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

arXiv:2609.01081 · cs.CL, cs.AI · Submitted 2026-09-01 · Read on arXiv

cs.CL, cs.AI

Submitted: 2026-09-01

Updated: 2026-09-07

Comments: Accepted to Findings of EMNLP 2026

Code: https://github.com/Cha0Ga0/SWAPSTATE

License: http://creativecommons.org/licenses/by/4.0/

The gist: Large language models often answer the same multiple-choice question inconsistently when it is posed under support-oriented and elimination-oriented framings.

Terminology

Abstract

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.

Sources

Related papers