World Models in Words: Auditing Physical State-Transition Commitments in Vision-Language Models

• URL: http://arxiv.org/abs/2605.29585v1
• Date: Thu, 28 May 2026 08:29:32 GMT
• Title: World Models in Words: Auditing Physical State-Transition Commitments in Vision-Language Models
• Abstract summary: Vision-language models (VLMs) are increasingly used to answer questions about physical scenes.<n>We introduce wmw, an evaluation framework for auditing the emphlanguage-expressed physical commitments ofVLMs.<n>A hybrid verifier then checks schema validity, state grounding, transition consistency, and answer-trace compatibility.
• Score: 3.1727619150610837
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Vision-language models (VLMs) are increasingly used to answer questions about physical scenes, yet most evaluations reduce performance to a final answer. This hides whether the model perceived the right objects, represented the right physical state, predicted a plausible transition, or merely selected the right option for the wrong reasons. We introduce \wmw, an evaluation framework for auditing the \emph{language-expressed physical commitments} of VLMs. Instead of scoring only $I,q\mapsto a$, we ask models to produce a typed trace $I,q\mapsto(s_0,Δs,s_1,a)$: an initial state, a state transition, a resulting state, and an answer. A hybrid verifier then checks schema validity, state grounding, transition consistency, and answer-trace compatibility, yielding typed error labels such as object, relation, force, transition, temporal, unit/scale, and faithfulness errors. We release \tracebank, a controlled trace resource with \nSeed schema- and recomputation-validated synthetic scenarios across \nFamilies physics families, \nPairs minimally perturbed contrastive preference pairs, verifier code, audit guidelines, and model outputs. We evaluate \nModels VLMs on both controlled and external physical-reasoning examples. \wmw reveals failures that answer-only evaluation misses: 35\% of correct answers from mid-tier models are backed by physically invalid traces. Verifier-guided reranking recovers up to 7 percentage points of trace validity without sacrificing answer accuracy, and trace-level preference tuning reduces hidden inconsistency by 41\% relative. The contribution is not another final-answer physics benchmark, but a reusable protocol for measuring whether a VLM's stated physical world can be true at the same time as its answer.

