---
title: "Mechanisms of Prompt-Induced Hallucination in"
description: "This paper studies that failure, called prompt-induced hallucination (PIH), using object counting. 56 references, cited by 1 paper here."
authors: ["William Rudman", "Michal Golovanevsky", "Dana Arad", "Yonatan Belinkov", "Carsten Eickhoff", "Ritambhara Singh", "Kyle Mahowald"]
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.1941/"
doi: "10.18653/v1/2026.acl-long.1941"
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/rudman-2026-mechanisms-prompt-induced-hallucination-vision/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/rudman-2026-mechanisms-prompt-induced-hallucination-vision/)

[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/8/) / Mechanisms of Prompt-Induced Hallucination in Vision–Language Models

# Mechanisms of Prompt-Induced Hallucination in Vision–Language Models

**Authors:** [William Rudman](https://md.research.elamur.ai/free/en/authors/william-rudman/), [Michal Golovanevsky](https://md.research.elamur.ai/free/en/authors/michal-golovanevsky/), [Dana Arad](https://md.research.elamur.ai/free/en/authors/dana-arad/), [Yonatan Belinkov](https://md.research.elamur.ai/free/en/authors/yonatan-belinkov/), [Carsten Eickhoff](https://md.research.elamur.ai/free/en/authors/carsten-eickhoff/), [Ritambhara Singh](https://md.research.elamur.ai/free/en/authors/ritambhara-singh/) and [Kyle Mahowald](https://md.research.elamur.ai/free/en/authors/kyle-mahowald/) · **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.1941/) · [PDF](https://aclanthology.org/2026.acl-long.1941.pdf) · [DOI 10.18653/v1/2026.acl-long.1941](https://doi.org/10.18653/v1/2026.acl-long.1941) · [OpenAlex](https://openalex.org/W7166809752)

**Citations:** 1 citations (OpenAlex, 2026-10-14) · 1 in our library, of the 149,439 papers whose bibliographies we read

## Summary

Vision–language models sometimes trust a user's wording more than what the image shows. This paper studies that failure, called prompt-induced hallucination (PIH), using object counting. Images from CountBench are first counted correctly by the model. The same images are then paired with a request to describe more objects than are actually present (N + k). The authors test LLaVA-OneVision-7B, Qwen2-VL-7B and Janus-Pro-7B.

With small scenes, the models usually correct the inflated number. Once more than four objects are present, they mostly accept the prompt's figure. This shift tracks the model's own falling confidence in its baseline count.

Mechanistic analysis locates a small group of attention heads, mostly in early language-model layers, that drive the copying. Switching them off requires no retraining. It cuts PIH substantially, moves attention toward the image, and carries over to a color task. However, the way copying is suppressed differs from model to model. The work matters because it points to cheap, targeted fixes for visual grounding.

## Key points

- **Error rates by scene size:** For 2–4 objects, PIH still appears in roughly 20–40% of cases, mostly when the prompt overshoots by one. Above four objects, prompt-matching answers climb to about 80–90%, even for very large offsets such as k = 50.
- **Link to confidence:** Per-sample Pearson correlation between baseline confidence and conforming to the prompt is 0.37 (Qwen-VL), 0.46 (Janus-Pro) and 0.1 (LLaVA-OneVision).
- **Effect of ablation:** Removing the PIH heads lowers hallucinations by at least 40% and restores correct counts by up to 54%. On a wrong-color prompt task, prompt-color copying drops by up to 94.25%.
- **Model-specific mechanisms:** Qwen-VL and LLaVA-OneVision share many PIH heads, pointing to their common language backbone. In Qwen-VL, ablation raises format copying (answering in the prompt's digit form) from 40.21% to 53.95%. LLaVA-OneVision suppresses all copying, shifts attention toward the image by about 12%, and also counts better on the baseline prompt.
- **Code:** Released on GitHub at michalg04/prompt-induced_hallucinations.

*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

- [Harvard University](https://md.research.elamur.ai/free/en/universities/harvard-university/) — University, Cambridge, United States · 126 works · 45,130 citations
- [The University of Texas at Austin](https://md.research.elamur.ai/free/en/universities/the-university-of-texas-at-austin/) — University, Austin, United States · 119 works · 57,950 citations
- [Brown University](https://md.research.elamur.ai/free/en/universities/brown-university/) — University, Providence, United States · 80 works · 26,036 citations
- [Technion – Israel Institute of Technology](https://md.research.elamur.ai/free/en/universities/technion-israel-institute-of-technology/) — University, Haifa, Israel · 54 works · 7,237 citations
- [University of Tübingen](https://md.research.elamur.ai/free/en/universities/university-of-tubingen/) — University, Tübingen, Germany · 78 works · 41,280 citations

📚 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 · [🖼️ Vision & language](https://md.research.elamur.ai/free/en/topics/multimodal/) · 811 works

