OpenAI Contract Workers Review ChatGPT Chats

Hundreds of contract workers are reading real ChatGPT conversations as part of OpenAI's effort to fine-tune the chatbot's behavior, according to a 404 Media investigation based on leaked internal documents and interviews with sources familiar with the work. These reviewers score ChatGPT's responses on a scale of one to seven, flagging things like excessive flattery or responses that come across as too human. The workers are recruited through a company called Crossing Hurdles, which markets itself as "connecting skilled professionals with the future of AI work," and paid through Mercor, an AI training firm. One North America-based reviewer told 404 Media they make more than $50 an hour for the work.

The prompts they review are anonymized, but that doesn't mean they're scrubbed of sensitive information. Some of the conversations 404 Media examined show users asking ChatGPT directly to keep what they're discussing private, seemingly unaware that a human being might end up reading it anyway. OpenAI does run a privacy filter over these chats, but the company itself acknowledges the filter isn't foolproof. That gap between what users assume and what actually happens is where the story starts.