How ChatGPT learns about the world while protecting privacy

OpenAI has outlined its practices regarding how ChatGPT learns about the world while safeguarding user privacy. To develop its frontier models, OpenAI utilizes a mix of information sources, including freely and openly accessible public internet content, partnership data, and information provided by users, contractors, and researchers. To protect individual privacy, OpenAI applies safeguards such as its internal Privacy Filter, designed to identify and mask personal information in text across training datasets and eligible user conversations. The company also offers the Privacy Filter for free to other developers. Furthermore, OpenAI provides multiple user controls to manage data. Users can turn off 'Improve the model for everyone' in their settings to prevent their chat history from being used for training. Temporary Chat allows users to have interactions that do not appear in chat history, do not create memories, and are deleted after 30 days. Additionally, the optional Memory feature allows ChatGPT to retain specific details across chats, though users can review, edit, delete, or entirely disable it. Users can also export data, delete accounts, or submit privacy requests through OpenAI's portal. While ChatGPT is designed to reject requests for private information, OpenAI acknowledges that mistakes can happen and provides channels for reporting inaccurate or inappropriate personal outputs. The company emphasizes that protecting privacy and addressing safety risks must work together as models continue to grow more capable.