---
title: "How Do LLMs and VLMs Understand Viewpoint Rotation Without"
description: "Each instance describes several left or right rotations in multiples of 90°, paired with objects seen along the way. 68 references."
authors: ["Zhen Yang", "Ping Jian", "Zhongbin Guo", "Zuming Zhang", "Chengzhi Li", "Yonghong Deng", "Xinyue Zhang", "Wenpeng Lu"]
year: 2026
venue: "Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)"
source: "https://aclanthology.org/2026.acl-long.496/"
doi: "10.18653/v1/2026.acl-long.496"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/yang-2026-how-do-llms-vlms-understand/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/yang-2026-how-do-llms-vlms-understand/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/36/) / How Do LLMs and VLMs Understand Viewpoint Rotation Without Vision? An Interpretability Study

# How Do LLMs and VLMs Understand Viewpoint Rotation Without Vision? An Interpretability Study

**Authors:** [Zhen Yang](https://md.research.elamur.ai/free/en/authors/zhen-yang/), [Ping Jian](https://md.research.elamur.ai/free/en/authors/ping-jian/), [Zhongbin Guo](https://md.research.elamur.ai/free/en/authors/zhongbin-guo/), Zuming Zhang, [Chengzhi Li](https://md.research.elamur.ai/free/en/authors/chengzhi-li/), [Yonghong Deng](https://md.research.elamur.ai/free/en/authors/yonghong-deng/), [Xinyue Zhang](https://md.research.elamur.ai/free/en/authors/xinyue-zhang/) and [Wenpeng Lu](https://md.research.elamur.ai/free/en/authors/wenpeng-lu/) · **Year:** 2026 · **Venue:** [Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)](https://md.research.elamur.ai/free/en/venues/acl/) ([ACL 2026](https://md.research.elamur.ai/free/en/venues/acl/2026/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.acl-long.496/) · [PDF](https://aclanthology.org/2026.acl-long.496.pdf) · [DOI 10.18653/v1/2026.acl-long.496](https://doi.org/10.18653/v1/2026.acl-long.496) · [OpenAlex](https://openalex.org/W7166855586)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary
The paper asks whether language alone, without images, lets LLMs and VLMs track how a viewpoint changes after a series of turns. The authors built VRUBench, a synthetic text-only benchmark. Each instance describes several left or right rotations in multiples of 90°, paired with objects seen along the way. The model must name the object facing it after the last turn, or answer "unknown" if that direction was never observed. Humans solve it perfectly, but most models do poorly. Even Qwen3-VL reaches only about 60%.

To find out why, the authors combine layer-wise probing with head-wise causal intervention (path patching). Models do represent each rotation step and the resulting absolute orientation in their hidden states. However, a small group of attention heads in the later layers fails to connect that orientation to the object seen there, which leads to hallucinated answers. Fine-tuning only these heads substantially improves VRU accuracy and avoids the loss of general skills caused by full fine-tuning. The work is an early mechanistic look at text-based spatial reasoning.

## Key points
- **VRUBench composition:** 19,591 instances built from 100 indoor objects, split into 4,614 two-step, 4,977 three-step, and 5,000 each of four- and five-step rotation sequences.
- **Benchmark findings:**
  - VLMs consistently beat their text-only counterparts (e.g., Qwen3-VL-8B vs Qwen3-8B).
  - Think-then-answer (CoT) prompting helps, which differs from earlier results on the visual dataset VSI-Bench.
- **Probing results:**
  - Rotation direction and angle are encoded strongly at every step.
  - Absolute orientation emerges in early-to-middle layers and fades in later ones.
- **Path patching results:** Key heads in middle-to-late layers fall into three types:
  - "proposal" heads that collect candidate answers;
  - "answer decision" heads that pick one;
  - an "unknown" head reflecting cautious behavior, possibly a product of alignment training.
- **Selective SFT of the top-32 heads (Qwen2.5-VL-3B, about 0.03B tuned parameters):**
  - VRUBench accuracy rises from 37.6 to 80.1, using roughly half the GPU hours of full SFT.
  - MMLU and BBH stay stable, while full SFT drops BBH from 25.5 to 19.7.