## Cite

Cite as: William Rudman, Michal Golovanevsky, Dana Arad, Yonatan Belinkov, Carsten Eickhoff, Ritambhara Singh and Kyle Mahowald. 2026. Mechanisms of Prompt-Induced Hallucination in Vision–Language Models. 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{rudman2026mechanisms,
  title = {{Mechanisms of Prompt-Induced Hallucination in Vision–Language Models}},
  author = {William Rudman, Michal Golovanevsky, Dana Arad, Yonatan Belinkov, Carsten Eickhoff, Ritambhara Singh and Kyle Mahowald},
  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.1941},
  url = {https://doi.org/10.18653/v1/2026.acl-long.1941}
}
```

- [← previous Maximizing Local Entropy Where It Matters: Prefix-Aware Localized LLM Unlearning](https://md.research.elamur.ai/free/en/papers/zhai-2026-maximizing-local-entropy-where-it/)
- [next → MedDCR: Learning to Design Agentic Workflows for Medical Coding](https://md.research.elamur.ai/free/en/papers/zheng-2026-meddcr-learning-design-agentic-workflows/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/8/)

## References

56 references, 20 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Dana Arad, Hadas Orgad, and Yonatan Belinkov. 2024. Refact: Updating text-to-image models by editing the text encoder. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), pages 2537… [find](https://scholar.google.com/scholar?q=Dana+Arad%2C+Hadas+Orgad%2C+and+Yonatan+Belinkov%2E+2024%2E+Refact%3A+Updating+text%2Dto%2Dimage+models+by+editing+the+text+encoder%2E+In+Proceedings+of+the+2024+Conference+of+the+North+American+Chapter+of+the+Associ)
2. Gino Brunner, Yang Liu, Damian Pascual, Oliver Richter, Massimiliano Ciaramita, and Roger Wattenhofer. 2019. On identifiability in transformers. arXiv preprint arXiv:1908.04211. — [arXiv:1908.04211](https://arxiv.org/abs/1908.04211) [find](https://scholar.google.com/scholar?q=Gino+Brunner%2C+Yang+Liu%2C+Damian+Pascual%2C+Oliver+Richter%2C+Massimiliano+Ciaramita%2C+and+Roger+Wattenhofer%2E+2019%2E+On+identifiability+in+transformers%2E+arXiv+preprint+arXiv%3A1908%2E04211%2E+%E2%80%94+%5BarXiv%3A1908%2E04211%5D%28h)
3. Xiaokang Chen, Zhiyu Wu, Xingchao Liu, Zizheng Pan, Wen Liu, Zhenda Xie, Xingkai Yu, and Chong Ruan. 2025. [Janus-pro: Unified multimodal understanding and generation with data and model scaling](https://md.research.elamur.ai/free/en/papers/chen-2025-janus-pro-unified-multimodal/). Preprint, arXiv:2501.17811.
4. Ailin Deng, Tri Cao, Zhirui Chen, and Bryan Hooi. 2025. Words or vision: Do vision-language models have blind faith in text? In Proceedings of the Computer Vision and Pattern Recognition Conference, pages 3867–3876. [find](https://scholar.google.com/scholar?q=Ailin+Deng%2C+Tri+Cao%2C+Zhirui+Chen%2C+and+Bryan+Hooi%2E+2025%2E+Words+or+vision%3A+Do+vision%2Dlanguage+models+have+blind+faith+in+text%3F+In+Proceedings+of+the+Computer+Vision+and+Pattern+Recognition+Conference%2C+p)
5. Jessica Maria Echterhoff, Yao Liu, Abeer Alessa, Julian McAuley, and Zexue He. 2024. Cognitive bias in decision-making with llms. In Findings of the association for computational linguistics: EMNLP 2024, pages 12640–12653. [find](https://scholar.google.com/scholar?q=Jessica+Maria+Echterhoff%2C+Yao+Liu%2C+Abeer+Alessa%2C+Julian+McAuley%2C+and+Zexue+He%2E+2024%2E+Cognitive+bias+in+decision%2Dmaking+with+llms%2E+In+Findings+of+the+association+for+computational+linguistics%3A+EMNLP+20)
6. Sheridan Feucht, Eric Todd, Byron Wallace, and David Bau. 2025. The dual-route model of induction. arXiv preprint arXiv:2504.03022. — [arXiv:2504.03022](https://arxiv.org/abs/2504.03022) [find](https://scholar.google.com/scholar?q=Sheridan+Feucht%2C+Eric+Todd%2C+Byron+Wallace%2C+and+David+Bau%2E+2025%2E+The+dual%2Droute+model+of+induction%2E+arXiv+preprint+arXiv%3A2504%2E03022%2E+%E2%80%94+%5BarXiv%3A2504%2E03022%5D%28https%3A%2F%2Farxiv%2Eorg%2Fabs%2F2504%2E03022%29)
