When Models Decide and When They Bind: A Two-Stage Computation for Multiple-Choice Question-Answering
Multiple-choice question answering (MCQA) is easy to evaluate but adds a meta-task: models must both solve the problem and output the symbol that *represents* the answer, conflating reasoning errors with symbol-binding failures. We study how language models implement MCQA internally using representational analyses (PCA, linear probes) as well as causal interventions. We find that option-boundary (newline) residual states often contain strong linearly decodable signals related to per-option correctness. Winner-identity probing reveals a two-stage progression: the winning *content position* becomes decodable immediately after the final option is processed, while the *output symbol* is represented closer to the answer emission position. Tests under symbol and content permutations support a two-stage mechanism in which models first select a winner in content space and then bind or route that winner to the appropriate symbol to emit.
Code (0)
등록된 구현이 없습니다.
Tasks
Question AnsweringSimilar Papers 제목 키워드 기반
Detailed Balance for Particle Models of Reversible Reactions in Bounded Domains
In particle-based stochastic reaction-diffusion models, reaction rate and placement kernels are used to decide the probability per time a reaction can occur between reactant particles, and to decide where product particl…
When Precedents Clash
Consistency of case bases is a way to avoid the problem of retrieving conflicting constraining precedents for new cases to be decided. However, in legal practice the consistency requirements for case bases may not be sat…
StemBind: When MLLMs Get Lost Between Rules and Instances in Abstract Visual Reasoning
Multimodal large language models (MLLMs) often know the rule but pick the wrong answer: on abstract visual reasoning (AVR) tasks, a model can describe what it sees and name the underlying pattern, yet still fail to choos…
Answer SelectionVisual ReasoningAttention Deficits in Language Models: Causal Explanations for Procedural Hallucinations
Large language models can follow complex procedures yet fail at a seemingly trivial final step: reporting a value they themselves computed moments earlier. We study this phenomenon as \emph{procedural hallucination}: fai…
On the emergence of single versus multi-state allostery
Several physical mechanisms have been proposed to explain allostery in proteins. They differ by the number of internal states that they assume a protein to occupy, leaving open the question of what controls the emergence…