Part of me agrees with the other comments that it sounds absurd, especially because of how involved/intricate your prompting was. But given you have been prompting this problem for a year, OpenAI could easily have seen your attempts/progress, and worked towards solving it with human intelligence that the agent was then trained on. Given the lengths these companies go to for training these things (Meta literally reorganizing their engineering department to provide training data for engineering problems), and also how much difficulty these models have with problems outside their dataset, I really have to wonder if it's the model being "intelligent" or if it has been trained on it