“…We believe retrieval augmented methods are crucial for LLMs. As future work, one idea is to combine passages and structured knowledge, such as databases (Vougiouklis et al, 2023) and/or knowledge graphs, possibly including uncertain knowledge graphs (Pan et al, 2005;Stoilos et al, 2006;Qi et al, 2007;Chen et al, 2019), knowledge graph with selected vocabulary (Wang et al, 2014) and temporal knowledge graphs (García-Durán et al, 2018;Bourgaux et al, 2021), for retrieval augmented generations. Furthermore, as suggested in (Pan et al, 2023a), there are a few pressing challenges in this space, such as unifying knowledge editing (Mitchell et al, 2022;Han et al, 2023) and retrieval augmentation, complex reasoning via retrieval augmentation and semi-parametric LLMs (Pan et al, 2023b) in general.…”