Imagine you have a relatively small (< 50 rows, < 10 columns) table with meaningful column names. This table could be a financial report for several companies, for example. What’s the best approach to derive a meaningful and insightful conclusions based on the data from this table?
I imagine it like that:
Input: “You are a data analyst… This data represents [context] Please describe and make overall conclusions from this table and compare presented companies: [table as text]”
Output: “This table represents revenue values for several large companies. The company with the most revenue is Apple, despite not profitable for the entire period except for Q1 2020…”
Table Question Answering models don’t fit because you should have a precise question and will get a precise answer.
Recent text generation/chat models are the closest to what I would like to get, but they are so terrible at math sometimes they convince me that 2 > 3 