7. Stella Frank, Emanuele Bugliarello, and Desmond Elliott. 2021. Vision-and-language or vision-for-language? on cross-modal influence in multimodal transformers. arXiv preprint arXiv:2109.04448. [find](https://scholar.google.com/scholar?q=Stella+Frank%2C+Emanuele+Bugliarello%2C+and+Desmond+Elliott%2E+2021%2E+Vision%2Dand%2Dlanguage+or+vision%2Dfor%2Dlanguage%3F+on+cross%2Dmodal+influence+in+multimodal+transformers%2E+arXiv+preprint+arXiv%3A2109%2E04448%2E)
8. Deqing Fu, Ruohao Guo, Ghazal Khalighinejad, Ollie Liu, Bhuwan Dhingra, Dani Yogatama, Robin Jia, and Willie Neiswanger. 2024. Isobench: Benchmarking multimodal foundation models on isomorphic representations. arXiv preprint arXiv:2404.01266. — [arXiv:2404.01266](https://arxiv.org/abs/2404.01266) [find](https://scholar.google.com/scholar?q=Deqing+Fu%2C+Ruohao+Guo%2C+Ghazal+Khalighinejad%2C+Ollie+Liu%2C+Bhuwan+Dhingra%2C+Dani+Yogatama%2C+Robin+Jia%2C+and+Willie+Neiswanger%2E+2024%2E+Isobench%3A+Benchmarking+multimodal+foundation+models+on+isomorphic+represe)
9. Michal Golovanevsky, William Rudman, Michael Lepori, Amir Bar, Ritambhara Singh, and Carsten Eickhoff. 2025a. [Pixels versus priors: Controlling knowledge priors in vision-language models through visual counterfacts](https://md.research.elamur.ai/free/en/papers/golovanevsky-2025-pixels-versus-priors-controlling-knowledge/). arXiv preprint arXiv:2505.17127.
10. Michal Golovanevsky, William Rudman, Vedant Palit, Ritambhara Singh, and Carsten Eickhoff. 2025b. What do vlms notice? a mechanistic interpretability pipeline for gaussian-noise-free text-image corruption and evaluation. Preprint, arXiv:2406.16320. [find](https://scholar.google.com/scholar?q=Michal+Golovanevsky%2C+William+Rudman%2C+Vedant+Palit%2C+Ritambhara+Singh%2C+and+Carsten+Eickhoff%2E+2025b%2E+What+do+vlms+notice%3F+a+mechanistic+interpretability+pipeline+for+gaussian%2Dnoise%2Dfree+text%2Dimage+corrup)
11. Christopher Grimsley, Elijah Mayfield, and Julia RS Bursten. 2020. Why [attention is not explanation](https://md.research.elamur.ai/free/en/papers/jain-2019-attention-is-not-explanation/): Surgical intervention and causal reasoning about neural models. In Proceedings of the Twelfth Language Resources and Evaluation Conference, pages 1780– 1790.
12. Tianze Hua, Tian Yun, and Ellie Pavlick. 2025. How do vision-language models process conflicting information across modalities? arXiv preprint arXiv:2507.01790. [find](https://scholar.google.com/scholar?q=Tianze+Hua%2C+Tian+Yun%2C+and+Ellie+Pavlick%2E+2025%2E+How+do+vision%2Dlanguage+models+process+conflicting+information+across+modalities%3F+arXiv+preprint+arXiv%3A2507%2E01790%2E)
13. Lei Huang, Weijiang Yu, Weitao Ma, Weihong Zhong, Zhangyin Feng, Haotian Wang, Qianglong Chen, Weihua Peng, Xiaocheng Feng, Bing Qin, et al. 2025. [A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions](https://md.research.elamur.ai/free/en/papers/huang-2023-survey-hallucination-large-language-models/). ACM Transactions on Information Systems, 43(2):…
14. Sarthak Jain and Byron C Wallace. 2019. [Attention is not explanation](https://md.research.elamur.ai/free/en/papers/jain-2019-attention-is-not-explanation/). arXiv preprint arXiv:1902.10186.
15. Ziwei Ji, Nayeon Lee, Rita Frieske, Tiezheng Yu, Dan Su, Yan Xu, Etsuko Ishii, Ye Jin Bang, Andrea Madotto, and Pascale Fung. 2023. [Survey of hallucination in natural language generation](https://md.research.elamur.ai/free/en/papers/ji-2022-survey-hallucination-natural-language-generation/). ACM computing surveys, 55(12):1–38.
