Evaluating fairness in ChatGPT
OpenAI Blog
Researchers evaluated ChatGPT's responses to determine whether the model treats users differently based on names that signal demographic characteristics. The study used AI research assistants to preserve privacy while testing how the model responded to identical prompts from different named users. The results will inform efforts to identify and reduce demographic bias in how the system generates responses.
Why it matters
We've analyzed how ChatGPT responds to users based on their name, using AI research assistants to protect privacy.