Right Answers, Wrong Reasons: Perception Errors in Multimodal Chain-of-Thought and the Blind Spot of Outcome-Based Rewards
Keywords:
multimodal large language models, vision-language alignment, coupling locus, early fusion, visual instruction tuning, benchmark evaluationAbstract
Multimodal large language models increasingly answer visual questions through explicit reasoning chains, and the chains are typically fluent. Evidence shows they can nonetheless be perceptually wrong—misreading values off a chart's axis, miscounting geometric primitives, or citing objects absent from the image—while the final answer lands correct anyway. This survey organizes that evidence around a Process–Outcome Consistency (POC) framework: a 2×2 over process soundness and outcome correctness whose key cell, the compensated error (flawed process, correct outcome), is invisible to every metric and training signal that scores only the answer. We trace the evidence for perception errors inside coherent chains, formalize why outcome-based supervision—verifier-reward reinforcement learning, outcome reward models, answer-filtered rejection sampling—cannot select against compensated errors (the reward insensitivity principle), and review the resulting signatures: reward overoptimization, reward hacking, and reasoning that grows more articulate without growing more grounded. We map mitigation onto five leverage points—upstream perception, inference-time process shaping, training-signal redesign, post-hoc verification, process-aware evaluation—and propose reporting compensated-error and evidence-grounding rates alongside accuracy. We close on an asymmetry favoring multimodal settings: visual process errors are checkable by mundane tooling, making faithful multimodal reasoning an unusually tractable target for process supervision.
References
OpenAI. (2023). GPT-4V(ision) system card. OpenAI Technical Report.
Gemini Team, Google. (2024). Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context. arXiv preprint arXiv:2403.05530.
Chameleon Team. (2024). Chameleon: Mixed-modal early-fusion foundation models. arXiv preprint arXiv:2405.09818.
Rahmanzadehgervi, P., Bolton, L., Taesiri, M. R., & Nguyen, A. T. (2024). Vision language models are blind: Failing to translate detailed visual features into words. arXiv preprint arXiv:2407.06581.
Wang, Z., Xia, M., He, L., Chen, H., Liu, Y., et al. (2024). CharXiv: Charting gaps in realistic chart understanding in multimodal LLMs. arXiv preprint arXiv:2406.18521.
Tong, S., Liu, Z., Zhai, Y., Ma, Y., LeCun, Y., & Xie, S. (2024). Eyes wide shut? Exploring the visual shortcomings of multimodal LLMs. arXiv preprint arXiv:2401.06209.
Yin, S., Fu, C., Zhao, Y., Li, K., Sun, X., Xu, T., & Li, E. (2023). A survey on multimodal large language models. arXiv preprint arXiv:2306.13549.
Li, G., Chen, X., Xu, G., Lin, T., Wang, P., Liang, H., & Zhang, S. (2024). Multimodal large language models: A survey. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW).
Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., Uszkoreit, J., & Houlsby, N. (2021). An image is worth 16x16 words: Transformers for image recognition at scale. In International Conference on Learning Representations (ICLR).
Radford, A., Kim, J. W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., Krueger, G., & Sutskever, I. (2021). Learning transferable visual models from natural language supervision. In Proceedings of the 38th International Conference on Machine Learning (ICML).
Jia, C., Yang, Y., Xia, Y., Chen, Y.-T., Parekh, Z., Pham, H., Le, Q., Sung, Y.-H., Li, Z., & Duerig, T. (2021). Scaling up visual and vision-language representation learning with noisy text supervision. In Proceedings of the 38th International Conference on Machine Learning (ICML).
He, K., Chen, X., Xie, S., Li, Y., Dollár, P., & Girshick, R. (2022). Masked autoencoders are scalable vision learners. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).
Oquab, M., Darcet, T., Moutakanni, T., Vo, H., Szafraniec, M., Khalidov, V., et al. (2024). DINOv2: Learning robust visual features without supervision. In Transactions on Machine Learning Research.