16. Erik Jones and Jacob Steinhardt. 2022. Capturing failures of large language models via human cognitive biases. Advances in Neural Information Processing Systems, 35:11785–11799. [find](https://scholar.google.com/scholar?q=Erik+Jones+and+Jacob+Steinhardt%2E+2022%2E+Capturing+failures+of+large+language+models+via+human+cognitive+biases%2E+Advances+in+Neural+Information+Processing+Systems%2C+35%3A11785%E2%80%9311799%2E)
17. Nitish Joshi, Javier Rando, Abulhair Saparov, Najoung Kim, and He He. 2024. Personas as a way to model truthfulness in language models. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 6346–6359. [find](https://scholar.google.com/scholar?q=Nitish+Joshi%2C+Javier+Rando%2C+Abulhair+Saparov%2C+Najoung+Kim%2C+and+He+He%2E+2024%2E+Personas+as+a+way+to+model+truthfulness+in+language+models%2E+In+Proceedings+of+the+2024+Conference+on+Empirical+Methods+in+Na)
18. Bo Li, Yuanhan Zhang, Dong Guo, Renrui Zhang, Feng Li, Hao Zhang, Kaichen Zhang, Peiyuan Zhang, Yanwei Li, Ziwei Liu, and Chunyuan Li. 2024a. [Llava-onevision: Easy visual task transfer](https://md.research.elamur.ai/free/en/papers/li-2024-llava-onevision-easy-visual-task/). Preprint, arXiv:2408.03326.
19. Fei-Fei Li, Marco Andreeto, Marc’Aurelio Ranzato, and Pietro Perona. 2022. Caltech 101. [find](https://scholar.google.com/scholar?q=Fei%2DFei+Li%2C+Marco+Andreeto%2C+Marc%E2%80%99Aurelio+Ranzato%2C+and+Pietro+Perona%2E+2022%2E+Caltech+101%2E)
20. Kenneth Li, Tianle Liu, Naomi Bashkansky, David Bau, Fernanda Viégas, Hanspeter Pfister, and Martin Wattenberg. 2024b. Measuring and controlling instruction (in) stability in language model dialogs. The Conference on Language Modeling COLM 2024. [find](https://scholar.google.com/scholar?q=Kenneth+Li%2C+Tianle+Liu%2C+Naomi+Bashkansky%2C+David+Bau%2C+Fernanda+Vi%C3%A9gas%2C+Hanspeter+Pfister%2C+and+Martin+Wattenberg%2E+2024b%2E+Measuring+and+controlling+instruction+%28in%29+stability+in+language+model+dialogs%2E+T)
21. Yifan Li, Yifan Du, Kun Zhou, Jinpeng Wang, Wayne Xin Zhao, and Ji-Rong Wen. 2023. [Evaluating object hallucination in large vision-language models](https://md.research.elamur.ai/free/en/papers/li-2023-evaluating-object-hallucination-large-vision/). In Proceedings of the 2023 conference on empirical methods in natural language processing, pages 292–305.
22. Simon Malberg, Roman Poletukhin, Carolin Schuster, and Georg Groh Groh. 2025. A comprehensive evaluation of cognitive biases in llms. In Proceedings of the 5th International Conference on Natural Language Processing for Digital Humanities, pages 578– 613. [find](https://scholar.google.com/scholar?q=Simon+Malberg%2C+Roman+Poletukhin%2C+Carolin+Schuster%2C+and+Georg+Groh+Groh%2E+2025%2E+A+comprehensive+evaluation+of+cognitive+biases+in+llms%2E+In+Proceedings+of+the+5th+International+Conference+on+Natural+Lang)
23. Kevin Meng, David Bau, Alex Andonian, and Yonatan Belinkov. 2023. Locating and editing factual associations in gpt. Preprint, arXiv:2202.05262. [find](https://scholar.google.com/scholar?q=Kevin+Meng%2C+David+Bau%2C+Alex+Andonian%2C+and+Yonatan+Belinkov%2E+2023%2E+Locating+and+editing+factual+associations+in+gpt%2E+Preprint%2C+arXiv%3A2202%2E05262%2E)
24. Aaron Mueller, Atticus Geiger, Sarah Wiegreffe, Dana Arad, Iván Arcuschin, Adam Belfki, Yik Siu Chan, Jaden Fiotto-Kaufman, Tal Haklay, Michael Hanna, et al. 2025. Mib: A mechanistic interpretability benchmark. arXiv preprint arXiv:2504.13151. — [arXiv:2504.13151](https://arxiv.org/abs/2504.13151) [find](https://scholar.google.com/scholar?q=Aaron+Mueller%2C+Atticus+Geiger%2C+Sarah+Wiegreffe%2C+Dana+Arad%2C+Iv%C3%A1n+Arcuschin%2C+Adam+Belfki%2C+Yik+Siu+Chan%2C+Jaden+Fiotto%2DKaufman%2C+Tal+Haklay%2C+Michael+Hanna%2C+et+al%2E+2025%2E+Mib%3A+A+mechanistic+interpretability+)
