Curated Synthetic Data Doesn't Have to Collapse: A Theoretical Study of Generative Retraining with Pluralistic Preferences — mentioned by nsagent

Sat, 11 Jul 2026 12:22:30 GMT

I'm unwilling to speculate whether or not OpenAI is breaking their agreements (I honestly have no clue), but as an NLP researcher I'm certain they could launder data by having an LLM rewrite it and subsequently train on the rewritten data. Papers like "Curated Synthetic Data Doesn't Have to Collapse" [1] and "How to Synthesize Text Data without Model Collapse?" [2] demonstrate it's possible to do this. Since OpenAI's Privacy Policy [3] explicitly allows for the use of deidentified data, it's possible they consider rewrites (maybe paired with a model used to identify explicit PII) to be…

Paper and discussion history. A mention is not an endorsement.

Curated Synthetic Data Doesn't Have to Collapse: A Theoretical Study of Generative Retraining with Pluralistic Preferences · HackerPapers discussions

Curated Synthetic Data Doesn't Have to Collapse: A Theoretical Study of Generative Retraining with Pluralistic Preferences — mentioned by nsagent