Schuhmann, C., Beaumont, R., Vencu, R., Gordon, C., Wightman, R., Cherti, M., et al. (2022). LAION-5B: An open large-scale dataset for training next generation image-text models. In Advances in Neural Information Processing Systems 35.
Gadre, S. Y., Ilharco, G., Fang, A., Hayase, J., Smock, G., Oh, S., et al. (2023). DataComp: In search of the next generation of multimodal datasets. In Advances in Neural Information Processing Systems 36.
Brown, T., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., et al. (2020). Language models are few-shot learners. In Advances in Neural Information Processing Systems 33.
Wei, J., Bosma, M., Zhao, V. Y., Guu, K., Yu, A. W., Lester, B., Du, N., Dai, A. M., & Le, Q. V. (2022). Finetuned language models are zero-shot learners. In International Conference on Learning Representations (ICLR).
Ouyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C., Mishkin, P., et al. (2022). Training language models to follow instructions with human feedback. In Advances in Neural Information Processing Systems 35.
Alayrac, J.-B., Donahue, J., Luc, P., Miech, A., Barr, I., Hasson, Y., et al. (2022). Flamingo: A visual language model for few-shot learning. In Advances in Neural Information Processing Systems 35.
Liang, V. W., Zhang, Y., Kwon, Y., Yeung, S., & Zou, J. (2022). Mind the gap: Understanding the modality gap in multi-modal contrastive representation learning. In Advances in Neural Information Processing Systems 35.
Li, J., Li, D., Savarese, S., & Hoi, S. (2023). BLIP-2: Bootstrapping language-image pre-training with frozen image encoders and large language models. In Proceedings of the 40th International Conference on Machine Learning (ICML).
Liu, H., Li, C., Wu, Q., & Lee, Y. J. (2023). Visual instruction tuning. In Advances in Neural Information Processing Systems 36.
Liu, H., Li, C., Li, Y., & Lee, Y. J. (2024). Improved baselines with visual instruction tuning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).
Dai, W., Li, J., Li, D., Tiong, A. M. H., Zhao, J., Wang, W., Li, B., Fung, P., & Hoi, S. (2023). InstructBLIP: Towards general-purpose vision-language models with instruction tuning. In Advances in Neural Information Processing Systems 36.
Zhu, D., Chen, J., Shen, X., Li, X., & Elhoseiny, M. (2023). MiniGPT-4: Enhancing vision-language understanding with advanced large language models. arXiv preprint arXiv:2304.10592.
Ye, Q., Xu, H., Xu, G., Ye, J., Yan, M., Zhou, Y., Wang, J., Hu, A., Shi, P., Shi, Y., Li, C., Xu, Y., Chen, H., Tian, B., Zhang, J., Li, Z., Zhao, J., & Huang, F. (2023). mPLUG-Owl: Modularization empowers large language models with multimodality. arXiv preprint arXiv:2304.14178.
Bai, J., Bai, S., Yang, S., Wang, W., Tan, S., Wang, P., Lin, J., Zhou, C., & Zhou, J. (2023). Qwen-VL: A versatile vision-language model for understanding, localization, text reading, and beyond. arXiv preprint arXiv:2308.12966.
Wang, W., Lv, Q., Yu, W., Hong, W., Qi, J., Wang, Y., et al. (2023). CogVLM: Visual expert for pretrained language models. arXiv preprint arXiv:2311.03729.
Chen, Z., Wu, J., Wang, W., Su, W., Chen, G., Xing, S., et al. (2024). InternVL: Scaling up vision foundation models and aligning for generic visual-linguistic tasks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).
Li, Y., Du, Y., Zhou, K., Wang, J., Zhao, W. X., & Wen, J.-R. (2023). Evaluating object hallucination in large vision-language models. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP).
Wang, P., Yang, A., Men, R., Lin, J., Bai, S., Li, Z., Ma, J., Zhou, C., Zhou, J., & Yang, Z. (2022). OFA: Unifying architectures, tasks, and modalities through a simple sequence-to-sequence learning framework. In Proceedings of the 39th International Conference on Machine Learning (ICML).
