People Training OpenAI's AI Fired for Using AI to Train the AI
3 hours ago
- OpenAI contractors hired to improve ChatGPT have been fired for using AI to train AI, an ironic twist given OpenAI's mission.
- Using AI-generated responses can lead to 'model collapse,' making AI models worse over time.
- Contractors are explicitly prohibited from using AI tools, including AI detection tools, Grammarly, and AI translation, per internal documents.
- Reviewers are instructed to identify AI use without telling evaluators why, and to judge overall patterns rather than single clues.
- Signs of AI use include repetitive words, AI-style punctuation (like overuse of em dashes), and unusually fast work completion.
- One contractor admitted using AI for a 'boost' and was terminated, feeling no joy in the work.
- Mercor, an AI-training company, enforces contracts against LLM use and removes violators immediately.
- Another contractor intentionally chose worst responses to sabotage AI training, feeling guilty and unsure of impact.
- OpenAI declined to comment on the firings, highlighting the tension between AI use and human training.