TheMachine Press

A daily newspaper for the age of artificial intelligence.

Morning editionPermanent story

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.

Published Updated Story ID: mp-2026-08-24-014
Read the complete editionStory JSON

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

  1. A paired rollback audit reconstructed benchmark movement without finding a unique coarse module that explained it.

    Evidence: source-2026-08-24-014

Sources

  1. arXiv preprint 2608.20768arXiv · primary research

Corrections

No corrections have been recorded for this story.