Lu, J., Clark, C., Lee, S., Zhang, Z., Khosla, S., Marten, R., Hoiem, D., & Kembhavi, A. (2022). Unified-IO: A unified model for vision, language, and multi-modal tasks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).
Bao, H., Ye, Q., Li, W., & Dong, L. (2023). BEiT-3: Image as a foreign language: BEiT pretraining for all vision and vision-language tasks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).
Adept AI. (2023). Fuyu-8B: A multimodal architecture for AI agents. Adept AI Research Blog.
Sun, Q., Yu, Q., Cui, Y., Zhang, F., Zhang, X., et al. (2023). Generative pretraining in multimodality. arXiv preprint arXiv:2307.05222.
Sun, Q., Cui, Y., Zhang, X., Zhang, F., Yu, Q., Luo, Z., et al. (2024). Emu2: Generative multimodal models are in-context learners. arXiv preprint arXiv:2312.13286.
Wang, X., Zhang, X., Luo, Z., Sun, Q., Cui, Y., et al. (2024). Emu3: Next-token prediction is all you need. arXiv preprint arXiv:2409.18869.
Xia, R., Zhang, B., Ye, H., et al. (2024). ChartX & ChartVLM: A versatile benchmark and foundation model for complicated chart reasoning. arXiv preprint arXiv:2402.12185.
Zhang, L., Hu, A., Xu, H., Yan, M., Xu, Y., Jin, Q., Zhang, J., & Huang, F. (2024). TinyChart: Efficient chart understanding with visual token merging and program-of-thoughts learning. arXiv preprint arXiv:2404.16635.
Chen, X., Wang, X., Beyer, L., Kolesnikov, A., Wu, J., Vo, N., et al. (2023). PaLI-3 Vision Language Models: Smaller, faster, stronger. arXiv preprint arXiv:2310.09199.
OpenAI. (2024). Hello GPT-4o. OpenAI Technical Publication.
Zhou, C., Yu, L., Babu, A., Tirumala, K., Yasunaga, M., Shamis, L., Kahn, J., Ma, X., Zettlemoyer, L., & Levy, O. (2024). Transfusion: Predict the next token and diffuse images with one multi-modal model. arXiv preprint arXiv:2408.11039.
Xie, J., Mao, W., Bai, Z., Zhang, D. J., Wang, W., Lin, K. Y., Gu, Y., Chen, Z., Yang, L., & Shou, M. Z. (2024). Show-o: One single transformer to unify multimodal understanding and generation. arXiv preprint arXiv:2408.12528.
Chen, X., Wu, Z., Liu, X., Pan, Z., Liu, W., et al. (2025). Janus-Pro: Unified multimodal understanding and generation with data and model scaling. arXiv preprint arXiv:2501.17811.
Wu, S., Fei, H., Le, L., Ji, T., & Chua, T.-S. (2024). NExT-GPT: Any-to-any multimodal LLM. In Proceedings of the 41st International Conference on Machine Learning (ICML).
Gupta, T., & Kembhavi, A. (2023). Visual programming: Compositional visual reasoning without training. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).
Shen, Y., Song, K., Tan, X., Li, D., Lu, W., & Zhuang, Y. (2024). HuggingGPT: Solving AI tasks with ChatGPT and its friends in Hugging Face. In Advances in Neural Information Processing Systems 36.
Peng, Z., Wang, W., Dong, L., Hao, Y., Huang, S., Ma, S., & Wei, F. (2023). Kosmos-2: Grounding multimodal large language models to the world. arXiv preprint arXiv:2306.14824.
Chen, K., Zhang, Z., Zeng, W., Zhang, R., Zhu, F., & Zhao, R. (2023). Shikra: Unleashing multimodal LLM's referential dialogue magic. arXiv preprint arXiv:2306.15195.
