OpenAI Fires Contractors for Using AI to Train Its AI

OpenAI's Project Lily hires contractors through Crossing Hurdles and Mercor to hand-score real ChatGPT conversations on a 1–7 scale, paying over $50 an hour. Using LLMs, Grammarly, or AI translation tools is strictly banned, and violators are removed from the project immediately. Training on user conversations is opt-in by default for Free, Plus, and Pro plans, with an opt-out in ChatGPT's data controls.

• Contractors earn over $50 an hour hand-scoring ChatGPT responses on a 1–7 scale
• LLMs, Grammarly, AI translators, and detectors like GPTZero are banned
• Training on user conversations is on by default for Free, Plus, and Pro
• Users can opt out via data controls in ChatGPT settings