As text-to-SQL systems move beyond demonstrations toward real-world deployment, ambiguity in user questions becomes a primary source of errors.
gpt-6-sol achieves only 51% end-to-end execution accuracy, and no open-source model exceeds 27%.
As text-to-SQL systems move beyond demonstrations toward real-world deployment, ambiguity in user questions becomes a primary source of errors.
gpt-6-sol achieves only 51% end-to-end execution accuracy, and no open-source model exceeds 27%.