Metraipolis › archive › dc_90471a49ea37

Rewired or Gated? How Instruction Tuning Shapes Knowledge-Conflict Circuits in LLMs — arXiv:2609.25602

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.25602
License (as archived)
CC0-1.0
Archived
2026-10-08T14:36:56.483Z
SHA-256 of the archived text
efe78ca8a3f3c2678a811af02525312158aa071a16b8c047f8ff6b5530a7fb4f
Stable URL of this page
https://metraipolis.com/documento/dc_90471a49ea37

How to cite

Rewired or Gated? How Instruction Tuning Shapes Knowledge-Conflict Circuits in LLMs — arXiv:2609.25602 — arXiv (en), https://arxiv.org/abs/2609.25602 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_90471a49ea37 (Metraipolis archive dc_90471a49ea37, sha256 efe78ca8a3f3c2678a811af02525312158aa071a16b8c047f8ff6b5530a7fb4f)
BibTeX
@misc{dc90471a49ea37,
  title = {Rewired or Gated? How Instruction Tuning Shapes Knowledge-Conflict Circuits in LLMs — arXiv:2609.25602},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.25602},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_90471a49ea37; sha256 efe78ca8a3f3c2678a811af02525312158aa071a16b8c047f8ff6b5530a7fb4f},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.25602. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

In language models, the choice between believing the prompt and believing the weights is made by a handful of identifiable attention heads. Instruction tuning changes how models behave under conflict, but whether it rewires the underlying circuit or merely gates/reweights already present components, remains unknown. We provide the first mechanistic base-vs-instruct comparison of conflict-resolution circuits, across three families (Llama-3.2-3B, Qwen-2.5-3B, Gemma-3-4B). Five independent methods, node and edge attribution, superposition role analysis, causal ablation, and path patching, converge on gating, with the same heads, in the same late-layers, are found to be reweighted rather than replaced with a high node overlap (0.60-0.82). Behaviorally, tuning shifts models toward parametric memory, making instruct models reject a terse counterfactual context far more than base ones, the opposite of a naive user-following expectation. Yet this added skepticism is a factor of framing since it disappears when the same false claim is delivered as a coherent, evidential passage. The robustness that instruction tuning buys against terse injection is therefore real but narrow. More broadly, we believe that because the conflict circuit is preserved rather than rebuilt, interpretability and control tools calibrated on base models should transfer directly to their deployed instruct siblings.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-23 in cs.CL).
Categorie arXiv: Machine Learning (cs.LG) ; Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2609.25602 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-08T13:46:15.787Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-23?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).