GPT-4: Researchers have developed BIRD, a new text-to-SQL benchmark that closely resembles real-world scenarios, enabling novice data analysts to extract information from large databases using natural language. BIRD contains 95 large databases and 12,751 complex instances, covering 37 professional disciplines. The benchmark considers efficiency, encouraging the use of more effective query techniques in large and noisy database environments. Current models, however, still need help with generalizing effectively, as their performance lags behind human capabilities.
Read more at MarkTechPost…