Various industries such as finance, meteorology, and energy produce vast amounts of heterogeneous data every day. There is a natural demand for humans to manage, process, and display data efficiently. However, it necessitates labor-intensive efforts and a high level of expertise for these data-related tasks. Considering large language models (LLMs) showcase promising capabilities in semantic understanding and reasoning, we advocate that the deployment of LLMs could autonomously manage and process massive amounts of data while interacting and displaying in a human-friendly manner. Based on this, we propose Data-Copilot, an LLM-based system that connects numerous data sources on one end and caters to diverse human demands on the other end. Acting as an experienced expert, Data-Copilot autonomously transforms raw data into multi-form output that best matches the user's intent. Specifically, it first designs multiple universal interfaces to satisfy diverse data-related requests, like querying, analysis, prediction, and visualization. In real-time response, it automatically deploys a concise workflow by invoking corresponding interfaces. The whole process is fully controlled by Data-Copilot, without human assistance. We release Data-Copilot-1.0 using massive Chinese financial data, e.g., stocks, funds, and news. Experiments indicate it achieves reliable performance with lower token consumption, showing promising application prospects.
View on arXiv