Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2602.04355 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866911421679271936 |
|---|---|
| author | Liang, Sichu Zhu, Hongyu Wang, Wenwen Zhou, Deyu |
| author_facet | Liang, Sichu Zhu, Hongyu Wang, Wenwen Zhou, Deyu |
| contents | Working memory is a central component of intelligent behavior, providing a dynamic workspace for maintaining and updating task-relevant information. Recent work has used n-back tasks to probe working-memory-like behavior in large language models, but it is unclear whether the same probe elicits comparable computations when information is carried in a visual rather than textual code in vision-language models. We evaluate Qwen2.5 and Qwen2.5-VL on a controlled spatial n-back task presented as matched text-rendered or image-rendered grids. Across conditions, models show reliably higher accuracy and d' with text than with vision. To interpret these differences at the process level, we use trial-wise log-probability evidence and find that nominal 2/3-back often fails to reflect the instructed lag and instead aligns with a recency-locked comparison. We further show that grid size alters recent-repeat structure in the stimulus stream, thereby changing interference and error patterns. These results motivate computation-sensitive interpretations of multimodal working memory. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_04355 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Can Vision Replace Text in Working Memory? Evidence from Spatial n-Back in Vision-Language Models Liang, Sichu Zhu, Hongyu Wang, Wenwen Zhou, Deyu Computation and Language Working memory is a central component of intelligent behavior, providing a dynamic workspace for maintaining and updating task-relevant information. Recent work has used n-back tasks to probe working-memory-like behavior in large language models, but it is unclear whether the same probe elicits comparable computations when information is carried in a visual rather than textual code in vision-language models. We evaluate Qwen2.5 and Qwen2.5-VL on a controlled spatial n-back task presented as matched text-rendered or image-rendered grids. Across conditions, models show reliably higher accuracy and d' with text than with vision. To interpret these differences at the process level, we use trial-wise log-probability evidence and find that nominal 2/3-back often fails to reflect the instructed lag and instead aligns with a recency-locked comparison. We further show that grid size alters recent-repeat structure in the stimulus stream, thereby changing interference and error patterns. These results motivate computation-sensitive interpretations of multimodal working memory. |
| title | Can Vision Replace Text in Working Memory? Evidence from Spatial n-Back in Vision-Language Models |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2602.04355 |