Hello OpenAI Community!
I would like to suggest an improvement to ChatGPT’s random-number selection behavior.
My observation
I tested the prompt “pick a number from 1 to 30” across fresh chats and guest mode. In 10 reported trials, the response was 17 every time. I have two screen recordings and a PDF evidence report.
I understand that this is a small experiment and does not prove that all ChatGPT sessions behave this way. However, I believe the observation deserves further investigation.
My proposal
When users explicitly request a uniformly random choice, ChatGPT should consider using a suitable random-number generator rather than relying solely on ordinary text generation.
The system should validate the requested range, be tested for unintended statistical bias, and undergo regression testing after relevant updates.
Why this matters
People use ChatGPT for games, classroom activities, and everyday decisions. A number that sounds random is not necessarily selected uniformly at random.
I hope OpenAI will consider evaluating this behavior and exploring a reliable solution that benefits all users.
I welcome feedback on how to improve the experiment and test the idea more rigorously.
Thank you!