OpenAI has hundreds of contract workers reading your ChatGPT conversations
decoder ·
OpenAI employs hundreds of contract workers to review real ChatGPT conversations in order to improve the chatbot's responses, according to an investigation by 404 Media. These reviewers rate interactions on a scale of one to seven and are specifically tasked with reducing excessive flattery and overly human-like behavior in the model's output. Although prompts are anonymized, they can still contain sensitive personal data, and some users have explicitly requested privacy within their chats. The contract workers are recruited through Crossing Hurdles and paid via Mercor, with some earning more than $50 an hour. By default, OpenAI enables a setting called "Improve the model for everyone," which allows human review unless users actively disable it. This disclosure is buried in an FAQ page that has been online since at least 2023, leading critics to argue the notice is too vague to ensure informed consent. Other major AI labs, including Anthropic and Google, utilize similar human review practices for their models when users opt into data sharing programs.
decoder