Related papers

• Beyond the Answer Key: Robustness Evaluation of Large Language Models for Step-Level Mathematical Verification [0.0]
  Large language models (LLMs) are increasingly used as graders, verifiers, and process auditors.<n>Most mathematical evaluations still emphasize final-answer accuracy.<n>We introduce a controlled linear-equation benchmark for evaluating LLMs in the evaluator role.
  arXiv  Detail & Related papers  (2026-08-28T16:17:11Z)
• Apple-$π$: Benchmarking Thinking with Video Towards Law-Grounded Physical Intelligence [54.58886283991723]
  Apple-PI is the first benchmark that anchors video-model evaluation explicitly in physical laws.<n>Apple-PI comprises three components.<n> Benchmarking 11 models shows that current video models remain far from reliable law-grounded world simulators.
  arXiv  Detail & Related papers  (2026-07-17T18:00:05Z)
• DriftScope: Measuring The Hidden Effects of Diffusion Model Adaptation [50.584834483403675]
  Adapting pre-trained text-to-image diffusion models, whether to learn new visual concepts or erase unwanted ones, is routinely evaluated on its intended effects alone.<n>We demonstrate that adaptation systematically damages semantically unrelated concepts in ways that aggregate metrics structurally cannot surface.<n>We introduce DriftScope, a prompt-level diagnostic tool that takes any two model checkpoints and returns a ranked list of tokens whose visual concepts have shifted most between them.
  arXiv  Detail & Related papers  (2026-06-30T20:57:51Z)
• ScratchWorld: Evaluating If World Models Compute Executable Consequences [2.3981743514438087]
  We introduce ScratchWorld, an offline diagnostic benchmark that treats Scratch projects as executable worlds.<n> ScratchWorld evaluates next-state prediction, long-horizon tracking, causal event attribution, and counterfactual prediction.
  arXiv  Detail & Related papers  (2026-06-30T14:02:25Z)
• VisualFLIP: Do Predictions Depend on Task-Critical Visual Evidence in Multimodal Reasoning? [79.44137632338062]
  When a multimodal large language model answers a visual reasoning question correctly, is the prediction actually supported by the task-critical visual evidence?<n>We introduce VisualFLIP, a paired benchmark with 1,374 images arranged as same-question pairs across cardinality, attribute, spatial, and logic tasks.<n>We evaluate 24 MLLMs with pair accuracy, which requires solving both sides of a pair, and Collapse Rate (CR), which measures how often a model that solves at least one side repeats the same non-empty answer for both images.
  arXiv  Detail & Related papers  (2026-06-05T22:06:07Z)
• When Does a Language Model Commit? A Finite-Answer Theory of Pre-Verbalization Commitment [6.48455847613423]
  Language models often generate reasoning before giving a final answer, but the visible answer does not reveal when the model's answer preference became stable.<n>We study this question through a narrow computable object: emphfinite-answer preference stabilization.<n>For a model state and specified answer verbalizers, we project the model's own continuation probabilities onto a finite answer set.<n>In binary tasks this yields an exact log-odds code, $()=S_(mathrmyesmid)-S_(mathrmnomid
  arXiv  Detail & Related papers  (2026-05-07T08:34:59Z)
• interwhen: A Generalizable Framework for Verifiable Reasoning with Test-time Monitors [47.363850513075356]
  We present a test-time verification framework, interwhen, that ensures that the output of a reasoning model is valid wrt. a given set of verifiers.<n> Verified reasoning is an important goal in high-stakes scenarios such as deploying agents in the physical world.
  arXiv  Detail & Related papers  (2026-02-05T08:35:01Z)
• Probing the Trajectories of Reasoning Traces in Large Language Models [4.599673637363014]
  We propose a protocol to probe the trajectories of reasoning traces in large language models.<n>We find that accuracy and decision commitment consistently increase as the percentage of provided reasoning tokens grows.<n>We show that trajectory probing provides diagnostics for efficient and safer deployment of reasoning models.
  arXiv  Detail & Related papers  (2026-01-30T16:45:16Z)
• Are Large Reasoning Models Interruptible? [77.53059044071107]
  Large Reasoning Models (LRMs) excel at complex reasoning but are traditionally evaluated in static, "frozen world" settings.<n>We show that even state-of-the-art LRMs, which achieve high accuracy in static settings, can fail unpredictably when interrupted or exposed to changing context.<n>Our analysis further reveals several novel failure modes, including reasoning leakage, panic, and self-doubt.
  arXiv  Detail & Related papers  (2025-10-13T17:59:35Z)
• CLUE: Non-parametric Verification from Experience via Hidden-State Clustering [64.50919789875233]
  We show that correctness of a solution is encoded as a geometrically separable signature within the trajectory of hidden activations.<n>ClUE consistently outperforms LLM-as-a-judge baselines and matches or exceeds modern confidence-based methods in reranking candidates.
  arXiv  Detail & Related papers  (2025-10-02T02:14:33Z)
• Reinforcement Learning with Verifiable yet Noisy Rewards under Imperfect Verifiers [90.50039419576807]
  Reinforcement Learning with Verifiable Rewards (RLVR) trains policies against automated verifiers to avoid costly human labeling.<n>To reduce vulnerability to verifier hacking, many RLVR systems collapse rewards to binary $0,1$ during training.<n>This choice carries a cost: it introduces textitfalse negatives (rejecting correct answers, FNs) and textitfalse positives (accepting incorrect ones, FPs)
  arXiv  Detail & Related papers  (2025-10-01T13:56:44Z)
• Rethinking Reward Models for Multi-Domain Test-Time Scaling [91.76069784586149]
  Prior work generally assumes that process reward models (PRMs) outperform outcome reward models (ORMs) that assess only the final answer.<n>We present the first unified evaluation of four reward model variants across 14 diverse domains.<n>We attribute this to PRM-style stepwise scoring, which inherits label noise from LLM auto-labeling and has difficulty evaluating long reasoning trajectories.
  arXiv  Detail & Related papers  (2025-10-01T04:21:14Z)
• Measuring Epistemic Humility in Multimodal Large Language Models [17.490955813494693]
  We present HumbleBench, a new hallucination benchmark designed to evaluate MLLMs' ability to reject plausible but incorrect answers.<n>We leverage fine-grained scene graph annotations to extract ground-truth entities and relations, and prompt GPT-4-Turbo to generate multiple-choice questions.<n>HumbleBench fills a key gap in current evaluation suites, providing a more realistic measure of MLLM reliability in safety-critical settings.
  arXiv  Detail & Related papers  (2025-09-11T17:54:00Z)
• AutoPSV: Automated Process-Supervised Verifier [10.283965168399158]
  textbfAutomated textbfProcess-textbfSupervised textbfVerifier (textbftextscAutoPSV) textscAutoPSV begins by training a verification model on the correctness of final answers. We experimentally validate that the step-level confidence changes learned by the verification model trained on the final answer correctness can effectively identify errors in the reasoning steps.
  arXiv  Detail & Related papers  (2024-05-27T03:44:24Z)

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.