Metraipolis › archive › dc_03ba1985dd36

Causal Syntax Circuits in CodeLMs: Budget and Probe Decide What Looks Localised — arXiv:2610.20683

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.20683
License (as archived)
CC0-1.0
Archived
2026-10-16T10:19:57.930Z
SHA-256 of the archived text
6f99a9ab92a7bc6a6623a5a40e60669508c3b8230f61d1ed600045bf29b94403
Stable URL of this page
https://metraipolis.com/documento/dc_03ba1985dd36

How to cite

Causal Syntax Circuits in CodeLMs: Budget and Probe Decide What Looks Localised — arXiv:2610.20683 — arXiv (en), https://arxiv.org/abs/2610.20683 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_03ba1985dd36 (Metraipolis archive dc_03ba1985dd36, sha256 6f99a9ab92a7bc6a6623a5a40e60669508c3b8230f61d1ed600045bf29b94403)
BibTeX
@misc{dc03ba1985dd36,
  title = {Causal Syntax Circuits in CodeLMs: Budget and Probe Decide What Looks Localised — arXiv:2610.20683},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.20683},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_03ba1985dd36; sha256 6f99a9ab92a7bc6a6623a5a40e60669508c3b8230f61d1ed600045bf29b94403},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.20683. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Code models write Python when the prompt does not say which language to use. We ask a mechanistic question about that habit: is it produced by a small, findable group of components inside the model? We run one pipeline---mean ablation, activation patching, greedy circuit extraction, a random-control significance test, and activation steering---over 55 model-language settings spanning ten open code models (five base and five instruction-tuned), three target languages (Rust, C#, PHP), two prompt suites, and two component budgets. We report five findings. First, the preference is large, grows with model size, and survives instruction tuning: it is larger on every instruction-tuned checkpoint than on its base version, by 12% on average. Second, its measured size is mostly a property of the probe; replacing our aligned prompts with XLCoST parallel functions takes the same measurement on the same models from about 1.6 to about 0.01. Third, whether the preference looks localised depends on how many components the search may keep: with a budget of 12, 1 of 15 settings yields a circuit that beats size-matched random controls after multiple-comparison correction, and with a budget of 40, 6 of 15 do. Fourth, interventions that a smaller-budget study reports as uniformly failing do sometimes succeed: patching makes a Python prompt prefer target syntax in 2 of 45 settings, and steering changes the most likely next token in 9 of 45. Fifth, every circuit whose effect is significant but points away from the target language comes from a setting where circuit extraction failed, which makes such results a diagnostic rather than a phenomenon. The practical lesson is that a claim about whether a behaviour is localised can turn on analysis choices that papers rarely report. We release the full grid so those choices can be varied.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.LG, nuovi lavori del giorno).
Categorie arXiv: Machine Learning (cs.LG) ; Software Engineering (cs.SE).
Pagina del lavoro: https://arxiv.org/abs/2610.20683 — 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 2026-10-16T08:48:02.643Z dalla pagina https://arxiv.org/list/cs.LG/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).