25. Yaniv Nikankin, Dana Arad, Yossi Gandelsman, and Yonatan Belinkov. 2025. Same task, different circuits: Disentangling modality-specific mechanisms in vlms. arXiv preprint arXiv:2506.09047. — [arXiv:2506.09047](https://arxiv.org/abs/2506.09047) [find](https://scholar.google.com/scholar?q=Yaniv+Nikankin%2C+Dana+Arad%2C+Yossi+Gandelsman%2C+and+Yonatan+Belinkov%2E+2025%2E+Same+task%2C+different+circuits%3A+Disentangling+modality%2Dspecific+mechanisms+in+vlms%2E+arXiv+preprint+arXiv%3A2506%2E09047%2E+%E2%80%94+%5BarXiv%3A25)
26. Catherine Olsson, Nelson Elhage, Neel Nanda, Nicholas Joseph, Nova DasSarma, Tom Henighan, Ben Mann, Amanda Askell, Yuntao Bai, Anna Chen, et al. 2022. [In-context learning and induction heads](https://md.research.elamur.ai/free/en/papers/olsson-2022-context-learning-induction-heads/). arXiv preprint arXiv:2209.11895.
27. Francesco Ortu, Zhijing Jin, Diego Doimo, and Alberto Cazzaniga. 2025. [When seeing overrides knowing: Disentangling knowledge conflicts in vision-language models](https://md.research.elamur.ai/free/en/papers/ortu-2026-when-seeing-overrides-knowing-disentangling/). arXiv preprint arXiv:2507.13868.
28. Roni Paiss, Ariel Ephrat, Omer Tov, Shiran Zada, Inbar Mosseri, Michal Irani, and Tali Dekel. 2023. Teaching clip to count to ten. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 3170–3180. [find](https://scholar.google.com/scholar?q=Roni+Paiss%2C+Ariel+Ephrat%2C+Omer+Tov%2C+Shiran+Zada%2C+Inbar+Mosseri%2C+Michal+Irani%2C+and+Tali+Dekel%2E+2023%2E+Teaching+clip+to+count+to+ten%2E+In+Proceedings+of+the+IEEE%2FCVF+International+Conference+on+Computer+V)
29. Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. 2019. [Pytorch: An imperative style, high-performance deep learning library](https://md.research.elamur.ai/free/en/papers/paszke-2019-pytorch-imperative-style-high-performance/). Advances in neural information processing systems, 32.
30. Renjie Pi, Kehao Miao, Li Peihang, Runtao Liu, Jiahui Gao, Jipeng Zhang, and Xiaofang Zhou. 2025. Pointing to a llama and call it a camel: On the sycophancy of multimodal large language models. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 20177–201… [find](https://scholar.google.com/scholar?q=Renjie+Pi%2C+Kehao+Miao%2C+Li+Peihang%2C+Runtao+Liu%2C+Jiahui+Gao%2C+Jipeng+Zhang%2C+and+Xiaofang+Zhou%2E+2025%2E+Pointing+to+a+llama+and+call+it+a+camel%3A+On+the+sycophancy+of+multimodal+large+language+models%2E+In+Pro)
31. Nikhil Prakash, Tamar Rott Shaham, Tal Haklay, Yonatan Belinkov, and David Bau. 2024. Fine-tuning enhances existing mechanisms: A case study on entity tracking. arXiv preprint arXiv:2402.14811. — [arXiv:2402.14811](https://arxiv.org/abs/2402.14811) [find](https://scholar.google.com/scholar?q=Nikhil+Prakash%2C+Tamar+Rott+Shaham%2C+Tal+Haklay%2C+Yonatan+Belinkov%2C+and+David+Bau%2E+2024%2E+Fine%2Dtuning+enhances+existing+mechanisms%3A+A+case+study+on+entity+tracking%2E+arXiv+preprint+arXiv%3A2402%2E14811%2E+%E2%80%94+%5BarX)
32. Nikhil Prakash, Natalie Shapira, Arnab Sen Sharma, Christoph Riedl, Yonatan Belinkov, Tamar Rott Shaham, David Bau, and Atticus Geiger. 2025. Language models use lookbacks to track beliefs. arXiv preprint arXiv:2505.14685. — [arXiv:2505.14685](https://arxiv.org/abs/2505.14685) [find](https://scholar.google.com/scholar?q=Nikhil+Prakash%2C+Natalie+Shapira%2C+Arnab+Sen+Sharma%2C+Christoph+Riedl%2C+Yonatan+Belinkov%2C+Tamar+Rott+Shaham%2C+David+Bau%2C+and+Atticus+Geiger%2E+2025%2E+Language+models+use+lookbacks+to+track+beliefs%2E+arXiv+prep)
