research
The Medical Gain Reappeared in the Weight Delta but Wouldn't Stay in One Component
A paired rollback audit reconstructed benchmark movement without finding a unique coarse module that explained it.
Summary
A paired rollback audit reconstructed benchmark movement without finding a unique coarse module that explained it.
Across aligned Gemma-to-MedGemma and Qwen-to-HuatuoGPT pairs, the full decoder update strongly reconstructed measured medical multiple-choice movement. MLPs were the strongest broad family, but off-domain changes and matched controls prevented a single-component explanation; the study makes no clinical-validation claim.
Why it matters
A paired rollback audit reconstructed benchmark movement without finding a unique coarse module that explained it.
Limits and context
No additional limitation was separately recorded.
Key claims
A paired rollback audit reconstructed benchmark movement without finding a unique coarse module that explained it.
Evidence: source-2026-08-24-014
Sources
- arXiv preprint 2608.20768arXiv · primary research
Corrections
No corrections have been recorded for this story.