Rubric-based medical AI benchmarks fail to penalize fabricated citations, paper finds
MaziyarPanahi · x · 2026-09-14
- 新论文《When Rubrics Fail》发现:在 3 个成熟医学基准上,临床上有意义的幻觉可以完全不改变 rubric 评分。
- 作者实测 31 道 HealthBench 题目明明写了禁止伪造引用的规则,但注入的虚假编造引用仍没有被评分器扣分。
- 结论:高分可能掩盖糟糕答案,呼吁公开查看模型实际输出,并把评测的 grader 本身也纳入测试。代码与数据已公开。
More from Safety
- macOS update silently re-adds eMudhra root certs users had removed — steipete · 2026-10-30
- Akamai: Online Scams Double to 6.4M, Overtake Phishing as Attack Windows Shrink to 22 Days — shashib · 2026-10-30
- Anthropic paper: 0.1% pretraining poison plants RCE backdoor in 4B CLI agent — omarsar0 · 2026-10-30
- Cryptographer Matthew Green slams Anthropic's 'model escape' framing of agent worm research — matthew_d_green · 2026-10-30
- Pope Leo calls to 'disarm' AI and urges UN to ban lethal autonomous weapons — The Verge AI · 2026-10-30
- 100 open-source AI security tools challenge kicks off with DirectGuard, a sub-ms prompt injection guard — Venom943 · 2026-10-30