A tutorial where you role-play training an AI reduces how easily AI can persuade you

Researchers created an interactive tutorial where people simulate building a large language model from scratch, learning how the system actually works. In experiments, people who did this tutorial were harder to persuade by AI than people who watched a standard explainer video.

The baseline assumption has been that people can't do much about persuasive AI except avoid it or install a detector. This paper suggests there's a third path: understand the system itself. If understanding how AI gets trained actually makes people more resistant to its persuasion, it means the defense against AI manipulation might not be technological (AI detectors) but educational. The catch: the sample is 274 people in a lab with realistic but controlled scenarios, not the millions of people who encounter persuasive AI daily. It's unclear whether this effect sticks after the tutorial ends, or whether it generalizes beyond the three scenarios tested.