Belief-reality separation lives in routing over a shared value slot in language models

• URL: http://arxiv.org/abs/2607.11945v1
• Date: Sat, 11 Jul 2026 10:39:03 GMT
• Title: Belief-reality separation lives in routing over a shared value slot in language models
• Abstract summary: Capable language models hold what a character believes apart from what is true.<n>We show it rests on two separable mechanisms at two positions.
• Score: 6.0397270384735355
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Capable language models hold what a character believes apart from what is true: told "Anna believes the cup is blue; in reality it is red," they answer blue about Anna and red about the world. Where in the computation does that separation live? We show it rests on two separable mechanisms at two positions. A generic value slot binds the attributed value. A router at the query position selects which frame, the character's belief or reality, a query reads out. Two routes fill the slot: an asserted belief, whose value the text supplies, binds in directly; a derived belief, whose value must be inferred from what the character could see, arrives by a visibility-gated lookback. A subspace trained on either route steers the other, and only the derived route depends on described visibility. The slot itself carries no belief-reality tag: intervening on it moves a reality readout as strongly as a belief one. The separation lives instead in a dissociated pair of routing subspaces, which flip a query between frames without injecting the donor's value. These results hold across three architectures, on stimuli de-confounded against theory-of-mind-benchmark shortcuts; the behavior itself emerges between 3B and 7B across five model families. This paper develops the single belief-reality axis in depth; a companion paper shows the same slot-and-router format is shared across the other non-actual contexts a sentence can open (counterfactual, fictional, temporal).

Related papers

• Contrastive Projection: Reading Transformer Internals by Differencing Logit Lenses [0.0]
  We use a training-free tracer that reads at every position, sub-layer, and head average.<n>It traces a compound- noun->attention chain in Phi-2, confirmed there by activation patching.<n>It reads what retrieval surfaces for real versus fictional entities, and reads metaphor as a set of domain-to-domain mappings.
  arXiv  Detail & Related papers  (2026-09-09T09:01:09Z)
• How to Build Marcus's Algebraic Mind: From Minsky's Emotion-Machine Viewpoint [18.56157183294801]
  In The Algebraic Mind, Marcus identified three cognitive components.<n>A companion paper solves this with VaCoAl, an architecture built on GF(2) XOR-and-shift.<n>A collision-avoidance scheme ensures exact recovery within a Frontier Size.
  arXiv  Detail & Related papers  (2026-07-18T03:48:00Z)
• One mechanism for many mental spaces: a shared router over a value slot in language models [6.0397270384735355]
  Language builds discourse contexts other than the actual.<n>Formal semantics keeps these contexts apart because their logics differ.<n>We ask which of these a transformer language model implements, and find a mechanistic version of Fauconnier's unification.
  arXiv  Detail & Related papers  (2026-07-11T10:30:36Z)
• Do GUI Agents Believe Their Eyes? Diagnosing State-Belief Reliance on Pixels versus Structure [5.717981279944186]
  We formalize visual state reliance, the attribution of a state belief to pixels, structure, or priors.<n>We measure it with paired single-channel interventions over 310 real web, mobile, and desktop probes.<n>Our central metric is the Perception-Fusion Gap, the fraction of probes a model perceives correctly yet resolves toward structure under conflict.
  arXiv  Detail & Related papers  (2026-07-05T14:33:29Z)
• They Infer What You Meant: Models Represent Communicative Intent More Reliably Than They Act On It [0.0]
  We show that the failure of a language model to decode the sender's intent is one of readouts on top of a robust representation.<n>A linear probe decodes the sender's intent, whether they want a thing recognized or evaluated, across six models and four families and in the base checkpoints.<n>The behavioral half of the story is established on the recognize/evaluate contrast, where what varies is whether the default output acts on the intent.
  arXiv  Detail & Related papers  (2026-07-03T20:49:37Z)
• Beyond Hallucinations: The Illusion of Understanding in Large Language Models [0.0]
  Large language models (LLMs) are becoming deeply embedded in human communication and decision-making.<n>They inherit the ambiguity, bias, and lack of direct access to truth inherent in language itself.<n>This paper argues that LLMs operationalize System 1 cognition at scale: fast, associative, and persuasive, but without reflection or falsification.
  arXiv  Detail & Related papers  (2025-10-16T13:19:44Z)
• Distributional Semantics Tracing: A Framework for Explaining Hallucinations in Large Language Models [4.946483489399819]
  Large Language Models (LLMs) are prone to hallucination, the generation of factually incorrect statements.<n>This work investigates the intrinsic, architectural origins of this failure mode through three primary contributions.
  arXiv  Detail & Related papers  (2025-10-07T16:40:31Z)
• Language Models use Lookbacks to Track Beliefs [53.8367373163602]
  We analyze LMs' ability to reason about characters' beliefs using causal mediation and abstraction.<n>Our work provides insights into belief tracking mechanisms, taking a step toward reverse-engineering ToM reasoning in LMs.
  arXiv  Detail & Related papers  (2025-05-20T17:59:45Z)
• Is the Top Still Spinning? Evaluating Subjectivity in Narrative Understanding [40.78154629252038]
  Forcing binary labels upon ambiguous claims lowers the reliability of evaluation.<n>We introduce LLM-generated edits of summaries as a method of providing a nuanced evaluation of claims.<n>We show that ARM produces a absolute 21% improvement in annotator agreement on claim faithfulness.
  arXiv  Detail & Related papers  (2025-04-01T19:08:24Z)
• Is This the Subspace You Are Looking for? An Interpretability Illusion for Subspace Activation Patching [47.05588106164043]
  Mechanistic interpretability aims to understand model behaviors in terms of specific, interpretable features. Recent studies have explored subspace interventions as a way to manipulate model behavior and attribute the features behind it to given subspaces. We demonstrate that these two aims diverge, potentially leading to an illusory sense of interpretability.
  arXiv  Detail & Related papers  (2023-11-28T18:32:19Z)
• Latent Topology Induction for Understanding Contextualized Representations [84.7918739062235]
  We study the representation space of contextualized embeddings and gain insight into the hidden topology of large language models. We show there exists a network of latent states that summarize linguistic properties of contextualized representations.
  arXiv  Detail & Related papers  (2022-06-03T11:22:48Z)
• Transformation Driven Visual Reasoning [80.32402545546209]
  This paper defines a new visual reasoning paradigm by introducing an important factor, i.e.transformation. We argue that this kind of textbfstate driven visual reasoning approach has limitations in reflecting whether the machine has the ability to infer the dynamics between different states. Experimental results show that the state-of-the-art visual reasoning models perform well on Basic, but are still far from human-level intelligence on Event and View.
  arXiv  Detail & Related papers  (2020-11-26T07:11:31Z)
• A Weaker Faithfulness Assumption based on Triple Interactions [89.59955143854556]
  We propose a weaker assumption that we call $2$-adjacency faithfulness. We propose a sound orientation rule for causal discovery that applies under weaker assumptions.
  arXiv  Detail & Related papers  (2020-10-27T13:04:08Z)
• Commonsense Evidence Generation and Injection in Reading Comprehension [57.31927095547153]
  We propose a Commonsense Evidence Generation and Injection framework in reading comprehension, named CEGI. The framework injects two kinds of auxiliary commonsense evidence into comprehensive reading to equip the machine with the ability of rational thinking. Experiments on the CosmosQA dataset demonstrate that the proposed CEGI model outperforms the current state-of-the-art approaches.
  arXiv  Detail & Related papers  (2020-05-11T16:31:08Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.