No. of Recommendations: 7
I'm guessing you're thinking of this paper:
arXiv: Prompt Architecture Determines Reasoning Quality:A Variable Isolation Study on the Car Wash ProblemThis was one independent researcher testing a prompt structure in his own app — not AI "overlords" fixing anything. Your Google answer was describing that paper, not AI in general.
When the car wash problem went viral, it landed on thousands of websites. All that text is what the next model gets trained on. So now it knows the answer — like a kid who aces the test because it got passed around. That's the likely reason it stopped being a problem.
This is still an active area of research. Not solved. That same researcher published a follow-up three weeks later where his fix stopped working once he put it into a real production prompt.
