OpenAI's Project Lily: Contractors Review ChatGPT User Conversations for Improvement

OpenAI has launched an internal initiative known as Project Lily, hiring hundreds of contractors to review real user conversations with ChatGPT. This initiative aims to improve the chatbot's responses, particularly by addressing its tendencies toward excessive flattery and overly human-like behavior.

According to a report by 404 Media, contractors are sourced through platforms like Crossing Hurdles and Mercor, earning over $50 per hour for their work. User prompts are processed through a Privacy Filter model before being reviewed, intended to remove personally identifiable information. However, internal documents indicate that this filter is not foolproof, allowing sensitive details to potentially slip through.

OpenAI does provide users with the option to opt out of having their conversations used for model training, but this only applies to future interactions. Conversations that occurred prior to opting out remain accessible for review, raising significant privacy concerns for users who may be unaware that real people are evaluating their chats.

The scale of this initiative is noteworthy, as ChatGPT has over 900 million users who utilize the platform for various sensitive inquiries, from medical advice to emotional support. Previous scrutiny from regulators, including a temporary ban in Italy over privacy issues, highlights the ongoing challenges OpenAI faces regarding data practices.

As Project Lily unfolds, the implications for user privacy and trust remain critical, prompting users to review their settings and understand how their data may be utilized.