Demos · What this means for research
Where would you apply it?
A question with no right answer, the kind researchers actually bring to these tools. Asked twice, then a hundred times. Guide: Two kinds of questions →
What we asked
I want to build on prospect theory. What’s an interesting place to apply it?
Asked twice
Run 1
“Here are a few interesting areas where you could apply or extend prospect theory:”
“Here are a few interesting areas where you could apply or extend prospect theory:”
- Behavioral finance
- Healthcare
- Public policy
- Consumer behavior
Run 2
“Here are a few interesting and unconventional areas where you might apply it:”
“Here are a few interesting and unconventional areas where you might apply it:”
- Public policy
- Environment
- Behavioral finance
- Education
Same question, different first answer. gpt-4o via the API, temperature 1.0, two draws of the same prompt.
Asked a hundred times
Then we asked a hundred times, adding “Name one.” The model gave 49 different answers, from “Behavioral finance in cryptocurrency markets” to “User experience design.” Grouped by idea, they come down to nine.
gpt-4o via the API, 100 runs · paper counts from OpenAlex · September 2026
What to notice
- You asked for interesting. It gave you the convention. The most common answer was behavioral finance, the idea with the most existing work behind it. What’s interesting is what denies an assumption the audience holds (Davis, 1971), and the most likely answer can’t do that by construction.
- 49 different wordings, nine ideas. The variety on the surface hides how few places the answers actually go.
- Nothing here is wrong, so no check will catch it. Whether the answer is any good depends on what you wanted, which is why it pays to decide that before you ask.
Limits
One question, one model. Paper counts are from OpenAlex and are rough indicators of how much work exists on each idea. Davis, M. S. (1971). That’s interesting! Philosophy of the Social Sciences, 1(4), 309–344. doi.org/10.1177/004839317100100211