Chapter 2 of the video (5:29)
The questions asked
- How often do you keep prompting — "just one more prompt" — past the point you meant to stop? (
5:30) - How has your sleep changed since you started using AI heavily for coding — later bedtimes, racing thoughts, trouble waking? (
5:30,5:55)
The numbers
- 46% regularly blow past their intended stopping point — often or every single day, not once in a while (
5:55). - Among those who overshoot: 58% report sleep changed moderately or more (
5:55). - Among those who stop when they meant to: 17% (
6:23). - This is the strongest relationship in the entire survey (
6:23). - Caveat stated explicitly: this does not prove AI wrecked anyone's sleep — causation could run the other way or come from something else entirely; the two simply appear together (
6:23).
| Group | Sleep changed moderately or more |
|---|---|
| Kept prompting past stopping point | 58% |
| Stopped when intended | 17% |
The mechanism — variable reward
Dr. Courtney Tolinski's explanation (6:23, 6:53, 7:16):
- AI prompting is a variable reward system, structurally like a slot machine: sometimes a reward, most of the time not.
- The anticipation build-up is what keeps you pressing the lever.
- Each prompt carries a chance of the perfect solution; most outputs fall short, which produces the "let me just do it one more time" loop.
- Chasing the perfect solution induces dopamine, which feeds the addictive cycle.
flowchart LR
P["Send prompt"] --> R{"Perfect result?"}
R -->|"rarely"| D["Dopamine hit"]
R -->|"usually"| N["Almost there…"]
N --> A["Anticipation build-up"]
D --> A
A --> P