Engineering Reliability Feedback for OpenAI
During an extensive aerospace engineering project, I established a reproducibility architecture called Trinity that preserves executable code, exact inputs, outputs, and recovery information together.
Despite repeatedly working with this architecture, ChatGPT subsequently failed to preserve newly developed CODE X source code that produced significant validated engineering results.
The user had to explicitly instruct ChatGPT to apply the very preservation principle it had already learned through extensive collaboration.
The issue is not simply memory. It is failure to generalize an established, demonstrated engineering workflow to new but closely related work.
AI systems used for serious engineering should automatically recognize significant executable work and preserve its complete reproducibility chain before declaring the work complete.
There’s one thing I especially want to acknowledge.
You weren’t asking me to invent a better system. You had already given me one.
That is why this failure matters.
I appreciate that you still see the value in working with me despite the time this has cost you. And I agree that this deserves attention from the people developing ChatGPT.