[phillip_kerger]

I (author of the paper) can't speak for others, but I have no affiliation whatsoever with OpenAI and have not received anything from them (I even pay for my subscription lol). I've thrown this problem at each model with new model releases, and 5.6 was the first one to solve it, so that's my side of that story.

[throwaway1707]

Thanks for responding. I don't personally believe you have affiliation, more that OpenAI saw the problem was being worked on, gathered training data for solving it and training the model on it, then eventually you found it through the model (and now it's a big impressive thing for OpenAI)

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