I am a UK-based, highly experienced real-world ChatGPT user, currently completing the final year of an Open University Bachelor of Laws (Honours) degree.
Through intensive use of ChatGPT, I have found that I am particularly effective at identifying model failures including factual errors, contextual regressions, inadequate source verification, failure to preserve established premises, generic responses that do not answer the precise question asked, and situations where lengthy AI interaction consumes more user time than it saves.
I tend not merely to identify that an answer is wrong, but to investigate why it failed, challenge the reasoning, preserve evidence and test whether subsequent responses actually resolve the underlying problem.
I have several recent, documented real-world examples involving materially different failure modes which I would be willing to provide for evaluation.
I believe this kind of critical real-world use could be useful to OpenAI in model evaluation, realistic prompt design, failure analysis and product improvement. I would therefore be interested in contributing evaluation prompts, documented failure cases and substantive qualitative feedback on an ongoing basis.
I am not principally seeking financial compensation. I would be happy to contribute alongside my university studies in return for appropriate continued access to ChatGPT. Equally, if OpenAI considered my experience useful for paid part-time or full-time evaluation, testing or related work, I would be open to discussing that.
I have already raised this proposition with OpenAI Support. They have documented it but have confirmed that Support has no internal referral route through which an individual proposal of this kind can be submitted to Product, User Research or another appropriate team.
My question is therefore simple: is there an appropriate OpenAI Human Data, Model Experience, Model Behavior, Evals, User Research or related route through which I could submit this proposition and some documented examples for consideration?
I would be grateful for guidance from an OpenAI staff member or anyone familiar with the appropriate route.