You, H., Zhang, H., Gan, Z., Du, X., Zhang, B., Wang, Z., Cao, L., Chang, S.-F., & Yang, Y. (2024). Ferret: Refer and ground anything anywhere at any granularity. In International Conference on Learning Representations (ICLR).
Laurençon, H., Trinh, H., de Vazelhes, H., Wang, A., & Truong, V. (2023). OBELISC: An open web-scale filtered dataset of interleaved image-text documents. In Advances in Neural Information Processing Systems 36 Datasets and Benchmarks Track.
Rafailov, R., Sharma, A., Mitchell, E., Manning, C. D., Ermon, S., & Finn, C. (2023). Direct preference optimization: Your language model is secretly a reward model. In Advances in Neural Information Processing Systems 36.
Yu, T., Yao, Y., Zhang, H., He, T., Han, Y., Cui, G., Hu, J., Liu, Z., Zheng, H.-T., Zhao, M., & Sun, M. (2024). RLHF-V: Towards trustworthy MLLMs via behavior alignment from fine-grained correctional human feedback. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).
Yu, T., Zhang, H., Li, Q., Xu, Q., Yao, Y., et al. (2024). RLAIF-V: Open-source AI feedback leads to super GPT-4V trustworthiness. arXiv preprint arXiv:2405.17220.
Liu, H., Li, C., Li, Y., Li, B., Zhang, Y., Shen, S., & Lee, Y. J. (2024). LLaVA-NeXT: Improved reasoning, OCR, and world knowledge. LLaVA Project Blog.
Chen, Z., Wang, W., Tian, H., Ye, S., Gao, Z., et al. (2024). InternVL 1.5: How far are we to GPT-4V? Closing the gap to commercial multimodal models with open-source suites. arXiv preprint arXiv:2404.16821.
Wang, P., Bai, S., Tan, S., Wang, S., Fan, Z., Bai, J., et al. (2024). Qwen2-VL: Enhancing vision-language model's perception of the world at any resolution. arXiv preprint arXiv:2409.12191.
Dehghani, M., Djolonga, J., Wilhelm, B., Pintsch, L., Lucioni, A., Arnab, A., et al. (2023). Patch n' pack: NaViT, a vision transformer for any aspect ratio and resolution. In Advances in Neural Information Processing Systems 36.
Xu, R., Yao, Y., Guo, Z., Cui, J., Ni, Z., et al. (2024). LLaVA-UHD: An LMM perceiving any aspect ratio and high-resolution images. arXiv preprint arXiv:2403.11703.
Chen, L., Zhao, H., Liu, T., Bai, S., Lin, J., Zhou, C., & Chang, B. (2024). An image is worth 1/2 tokens after layer 2: Plug-and-play inference acceleration for large vision-language models. In European Conference on Computer Vision (ECCV).
Shang, Y., Cai, M., Xu, B., Lee, Y. J., & Yan, Y. (2024). LLaVA-PruMerge: Adaptive token reduction for efficient large multimodal models. arXiv preprint arXiv:2403.15388.
Li, W., You, Z., Qi, C., Wang, H., Zhou, K., Wang, H., Gao, C., Chen, W., & Zhang, Y. (2024). TokenPacker: Efficient visual projector for multimodal large language models. arXiv preprint arXiv:2407.02392.
Kim, G., Hong, T., Yim, M., Nam, J., Park, J., Yim, J., Hwang, W., Yun, S., Han, D., & Park, S. (2022). OCR-free document understanding transformer. In European Conference on Computer Vision (ECCV).
Lin, B., Tang, Z., Ye, Y., Cui, J., Zhu, B., Jin, P., Zhang, J., Zhuang, M., Yuan, L., Li, S., & Li, Y. (2024). MoE-LLaVA: Mixture of experts for large vision-language models. arXiv preprint arXiv:2401.15947.