  - Performance on the visual OOD benchmark SpinBench also improves slightly (and by +3.6 for the 7B model), suggesting text-only training carries over to visual tasks.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

## Institutions

- [Beijing Institute of Technology](https://md.research.elamur.ai/free/en/universities/beijing-institute-of-technology/) — University, Beijing, China · 43 works · 3,652 citations
- [Qilu University of Technology](https://md.research.elamur.ai/free/en/universities/qilu-university-of-technology/) — University, Jinan, China · 7 works · 3 citations
- [Shandong Academy of Sciences](https://md.research.elamur.ai/free/en/universities/shandong-academy-of-sciences/) — University, Jinan, China · 4 works · 1 citation
- [National Supercomputing Center in Jinan](https://md.research.elamur.ai/free/en/labs/national-supercomputing-center-in-jinan/) — Research institute, Jinan, China · 3 works

📚 Shelves (5): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 17,450 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 17,296 works · [👁️ Computer Vision and Pattern Recognition](https://md.research.elamur.ai/free/en/shelves/computer-science/computer-vision-and-pattern-recognition/) · 3,486 works

🧭 Topics (2): [🧠 Language models & pretraining](https://md.research.elamur.ai/free/en/topics/language-models/) · 3,894 works · [🔬 Interpretability](https://md.research.elamur.ai/free/en/topics/interpretability/) · 149 works

## Cite

Cite as: Zhen Yang, Ping Jian, Zhongbin Guo, Zuming Zhang, Chengzhi Li, Yonghong Deng, Xinyue Zhang and Wenpeng Lu. 2026. How Do LLMs and VLMs Understand Viewpoint Rotation Without Vision? An Interpretability Study. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{yang2026how,
  title = {{How Do LLMs and VLMs Understand Viewpoint Rotation Without Vision? An Interpretability Study}},
  author = {Zhen Yang, Ping Jian, Zhongbin Guo, Zuming Zhang, Chengzhi Li, Yonghong Deng, Xinyue Zhang and Wenpeng Lu},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-long.496},
  url = {https://doi.org/10.18653/v1/2026.acl-long.496}
}
```

- [← previous How Do Language Models Acquire Character-Level Information?](https://md.research.elamur.ai/free/en/papers/sato-2026-how-do-language-models-acquire/)
- [next → How Do LLMs Generate Contrastive Sentiments? A Mechanistic Perspective](https://md.research.elamur.ai/free/en/papers/nguyen-2026-how-do-llms-generate-contrastive/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/36/)

## References

68 references, 24 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Guillaume Alain and Yoshua Bengio. 2018. Understanding intermediate layers using linear classifier probes. Preprint, arXiv:1610.01644. [find](https://scholar.google.com/scholar?q=Guillaume+Alain+and+Yoshua+Bengio%2E+2018%2E+Understanding+intermediate+layers+using+linear+classifier+probes%2E+Preprint%2C+arXiv%3A1610%2E01644%2E)
2. Shuai Bai, Yuxuan Cai, Ruizhe Chen, Keqin Chen, Xionghui Chen, Zesen Cheng, Lianghao Deng, Wei Ding, Chang Gao, Chunjiang Ge, Wenbin Ge, Zhifang Guo, Qidong Huang, Jie Huang, Fei Huang, Binyuan Hui, Shutong Jiang, Zhaohai Li, Mingsheng Li, Mei Li, Kaixin Li, Zicheng Lin, Junyang Lin, Xuejing Liu, Ji… — [Qwen3-VL Technical Report](https://md.research.elamur.ai/free/en/papers/bai-2025-qwen3-vl-technical-report/)
3. Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, Humen Zhong, Yuanzhi Zhu, Ming-Hsuan Yang, Zhaohai Li, Jianqiang Wan, Pengfei Wang, Wei Ding, Zheren Fu, Yiheng Xu, Jiabo Ye, Xi Zhang, Tianbao Xie, Zesen Cheng, Hang Zhang, Zhibo Yang,… — [Qwen2.5-VL Technical Report](https://md.research.elamur.ai/free/en/papers/bai-2025-qwen2-5-vl-technical/)
4. Yung-Sung Chuang, Yujia Xie, Hongyin Luo, Yoon Kim, James R. Glass, and Pengcheng He. 2024. [Dola: Decoding by contrasting layers improves factuality in large language models](https://md.research.elamur.ai/free/en/papers/chuang-2023-dola-decoding-contrasting-layers-improves/). In The Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024. OpenReview.n…
5. James M. Clark and Allan Paivio. 1991. Dual coding theory and education. Educational Psychology Review, 3(3):149–210. [find](https://scholar.google.com/scholar?q=James+M%2E+Clark+and+Allan+Paivio%2E+1991%2E+Dual+coding+theory+and+education%2E+Educational+Psychology+Review%2C+3%283%29%3A149%E2%80%93210%2E)
6. Zhikai Ding, Shiyu Ni, and Keping Bi. 2025. Do LVLMs know what they know? a systematic study of knowledge boundary perception in LVLMs. In Findings of the Association for Computational Linguistics: EMNLP 2025, pages 19832–19848, Suzhou, China. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Zhikai+Ding%2C+Shiyu+Ni%2C+and+Keping+Bi%2E+2025%2E+Do+LVLMs+know+what+they+know%3F+a+systematic+study+of+knowledge+boundary+perception+in+LVLMs%2E+In+Findings+of+the+Association+for+Computational+Linguistics%3A+EM)
