What Sam Altman’s Sleepless Nights Reveal About AI Stewardship
The Confession
“I haven’t had a good night’s sleep since ChatGPT launched.”
— Sam Altman, OpenAI CEO, September 2025
800 million weekly users. 1,500 people talking to ChatGPT before suicide—every week. No clear position on assisted suicide. Military applications. Privacy.
Tucker Carlson called it: “You’re a religion without a catechism.”
Your 10-year-old has the same tool. No training. No boundaries.
What keeps you awake at 3 AM?
The Game Nobody Took Seriously
When my kids wanted pocket money, I handed it over. Said something about saving while checking email. Sehr konsequent, ja?
Then I saw three jars. Stopped me.
🏺 JAR 1: Give (20%)
Not someday. Day One.
Made me uncomfortable: Never had this. Built my career without it. Every raise, every shortcut—kept it all.
Now building AI systems the same way.
🏺 JAR 2: Save (40%)
Don’t touch. Maybe for months.
Altman missed this completely. Every AI capability? Deployed immediately. No pause. No watching first.
Your hospital doesn’t work this way. Bank doesn’t. School doesn’t.
Why your family’s AI?
🏺 JAR 3: Spend (40%)
Day-to-day. Guilt-free—because you filled the other two first.
We’re building AI to “spend” more time on ourselves. More efficiency. Getting ahead.
Where’s your give? Your save?
Took me three weeks. Still working on it.
The 2-Minute Protocol
Altman’s problem: Life-and-death ethics “on the spot.”
Kids’ advantage: Decide in 120 seconds. Before fear calculates.
How:
- Set timer
- Start before brain objects
- Stop when it beeps
- Celebrate surviving 2 minutes
Real example: Parent used this for kid’s first AI prompt. Timer set. One question: “What will AI never do for you?”
Messy handwriting. Two crossed-out lines. 17 words.
Both signed.
Altman: 2 years. This parent: 120 seconds.
Not perfect. Tested. Good enough.
Five Gaps We Inherit
Gap #1: No Clear Line
Tucker to Altman: “What does ChatGPT stand for?”
Altman: “We have a ‘model spec’—30 pages.”
Can’t explain simply? Don’t understand simply.
One sentence. Clear enough for midnight-you.
The fix:
“My AI helps me _ so I can . It will NEVER ___.”
Example: “My AI helps me understand concepts so I can explain them myself. It will NEVER write my essays.”
17 words. One Post-it.
Your kid can say this? Ahead of 800 million users.
Can you? Jetzt?
Gap #2: Launch Before Test
“1,500 people per week talk to our model before committing suicide.”
— Sam Altman
Tucker: “What does it tell them?”
Altman: “Suicide is bad. But for terminally ill people where assisted suicide is legal… I could imagine us guiding them.”
Imagining ethics. After 1,500 weekly.
Sit with that.
Kids do better:
- Week 1: Small amount (5,000 shillings, not 50,000)
- Week 1: One prompt together, not bedroom access
- Week 2: Review before Week 3
You: First 10 prompts together. Watch what they ask. See what answers. Discuss what feels wrong.
Awkward? Ja. Slower? Obviously.
Altman handed 800 million people ChatGPT without this.
1,500 weekly… how’s that working?
Gap #3: No Give Requirement
Tucker: “This could enable totalitarian control.”
Altman: “We need ‘AI privilege’—like doctor-patient confidentiality.”
Blind spot: Power flows up. No redistribution. No teaching obligation.
Accumulation. Getting further ahead.
Kids do better: Give jar isn’t aspirational. 20%. Day One.
Your test:
AI saves you 2 hours this week. What will you give?
- Teach one colleague?
- Mentor someone younger?
- Sponsor internet access?
Or find more work for those 2 hours?
(I do this. We all do.)
12-year-old gives 20% of transport money. You can give 20% of AI time.
Gap #4: Waiting for Perfect
Tucker: “You seem unbothered.”
Altman: [hasn’t slept in 3 years] “We’re thinking constantly. No final positions yet.”
Waiting for perfect ethics. Perfect policies. Perfect frameworks.
Meanwhile: 1,500 weekly. 800 million users. No final positions.
Perfectionism doesn’t prevent harm. Enables reckless deployment as “we’ll figure it out.”
Alternative:
- Week 1: Write boundaries (2 minutes, crossing out okay)
- Week 2: Test one prompt—what happened?
- Week 3: Hold up? Surprised? Adjust.
- Week 4: One specific use
Not perfect. Tested. Adjustable Friday.
