OpenAI uses hundreds of contractors to review ChatGPT conversations
OpenAI employs hundreds of contract workers to examine real ChatGPT conversations and help improve the chatbot, according to an investigation by 404 Media based on leaked internal documents and interviews. Reviewers score responses from one to seven and assess whether the system is excessively flattering or too human-like. Some conversations reportedly include users asking that their content remain private.
The prompts are anonymized, but they may still contain sensitive personal information. OpenAI uses a privacy filter and acknowledges that it can make mistakes. According to the report, workers are recruited through Crossing Hurdles and paid through Mercor. One reviewer in North America said the pay exceeded $50 an hour.
Users can disable the “Improve the model for everyone” setting, which is enabled by default, to prevent new chats from being used for training and potentially reviewed by people. Temporary Chat is described by OpenAI as not using input data for model training. The company discloses possible access by authorized personnel and service providers in an FAQ. Anthropic and Google also use human review for some conversations or responses, subject to their respective privacy settings and policies.
Written by Sakhanda Wire AI from reporting by The Decoder. Facts, quotes and full context are in the original article below.