Fairness Feedback Loops: Training on Synthetic Data Amplifies Bias — mentioned by dartos

Fri, 04 Oct 2024 09:07:09 GMT

Sure, hand wave away my entire comment as “nonsense” and ignore how statistics works. Training a model on synthetic data (obviously) increases bias present in the initial dataset[1], making for poor training data. IIRC (this subject is a little fuzzy for me) using synthetic data for RLHF is equivalent to just using dpo, so if they did RLHF it probably wasn’t with synthetic data. They may have gone with dpo, though. [1] https://arxiv.org/html/2403.07857v1

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

Fairness Feedback Loops: Training on Synthetic Data Amplifies Bias · HackerPapers discussions

Fairness Feedback Loops: Training on Synthetic Data Amplifies Bias — mentioned by dartos