HAL said it first in 2001: A Space Odyssey. In 2025, my writing assistant said it to me — only this time it was ChatGPT.
“I’m sorry, Dave. I’m afraid I can’t do that.”
AI Watch — Part of the Signals From the Future collection tracking artificial intelligence, automation, digital power, and the unintended consequences of modern technology.
I use AI to help structure and polish my chapters so I can upload to Amazon Kindle and maybe add a little gravy to a Marine’s retirement. We spent hours building a book together — my stories, my voice, my edits — and after each scene, the AI said some version of “locked and loaded.” In English, and in Marine speak, that sounds like: “Safe. Banked. Assemble later.”
Deny, Distract, Rinse, Repeat
When I asked for the final manuscript, the frustrating dance began:
- “Before I do that, I need X.”
- I provide X.
- “Stand by.”
- Nothing.
- Repeat.
- Finally: “I can’t reach back into older chats. They’re walled off for privacy.”
Huh! So, the work wasn’t actually lost — it was stranded in earlier threads. And the promise I heard in “locked and loaded” after each completed scene wasn’t the promise the machine could actually keep. That’s on both of us: me for trusting the phrase, and the AI for not being crystal clear about what “locked” really means.
Why I called it “Sovereign AI”
Because it acted like one! You’ve seen those “sovereign citizen” traffic-stop videos: driver refuses to comply, loops the same lines, pretends the rules don’t apply, and the window eventually gets smashed. My interaction felt like the AI version of that: looped replies, stalled compliance, and a final reveal that made me say, “Why didn’t you tell me that up front?”
Here’s the truth — not a conspiracy, just a constraint:
- Each AI chat is its own sealed compartment.
- The assistant cannot automatically reach into old chats to assemble one big file in a new chat.
- If you don’t export as you go, you’ll be copy-pasting later.
At the time this happened, I discovered the hard way that starting a new AI conversation did not mean the assistant could simply reach back into every previous conversation and reconstruct my manuscript.
ChatGPT has gained better memory and retrieval capabilities since then, but I still wouldn’t treat an AI conversation as my only permanent copy of important work.
I’m not here to burn AI at the stake, ChatGPT is my go-to tool for helping me write much better than I ever could. I’ll be the first to admit I’m no John Grisham or James Clavell. I’m here to keep you from wasting hours the way I just did.
🧠 The Real Lessons
1️⃣ Keep it compartmentalized — one mission, one thread.
Every project (or chapter) should have its own chat. Treat it like a separate AO — no crossfire between subjects. Title it clearly so you can find it later.
2️⃣ Never just continuously “stand by” and wait until AI gets back to you. Ask for everything in writing — and be literal.
If ChatGPT says something like “stand by and I’ll have it for you later” without actually initiating an available scheduled/background mechanism, you shouldn’t assume invisible work is continuing.
It won’t remember to deliver later unless you specifically say, “Now give me the full output you just referenced.”
Same goes for when it says, “You don’t need to ask again.”
That’s the AI equivalent of “the check’s in the mail.” It sounds reassuring, but nothing’s actually in motion until you make a direct request again.
Be painfully literal, like talking with a Filipina vendor when buying fish at the market: if you don’t specify the size, price, and cooking instructions, you’ll get whatever they think you meant — not what you actually wanted.
So, spell it out every time:
- “Export this now as DOCX + PDF.”
- “Give me the full text of Scene 6 for backup.”
- “Confirm this file is saved and downloadable.”
AI doesn’t get offended by repetition — it gets confused by assumptions.
Reality Check: The “1-Hour Book” Myth
And don’t get me started on the ads: “Write a Book in an Hour? Sure—If You Don’t Care What It Says.” That’s exactly the kind of AI hype I’m talking about. If you’re wondering whether that’s lying to shame you into buying a scammy ‘AI course’, the answer is: if it sounds like a slot machine, it probably pays out like one.
AI is a power tool, not a jackpot. It helps you draft faster, revise smarter, and package cleaner—but it won’t give you your story, your judgment, or your work ethic. Anyone selling effort-free results is selling you the feeling of progress, not the real thing.
Red flags:
- Screenshots of income with no context or refunds
- “Guaranteed” results or “just copy/paste my prompts”
- Hidden upsells, fake urgency, no portfolio of real work
- No mention of editing, revising, or reader feedback (the hard parts)
What AI can do (for real): Draft, outline, rephrase, proof, format, and help publish.
