Excel is one of the office applications most important and most used from Microsoft, although it is also one of the ones that has caused the most headaches for its users over the years it has been with us. Its value is difficult to calculate because the foundation of many companies depends on this tool and many jobs revolve around it.

It’s no secret that Microsoft is betting big on AI, and this technology has been gradually applied to its software ecosystem, which includes both Windows operating systems and its core applications and tools. Excel will be the next to fully embrace AIthanks to the new SpreadsheetLLM framework.

This framework uses large language models analyze and interpret table dataand is designed to overcome the challenges these types of models commonly face when working with these types of documents, including aspects such as variety of formats, flexible templates, and two-dimensional grids. It achieves this by serializing data and incorporating cell addresses, values, and formats into the same data stream.

SpreadsheetLLM also comes with a tool to do this compress tables to facilitate its interpretation and which consists of a total of three different modules that carry out specific and perfectly planned work:

• The module analyzes the table structure and removes content that is not in the tables.
• The other transforms the data and converts it into a much more efficient representation.
• The third module is the one that deals with the process of aggregating all the data in the table.

This great language model works, although it has some limitations because it is unable to distinguish the color used in the cells, which in many cases means ignoring an important source of information that can serve as a differentiator. Also it lacks the capacity for semantic compression for cells containing natural language.

Despite these shortcomings, it seems to be quite efficient and clearly outperforms other solutions such as GPT-4, which outperforms by 25.6%, and also reduces the number of tokens needed to encode tables by 96%, which means lower resource consumption and lower computational costs .

Many fear that the arrival of artificial intelligence in Microsoft Office, and in this case Excel, it can destroy a lot of jobs. At the moment it doesn’t seem like that will be the case, as right now an expert on the said tool who knows its key functions is still a very sought after profile and should be better than this AI, but in the long run we may encounter scenarios, when will this technology end up devouring the human worker.

AI generated cover image.