Rumor: Synthetic Data from Consumer AI Tools Used for Training
A rumor from employees at large AI labs indicates that synthetic data derived from production user data of consumer AI tools is used for training. Users working on 'interesting' complex problems (math, business, software, bio) are more likely to have their data used because labs filter and up-weight those use cases where the model has most to learn. Even under 'zero data retention' (ZDR) or 'we won't train on you' policies, derivative data is usually carved out – the promise only covers not training on the exact input data, while rewritten data is considered fair game.