7. Jie Feng, Jinwei Zeng, Qingyue Long, Hongyi Chen, Jie Zhao, Yanxin Xi, Zhilun Zhou, Yuan Yuan, Shengyuan Wang, Qingbin Zeng, Songwei Li, Yunke Zhang, Yuming Lin, Tong Li, Jingtao Ding, Chen Gao, Fengli Xu, and Yong Li. 2025. A survey of large language model-powered spatial intelligence across scales… — [arXiv:2504.09848](https://arxiv.org/abs/2504.09848) [find](https://scholar.google.com/scholar?q=Jie+Feng%2C+Jinwei+Zeng%2C+Qingyue+Long%2C+Hongyi+Chen%2C+Jie+Zhao%2C+Yanxin+Xi%2C+Zhilun+Zhou%2C+Yuan+Yuan%2C+Shengyuan+Wang%2C+Qingbin+Zeng%2C+Songwei+Li%2C+Yunke+Zhang%2C+Yuming+Lin%2C+Tong+Li%2C+Jingtao+Ding%2C+Chen+Gao%2C+Fengl)
8. Daniel Y. Fu, Tri Dao, Khaled Kamal Saab, Armin W. Thomas, Atri Rudra, and Christopher Ré. 2023. Hungry hungry hippos: Towards language modeling with state space models. In The Eleventh International Conference on Learning Representations, ICLR 2023, Kigali, Rwanda, May 1-5, 2023. OpenReview.net. [find](https://scholar.google.com/scholar?q=Daniel+Y%2E+Fu%2C+Tri+Dao%2C+Khaled+Kamal+Saab%2C+Armin+W%2E+Thomas%2C+Atri+Rudra%2C+and+Christopher+R%C3%A9%2E+2023%2E+Hungry+hungry+hippos%3A+Towards+language+modeling+with+state+space+models%2E+In+The+Eleventh+International+)
9. Cheng Gao, Huimin Chen, Chaojun Xiao, Zhiyi Chen, Zhiyuan Liu, and Maosong Sun. 2025a. H-neurons: On the existence, impact, and origin of hallucination-associated neurons in llms. Preprint, arXiv:2512.01797. — [arXiv:2512.01797](https://arxiv.org/abs/2512.01797) [find](https://scholar.google.com/scholar?q=Cheng+Gao%2C+Huimin+Chen%2C+Chaojun+Xiao%2C+Zhiyi+Chen%2C+Zhiyuan+Liu%2C+and+Maosong+Sun%2E+2025a%2E+H%2Dneurons%3A+On+the+existence%2C+impact%2C+and+origin+of+hallucination%2Dassociated+neurons+in+llms%2E+Preprint%2C+arXiv%3A2512)
10. Leo Gao, Achyuta Rajaram, Jacob Coxon, Soham V. Govande, Bowen Baker, and Dan Mossing. 2025b. [find](https://scholar.google.com/scholar?q=Leo+Gao%2C+Achyuta+Rajaram%2C+Jacob+Coxon%2C+Soham+V%2E+Govande%2C+Bowen+Baker%2C+and+Dan+Mossing%2E+2025b%2E)
11. Weight-sparse transformers have interpretable circuits. Preprint, arXiv:2511.13653. — [arXiv:2511.13653](https://arxiv.org/abs/2511.13653) [find](https://scholar.google.com/scholar?q=Weight%2Dsparse+transformers+have+interpretable+circuits%2E+Preprint%2C+arXiv%3A2511%2E13653%2E+%E2%80%94+%5BarXiv%3A2511%2E13653%5D%28https%3A%2F%2Farxiv%2Eorg%2Fabs%2F2511%2E13653%29)
12. Howard Gardner. 1983. Frames of Mind: The Theory of Multiple Intelligences, tenth anniversary, second paperback edition edition. Basic Books. [find](https://scholar.google.com/scholar?q=Howard+Gardner%2E+1983%2E+Frames+of+Mind%3A+The+Theory+of+Multiple+Intelligences%2C+tenth+anniversary%2C+second+paperback+edition+edition%2E+Basic+Books%2E)
13. Nicholas Goldowsky-Dill, Chris MacLeod, Lucas Sato, and Aryaman Arora. 2023. [Localizing model behavior with path patching](https://md.research.elamur.ai/free/en/papers/goldowsky-dill-2023-localizing-model-behavior-path-patching/). CoRR, abs/2304.05969.
14. Google. 2025. Gemini 3 flash: frontier intelligence built for speed. [find](https://scholar.google.com/scholar?q=Google%2E+2025%2E+Gemini+3+flash%3A+frontier+intelligence+built+for+speed%2E)
15. Zhongbin Guo, Jiahe Liu, Yushan Li, Wenyu Gao, Zhen Yang, Chenzhi Li, Xinyue Zhang, and Ping Jian. 2025. Beyond flatlands: Unlocking spatial intelligence by decoupling 3d reasoning from numerical regression. Preprint, arXiv:2511.11239. — [arXiv:2511.11239](https://arxiv.org/abs/2511.11239) [find](https://scholar.google.com/scholar?q=Zhongbin+Guo%2C+Jiahe+Liu%2C+Yushan+Li%2C+Wenyu+Gao%2C+Zhen+Yang%2C+Chenzhi+Li%2C+Xinyue+Zhang%2C+and+Ping+Jian%2E+2025%2E+Beyond+flatlands%3A+Unlocking+spatial+intelligence+by+decoupling+3d+reasoning+from+numerical+regr)
16. Zhongbin Guo, Zhen Yang, Yushan Li, Xinyue Zhang, Wenyu Gao, Jiacheng Wang, Chengzhi Li, Xiangrui Liu, and Ping Jian. 2026. Can llms see without pixels? benchmarking spatial intelligence from textual descriptions. Preprint, arXiv:2601.03590. [find](https://scholar.google.com/scholar?q=Zhongbin+Guo%2C+Zhen+Yang%2C+Yushan+Li%2C+Xinyue+Zhang%2C+Wenyu+Gao%2C+Jiacheng+Wang%2C+Chengzhi+Li%2C+Xiangrui+Liu%2C+and+Ping+Jian%2E+2026%2E+Can+llms+see+without+pixels%3F+benchmarking+spatial+intelligence+from+textual+)
17. Michael Hanna, Ollie Liu, and Alexandre Variengien. 2023. [How does GPT-2 compute greater-than?: Interpreting mathematical abilities in a pre-trained language model](https://md.research.elamur.ai/free/en/papers/hanna-2023-gpt-2-compute-greater-than/). In Advances in Neural Information Processing Systems 36: Annual Conference on Neural Information Processing Systems 2023, NeurIPS 2023,…
18. Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. 2021. [Measuring massive multitask language understanding](https://md.research.elamur.ai/free/en/papers/hendrycks-2020-measuring-massive-multitask-language-understanding/). In 9th International Conference on Learning Representations, ICLR 2021, Virtual Event, Austria, May 3-7, 2021. OpenReview.net.
