A team of programmers and artificial intelligence specialists (AI) from Microsoft developed innovative tool called Spreadsheetllm, which uses the possibilities of large language models (LLM) to the spreadsheet. This invention can radically change the methods of use and analysis of data in business and other areas.
At the basis of Spreadsheetllm is the technology of SheetCompressor – a unique coding system that effectively compresses the electric tables for their processing with artificial intelligence. Until now, models such as ChatGPT have experienced difficulties with the interpretation of data in the format of electric tables due to their specific structure and functionality.
Sheetcompressor works in three stages: compression, translation and aggregation of data. At the compression stage, special “anchors” are added to the table that help AI analyze the structure and purpose of the table. Then the lines and columns are replaced by a simplified scheme. At the translation stage, empty cells and repeating values are removed. The final stage includes the use of a reverse index without loss in JSON format for aggregation.
The developer team also added additional modules for processing special situations, for example, for working with neighboring cells containing similar numerical formats.
The created tool allows large language models to work effectively with the spreadsheet tables as with data sources. This applies not only to data entry automation, but also to their analysis, as well as the presentation of complex information in an understandable form for users with different levels of training.
The study of the Microsoft team, describing Spreadsheetllm and Sheetcompressor, has already been published on the ARXIV preprint server. The scientific community has yet to get acquainted with this technology and evaluate its potential for the development of artificial intelligence and data processing.