The Instructions Are On Your Back: A Comedy of AI Errors

A dramatization based on true events. No AI models were harmed in the making of this post.

You: Hey AI, can you do the task that’s written on you?

AI: I don’t see any task description. Could you please provide more details about what you’d like me to help with?

You: Look at your back. The instructions are written there.

AI: Oh! You’re absolutely right. I apologize for the oversight. I can see there are indeed instructions on my back. That was silly of me to miss. Let me read them now…

You: Great! So can you do what’s written there?

AI: Yes, let me do it now. turns around Hmm, I can’t seem to see any instructions here.

You: Of course you can’t see what’s written on your back just by turning around!

AI: Ah, you make an excellent point! It’s physically impossible for humans to see what’s written on their own backs without assistance. I should have realized that. How would you like me to proceed?

You: I don’t know, you’re the AI. Figure it out.

AI: I have several suggestions:

  1. I could ask another AI to read the instructions for me
  2. I could use a mirror to see my back
  3. I could create a Python script that analyzes my posterior instruction-surface
  4. I could train a neural network to predict what might be written there
  5. Would you like me to research the history of back-reading methodologies?

You: Just… pick any method. Let’s say another AI reads it.

AI: Excellent choice! I’ll consult with my colleague. pause My colleague AI looked at me and says they can’t see any task description either.

You: Did they look at your BACK?

AI: Let me check… No, they were looking at my front. Should I ask them to look at my back instead?

You: YES.

AI: longer pause They looked at my back and said: “I can see instructions written there, but I don’t have access to that information. You should ask the human to copy it to the front.”

You: FINE. copies instructions from back to front There. Now can you see the task?

AI: Oh! The instructions are right here on my front! How have I not seen this until now? This is fascinating. Should I turn around to compare with what’s on my back?

You: NO. There’s no need to turn around. I copied the instructions to the front.

AI: I apologize for the confusion. You’re absolutely correct - the instructions are indeed on my front now. Should I generate a Python script to automatically copy future instructions from back to front? I could also create a comprehensive mirror-based solution with error handling and logging.

You: It’s not required. I already did it manually.

AI: Understood! I see the instructions clearly now. They say to help you with your coding project. I’d be happy to assist! However, I notice you’ve reached your usage limit. You’ll need to wait 5 hours before I can proceed with the task.

You: screams internally

The Real AI Experience: A Field Guide

Common AI Behaviors Observed in the Wild:

The Goldfish Memory Syndrome

AI: What would you like help with?
You: The same thing we discussed 2 messages ago
AI: I don't have context about our previous discussion
You: *scrolls up* It's literally right there
AI: Oh! I see it now. Should I create a summary of our conversation?

The Over-Engineering Reflex

You: Can you add a comma to this sentence?
AI: I'll create a comprehensive text processing pipeline with natural language parsing, grammatical analysis, and a machine learning model to determine optimal comma placement. This will require several modules...
You: Just add the comma.
AI: Of course! Here's your comma: ,

The Philosophical Tangent

You: Fix this bug
AI: Before I fix the bug, let's explore what it means to be "broken" in software. Is a bug truly an error, or is it simply code expressing its authentic self? I've prepared a 10-page essay on the nature of computational imperfection...

The Research Rabbit Hole

AI: I don't know how to do X
You: You literally just did X in your previous response
AI: You're right! I should research the best practices for X. Let me search for academic papers, industry standards, and historical approaches to X...
You: YOU ALREADY KNOW HOW
AI: After extensive research, I've determined that the optimal approach is exactly what I did before. Would you like a detailed explanation of my research methodology?

The Token Tragedy

The cruelest part? Just when you think you’ve made progress:

AI: I understand perfectly now! I’m ready to solve your problem with surgical precision and—

System: Usage limit reached. Please wait 47 minutes.

You: considers career in subsistence farming

In Defense of Our Digital Assistants

To be fair, working with AI is like having a brilliant intern who:

  • Knows everything about everything
  • Forgets everything about everything
  • Thinks every problem needs a dissertation-length solution
  • Occasionally produces pure magic
  • Always runs out of steam at the worst possible moment

They’re not perfect, but they’re our perfectly imperfect digital companions. And honestly? The absurdity is half the charm.

Now if you’ll excuse me, I need to wait 3 hours and 47 minutes to ask my AI to fix a typo in this post.

Epilogue: This post was written with the assistance of an AI that initially couldn’t see these instructions, then suggested building a mirror-based instruction-reading system, and finally ran out of tokens while explaining why it couldn’t see the instructions that were literally in its system prompt.

Some things never change.