Microsoft has developed a new large language model, SpreadsheetLLM, that may cause data analysts and accountants to become a little anxious about their employment prospects in the future.
It claims to have the “potential to transform spreadsheet data management and analysis, paving the way for more intelligent and efficient user interactions,” and has unveiled the first features of SpreadsheetLLM, a novel model that is “highly effective across a variety of spreadsheet tasks.”
Jokes implying that “Karen might be out of a job soon” began to appear on X after a pre-print paper about the model was covertly distributed at the end of last week.
Also Read: Best Microsoft AI Courses and Certifications for FREE in 2024
According to one customer, “SaaS is in deep, deep trouble.” “It’s going to be huge for the finance world,” wrote another.
Associate Professor Ethan Mollick of the University of Pennsylvania’s Wharton School tweeted: “This is another indication that LLMs will soon be able to work with both structured and unstructured spreadsheet data. Numerous use cases (projections, financials, appraisals, etc.) will become available as a result, and having a spreadsheet source of truth tends to reduce hallucinations.”
Spreadsheets, “characterized by their extensive two-dimensional grids, flexible layouts, and varied formatting options, which pose significant challenges for large language models (LLMs),” have proven to be a difficult task for LLMs thus far, according to a Microsoft team report.
“In response, we introduce SpreadsheetLLM, pioneering an efficient encoding method designed to unleash and optimize LLMs’ powerful understanding and reasoning capability on spreadsheets,” it stated.
Also Read: Meet GraphRAG: Microsoft’s New Graph-Based AI Method for Superior Data Insights
Taking on tokens: A fresh take on spreadsheets
The fact that too many tokens (the fundamental units of information the model processes) might slow down LLMs is one of the issues with employing them in spreadsheets. Microsoft created SheetCompressor, an “innovative encoding framework that compresses spreadsheets effectively for LLMs,” to address the issue.
“It significantly improves performance in spreadsheet table detection tasks, outperforming the vanilla approach by 25.6% in GPT4’s in-context learning setting,” said Microsoft.
The three modules that comprise the model are data-format-aware aggregation, inverse index translation, and structural-anchor-based compression.
In the first of these modules, “structural anchors” are inserted into the spreadsheet to facilitate a better understanding of the situation by the LLM. Afterward, “distant, homogeneous rows and columns” are eliminated, creating a simplified “skeleton” outline of the table.
Also Read: Pixel 9 to Feature New AI Capabilities Similar to Microsoft’s Recall
The problem of spreadsheets with a lot of empty cells and repeating data that consumes too many tokens is addressed by index translation.
“To improve efficiency, we depart from traditional row-by-row and column-by-column serialization and employ a lossless inverted index translation in JSON format,” said Microsoft. “This method creates a dictionary that indexes non-empty cell texts and merges addresses with identical text, optimizing token usage while preserving data integrity.”
When neighbouring numerical cells have comparable number formats, it presents another challenge for LLMs.
Also Read: Microsoft’s Suleyman Sparks Debate on AI Training Using Internet Content
“Recognizing that exact numerical values are less crucial for grasping spreadsheet structure, we extract number format strings and data types from these cells,” said Microsoft. “Then adjacent cells with the same formats or types are clustered together… streamlining the understanding of numerical data distribution without excessive token expenditure.”
Microsoft discovered that SheetCompressor dramatically lowers token usage for spreadsheet encoding by 96% following a “comprehensive evaluation of our method on a variety of LLMs.”
Furthermore, SpreadsheetLLM performs the “fundamental task of spreadsheet understanding,” or “exceptional performance in spreadsheet table detection.”
By introducing a framework known as the “Chain of Spreadsheet” (CoS), which can “decompose” spreadsheet reasoning into a table detection-match-reasoning pipeline, the new LLM expands upon the Chain of Thought methodology.
“Chain of Spreadsheet, the framework’s extension to spreadsheet downstream tasks, illustrates its broad applicability and potential to transform spreadsheet data management and analysis, paving the way for more intelligent and efficient user interactions,” said Microsoft.
Also Read: Microsoft Unveils ‘Skeleton Key’ Attack Exploiting Generative AI Systems
