# Beyond Endpoint Gains: A Weight-Delta Audit of Medical Specialization

arXiv: https://arxiv.org/abs/2608.20768 · HTML: https://mlsanity.com/paper/2608.20768/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 25 of its 40 references that are in the map.
- First posted on arXiv 21 Aug 2026.

## Builds on (most cited first, 25 of 25)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Measuring Massive Multitask Language Understanding](https://mlsanity.com/paper/2009.03300/) (2020)
- [On the Opportunities and Risks of Foundation Models](https://mlsanity.com/paper/2108.07258/) (2021)
- [Locating and Editing Factual Associations in GPT](https://mlsanity.com/paper/2202.05262/) (2022)
- [Language Models as Knowledge Bases?](https://mlsanity.com/paper/1909.01066/) (2019)
- [Sparse Autoencoders Find Highly Interpretable Features in Language Models](https://mlsanity.com/paper/2309.08600/) (2023)
- [Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing inference time](https://mlsanity.com/paper/2203.05482/) (2022)
- [What Disease does this Patient Have? A Large-scale Open Domain Question Answering Dataset from Medical Exams](https://mlsanity.com/paper/2009.13081/) (2020)
- [Transformer Feed-Forward Layers Are Key-Value Memories](https://mlsanity.com/paper/2012.14913/) (2020)
- [Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 small](https://mlsanity.com/paper/2211.00593/) (2022)
- [Steering Llama 2 via Contrastive Activation Addition](https://mlsanity.com/paper/2312.06681/) (2023)
- [Mass-Editing Memory in a Transformer](https://mlsanity.com/paper/2210.07229/) (2022)
- [Steering Language Models With Activation Engineering](https://mlsanity.com/paper/2308.10248/) (2023)
- [TIES-Merging: Resolving Interference When Merging Models](https://mlsanity.com/paper/2306.01708/) (2023)
- [Towards Automated Circuit Discovery for Mechanistic Interpretability](https://mlsanity.com/paper/2304.14997/) (2023)
- [How Much Knowledge Can You Pack Into the Parameters of a Language Model?](https://mlsanity.com/paper/2002.08910/) (2020)
- [Capabilities of GPT-4 on Medical Challenge Problems](https://mlsanity.com/paper/2303.13375/) (2023)
- [Merging Models with Fisher-Weighted Averaging](https://mlsanity.com/paper/2111.09832/) (2021)
- [Dissecting Recall of Factual Associations in Auto-Regressive Language Models](https://mlsanity.com/paper/2304.14767/) (2023)
- [Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Models](https://mlsanity.com/paper/2403.19647/) (2024)
- [HuatuoGPT-o1, Towards Medical Complex Reasoning with LLMs](https://mlsanity.com/paper/2412.18925/) (2024)
- [Attribution Patching Outperforms Automated Circuit Discovery](https://mlsanity.com/paper/2310.10348/) (2023)
- [Have Faith in Faithfulness: Going Beyond Circuit Overlap When Finding Model Mechanisms](https://mlsanity.com/paper/2403.17806/) (2024)
- [Medical Reasoning with Large Language Models: A Survey and MR-Bench](https://mlsanity.com/paper/2604.08559/) (2026)
- [Weight Patching: Toward Source-Level Mechanistic Localization in LLMs](https://mlsanity.com/paper/2604.13694/) (2026)

## Abstract (arXiv)

Specialist language models are usually understood through endpoint gains: the generalist scores lower, the specialist scores higher, and the difference is treated as evidence of specialization. This leaves the released update itself largely unexamined. We propose a paired weight-delta path audit and apply it to two public, aligned generalist-to-medical-specialist checkpoint pairs: Gemma-3-4B-IT to MedGemma-4B-IT and Qwen2.5-7B-Instruct to HuatuoGPT-o1-7B. In both pairs, the full decoder-side update strongly reconstructs measured medical benchmark movement (0.974 and 1.183 endpoint-normalized retention), making each decoder delta an appropriate substrate for the audit. Yet the movement is not cleanly localized. MLP is the strongest broad component family in both pairs, but mixed off-domain movements, 10-seed matched controls, and endpoint-anchored rollbacks prevent a unique coarse-family explanation. The audit therefore separates update-level reconstruction from component-level explanation. Its claims concern text-only multiple-choice benchmark movement, not clinical validation, repair, or circuit-level mechanism.