Mart: Improving LLM Safety with Multi-Round Automatic Red-Teaming

Nov 15, 2023 ·

This paper presents a method for efficient retrieval of large-scale documents from a massive corpus. It uses a novel probabilistic method for document retrieval, which combines the advantages of traditional vector space models and deep learning models. The model utilizes a new concept of "query expansion" to improve relevance scores for information retrieval tasks. This involves expanding the query terms into related topics, which can lead to better recall and improved precision. Furthermore, the model is scalable and can be used to efficiently search large datasets. In experiments, the model was able to achieve significantly higher effectiveness than conventional methods. This opens up possibilities for efficient retrieval of documents from massive collections of text.