Loading page...Загружаем страницу...正在加载页面...
AtlasRepo

We ran HealthBench on our health AI's safety layer. It scored lower than the bare model

说明与用途

在 HealthBench 上测试 Tabibu 的安全层显示,安全层降低了基准得分,但通过策略调整可提升得分并发现剂量引用错误。

问题

Assessing the impact of a safety layer on a health AI’s performance on a public benchmark.

结果

The safety layer reduced benchmark scores due to its conservative policies, but policy tweaks improved performance while uncovering a safety bug involving prescription dose quoting.

工作流

  • Run HealthBench on a 150‑conversation subset with a gpt‑4.1‑mini grader
  • Compare scores of (1) full Tabibu pipeline, (2) bare gpt‑6‑luna, (3) bare gpt‑4o‑mini
  • Identify per‑criterion failures
  • Modify safety policies to allow general answers and tiered doses
  • Re‑run benchmark
  • Implement safety fix for prescription dose quoting
  • Re‑run benchmark

证据

The full pipeline scored 0.362 on 150 HealthBench conversations

来源

The pipeline was 0.158 below the bare model on the same conversations (paired difference, ±0.039).

来源

I changed one policy: let the assistant answer general, low‑risk questions from the model's own knowledge, with a clear label saying the answer is not from its curated sources, instead of refusing.

来源

With the original rules, the pipeline could quote prescription doses straight out of a retrieved national formulary

来源

  • 记录日期: 2026-10-06T06:41:11.777Z