open-compass / open-compass/opencompass

MedFailBench integration proposal for clinical AI safety evaluation

Open
#2,516 2 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
7.5k
Forks
869
Avg merge
17h 52m
Merged PRs (30d)
13

Description

Hello OpenCompass team,

I am a physician from Turkey building MedFailBench, an open source clinical AI safety benchmark based on synthetic clinician authored cases.

Project: https://github.com/goktugozkanmd/medical-ai-failure-atlas

OpenCompass looks like the right technical route if MedFailBench should become a reusable clinical safety evaluation task rather than only a standalone repo.

I want to explore a China Turkey collaboration around multilingual clinical AI safety evaluation. A concrete first step could be a small MedFailBench adapter or task, followed by a shared model evaluation report and coauthored paper.

I can contribute the initial cases, safety taxonomy, scoring files, and integration work.

No patient data is involved. This is research infrastructure for safer medical LLM evaluation, not clinical advice or deployment.

Would your team be open to discussing a joint pilot?

Best,
Goktug Ozkan

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by reviewing OpenCompass's task and benchmark integration conventions, then inspect the linked MedFailBench repository and its cases, safety taxonomy, and scoring files. Clarify the pilot's scope with the issue author before implementation. Done would mean an agreed adapter or task integrated into OpenCompass and a reproducible clinical safety evaluation.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning, testing-qa
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Quiet
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.