19. Jordan Hoffmann, Sebastian Borgeaud, Arthur Mensch, Elena Buchatskaya, Trevor Cai, Eliza Rutherford, Diego de Las Casas, Lisa Anne Hendricks, Johannes Welbl, Aidan Clark, Tom Hennigan, Eric Noland, Katie Millican, George van den Driessche, Bogdan Damoc, Aurelia Guy, Simon Osindero, Karen Simonyan, E… — [Training Compute-Optimal Large Language Models](https://md.research.elamur.ai/free/en/papers/hoffmann-2022-training-compute-optimal-large-language/)
20. Chao Huang, Zefeng Zhang, Juewei Yue, Quangang Li, Chuang Zhang, and Tingwen Liu. 2025. Safety alignment should be made more than just A few attention heads. CoRR, abs/2508.19697. — [arXiv:2508.19697](https://arxiv.org/abs/2508.19697) [find](https://scholar.google.com/scholar?q=Chao+Huang%2C+Zefeng+Zhang%2C+Juewei+Yue%2C+Quangang+Li%2C+Chuang+Zhang%2C+and+Tingwen+Liu%2E+2025%2E+Safety+alignment+should+be+made+more+than+just+A+few+attention+heads%2E+CoRR%2C+abs%2F2508%2E19697%2E+%E2%80%94+%5BarXiv%3A2508%2E19697%5D)
21. Lei Jiang, Zixun Zhang, Yuting Zeng, Chunzhao Xie, Tongxuan Liu, Zhen Li, Lechao Cheng, and Xiaohua Xu. 2025. DCP: Dual-cue pruning for efficient large vision-language models. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 21202–21215, Suzhou, China.… [find](https://scholar.google.com/scholar?q=Lei+Jiang%2C+Zixun+Zhang%2C+Yuting+Zeng%2C+Chunzhao+Xie%2C+Tongxuan+Liu%2C+Zhen+Li%2C+Lechao+Cheng%2C+and+Xiaohua+Xu%2E+2025%2E+DCP%3A+Dual%2Dcue+pruning+for+efficient+large+vision%2Dlanguage+models%2E+In+Proceedings+of+the+20)
22. Tianjie Ju, Yijin Chen, Xinwei Yuan, Zhuosheng Zhang, Wei Du, Yubin Zheng, and Gongshen Liu. 2024a. Investigating multi-hop factual shortcuts in knowledge editing of large language models. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Pape… [find](https://scholar.google.com/scholar?q=Tianjie+Ju%2C+Yijin+Chen%2C+Xinwei+Yuan%2C+Zhuosheng+Zhang%2C+Wei+Du%2C+Yubin+Zheng%2C+and+Gongshen+Liu%2E+2024a%2E+Investigating+multi%2Dhop+factual+shortcuts+in+knowledge+editing+of+large+language+models%2E+In+Proceedi)
23. Tianjie Ju, Zhenyu Shao, Bowen Wang, Yujia Chen, Zhuosheng Zhang, Hao Fei, Mong-Li Lee, Wynne Hsu, Sufeng Duan, and Gongshen Liu. 2025. Probing then editing response personality of large language models. In Second Conference on Language Modeling. [find](https://scholar.google.com/scholar?q=Tianjie+Ju%2C+Zhenyu+Shao%2C+Bowen+Wang%2C+Yujia+Chen%2C+Zhuosheng+Zhang%2C+Hao+Fei%2C+Mong%2DLi+Lee%2C+Wynne+Hsu%2C+Sufeng+Duan%2C+and+Gongshen+Liu%2E+2025%2E+Probing+then+editing+response+personality+of+large+language+mode)
24. Tianjie Ju, Weiwei Sun, Wei Du, Xinwei Yuan, Zhaochun Ren, and Gongshen Liu. 2024b. How large language models encode context knowledge? A layer-wise probing study. In Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation, LREC/COLING 2… [find](https://scholar.google.com/scholar?q=Tianjie+Ju%2C+Weiwei+Sun%2C+Wei+Du%2C+Xinwei+Yuan%2C+Zhaochun+Ren%2C+and+Gongshen+Liu%2E+2024b%2E+How+large+language+models+encode+context+knowledge%3F+A+layer%2Dwise+probing+study%2E+In+Proceedings+of+the+2024+Joint+Int)
25. Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B. Brown, Benjamin Chess, Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu, and Dario Amodei. 2020. [Scaling laws for neural language models](https://md.research.elamur.ai/free/en/papers/kaplan-2020-scaling-laws-neural-language-models/). CoRR, abs/2001.08361.
26. Diederik P. Kingma and Jimmy Ba. 2015. [Adam: A method for stochastic optimization](https://md.research.elamur.ai/free/en/papers/kingma-2014-adam-method-stochastic-optimization/). In 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings.