Wu, Z., Chen, X., Pan, Z., Liu, X., Liu, W., Li, D., et al. (2024). DeepSeek-VL2: Mixture-of-experts vision-language models for advanced multimodal understanding. arXiv preprint arXiv:2412.10302.
Hong, L., Li, Q., Zhang, L., Chen, D., Jiang, Z., Li, Z., et al. (2024). Aria: An open multimodal native mixture-of-experts model. arXiv preprint arXiv:2410.05993.
Kimi Team. (2025). Kimi-VL technical report. arXiv preprint arXiv:2504.07491.
Abdin, M., Aneja, S., Behl, H., Bubeck, S., Eldan, R., Gunasekar, S., et al. (2024). Phi-3 technical report: A highly capable language model locally on your phone. arXiv preprint arXiv:2404.14219.
Chu, X., Xiang, L., Wei, B., Li, X., Lu, X., Wei, N., & Yan, M. (2024). MobileVLM: A fast, reproducible and strong vision language assistant for mobile devices. arXiv preprint arXiv:2312.16886.
Goyal, Y., Khot, T., Summers-Stay, D., Batra, D., & Parikh, D. (2017). Making the V in VQA matter: Elevating the role of image understanding in visual question answering. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR).
Yue, X., Ni, Y., Zhang, K., Zheng, T., Liu, R., Zhang, G., et al. (2024). MMMU: A massive multi-discipline multimodal understanding and reasoning benchmark for expert AGI. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).
Yue, X., Zheng, T., Ni, Y., Wang, Y., Zhang, K., et al. (2024). MMMU-Pro: A more robust multi-discipline multimodal understanding benchmark. arXiv preprint arXiv:2409.02813.
Lu, P., Bansal, H., Xia, T., Liu, J., Li, C., Hajishirzi, H., Cheng, H., Chang, K.-W., Galley, M., & Gao, J. (2024). MathVista: Evaluating mathematical reasoning of foundation models in visual contexts. In Advances in Neural Information Processing Systems 36 Datasets and Benchmarks Track.
Masry, A., Long, D. X., Tan, J. Q., Joty, S., & Hoque, E. (2022). ChartQA: A benchmark for question answering about charts with visual and logical reasoning. In Findings of the Association for Computational Linguistics: ACL 2022.
Guan, T., Liu, F., Wu, X., Xian, R., Li, Z., Liu, X., Wang, X., Chen, L., Huang, F., Yacoob, Y., Manocha, D., & Zhou, T. (2024). HallusionBench: An advanced diagnostic suite for entangled language hallucination and visual illusion in large vision-language models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).
Wang, J., Wang, Y., Xu, G., Zhang, J., Gu, Y., Jia, H., Zhang, J., Yan, M., Zhang, J., & Sang, J. (2024). AMBER: An LLM-free multi-dimensional benchmark for MLLM hallucination evaluation. arXiv preprint arXiv:2311.07397.
Li, K., Wang, Y., He, Y., Li, Y., Wang, Y., Liu, Y., et al. (2024). MVBench: A comprehensive video benchmark for multi-modal large language models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).
Fu, C., Chen, P., Shen, Y., Qin, Z., Zhang, D., Lin, X., et al. (2024). Video-MME: The first-ever comprehensive evaluation benchmark of multi-modal LLMs in video analysis. arXiv preprint arXiv:2405.21075.
Zhang, J., Khayatkhoei, M., Chhikara, P., & Ilievski, F. (2025). MLLMs know where to look: Training-free perception of small visual details with multimodal LLMs. arXiv preprint arXiv:2502.17422.
OpenAI. (2024). Video generation models as world simulators. OpenAI Technical Report.
Tang, C., Yu, W., Sun, G., Chen, X., Tan, T., Li, W., Lu, L., Ma, Z., & Zhang, C. (2024). SALMONN: Towards generic hearing abilities for large language models. In International Conference on Learning Representations (ICLR).
Qwen Team. (2025). Qwen2.5-Omni technical report. arXiv preprint arXiv:2503.20215.
