/sunbreaks
SUNBREAKS
A family weather report for the AI years
This autumn I gave my family a talk about AI: animated pixel-art slides, each set in a Portland place, and a live voice assistant they could question.
Portland weather was the frame. The mountain is there even when it's not out. You dress for rain instead of predicting it. And sunbreaks do happen.
/rules
How I talk about AI with people who didn't ask
Ten rules I set before the talk. The goal isn't to win an argument about AI. It's to leave the room closer, calmer and a little more prepared.
- Care before claims. Open with why you're sharing, not with how big AI is.
- Weather words, not prophecy words. Forecast, update, dress for it, check back in December. Never “inevitable.”
- No fear. Every caution comes with a handhold: an action, a comfort or a hopeful fact.
- Allow every feeling. Excited, scared, skeptical, tired: each one is a reasonable response to this year.
- Talk about tasks, not people. “Routine thinking work is getting cheap” is true and survivable. “Human intelligence is worthless” is neither.
- Receipts over opinions, misses included. Every number gets a date and a source. Showing the misses makes the hits believable.
- Touch money lightly. Talk about the economy, never about anyone's own money.
- Leave room for faith and values. The biggest questions AI raises are older than any lab.
- Sell the good parts honestly. Say “could” and “worth building,” not “will.”
- Point people to each other. Help should run in every direction, not through you.
Say this instead
| Instead of | Try |
|---|---|
| The singularity is coming. | Things are changing faster than usual. Let's watch it together. |
| Intelligence is becoming worthless. | Answers are getting cheap. Judgment, trust and care are getting precious. |
| AI will take our jobs. | Tasks are moving. Learning is our safety net, and we'll do it together. |
| You'll be left behind. | The earlier we start, the more choices we keep. |
| Trust me. | Here's my list. Grade me in December. |
/june-list
The June list, graded
On June 19 I wrote down 18 signs to watch for over six months. At month three, on September 25, I graded each one strictly: an announcement is not a deployment.
Happened 5
- A government restricts or licenses access to a model
- A major AI-authored math or science result, verified
- A big company credits AI for a leaner headcount while growing
- Data centers visibly straining the power grid
- An explicit national AI project
Partly 6
- AI doing workday-length (8-hour) tasks
- AI measurably speeding up AI research
- A deepfake draining an institution or changing a policy
- Large repeat orders for robots
- An AI tutor becoming the default somewhere big
- An AI-made culture hit
Not yet 7
- Nasdaq at 40,000
- The first $10 trillion company
- An AI-designed drug approved, or changing survival
- A wave of college closures, or elite colleges repositioning
- National income support for people displaced by AI
- Courts or newsrooms requiring proof of provenance
- Model weights stolen
Went the other way 0
None so far.
Two of the five that happened are shaky: one restriction was lifted weeks later, and the math result is still being checked. Final grades on December 19, 2026, on this page.
Use it
If you want to give a talk like this to your own family, take the rules. If you'd like the slides, write to me.
Made with Claude, which did the research, drew every frame in code and helped with the words. I chose what to say and gave the talk. The art builds on Portland in Pixels by VYBSRF, LLC. The captions in the pictures are set in Galmuri by Lee Minseo (SIL Open Font License).