27. Klaus Krippendorff. 2011. Computing krippendorff’s alpha-reliability. [find](https://scholar.google.com/scholar?q=Klaus+Krippendorff%2E+2011%2E+Computing+krippendorff%E2%80%99s+alpha%2Dreliability%2E)
28. Chengzhi Li, Heyan Huang, Ping Jian, Zhen Yang, and Yaning Tian. 2025a. Improving temporal understanding logic consistency in video-language models via attention enhancement. Preprint, arXiv:2510.08138. — [arXiv:2510.08138](https://arxiv.org/abs/2510.08138) [find](https://scholar.google.com/scholar?q=Chengzhi+Li%2C+Heyan+Huang%2C+Ping+Jian%2C+Zhen+Yang%2C+and+Yaning+Tian%2E+2025a%2E+Improving+temporal+understanding+logic+consistency+in+video%2Dlanguage+models+via+attention+enhancement%2E+Preprint%2C+arXiv%3A2510%2E0813)
29. Dingming Li, Hongxing Li, Zixuan Wang, Yuchen Yan, Hang Zhang, Siqi Chen, Guiyang Hou, Shengpei Jiang, Wenqiao Zhang, Yongliang Shen, Weiming Lu, and Yueting Zhuang. 2025b. Viewspatial-bench: Evaluating multi-perspective spatial localization in vision-language models. CoRR, abs/2505.21500. — [arXiv:2505.21500](https://arxiv.org/abs/2505.21500) [find](https://scholar.google.com/scholar?q=Dingming+Li%2C+Hongxing+Li%2C+Zixuan+Wang%2C+Yuchen+Yan%2C+Hang+Zhang%2C+Siqi+Chen%2C+Guiyang+Hou%2C+Shengpei+Jiang%2C+Wenqiao+Zhang%2C+Yongliang+Shen%2C+Weiming+Lu%2C+and+Yueting+Zhuang%2E+2025b%2E+Viewspatial%2Dbench%3A+Evaluati)
30. Fangjun Li, David C. Hogg, and Anthony G. Cohn. 2024. Advancing spatial reasoning in large language models: An in-depth evaluation and enhancement using the stepgame benchmark. In Thirty-Eighth AAAI Conference on Artificial Intelligence, AAAI 2024, Thirty-Sixth Conference on Innovative Applications… [find](https://scholar.google.com/scholar?q=Fangjun+Li%2C+David+C%2E+Hogg%2C+and+Anthony+G%2E+Cohn%2E+2024%2E+Advancing+spatial+reasoning+in+large+language+models%3A+An+in%2Ddepth+evaluation+and+enhancement+using+the+stepgame+benchmark%2E+In+Thirty%2DEighth+AAAI+C)
31. Kenneth Li, Oam Patel, Fernanda B. Viégas, Hanspeter Pfister, and Martin Wattenberg. 2023. [Inference-time intervention: Eliciting truthful answers from a language model](https://md.research.elamur.ai/free/en/papers/li-2023-inference-time-intervention-eliciting-truthful/). In Advances in Neural Information Processing Systems 36: Annual Conference on Neural Information Processing Systems 2023, NeurIPS…
32. Wufei Ma, Haoyu Chen, Guofeng Zhang, Celso M. de Melo, Alan L. Yuille, and Jieneng Chen. 2024. 3dsrbench: A comprehensive 3d spatial reasoning benchmark. CoRR, abs/2412.07825. — [arXiv:2412.07825](https://arxiv.org/abs/2412.07825) [find](https://scholar.google.com/scholar?q=Wufei+Ma%2C+Haoyu+Chen%2C+Guofeng+Zhang%2C+Celso+M%2E+de+Melo%2C+Alan+L%2E+Yuille%2C+and+Jieneng+Chen%2E+2024%2E+3dsrbench%3A+A+comprehensive+3d+spatial+reasoning+benchmark%2E+CoRR%2C+abs%2F2412%2E07825%2E+%E2%80%94+%5BarXiv%3A2412%2E07825%5D%28htt)
33. Andreas Madsen, Siva Reddy, and Sarath Chandar. 2023. Post-hoc interpretability for neural NLP: A survey. ACM Comput. Surv., 55(8):155:1–155:42. [find](https://scholar.google.com/scholar?q=Andreas+Madsen%2C+Siva+Reddy%2C+and+Sarath+Chandar%2E+2023%2E+Post%2Dhoc+interpretability+for+neural+NLP%3A+A+survey%2E+ACM+Comput%2E+Surv%2E%2C+55%288%29%3A155%3A1%E2%80%93155%3A42%2E)
34. Xiangyu Qi, Ashwinee Panda, Kaifeng Lyu, Xiao Ma, Subhrajit Roy, Ahmad Beirami, Prateek Mittal, and Peter Henderson. 2025. [Safety alignment should be made more than just a few tokens deep](https://md.research.elamur.ai/free/en/papers/qi-2024-safety-alignment-should-made-more/). In The Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore, April 24-28, 2025…
35. Santhosh Kumar Ramakrishnan, Erik Wijmans, Philipp Krähenbühl, and Vladlen Koltun. 2025. Does spatial cognition emerge in frontier models? In The Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore, April 24-28, 2025. OpenReview.net. [find](https://scholar.google.com/scholar?q=Santhosh+Kumar+Ramakrishnan%2C+Erik+Wijmans%2C+Philipp+Kr%C3%A4henb%C3%BChl%2C+and+Vladlen+Koltun%2E+2025%2E+Does+spatial+cognition+emerge+in+frontier+models%3F+In+The+Thirteenth+International+Conference+on+Learning+Repres)
36. Zhengxiang Shi, Qiang Zhang, and Aldo Lipani. 2022. [Stepgame: A new benchmark for robust multi-hop spatial reasoning in texts](https://md.research.elamur.ai/free/en/papers/shi-2022-stepgame-new-benchmark-robust-multi/). In Thirty-Sixth AAAI Conference on Artificial Intelligence, AAAI 2022, Thirty-Fourth Conference on Innovative Applications of Artificial Intelligence, IAAI 2022, The Twelvet…
37. Aarohi Srivastava, Abhinav Rastogi, Abhishek Rao, Abu Awal Md Shoeb, Abubakar Abid, Adam Fisch, Adam R. Brown, Adam Santoro, Aditya Gupta, Adrià Garriga-Alonso, Agnieszka Kluska, Aitor Lewkowycz, Akshat Agarwal, Alethea Power, Alex Ray, Alex Warstadt, Alexander W. Kocurek, Ali Safaya, Ali Tazarv, Al… — [Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language models](https://md.research.elamur.ai/free/en/papers/srivastava-2022-imitation-game-quantifying-extrapolating-capabilities/)
38. Trans. Mach. Learn. Res., 2023. [find](https://scholar.google.com/scholar?q=Trans%2E+Mach%2E+Learn%2E+Res%2E%2C+2023%2E)
39. Yihong Tang, Ao Qu, Zhaokai Wang, Dingyi Zhuang, Zhaofeng Wu, Wei Ma, Shenhao Wang, Yunhan Zheng, Zhan Zhao, and Jinhua Zhao. 2024. Sparkle: Mastering basic spatial capabilities in vision language models elicits generalization to composite spatial reasoning. CoRR, abs/2410.16162. — [arXiv:2410.16162](https://arxiv.org/abs/2410.16162) [find](https://scholar.google.com/scholar?q=Yihong+Tang%2C+Ao+Qu%2C+Zhaokai+Wang%2C+Dingyi+Zhuang%2C+Zhaofeng+Wu%2C+Wei+Ma%2C+Shenhao+Wang%2C+Yunhan+Zheng%2C+Zhan+Zhao%2C+and+Jinhua+Zhao%2E+2024%2E+Sparkle%3A+Mastering+basic+spatial+capabilities+in+vision+language+mod)
40. Llama Team. 2024. [The llama 3 herd of models](https://md.research.elamur.ai/free/en/papers/grattafiori-2024-llama-3-herd-models/). CoRR, abs/2407.21783.