33. Pooyan Rahmanzadehgervi, Logan Bolton, Mohammad Reza Taesiri, and Anh Totti Nguyen. 2024. Vision language models are blind. In Proceedings of the Asian Conference on Computer Vision, pages 18– 34. [find](https://scholar.google.com/scholar?q=Pooyan+Rahmanzadehgervi%2C+Logan+Bolton%2C+Mohammad+Reza+Taesiri%2C+and+Anh+Totti+Nguyen%2E+2024%2E+Vision+language+models+are+blind%2E+In+Proceedings+of+the+Asian+Conference+on+Computer+Vision%2C+pages+18%E2%80%93+34%2E)
34. William Rudman, Michal Golovanevsky, Amir Bar, Vedant Palit, Yann LeCun, Carsten Eickhoff, and Ritambhara Singh. 2025. Forgotten polygons: Multimodal large language models are shape-blind. arXiv preprint arXiv:2502.15969. [find](https://scholar.google.com/scholar?q=William+Rudman%2C+Michal+Golovanevsky%2C+Amir+Bar%2C+Vedant+Palit%2C+Yann+LeCun%2C+Carsten+Eickhoff%2C+and+Ritambhara+Singh%2E+2025%2E+Forgotten+polygons%3A+Multimodal+large+language+models+are+shape%2Dblind%2E+arXiv+prepr)
35. Emmanuelle Salin, Badreddine Farah, Stéphane Ayache, and Benoit Favre. 2022. Are vision-language transformers learning multimodal representations? a probing perspective. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 36, pages 11248–11257. [find](https://scholar.google.com/scholar?q=Emmanuelle+Salin%2C+Badreddine+Farah%2C+St%C3%A9phane+Ayache%2C+and+Benoit+Favre%2E+2022%2E+Are+vision%2Dlanguage+transformers+learning+multimodal+representations%3F+a+probing+perspective%2E+In+Proceedings+of+the+AAAI+Con)
36. Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pier-ric Cistac, Tim Rault, Rémi Louf, Morgan Funtowicz, et al. 2019. [Huggingface’s transformers: State-ofthe-art natural language processing](https://md.research.elamur.ai/free/en/papers/wolf-2020-transformers-state-art-natural-language/). arXiv preprint arXiv:1910.03771.
37. Sofia Serrano and Noah A Smith. 2019. [Is attention interpretable?](https://md.research.elamur.ai/free/en/papers/serrano-2019-is-attention-interpretable/) arXiv preprint arXiv:1906.03731.
38. Jian Xie, Kai Zhang, Jiangjie Chen, Renze Lou, and Yu Su. 2023. [Adaptive chameleon or stubborn sloth: Revealing the behavior of large language models in knowledge conflicts](https://md.research.elamur.ai/free/en/papers/xie-2023-adaptive-chameleon-stubborn-sloth-revealing/). In The Twelfth International Conference on Learning Representations.
39. Mrinank Sharma, Meg Tong, Tomasz Korbak, David Duvenaud, Amanda Askell, Samuel R. Bowman, Esin Durmus, Zac Hatfield-Dodds, Scott R. Johnston, Shauna Kravec, Timothy Maxwell, Sam McCandlish, Kamal Ndousse, Oliver Rausch, Nicholas Schiefer, Da Yan, Miranda Zhang, and Ethan Perez. 2024. [Towards underst…](https://md.research.elamur.ai/free/en/papers/sharma-2023-towards-understanding-sycophancy-language-models/)
40. Adi Simhi, Jonathan Herzig, Itay Itzhak, Dana Arad, Zorik Gekhman, Roi Reichart, Fazl Barez, Gabriel Stanovsky, Idan Szpektor, and Yonatan Belinkov. 2025. Hack: Hallucinations along certainty and knowledge axes. arXiv preprint arXiv:2510.24222. — [arXiv:2510.24222](https://arxiv.org/abs/2510.24222) [find](https://scholar.google.com/scholar?q=Adi+Simhi%2C+Jonathan+Herzig%2C+Itay+Itzhak%2C+Dana+Arad%2C+Zorik+Gekhman%2C+Roi+Reichart%2C+Fazl+Barez%2C+Gabriel+Stanovsky%2C+Idan+Szpektor%2C+and+Yonatan+Belinkov%2E+2025%2E+Hack%3A+Hallucinations+along+certainty+and+know)
