跳到正文
arXiv cs.AI· Ons Zammel, Carlos Soares, Fl\'avia Carvalhido, Gerd Kortemeyer·· 2 小时前AI 评分33

多模态大语言模型批改助手运动学图像读图能力压力测试

Stress Testing a Multimodal Large Language Model Grading Assistant on Kinematics Graph Reading

AI 导读

一项探索性研究对多模态大语言模型(MLLM)批改助手在四道运动学题目合成答案的图像读图评分能力进行压力测试,使用视觉畸变和逐步变换的图像及多种模型与提示词配置。部分判断能正确区分瞬时零速度与有限区间静止、识别正曲率,但严重畸变的速度-时间图竟获满分,而同配置下其他答卷被扣大量分数;无刻度图的数值判定也缺乏坐标约定。研究者建议教师审阅评分(包括满分答卷)前不应将其作为最终成绩。

正文 · 原文

View PDF HTML (experimental)

Abstract:Multimodal large language models (MLLMs) can recognize graph features yet award credit without sufficient graphical support. This exploratory study examines selected grading outputs for synthetic responses to four kinematics exercises, generated through visual distortion and progressive transformation toward reference graphs. Different model and prompt configurations were used. Some judgments correctly distinguished instantaneous zero velocity from rest over a finite interval and recognized positive curvature. Yet a severely distorted velocity--time graph received full credit despite disconnected and branching traces and an unclear zero-velocity reference. Other submissions graded under the same configuration lost substantial credit. Numerical cutoffs intended to distinguish meaningful features from drawing irregularities were supplied without a coordinate convention for unnumbered graphs, leaving the basis of exact partial-credit awards unresolved. Stress testing can expose unsupported awards before classroom use. For the configurations examined, we do not recommend accepting scores as final grades without instructor review, including review of full-credit responses.
Subjects: Physics Education (physics.ed-ph); Artificial Intelligence (cs.AI)
Cite as: arXiv:2611.22726 [physics.ed-ph]
  (or arXiv:2611.22726v1 [physics.ed-ph] for this version)
  https://doi.org/10.48550/arXiv.2611.22726

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Gerd Kortemeyer [view email]
[v1] Tue, 17 Nov 2026 13:00:23 UTC (2,329 KB)

来源:arXiv cs.AI · arxiv.org