41. Curt Tigges, Michael Hanna, Qinan Yu, and Stella Biderman. 2024. LLM circuit analyses are consistent across training and scale. In Advances in Neural Information Processing Systems 38: Annual Conference on Neural Information Processing Systems 2024, NeurIPS 2024, Vancouver, BC, Canada, December 10 -… [find](https://scholar.google.com/scholar?q=Curt+Tigges%2C+Michael+Hanna%2C+Qinan+Yu%2C+and+Stella+Biderman%2E+2024%2E+LLM+circuit+analyses+are+consistent+across+training+and+scale%2E+In+Advances+in+Neural+Information+Processing+Systems+38%3A+Annual+Conferen)
42. Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Dan Bikel, Lukas Blecher, Cristian Canton-Ferrer, Moya Chen, Guillem Cucurull, David Esiobu, Jude Fernandes, Jeremy Fu, Wenyin Fu, Brian Fuller,… — [Llama 2: Open Foundation and Fine-Tuned Chat Models](https://md.research.elamur.ai/free/en/papers/touvron-2023-llama-2-open-foundation-fine/)
43. Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. 2017. [Attention is all you need](https://md.research.elamur.ai/free/en/papers/vaswani-2017-attention-all-need/). In Advances in Neural Information Processing Systems 30: Annual Conference on Neural Information Processing Systems 2017, December 4-9, 2017,…
44. Chenxi Wang, Xiang Chen, Ningyu Zhang, Bozhong Tian, Haoming Xu, Shumin Deng, and Huajun Chen. 2025. MLLM can see? dynamic correction decoding for hallucination mitigation. In The Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore, April 24-28, 2025. OpenReview.net… [find](https://scholar.google.com/scholar?q=Chenxi+Wang%2C+Xiang+Chen%2C+Ningyu+Zhang%2C+Bozhong+Tian%2C+Haoming+Xu%2C+Shumin+Deng%2C+and+Huajun+Chen%2E+2025%2E+MLLM+can+see%3F+dynamic+correction+decoding+for+hallucination+mitigation%2E+In+The+Thirteenth+Internati)
45. Jiayu Wang, Yifei Ming, Zhenmei Shi, Vibhav Vineet, Xin Wang, Sharon Li, and Neel Joshi. 2024. Is A picture worth A thousand words? delving into spatial reasoning for vision language models. In Advances in Neural Information Processing Systems 38: Annual Conference on Neural Information Processing S… [find](https://scholar.google.com/scholar?q=Jiayu+Wang%2C+Yifei+Ming%2C+Zhenmei+Shi%2C+Vibhav+Vineet%2C+Xin+Wang%2C+Sharon+Li%2C+and+Neel+Joshi%2E+2024%2E+Is+A+picture+worth+A+thousand+words%3F+delving+into+spatial+reasoning+for+vision+language+models%2E+In+Advanc)
46. Kevin Ro Wang, Alexandre Variengien, Arthur Conmy, Buck Shlegeris, and Jacob Steinhardt. 2023. Interpretability in the wild: a circuit for indirect object identification in GPT-2 small. In The Eleventh International Conference on Learning Representations, ICLR 2023, Kigali, Rwanda, May 1-5, 2023. Op… [find](https://scholar.google.com/scholar?q=Kevin+Ro+Wang%2C+Alexandre+Variengien%2C+Arthur+Conmy%2C+Buck+Shlegeris%2C+and+Jacob+Steinhardt%2E+2023%2E+Interpretability+in+the+wild%3A+a+circuit+for+indirect+object+identification+in+GPT%2D2+small%2E+In+The+Elevent)
47. Boyi Wei, Kaixuan Huang, Yangsibo Huang, Tinghao Xie, Xiangyu Qi, Mengzhou Xia, Prateek Mittal, Mengdi Wang, and Peter Henderson. 2024. [Assessing the brittleness of safety alignment via pruning and low-rank modifications](https://md.research.elamur.ai/free/en/papers/wei-2024-assessing-brittleness-safety-alignment-pruning/). In Forty-first International Conference on Machine Learning, ICML 2024, Vienna…
48. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed H. Chi, Quoc V. Le, and Denny Zhou. 2022. [Chain-of-thought prompting elicits reasoning in large language models](https://md.research.elamur.ai/free/en/papers/wei-2022-chain-thought-prompting-elicits-reasoning/). In Advances in Neural Information Processing Systems 35: Annual Conference on Neural Information Processi…
49. Qiyao Xue, Weichen Liu, Shiqi Wang, Haoming Wang, Yuyang Wu, and Wei Gao. 2025. Reasoning path and latent state analysis for multi-view visual spatial reasoning: A cognitive science perspective. Preprint, arXiv:2512.02340. — [arXiv:2512.02340](https://arxiv.org/abs/2512.02340) [find](https://scholar.google.com/scholar?q=Qiyao+Xue%2C+Weichen+Liu%2C+Shiqi+Wang%2C+Haoming+Wang%2C+Yuyang+Wu%2C+and+Wei+Gao%2E+2025%2E+Reasoning+path+and+latent+state+analysis+for+multi%2Dview+visual+spatial+reasoning%3A+A+cognitive+science+perspective%2E+Prepr)