41. Yingtian Tang, Yutaro Yamada, Yoyo Zhang, and Ilker Yildirim. 2023. When are lemons purple? the concept association bias of vision-language models. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, EMNLP 2023, Singapore, December 6-10, 2023, pages 14333– 1434… [find](https://scholar.google.com/scholar?q=Yingtian+Tang%2C+Yutaro+Yamada%2C+Yoyo+Zhang%2C+and+Ilker+Yildirim%2E+2023%2E+When+are+lemons+purple%3F+the+concept+association+bias+of+vision%2Dlanguage+models%2E+In+Proceedings+of+the+2023+Conference+on+Empirical+M)
42. SMTI Tonmoy, SM Zaman, Vinija Jain, Anku Rani, Vipula Rawte, Aman Chadha, and Amitava Das. 2024. [A comprehensive survey of hallucination mitigation techniques in large language models](https://md.research.elamur.ai/free/en/papers/tonmoy-2024-comprehensive-survey-hallucination-mitigation-techniques/). arXiv preprint arXiv:2401.01313, 6.
43. Angela van Sprang, Laurens Samson, Ana Lucic, Erman Acar, Sennay Ghebreab, and Yuki M Asano. 2025. Same content, different answers: Cross-modal inconsistency in mllms. arXiv preprint arXiv:2512.08923. — [arXiv:2512.08923](https://arxiv.org/abs/2512.08923) [find](https://scholar.google.com/scholar?q=Angela+van+Sprang%2C+Laurens+Samson%2C+Ana+Lucic%2C+Erman+Acar%2C+Sennay+Ghebreab%2C+and+Yuki+M+Asano%2E+2025%2E+Same+content%2C+different+answers%3A+Cross%2Dmodal+inconsistency+in+mllms%2E+arXiv+preprint+arXiv%3A2512%2E08923%2E)
44. An Vo, Khai-Nguyen Nguyen, Mohammad Reza Taesiri, Vy Tuong Dang, Anh Totti Nguyen, and Daeyoung Kim. 2025. Vision language models are biased. arXiv preprint arXiv:2505.23941. — [arXiv:2505.23941](https://arxiv.org/abs/2505.23941) [find](https://scholar.google.com/scholar?q=An+Vo%2C+Khai%2DNguyen+Nguyen%2C+Mohammad+Reza+Taesiri%2C+Vy+Tuong+Dang%2C+Anh+Totti+Nguyen%2C+and+Daeyoung+Kim%2E+2025%2E+Vision+language+models+are+biased%2E+arXiv+preprint+arXiv%3A2505%2E23941%2E+%E2%80%94+%5BarXiv%3A2505%2E23941%5D%28http)
45. Kevin Wang, Alexandre Variengien, Arthur Conmy, Buck Shlegeris, and Jacob Steinhardt. 2022. Interpretability in the wild: a circuit for indirect object identification in gpt-2 small. arXiv preprint arXiv:2211.00593. [find](https://scholar.google.com/scholar?q=Kevin+Wang%2C+Alexandre+Variengien%2C+Arthur+Conmy%2C+Buck+Shlegeris%2C+and+Jacob+Steinhardt%2E+2022%2E+Interpretability+in+the+wild%3A+a+circuit+for+indirect+object+identification+in+gpt%2D2+small%2E+arXiv+preprint+ar)
46. Peng Wang, Shuai Bai, Sinan Tan, Shijie Wang, Zhihao Fan, Jinze Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Yang Fan, Kai Dang, Mengfei Du, Xuancheng Ren, Rui Men, Dayiheng Liu, Chang Zhou, Jingren Zhou, and Junyang Lin. 2024. [Qwen2-vl: Enhancing vision-language model’s perception of the w…](https://md.research.elamur.ai/free/en/papers/wang-2024-qwen2-vl-enhancing-vision/)
47. Rongwu Xu, Brian Lin, Shujian Yang, Tianqi Zhang, Weiyan Shi, Tianwei Zhang, Zhixuan Fang, Wei Xu, and Han Qiu. 2024a. The earth is flat because...: Investigating llms’ belief towards misinformation via persuasive conversation. In Proceedings of the 62nd Annual Meeting of the Association for Computa… [find](https://scholar.google.com/scholar?q=Rongwu+Xu%2C+Brian+Lin%2C+Shujian+Yang%2C+Tianqi+Zhang%2C+Weiyan+Shi%2C+Tianwei+Zhang%2C+Zhixuan+Fang%2C+Wei+Xu%2C+and+Han+Qiu%2E+2024a%2E+The+earth+is+flat+because%2E%2E%2E%3A+Investigating+llms%E2%80%99+belief+towards+misinformation+v)
48. Rongwu Xu, Zehan Qi, Zhijiang Guo, Cunxiang Wang, Hongru Wang, Yue Zhang, and Wei Xu. 2024b. [Knowledge conflicts for llms: A survey](https://md.research.elamur.ai/free/en/papers/xu-2024-knowledge-conflicts-llms-survey/). arXiv preprint arXiv:2403.08319.
