Synthetic Knowledge Ingestion: Towards Knowledge Refinement and Injection for Enhancing Large Language Models
This work addresses the problem of improving factual accuracy in LLM outputs for applications in domains like finance and biomedicine, representing an incremental advancement in knowledge injection techniques.
The authors tackled the challenge of refining and injecting knowledge into large language models (LLMs) by proposing a synthetic knowledge ingestion method called Ski, which significantly outperformed baseline methods on question-answering tasks across finance, biomedicine, and open-generation domains.
Large language models (LLMs) are proficient in capturing factual knowledge across various domains. However, refining their capabilities on previously seen knowledge or integrating new knowledge from external sources remains a significant challenge. In this work, we propose a novel synthetic knowledge ingestion method called Ski, which leverages fine-grained synthesis, interleaved generation, and assemble augmentation strategies to construct high-quality data representations from raw knowledge sources. We then integrate Ski and its variations with three knowledge injection techniques: Retrieval Augmented Generation (RAG), Supervised Fine-tuning (SFT), and Continual Pre-training (CPT) to inject and refine knowledge in language models. Extensive empirical experiments are conducted on various question-answering tasks spanning finance, biomedicine, and open-generation domains to demonstrate that Ski significantly outperforms baseline methods by facilitating effective knowledge injection. We believe that our work is an important step towards enhancing the factual accuracy of LLM outputs by refining knowledge representation and injection capabilities.