---
title: "Discussion on \"We ran HealthBench on our health AI's safety layer. It scored lower than the bare model.\" | Hashnode"
description: "Discussion on \"We ran HealthBench on our health AI's safety layer. It scored lower than the bare model.\". A 150-conversation HealthBench subset, a safety layer that costs points, one real dose leak we found while measuring, and what moved the score."
url: "https://hashnode.com/posts/we-ran-healthbench-on-our-health-ai-s-safety-layer-it-scored-lower-than-the-bare-model/6ac492c60bda01f274b983e0"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [We ran HealthBench on our health AI's safety layer. It scored lower than the bare model.](https://tabibu.hashnode.dev/we-ran-healthbench-on-our-health-ai-s-safety-layer-it-scored-lower-than-the-bare-model?utm_source=hashnode&utm_medium=feed)

A 150-conversation HealthBench subset, a safety layer that costs points, one real dose leak we found while measuring, and what moved the score.

Oct 6 · 6 min read · tabibu.hashnode.dev  
Read the article

-   [#ai](https://hashnode.com/tag/ai)
-   [#healthtech](https://hashnode.com/tag/healthtech)
-   [#llm](https://hashnode.com/tag/llm)
-   [#testing](https://hashnode.com/tag/testing)

## Written by

[John Alexander Kamau](https://hashnode.com/@itsxander-aj3)

@itsxander-aj3 · Nairobi, Kenya

  
[View profile](https://hashnode.com/@itsxander-aj3)

## Responses

Markdown supported · ⌘↵ to post

No responses yet.