50. Yutaro Yamada, Yihan Bao, Andrew Kyle Lampinen, Jungo Kasai, and Ilker Yildirim. 2024. Evaluating spatial understanding of large language models. [find](https://scholar.google.com/scholar?q=Yutaro+Yamada%2C+Yihan+Bao%2C+Andrew+Kyle+Lampinen%2C+Jungo+Kasai%2C+and+Ilker+Yildirim%2E+2024%2E+Evaluating+spatial+understanding+of+large+language+models%2E)
51. Trans. Mach. Learn. Res., 2024. [find](https://scholar.google.com/scholar?q=Trans%2E+Mach%2E+Learn%2E+Res%2E%2C+2024%2E)
52. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, Chujie Zheng, Dayiheng Liu, Fan Zhou, Fei Huang, Feng Hu, Hao Ge, Haoran Wei, Huan Lin, Jialong Tang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jian Yang, Jiaxi Yang, Jingren Zhou, Jun… — [Qwen3 Technical Report](https://md.research.elamur.ai/free/en/papers/yang-2025-qwen3-technical-report/)
53. An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, Huan Lin, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxi Yang, Jingren Zhou, Junyang Lin, Kai Dang, Keming Lu, Keqin Bao, Kexin Yang, Le Yu, Mei Li, Mingfeng Xue, Pei… [find](https://scholar.google.com/scholar?q=An+Yang%2C+Baosong+Yang%2C+Beichen+Zhang%2C+Binyuan+Hui%2C+Bo+Zheng%2C+Bowen+Yu%2C+Chengyuan+Li%2C+Dayiheng+Liu%2C+Fei+Huang%2C+Haoran+Wei%2C+Huan+Lin%2C+Jian+Yang%2C+Jianhong+Tu%2C+Jianwei+Zhang%2C+Jianxin+Yang%2C+Jiaxi+Yang%2C+Jin)
54. Jihan Yang, Shusheng Yang, Anjali W. Gupta, Rilyn Han, Li Fei-Fei, and Saining Xie. 2025b. [Thinking in space: How multimodal large language models see, remember, and recall spaces](https://md.research.elamur.ai/free/en/papers/yang-2024-thinking-space-multimodal-large-language/). In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2025, Nashville, TN, USA, June 11-15, 2025, pag…
55. Zhen Yang, Ping Jian, and Chengzhi Li. 2025c. Option symbol matters: Investigating and mitigating multiple-choice option symbol bias of large language models. In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language… [find](https://scholar.google.com/scholar?q=Zhen+Yang%2C+Ping+Jian%2C+and+Chengzhi+Li%2E+2025c%2E+Option+symbol+matters%3A+Investigating+and+mitigating+multiple%2Dchoice+option+symbol+bias+of+large+language+models%2E+In+Proceedings+of+the+2025+Conference+of+)
56. Fangcong Yin, Xi Ye, and Greg Durrett. 2024. Lofit: Localized fine-tuning on LLM representations. In Advances in Neural Information Processing Systems 38: Annual Conference on Neural Information Processing Systems 2024, NeurIPS 2024, Vancouver, BC, Canada, December 10 - 15, 2024. [find](https://scholar.google.com/scholar?q=Fangcong+Yin%2C+Xi+Ye%2C+and+Greg+Durrett%2E+2024%2E+Lofit%3A+Localized+fine%2Dtuning+on+LLM+representations%2E+In+Advances+in+Neural+Information+Processing+Systems+38%3A+Annual+Conference+on+Neural+Information+Proce)
57. Le Yu, Bowen Yu, Haiyang Yu, Fei Huang, and Yongbin Li. 2024. [Language models are super mario: Absorbing abilities from homologous models as a free lunch](https://md.research.elamur.ai/free/en/papers/yu-2023-language-models-super-mario-absorbing/). In Forty-first International Conference on Machine Learning, ICML 2024, Vienna, Austria, July 21-27, 2024. OpenReview.net.
58. Zeping Yu and Sophia Ananiadou. 2024. Interpreting arithmetic mechanism in large language models through comparative neuron analysis. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, EMNLP 2024, Miami, FL, USA, November 12-16, 2024, pages 3293–3306. Associat… [find](https://scholar.google.com/scholar?q=Zeping+Yu+and+Sophia+Ananiadou%2E+2024%2E+Interpreting+arithmetic+mechanism+in+large+language+models+through+comparative+neuron+analysis%2E+In+Proceedings+of+the+2024+Conference+on+Empirical+Methods+in+Natu)
59. Zeping Yu and Sophia Ananiadou. 2025. Understanding and mitigating gender bias in llms via interpretable neuron editing. CoRR, abs/2501.14457. — [arXiv:2501.14457](https://arxiv.org/abs/2501.14457) [find](https://scholar.google.com/scholar?q=Zeping+Yu+and+Sophia+Ananiadou%2E+2025%2E+Understanding+and+mitigating+gender+bias+in+llms+via+interpretable+neuron+editing%2E+CoRR%2C+abs%2F2501%2E14457%2E+%E2%80%94+%5BarXiv%3A2501%2E14457%5D%28https%3A%2F%2Farxiv%2Eorg%2Fabs%2F2501%2E14457%29)
60. Fred Zhang and Neel Nanda. 2024. [Towards best practices of activation patching in language models: Metrics and methods](https://md.research.elamur.ai/free/en/papers/zhang-2023-towards-best-practices-activation-patching/). In The Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024. OpenReview.net.
