Primary sourcearXiv: DeepSeek
View sourceEvent ·Official first-hand
VirtualSet:类型化本体世界作为LLM生成目标,实现有依据查询和受保护决策
VirtualSet提出将LLM的生成目标从SQL替换为类型化的set expressions,通过Generic Constraint Projection在执行前检测类型错误。在BIRD基准的1,072题测试集上,使用deepseek-reasoner模型,VirtualSet达到67.5%准确率,高于直接SQL的63.5%(McNemar exact p=0.00117)。在一个30个动作的防护测试中,VirtualSet成功拦截全部20个幻觉动作且零误报。该方法在保持SQL基准竞争力的同时,提供了写操作前的语义决策保障。
Current read
VirtualSet让LLM不用冒SQL幻觉的风险,改天用类型安全的set表达式,BIRD上准确率反超4个百分点,写操作还能拦截瞎编的数据输入。
2 sources47° AI heatLast updated 2026/7/21 07:57:17
Evidence
2 sourcesRelated source 1marktechpost
View sourceConflict check
Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.