AI News HubLIVE
原文

Mechanistic origins of catastrophic forgetting: why RL preserves circuits better than SFT?

Recent work shows RL retains prior capabilities more effectively than SFT. This paper extends to the mechanistic level, introducing differential circuit vulnerability to measure circuit degradation. On Qwen2.5-3B-Instruct for scientific QA, SFT adapts faster but causes greater circuit disruption and forgetting, while RL preserves circuits at the cost of slower adaptation. Results suggest circuit preservation explains RL's robustness against catastrophic forgetting.

Article intelligence

EngineersAdvanced

Key points

  • SFT adapts quickly but disrupts internal circuits, leading to catastrophic forgetting.
  • RL preserves more of the base model's circuits, resulting in less forgetting but slower task adaptation.
  • New metric, differential circuit vulnerability, quantifies circuit degradation during fine-tuning.
  • Study on Qwen2.5-3B-Instruct scientific QA model confirms mechanistic trade-off.

Why it matters

This matters because SFT adapts quickly but disrupts internal circuits, leading to catastrophic forgetting.

Technical impact

May affect model selection, inference cost, product capability, and evaluation benchmarks.

[2605.28860] Mechanistic origins of catastrophic forgetting: why RL preserves circuits better than SFT?

[Submitted on 21 May 2026]

Title:Mechanistic origins of catastrophic forgetting: why RL preserves circuits better than SFT?

View a PDF of the paper titled Mechanistic origins of catastrophic forgetting: why RL preserves circuits better than SFT?, by Jeanmely Rojas Nunez and 6 other authors

View PDF HTML (experimental)

Abstract:Fine-tuning large language models (LLMs) frequently induces catastrophic forgetting of prior capabilities. Recent work has shown that reinforcement learning (RL) retains prior capabilities more effectively than supervised fine-tuning (SFT), attributing this to policy-gradient updates remaining closer to the base policy \cite{shenfeld2025rl}. We extend this behavioral account to the mechanistic level and ask whether RL's advantage is mirrored by stronger preservation of internal computational circuits. We introduce differential circuit vulnerability, a head-level measure of how much a circuit degrades under fine-tuning, and use it to compare RL and SFT on Qwen2.5-3B-Instruct adapted to scientific question-answering. We find a clear mechanistic trade-off: SFT adapts more rapidly to the target task but produces substantially greater circuit disruption and forgetting of prior capabilities, whereas RL preserves a larger fraction of the base circuit at the cost of slower task adaptation. These findings suggest that circuit preservation may help explain why RL is more robust to catastrophic forgetting. We released our code here: this https URL.

Subjects:

Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)

Cite as: arXiv:2605.28860 [cs.LG]

(or arXiv:2605.28860v1 [cs.LG] for this version)

https://doi.org/10.48550/arXiv.2605.28860

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Maheep Chaudhary [view email] [v1] Thu, 21 May 2026 19:03:57 UTC (919 KB)

Full-text links:

Access Paper:

View a PDF of the paper titled Mechanistic origins of catastrophic forgetting: why RL preserves circuits better than SFT?, by Jeanmely Rojas Nunez and 6 other authors

View PDF

HTML (experimental)

TeX Source

view license

Current browse context:

cs.LG

new | recent | 2026-05

Change to browse by:

cs cs.AI cs.CL

References & Citations

NASA ADS

Google Scholar

Semantic Scholar

Loading...

Data provided by:

Bibliographic Tools

Bibliographic and Citation Tools

Bibliographic Explorer Toggle

Bibliographic Explorer (What is the Explorer?)

Connected Papers Toggle

Connected Papers (What is Connected Papers?)

Litmaps Toggle

Litmaps (What is Litmaps?)

scite.ai Toggle

scite Smart Citations (What are Smart Citations?)

Code, Data, Media

Code, Data and Media Associated with this Article

alphaXiv Toggle

alphaXiv (What is alphaXiv?)

Links to Code Toggle

CatalyzeX Code Finder for Papers (What is CatalyzeX?)

DagsHub Toggle

DagsHub (What is DagsHub?)

GotitPub Toggle

Gotit.pub (What is GotitPub?)

Huggingface Toggle

Hugging Face (What is Huggingface?)

ScienceCast Toggle

ScienceCast (What is ScienceCast?)

Demos

Demos

Replicate Toggle

Replicate (What is Replicate?)

Spaces Toggle

Hugging Face Spaces (What is Spaces?)

Spaces Toggle

TXYZ.AI (What is TXYZ.AI?)

Related Papers

Recommenders and Search Tools

Link to Influence Flower

Influence Flower (What are Influence Flowers?)

Core recommender toggle

CORE Recommender (What is CORE?)

IArxiv recommender toggle

IArxiv Recommender (What is IArxiv?)

Author

Venue

Institution

Topic

About arXivLabs

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.

Which authors of this paper are endorsers? | Disable MathJax (What is MathJax?)