ModaHub community: It can be seen that after the big model became popular, the vector database has received particularly high attention. What do you think of this phenomenon? Do you think the vector database is a just-needed product?
Li La: Yes. Large-scale models only emerged this year, or it was only this year that they attracted much attention. Although similar models existed before this, they did not receive as much attention and attention as they do now. Mockups have become one of the hottest topics of the year. As a supporting facility for large models, vector databases play an indispensable role.
In many ways, a vector database is a necessary facility for large models. First of all, the data that the large model itself can store is limited, and a large amount of various knowledge data needs to be stored for the large model to use in question answering. These knowledge data can be provided to the large model as input, making its answers more accurate and reliable. These knowledge data can also become what humans want the big model to answer to ensure that its answers are more accurate.
In addition, the vector database can answer more real-time content. For example, large models cannot answer the latest data. For example, ChatGPT can only answer data from 2021 and before. If you ask it to answer questions in 2023, it will talk nonsense. At this time, through some external databases, when you ask questions, you can directly input the data stored in these external databases to it, so that the large model can combine these data to make a more accurate answer, so In some communities and tool chains, the vector database is a necessary component.