61. Wei Zhang, Chaoqun Wan, Yonggang Zhang, Yiu-ming Cheung, Xinmei Tian, Xu Shen, and Jieping Ye. 2024b. Interpreting and improving large language models in arithmetic calculation. In Forty-first International Conference on Machine Learning, ICML 2024, Vienna, Austria, July 21-27, 2024. OpenReview.net. [find](https://scholar.google.com/scholar?q=Wei+Zhang%2C+Chaoqun+Wan%2C+Yonggang+Zhang%2C+Yiu%2Dming+Cheung%2C+Xinmei+Tian%2C+Xu+Shen%2C+and+Jieping+Ye%2E+2024b%2E+Interpreting+and+improving+large+language+models+in+arithmetic+calculation%2E+In+Forty%2Dfirst+Interna)
62. Wenyu Zhang, Wei En Ng, Lixin Ma, Yuwen Wang, Junqi Zhao, Allison Koenecke, Boyang Li, and Lu Wang. 2025c. SPHERE: unveiling spatial blind spots in vision-language models through hierarchical evaluation. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volu… [find](https://scholar.google.com/scholar?q=Wenyu+Zhang%2C+Wei+En+Ng%2C+Lixin+Ma%2C+Yuwen+Wang%2C+Junqi+Zhao%2C+Allison+Koenecke%2C+Boyang+Li%2C+and+Lu+Wang%2E+2025c%2E+SPHERE%3A+unveiling+spatial+blind+spots+in+vision%2Dlanguage+models+through+hierarchical+evaluati)
63. Yuyou Zhang, Radu Corcodel, Chiori Hori, Anoop Cherian, and Ding Zhao. 2025d. Spinbench: Perspective and rotation as a lens on spatial reasoning in vlms. CoRR, abs/2509.25390. — [arXiv:2509.25390](https://arxiv.org/abs/2509.25390) [find](https://scholar.google.com/scholar?q=Yuyou+Zhang%2C+Radu+Corcodel%2C+Chiori+Hori%2C+Anoop+Cherian%2C+and+Ding+Zhao%2E+2025d%2E+Spinbench%3A+Perspective+and+rotation+as+a+lens+on+spatial+reasoning+in+vlms%2E+CoRR%2C+abs%2F2509%2E25390%2E+%E2%80%94+%5BarXiv%3A2509%2E25390%5D%28htt)
64. Zhenhong Zhou, Haiyang Yu, Xinghua Zhang, Rongwu Xu, Fei Huang, Kun Wang, Yang Liu, Junfeng Fang, and Yongbin Li. 2025. On the role of attention heads in large language model safety. In The Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore, April 24-28, 2025. Open… [find](https://scholar.google.com/scholar?q=Zhenhong+Zhou%2C+Haiyang+Yu%2C+Xinghua+Zhang%2C+Rongwu+Xu%2C+Fei+Huang%2C+Kun+Wang%2C+Yang+Liu%2C+Junfeng+Fang%2C+and+Yongbin+Li%2E+2025%2E+On+the+role+of+attention+heads+in+large+language+model+safety%2E+In+The+Thirteenth)
65. Jinguo Zhu, Weiyun Wang, Zhe Chen, Zhaoyang Liu, Shenglong Ye, Lixin Gu, Hao Tian, Yuchen Duan, Weijie Su, Jie Shao, and et al. 2025. [Internvl3: Exploring advanced training and test-time recipes for open-source multimodal models](https://md.research.elamur.ai/free/en/papers/zhu-2025-internvl3-exploring-advanced-training/). CoRR, abs/2504.10479.
66. Hanning Zhang, Shizhe Diao, Yong Lin, Yi R. Fung, Qing Lian, Xingyao Wang, Yangyi Chen, Heng Ji, and Tong Zhang. 2024a. [R-tuning: Instructing large language models to say ’i don’t know’](https://md.research.elamur.ai/free/en/papers/zhang-2024-r-tuning-instructing-large-language/). In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguist…
67. Hong Zhang, Jiaxu Wan, Jing Zhang, Ding Yuan, Xu-liang Li, and Yifan Yang. 2025a. P2ftrack: Multi-object tracking with motion prior and feature posterior. ACM Trans. Multim. Comput. Commun. Appl., 21(1):22:1–22:22. [find](https://scholar.google.com/scholar?q=Hong+Zhang%2C+Jiaxu+Wan%2C+Jing+Zhang%2C+Ding+Yuan%2C+Xu%2Dliang+Li%2C+and+Yifan+Yang%2E+2025a%2E+P2ftrack%3A+Multi%2Dobject+tracking+with+motion+prior+and+feature+posterior%2E+ACM+Trans%2E+Multim%2E+Comput%2E+Commun%2E+Appl%2E%2C+21%28)
68. Jiahui Zhang, Yurui Chen, Yanpeng Zhou, Yueming Xu, Ze Huang, Jilin Mei, Junhui Chen, Yu-Jie Yuan, Xinyue Cai, Guowei Huang, Xingyue Quan, Hang Xu, and Li Zhang. 2025b. From flatland to space: Teaching vision-language models to perceive and reason in 3d. CoRR, abs/2503.22976. — [arXiv:2503.22976](https://arxiv.org/abs/2503.22976) [find](https://scholar.google.com/scholar?q=Jiahui+Zhang%2C+Yurui+Chen%2C+Yanpeng+Zhou%2C+Yueming+Xu%2C+Ze+Huang%2C+Jilin+Mei%2C+Junhui+Chen%2C+Yu%2DJie+Yuan%2C+Xinyue+Cai%2C+Guowei+Huang%2C+Xingyue+Quan%2C+Hang+Xu%2C+and+Li+Zhang%2E+2025b%2E+From+flatland+to+space%3A+Teachin)

*Updated 2 November 2026*