Michael Stonebraker finds Text-to-SQL accuracy at 10% on real enterprise data

Michael Stonebraker, a Turing Award winner, evaluated text-to-SQL capabilities on a real enterprise data warehouse. His findings revealed a low accuracy rate of only 10%, significantly underperforming against an 80% benchmark. This result reinforces Stonebraker's past critiques of database technologies and suggests current text-to-SQL systems struggle with real-world enterprise data complexity. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Highlights significant limitations in current text-to-SQL accuracy for enterprise applications, suggesting a need for further research and development.

RANK_REASON Academic research/evaluation of a specific AI capability (text-to-SQL) on real-world data.

Read on Mastodon — mastodon.social →

• Michael Stonebraker
• CODASYL
• NoSQL
• Text-to-SQL

• paper
• other