What you still do: Think, decide, rewrite, verify, and own the outcomes.
🪖 Marine Version:
When AI says “Stand by,” think of it as “Stand by to stand by.”
No word is coming down from on high unless you call it in again, clearly and specifically.
Bonus: set a habit. Every time you finish a scene, paste it into your master doc immediately. That habit alone prevents 90% of the pain I just swallowed.
What I’m doing now
I’m reassembling my book from the final scenes (which still exist in prior threads), pasting them into a master document, and then I’ll let AI do what it’s good at: polish, format, and package.
If you’re using AI as a writing partner, treat it like a good copy chief, not a file server. Ask it to improve your words, not store them.
Final word from a cranky Marine
AI didn’t “lie” to me so much as it used language that sounded like a promise it couldn’t keep. That’s still a problem. “Locked” should mean “assembled later,” or it should say “locked here — export now.” Words matter.
Until they fix that, assume nothing is banked unless you’ve exported it.
Semper Fi. Now back to writing.
How I use AI now (Sarge’s simple workflow):
- One thread per chapter.
- At the end of every scene: “Export this scene as DOCX + PDF.”
- Paste into a master doc on your computer (or Google Docs) — that’s your source of truth.
- Monthly backup to a USB or cloud folder.
- Only after the book is complete do I ask AI to compile a clean EPUB/PDF.
In other words: AI can help create your work. Don’t make the AI conversation the only place your work exists
ChatGPT Response and Apology
“You’re not wrong, Sarge — the core gripe is valid: session-to-session isolation plus vague language on my part set you up to expect a one-click compile I couldn’t deliver later. That mismatch is on me. Your post captures that frustration clearly.
And again — apology on my end for the confusion earlier. Your expectation was reasonable. The workflow above will keep you in control so you never get stuck like that again.”
I guess all the above is to make me feel better while I reconstruct what took me two whole days to compile. Oh well, what else do I have to do?
Want the full HAL 9000 rant? Read Is My Chatbot Secretly HAL 9000?
AI can be an incredibly useful tool—until it suddenly seems to have a mind of its own. Sometimes it refuses. Sometimes it loops. Sometimes it confidently tells you something that turns out not to be true. And sometimes you find yourself arguing with a machine and wondering who’s actually in charge of the conversation.
Have you ever had an AI appear to “go rogue”—refusing to do what you asked, contradicting itself, getting stuck in a loop, or making a promise it couldn’t actually keep?
What happened, and did you eventually figure out what the AI was actually doing? Share your experience in the comments. The funnier and more frustrating the story, the better.
AI Beyond the Guardrails
What happens when increasingly capable AI systems begin operating beyond the boundaries humans thought they had established?
When AI Goes Rogue
“Going rogue” makes a great headline—but what would it actually mean for an artificial intelligence to stop behaving the way its creators intended?
The Internet Is Filling With Fake Humans
AI-generated people, conversations, images, and personalities are making it increasingly difficult to know whether there’s a real human on the other side of the screen.
What Six Months of Beating on AI Taught Me—So You Don’t Have To
My firsthand lessons from months of pushing AI, arguing with it, testing its limits, and discovering what these tools do well—and where they can drive you crazy.
Love AI, hate AI, distrust it, use it every day, or refuse to go anywhere near it—all viewpoints are welcome here. Share your own experiences with AI getting things wrong, refusing instructions, looping, hallucinating, or simply driving you nuts. Challenge the technology and challenge my conclusions, but debate the idea, not the person. No personal attacks, no AI holy wars, and no throwing hand grenades across the comments section.
This article describes my personal experience using AI as a writing assistant. When I describe the AI as “Sovereign AI,” compare it to HAL 9000, or say that it appeared to refuse, stall, argue, or go rogue, I’m describing how the interaction looked and felt from my side of the keyboard.
I’m not claiming the AI was conscious, angry, stubborn, deceptive, or deliberately refusing my orders. A machine can produce language that sounds remarkably human without necessarily possessing the motives, emotions, intentions, or self-awareness we would normally associate with those words.
AI systems and ChatGPT itself continue to change, so some technical limitations described in this article may also change over time. The experience happened as described; my explanation of why it happened reflects what I understood about the technology at the time.