49. Qianqi Yan, Yue Fan, Hongquan Li, Shan Jiang, Yang Zhao, Xinze Guan, Ching-Chen Kuo, and Xin Eric Wang. 2025. Multimodal inconsistency reasoning (mmir): A new benchmark for multimodal reasoning models. arXiv preprint arXiv:2502.16033. [find](https://scholar.google.com/scholar?q=Qianqi+Yan%2C+Yue+Fan%2C+Hongquan+Li%2C+Shan+Jiang%2C+Yang+Zhao%2C+Xinze+Guan%2C+Ching%2DChen+Kuo%2C+and+Xin+Eric+Wang%2E+2025%2E+Multimodal+inconsistency+reasoning+%28mmir%29%3A+A+new+benchmark+for+multimodal+reasoning+models)
50. Jia-Yu Yao, Kun-Peng Ning, Zhen-Hui Liu, Mu-Nan Ning, Yu-Yang Liu, and Li Yuan. 2023. Llm lies: Hallucinations are not bugs, but features as adversarial examples. arXiv preprint arXiv:2310.01469. — [arXiv:2310.01469](https://arxiv.org/abs/2310.01469) [find](https://scholar.google.com/scholar?q=Jia%2DYu+Yao%2C+Kun%2DPeng+Ning%2C+Zhen%2DHui+Liu%2C+Mu%2DNan+Ning%2C+Yu%2DYang+Liu%2C+and+Li+Yuan%2E+2023%2E+Llm+lies%3A+Hallucinations+are+not+bugs%2C+but+features+as+adversarial+examples%2E+arXiv+preprint+arXiv%3A2310%2E01469%2E+%E2%80%94+%5Ba)
51. Weihao Yu, Zhengyuan Yang, Linjie Li, Jianfeng Wang, Kevin Lin, Zicheng Liu, Xinchao Wang, and Lijuan Wang. 2023. [Mm-vet: Evaluating large multimodal models for integrated capabilities](https://md.research.elamur.ai/free/en/papers/yu-2023-mm-vet-evaluating-large-multimodal/). arXiv preprint arXiv:2308.02490.
52. Huixuan Zhang, Junzhe Zhang, and Xiaojun Wan. 2023a. Quantity matters: Towards assessing and mitigating number hallucination in large vision-language models. arXiv preprint arXiv:2403.01373. — [arXiv:2403.01373](https://arxiv.org/abs/2403.01373) [find](https://scholar.google.com/scholar?q=Huixuan+Zhang%2C+Junzhe+Zhang%2C+and+Xiaojun+Wan%2E+2023a%2E+Quantity+matters%3A+Towards+assessing+and+mitigating+number+hallucination+in+large+vision%2Dlanguage+models%2E+arXiv+preprint+arXiv%3A2403%2E01373%2E+%E2%80%94+%5BarXiv%3A)
53. Muru Zhang, Ofir Press, William Merrill, Alisa Liu, and Noah A Smith. 2023b. [How language model hallucinations can snowball](https://md.research.elamur.ai/free/en/papers/zhang-2023-language-model-hallucinations-snowball/). arXiv preprint arXiv:2305.13534.
54. Zhuoran Zhang, Tengyue Wang, Xilin Gong, Yang Shi, Haotian Wang, Di Wang, and Lijie Hu. 2025. When modalities conflict: How unimodal reasoning uncertainty governs preference dynamics in mllms. arXiv preprint arXiv:2511.02243. — [arXiv:2511.02243](https://arxiv.org/abs/2511.02243) [find](https://scholar.google.com/scholar?q=Zhuoran+Zhang%2C+Tengyue+Wang%2C+Xilin+Gong%2C+Yang+Shi%2C+Haotian+Wang%2C+Di+Wang%2C+and+Lijie+Hu%2E+2025%2E+When+modalities+conflict%3A+How+unimodal+reasoning+uncertainty+governs+preference+dynamics+in+mllms%2E+arXiv+p)
55. Jinkun Zhao, Lei Huang, and Wenjun Wu. 2025. What color is it? a text-interference multimodal hallucination benchmark. arXiv preprint arXiv:2511.13400. — [arXiv:2511.13400](https://arxiv.org/abs/2511.13400) [find](https://scholar.google.com/scholar?q=Jinkun+Zhao%2C+Lei+Huang%2C+and+Wenjun+Wu%2E+2025%2E+What+color+is+it%3F+a+text%2Dinterference+multimodal+hallucination+benchmark%2E+arXiv+preprint+arXiv%3A2511%2E13400%2E+%E2%80%94+%5BarXiv%3A2511%2E13400%5D%28https%3A%2F%2Farxiv%2Eorg%2Fabs%2F2511%2E)
56. Tinghui Zhu, Qin Liu, Fei Wang, Zhengzhong Tu, and Muhao Chen. 2024. Unraveling cross-modality knowledge conflicts in large vision-language models. arXiv preprint arXiv:2410.03659. — [arXiv:2410.03659](https://arxiv.org/abs/2410.03659) [find](https://scholar.google.com/scholar?q=Tinghui+Zhu%2C+Qin+Liu%2C+Fei+Wang%2C+Zhengzhong+Tu%2C+and+Muhao+Chen%2E+2024%2E+Unraveling+cross%2Dmodality+knowledge+conflicts+in+large+vision%2Dlanguage+models%2E+arXiv+preprint+arXiv%3A2410%2E03659%2E+%E2%80%94+%5BarXiv%3A2410%2E03659)

*Updated 2 November 2026*