DC Decoded
One Small Default Changed the Outcome, 65% Shorter Answers
Early results from a study testing whether a single instruction can make AI get to the point faster, without switching models or learning prompt engineering.
Most AI answers are longer than they need to be. So I ran an experiment.
For a UCLA Anderson and UCLA Environment sustainability project, I tested whether a single instruction could make AI get to the point faster. Eleven people. ChatGPT, Claude, and Gemini. The responses became dramatically, 65%, shorter while still answering the question.
That surprised me. Because nobody changed models. Nobody learned prompt engineering. Nobody changed their habits. One small default changed the outcome. As a side effect, the systems used substantially fewer tokens as well.
Now I'm expanding the study to 100 participants. I'm curious whether we can apply ideas from behavioral science and choice architecture to help people use AI more efficiently, without asking them to do extra work.