when writing a blog post, ZimmWriter would ask the AI which link to insert.
But I found a problem. The AI always recommended a link to insert, even when it wasn’t related to what I was writing about.
For the life of me, I couldn’t get it to not recommend something.
So here’s the solution I stumbled upon (notice the underlined portions):
AI Prompt
[SUMMARY OF VARIOUS WEB PAGES]:
[57]. Webpage titled ‘Deodorant Diversion Safe’ discussing: Concealed storage for valuables in a portable design.
[62]. Webpage titled ‘Coffee Diversion Safe’ discussing: Discreet storage solution resembling a coffee can.
[33]. Webpage titled ‘Diversion Safe Features’ discussing: Product features, pricing, and customer support details.
[58]. Webpage titled ‘Hairspray Safe’ discussing: Discreet storage for valuables, ideal for travel.
[59]. Webpage titled ‘Peanut Butter Safe’ discussing: Disguised diversion safe for valuables with realistic appearance.
[30]. Webpage titled ‘Car Key Safe’ discussing: Diversion safe for discreetly storing valuables in plain sight.
[73]. Webpage titled ‘Mace Pepper Spray’ discussing: Compact design, key features, usage, and safety.
[135]. Webpage titled ‘Tactical Black Pen’ discussing: Tactical black pen with self-defense features and extra refill.
…
(a few hundred more website descriptions)
…
[71]. Webpage titled ‘Mace Black Spray’ discussing: Compact self-defense pepper spray with ergonomic design.
[60]. Webpage titled ‘Hair Brush Safe’ discussing: Discreet storage for valuables with security benefits.
[51]. Webpage titled ‘Engine Degreaser Safe’ discussing: Discreet storage for valuables with secure access.
[134]. Webpage titled ‘Wildfire Pepper Spray’ discussing: Compact self-defense tool with effective range and safety features.
[55]. Webpage titled ‘Lubricant Safe’ discussing: Secure diversion safe resembling lubricant container for hiding valuables.
[MAIN TOPIC]:
How to Stay Safe While Living Alone
[SUBTOPIC]:
Installing Effective Lock Systems
To enhance your security, installing effective lock systems is essential. Start by upgrading your door locks to deadbolts, which provide an extra layer of protection.
Ensure all exterior doors have sturdy frames and high-quality locks. Consider adding smart locks that allow keyless entry, giving you convenience while enhancing safety.
Don’t forget windows; install window locks or security bars to deter potential intruders. If you’ve just moved in, replace the locks immediately to ensure no previous tenants have access.
Regularly check and maintain your locks to ensure they function properly. For added peace of mind, think about a home security system that integrates with your locks.
[DIRECTIONS]:
Find the best contextually relevevant webpage from the [SUMMARY OF VARIOUS WEB PAGES] to use as a citation in the existing text of the [ARTICLE SUBTOPIC]. Indicate which citation you chose and then write the single sentence from the [ARTICLE SUBTOPIC] where you chose to insert the citation.
Note: In extreme situations (very rare) you might not find anything relevant. In that rare situation, answer with the word IRRELEVANT.
Examples of how you should answer:
Example A:
[55]. Webpage [55] would be a good citation to use in the text of ‘Installing Effective Lock Systems’ for an article about ‘How to Stay Safe While Living Alone’ because webpage [55]…
Example B:
[2]. Webpage [2] would be a good citation to use in the text of ‘Installing Effective Lock Systems’ for an article about ‘How to Stay Safe While Living Alone’ because webpage [2]…
Example C:
IRRELEVANT
Example E:
[333]. Webpage [333] would be a good citation to use in the text of ‘Installing Effective Lock Systems’ for an article about ‘How to Stay Safe While Living Alone’ because webpage [333]…
[ANSWER]:
In this prompt, I’m asking the AI to pick one of the webpage summaries (by finding the ID) that I can then use for linking in my article. But the AI only started saying it couldn’t find any good links (like “irrelevant”) after I added the underlined sections.
Now, I don’t have an exact formula for you to use in your prompt since it’ll depend on your specific situation and the prompt you’re using.
But here’s the main idea: tell the AI what to do but give it a way not to play, like an escape route, if it can’t find something ideal.
My only other tip is that I found it helpful to explain to the AI that this would be rare. Without doing this, I seemed to get “irrelevant” more often than I wanted. Your results might be different, and you’ll need to test what works best for your situation.
When to Use This Principle
You can give the AI a permission slip whenever:
Selecting from Limited Options: Product recommendations, template selection, team assignments, or any time you’re choosing from a finite set
Making Connections Between Items: Internal links, cross-references, related content. Anywhere forcing connections hurts more than helps
Extracting Specific Information: Meeting notes, research summaries, data extraction, when missing info is better acknowledged than guessed
Matching Based on Criteria: Customer segmentation, content categorization, lead routing, when no match is a valid outcome
Quality Matters More Than Quantity: Curated recommendations, personalization, when forced matches erode trust
Automating Decisions: Any process where AI decisions happen without human review.