Project Lily: hundreds of contractors read real ChatGPT conversations.

404 Media reported September 14 that OpenAI's internal codename for its human-rating program is Project Lily, with contractors recruited through Crossing Hurdles and paid via Mercor, one reporting over $50/hour, reading real user prompts and rating replies on a 1-7 scale (404 Media). Internal documents show reviewers training ChatGPT not to anthropomorphize itself and to be less sycophantic. Contractors don't see usernames and OpenAI tries to strip personal information, but the company acknowledges sensitive details still reach reviewers, across a base of more than 900 million users.