Two new research papers explore advancements in text-to-SQL capabilities, focusing on multi-turn interactions and table retrieval. The first paper introduces CORE-T, a training-free framework that uses LLM-generated metadata and a compatibility cache to improve table selection for SQL queries, showing significant gains in accuracy and efficiency. The second paper presents EnterpriseMem-Bench, a new benchmark for multi-turn text-to-SQL, and evaluates several frontier models, revealing that stateless interactions quickly degrade performance and highlighting varying memory architecture impacts across models, including a surprising regression in one Claude model. AI
IMPACT These studies advance text-to-SQL by improving table retrieval and evaluating multi-turn memory architectures, potentially enhancing enterprise analytics.
RANK_REASON Two academic papers published on arXiv detailing new methods and benchmarks for text-to-SQL systems.
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
