Never the Number: Structural Abstention for AI Systems Whose Answers Are Consumed as Fact
绝非数字:将答案作为事实使用的AI系统的结构弃权
机构 * Apple Inc.(苹果公司)
AI总结 针对LLM文本转SQL系统返回错误答案且无法区分的问题,提出带生成式外壳的可信内核架构,即结构弃权,在生产案例中验证其可靠性优于两种生成式替代方案。
Comments 26 pages, 5 figures, 5 tables. Technical report. Describes architecture and design principles only; contains no code, schemas, datasets, or performance metrics