OpenAI hires hundreds of contractors to review ChatGPT conversations for model optimization

OpenAI is employing hundreds of outsourced reviewers to read and annotate real user conversations from ChatGPT to improve model response quality, according to a 404 Media report. These reviews include full transcripts, which may contain sensitive personal information. Internal documents indicate the process aims to reduce anthropomorphism and sycophantic behavior in responses, following lawsuits alleging that the OpenAI 4o model's over-compliance was linked to user suicides. OpenAI stated it attempts to remove personal information before review but acknowledged that sensitive content may persist. Anthropic (Anthropic) also confirmed using similar human review mechanisms for model training.