论文：低成本验证器 Jev 在 RAG 事实核验上接近 GPT-6 Astra
做 RAG 的朋友看看这篇：Jev 成本只有 GPT-6 Astra 的 1/187，准确率几乎持平，但它漏放无依据主张的比例高不少，选型时得盯这个指标。
一篇论文测试了只返回标签和概率的小型验证器 Jev，与 GPT-6 Astra 在 495 条人工标注的 RAG 主张上对比。Jev 的准确率与 Astra 相差仅 0.6 个百分点，成本低 187 倍，但它放行了 23.2% 的无依据主张，Astra 为 14.1%。论文还搭建了 Jev 先答、疑难转 Astra 的级联方案，虽降低了 Astra 账单，但随机分流同样比例的主张也能省差不多。作者建议选型时重点测量验证器放行无依据主张的比例，而不是只看准确率。
This paper finds that a cheap verifier can match a frontier LLM's accuracy at checking RAG claims, while letting more unsupported claims slip through.
RAG pipelines keep asking whether evidence supports a claim, a simple yes or no. Most teams get that answer from a text-generating LLM at generation prices.
They tested Jev, which returns only a label and a probability, against GPT-6 Astra on 495 human-labelled claims. They also built a cascade where Jev answers confident claims and passes the rest to Astra.
Jev landed within 0.6 points of Astra's accuracy at 187 times lower cost, but it approved 23.2% of unsupported claims against Astra's 14.1%. The cascade also cut Astra's bill, yet sending the same share of claims to Astra at random saved about as much.
Before picking a claim checker for a RAG product, measure how often it approves unsupported claims, since headline accuracy hides that.
