Large Language Model Can Continue Evolving From Mistakes

arXiv:2404.08707v1 Announce Type: new
Abstract: Large Language Models (LLMs) demonstrate impressive performance in various downstream tasks. However, they may still generate incorrect responses in certain scenarios due to the knowledge deficiencies and the flawed pre-training data. Continual Learning (CL) is a commonly used method to address this issue. Traditional CL is task-oriented, using novel or factually accurate data to retrain LLMs from scratch. However, this method requires more task-related training data and incurs expensive training costs. To address this challenge, …

abstract arxiv continual cs.ai cs.cl cs.lg data generate however issue knowledge language language model language models large language large language model large language models llms mistakes performance pre-training responses tasks training training data type