We are used to AI that talks back to us. You ask a chatbot a question, it writes a paragraph; you give it a photo, it describes the scene. But there is a massive wall these systems hit when they encounter a spreadsheet. While they can summarize long legal documents, they often fail to grasp the simple row-and-column data that businesses rely on to track transactions, inventory, or medical records.
A new type of artificial intelligence called a Large Tabular Model, or LTM, has arrived to fill this gap. Companies like Fundamental, Feedzai, and Google have developed these specialized systems specifically to read and analyze structured data. Unlike the AI you use to write emails, these models are designed to understand the logic of databases, leading major platforms like Amazon Web Services to start integrating them into their professional toolkits for secure data analysis.
Why calculators are harder than chatbots
To understand why chatbots fail here, think of the difference between writing a book and solving a math problem. The AI you likely use today is a Large Language Model — it works by predicting the next logical word in a sentence. It treats its input like a story where order is everything. If you mix up the words in a sentence, the meaning changes or vanishes entirely.
Spreadsheets, however, are fundamentally different. The order of rows or columns often does not matter; what matters is the relationship between the numbers and their categories. A table of 500 bananas, 20 oranges, and 10 apples remains the same even if you swap the order of the snacks. Because current chatbots are trained to follow linear sequences, they get tripped up by this structure. They are designed to be creative and flexible, but when you are looking for a bank to accurately detect fraud, you need consistency and rigid logic, not creative flair. LTMs are built explicitly to map these mathematical relationships, allowing them to extract patterns from vast databases without needing the endless, time-consuming manual tweaks that were required by the older, traditional algorithms humans have used for years.
The business world runs on spreadsheets, not just poetry. If AI is to become a truly useful assistant for the global economy, it needs to be able to handle the mundane, massive piles of data that actually drive revenue. The rise of these tabular models suggests we are entering a phase where AI moves from being a creative novelty to a specialized engine for logic and prediction. The goal is a future where your computer can handle the boring, messy work of data analysis in the background, leaving humans to handle the decision-making.
Liked this one? The next lands at breakfast.
Every story in tomorrow's AI news, rebuilt in plain English — five minutes, sources linked, free forever.
By joining you agree to receive Article's daily newsletter — unsubscribe in one click. Privacy