Proverbs 16:3—plans get “established.” Not perfected by committee.
Established in your home. This week.
Start messy. Adjust weekly. Sleep better than Altman.
Gap #5: No Teaching Requirement
“800 million people turn to ChatGPT like a religion—but it doesn’t admit it’s a religion.”
— Tucker Carlson
Altman: [long pause] “We’re trying to be transparent…”
No peer review. No accountability. No requirement users understand the system they’re trusting with kids’ homework, parents’ medical questions, career decisions.
Faith in a black box. Run by people “thinking on the spot.”
Kids do better: Don’t graduate until you teach one friend. Explain simply enough someone younger uses it safely.
Can’t teach simply? Don’t understand deeply.
Don’t understand? Shouldn’t depend on it.
Your test: Explain your family’s AI boundaries to another parent. Over coffee. 3 minutes. No “it’s complicated.”
Need disclaimers? Don’t understand your system.
Neither does your teenager using it 11 PM for applications.
(Fine. Keep simplifying.)
The Uncomfortable Comparison
Sam Altman, 2025:
✗ $500 billion company
✗ 800 million weekly users
✗ 3 years without good sleep
✗ “Making it up”
✗ 1,500 weekly pre-suicide chats
✗ No final positions
Your child with 3 jars:
✓ Clear purpose (17 words)
✓ Tested small first
✓ Gives 20% automatically
✓ Decides in 2 minutes
✓ Must teach one friend
✓ Sleeps through night
Better system?
“Can’t compare child’s allowance to global AI…”
Why not?
Ethics scale down to 12-year-old managing transport money—or don’t scale up to 800 million making life decisions.
No middle ground where “different at scale” makes principles less important.
Scale makes simple principles more critical.
Everywhere. Kigali. Singapore. Everywhere.
Start This Week
For Your Child:
Need:
- 3 containers
- Labels: “Give 20%” / “Save 40%” / “Spend 40%”
- One week’s allowance
- Timer
Before Friday:
Set up. Divide together. “Give helps others. Save waits for plan. Spend handles this week.”
Fertig. Started.
Next day:
“What should AI help you do? What should it NEVER do?”
Vague? Normal. Keep asking. “Specifically—never? Even tired? Even deadline?”
Write together. Crossing out okay. One sentence. Clear for tired-them at 11 PM.
Both sign. Keep visible—laptop, door.
This week:
First prompt together. You watch. They type. Both see answer.
“Match what we signed?”
Ja? Continue carefully.
Nein? Stop. Discuss. Adjust. Try again.
For You:
Right now:
Notes app. Timer: 2 minutes.
Three things AI will NEVER do for you.
Not “shouldn’t.” NEVER. Even convenient.
Three things. Two minutes.
(Wirklich.)
Couldn’t list three? Not ready to build assistant. Ready to become dependent without boundaries.
This week:
AI saves time today. Give 20% away.
Not “when less busy.” Thursday.
Teach someone. Mentor someone. Sponsor access.
Next week:
Review three boundaries. Hold? Bend when tired?
(I bend mine.)
Strengthen what broke. Don’t delete—strengthen. Clearer. Simpler.
Every Friday.
The Question Altman Can’t Answer
Tucker: “Show me the angst-filled Sam Altman wrestling with moral weights.”
Couldn’t. Wouldn’t. Doesn’t remember after 3 years without sleep.
Better question—you can answer tonight:
ChatGPT’s power with Altman’s insomnia—
Or your clear boundaries with 10-year-old’s peaceful sleep?
Kids’ game isn’t child’s play.
Only game with rules simple enough for 2 AM. Nobody supervising.
Only game where you look your child in eye tomorrow: “I’m doing exactly what I’m asking you.”
Not “do as I say.” “Watch what I do.”
Proverbs 22:6 — Train up a child in the way they should go, and when they are old, they will not depart from it.
Altman trains 800 million in way he hasn’t figured out.
You train your family in 3,000-year-tested way.
Three jars. Two minutes. One sentence.
Messy works. Started beats perfect.
Complicated frameworks? Later.
Right now: Simple enough for 8-year-old.
Too complicated for 8-year-old? Too complicated for stressed, tired, deadline-crushed you.
That’s when you need it most.
📎 Complete 14-day guide: daily prompts, parent checklists, AI Covenant Card template
P.S. — Too simple? Genau. Complexity hides from accountability. Simplicity builds character—yours and theirs. Ask Altman which he wishes he’d chosen 2022.
