Text-to-SQL generator takes top place on benchmark

The solution, based on IBM’s Granite code model, is part of a larger effort to bring generative AI into data services to help businesses extract fresh insights from large databases

From website clicks to sales reports, organizations are gathering and storing more data than ever. But tools for finding the information you need across databases, data warehouses, and data lakehouses, and transforming it into something useful, have not kept pace.

Many businesses fail to unlock the full value of their data because employees either can’t find what they’re looking for or can’t translate their questions into the code required to unlock the answers.

Generative AI is set to simplify the process, with large language models (LLMs) removing key roadblocks that currently make finding, retrieving, and transforming tabular data so difficult. SQL is the dominant language for interacting with databases. But in any given enterprise there are a limited number of people who understand how large databases are laid out and can query them in SQL, effectively limiting who can access the data to uncover insights to improve the business.

To open enterprise data to more users, IBM and others in tech have focused on teaching LLMs to write SQL, or structured query language. In a recent milestone, IBM’s Granite code model jumped to the top of the BIRD leaderboard that measures how well LLMs can parse a question in natural language and translate it to SQL, to be run on the real data to answer the question.

More information here