Metraipolis › archive › dc_2bcdd60b2b91

StateSwap: Probing Support-Elimination Hidden States in Multiple-Choice Questions — arXiv:2609.01081

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.01081
License (as archived)
CC0-1.0
Archived
2026-10-17T06:33:54.932Z
SHA-256 of the archived text
27aaba121d68b8a360b12bc4da7e8efd2f1cfd2873b5dfd920bde64705f4dce8
Stable URL of this page
https://metraipolis.com/documento/dc_2bcdd60b2b91

How to cite

StateSwap: Probing Support-Elimination Hidden States in Multiple-Choice Questions — arXiv:2609.01081 — arXiv (en), https://arxiv.org/abs/2609.01081 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_2bcdd60b2b91 (Metraipolis archive dc_2bcdd60b2b91, sha256 27aaba121d68b8a360b12bc4da7e8efd2f1cfd2873b5dfd920bde64705f4dce8)
BibTeX
@misc{dc2bcdd60b2b91,
  title = {StateSwap: Probing Support-Elimination Hidden States in Multiple-Choice Questions — arXiv:2609.01081},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.01081},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_2bcdd60b2b91; sha256 27aaba121d68b8a360b12bc4da7e8efd2f1cfd2873b5dfd920bde64705f4dce8},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.01081. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Large language models often answer the same multiple-choice question inconsistently when it is posed under support-oriented and elimination-oriented framings. We investigate whether these discrepancies arise from different internal representations induced by the two framings. We introduce a dual-framing protocol with minimally varied prompts that use either support- or elimination-oriented framing while keeping the evaluation target fixed. To probe the internal computation, we append an untrained special token, [STATE], and treat its residual-stream activation as an intervention interface. Across both models, the two framings induce separable [STATE] activations concentrated in intermediate layers. Swapping these activations between paired prompts systematically changes predictions and improves cross-framing agreement, providing intervention-based evidence that the activations are behaviorally relevant. Beyond instance-level substitution, mean-difference steering directions derived from the dual-framing contrast exhibit more bounded layer-wise responses than matched contrastive activation addition directions under the evaluated protocol.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-02 in cs.AI).
Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.01081 — 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 2026-10-17T06:15:12.858Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-02?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).