OpenAI has hired hundreds of contractors to read real conversations in ChatGPT and rate them on a scale from one to seven. This is reported by The Decoder, citing 404 Media.
According to the account, this evaluation is used, among other things, to reduce sycophancy and excessively human-like behavior by the system. Requests are anonymized but may still contain sensitive information.
Users who do not want their conversations reviewed by humans must manually disable the 'Improve the model for everyone' setting, which the source description states is enabled by default.
editorial commentary
Why it matters
If the description is confirmed, users will need to more carefully check settings related to the use of their conversations for system improvement. The nearest observable signal is a public comment from OpenAI regarding human review, opt-out settings, and the protection of sensitive data. Significant uncertainty remains due to the single source and the lack of a primary statement in the package.