Observations and Remedies for Large Language Model Bias in Self-Consuming Performative Loop

Yaxuan Wang, Zhongteng Cai, Yujia Bao +2

The rapid advancement of large language models (LLMs) has led to growing interest in using synthetic data to train future models. However, this creates a self-consuming retraining…