Elamur Research⌕
☰
home / papers / 2026 / Blind Spot Navigation in Large Language Model Reasoning with Thought Space Explorer

Blind Spot Navigation in Large Language Model Reasoning with Thought Space Explorer

Paper · 2026 · Jinghan Zhang, Fengran Mo, Tharindu Cyril Weerasooriya et al.

Summary

Tree-like and multi-chain prompting methods such as ToT usually refine or extend reasoning paths the model has already produced. As a result, they keep returning to familiar directions and miss parts of the solution space the authors call "blind spots." This paper proposes Thought Space Explorer (TSE) to push LLMs into those unexplored regions.

TSE works in three stages:

  1. It scores each reasoning step's influence on the conclusion, using relative gradients, and keeps the most important ones as key nodes.
  2. It creates a new node by fusing information from two key nodes in different chains, then grows fresh branches from that node.
  3. It reasons jointly over the whole expanded structure to produce the answer.

For black-box models, the gradient-based steps are replaced by the LLM's own semantic judgment and self-evaluation.

The method was tested with Qwen3-4B and Qwen3-8B on math and QA benchmarks. TSE achieves the best average accuracy, more diverse and more correct intermediate paths, and a favourable token–accuracy balance. This makes structured reasoning more practical under limited compute.

Key points

  • Benchmarks are GSM8K, AIME24, AIME25 and GPQA-Diamond. Baselines are Direct output, Think, ToT, RATT and Self-Route.
  • On average accuracy, TSE scores 64.8% with Qwen3-4B (next best is Think at 61.5%) and 65.5% with Qwen3-8B (next best is Self-Route at 64.4%). TSE does not win every single benchmark.
  • With Qwen3-4B, TSE reaches 66.7% on AIME24 and 50.0% on AIME25. With Qwen3-8B, it reaches 53.3% on AIME25, compared with 23.3% for Direct output.
  • Path diversity was measured as 1 minus the mean Jaccard similarity between reasoning steps. On GPQA-D, TSE produced the most diverse paths, ahead of the tree-based ToT and RATT.
  • On AIME24 with Qwen3-8B, TSE shows the best token-versus-accuracy trade-off. ToT's exponential token growth put it off the chart. Stated limitations include:
    • extra cost compared with direct decoding;
    • dependence on prompt quality in the black-box variant;
    • no tests yet in domains such as legal or medical reasoning.

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

🎓Université de MontréalMontreal · Canada68 works · 79,382 citations🎓Arizona State UniversityTempe · United States50 works · 2,032 citations🏢AccentureZurich · Switzerland6 works · 27 citations🎓University of KansasLawrence · United States8 works · 32 citations🎓University of AlabamaTuscaloosa · United States6 works · 403 citations🎓Clemson UniversityClemson · United States2 works · 28 citations🏢Accenture (United States)New York · United States5 works · 261 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksComputer Vision and Pattern Recognition · 2,869 works

🧭 Topics 2Research subjects and languages the work is about.Language models & pretraining · 3,320 worksReasoning · 1,426 works

Cite

Cite as Jinghan Zhang, Fengran Mo, Tharindu Cyril Weerasooriya, Xinyue Ye, Dongjie Wang, Yanjie Fu and Kunpeng Liu. 2026. Blind Spot Navigation in Large Language Model Reasoning with Thought Space Explorer. Findings of the Association for Computational Linguistics: EACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-eacl.191

@inproceedings{zhang2026blind,
  title = {{Blind Spot Navigation in Large Language Model Reasoning with Thought Space Explorer}},
  author = {Jinghan Zhang, Fengran Mo, Tharindu Cyril Weerasooriya, Xinyue Ye, Dongjie Wang, Yanjie Fu and Kunpeng Liu},
  booktitle = {Findings of the Association for Computational Linguistics: EACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-eacl.191},
  url = {https://doi.org/10.18653/v1/2026.findings-eacl.191}
}

References

54 references, 17 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, and 1 others. 2023. Gpt-4 technical report. arXiv preprint arXiv:2303.08774.
  2. Maciej Besta, Nils Blach, Ales Kubicek, Robert Gerstenberger, Michal Podstawski, Lukas Gianinazzi, Joanna Gajda, Tomasz Lehmann, Hubert Niewiadomski, Piotr Nyczyk, and 1 others. 2024. Graph of thoughts: Solving elaborate problems with large language models. In Proceedings of the AAAI Conference on A… find
  3. Qiguang Chen, Libo Qin, Jinhao Liu, Dengyun Peng, Jiannan Guan, Peng Wang, Mengkang Hu, Yuhang Zhou, Te Gao, and Wanxiang Che. 2025. Towards reasoning era: A survey of long chain-of-thought for reasoning large language models. arXiv preprint arXiv:2503.09567.
  4. Zheng Chu, Jingchang Chen, Qianglong Chen, Weijiang Yu, Tao He, Haotian Wang, Weihua Peng, Ming Liu, Bing Qin, and Ting Liu. 2023. A survey of chain of thought reasoning: Advances, frontiers and future. arXiv preprint arXiv:2309.15402. find
  5. Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, and 1 others. 2021. Training verifiers to solve math word problems. arXiv preprint arXiv:2110.14168.
  6. Sujan Dutta, Deepak Pandita, Tharindu Cyril Weerasooriya, Marcos Zampieri, Christopher M Homan, and Ashiqur R KhudaBukhsh. 2025. Annotator reliability through in-context learning (student abstract). In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 29356–29358. find
  7. Xidong Feng, Ziyu Wan, Muning Wen, Stephen Marcus McAleer, Ying Wen, Weinan Zhang, and Jun Wang. 2023. Alphazero-like tree-search can guide large language model decoding and training. arXiv preprint arXiv:2309.17179. find
  8. Zitian Gao, Boye Niu, Xuzheng He, Haotian Xu, Hongzhang Liu, Aiwei Liu, Xuming Hu, and Lijie Wen. 2024. Interpretable contrastive monte carlo tree search reasoning. arXiv preprint arXiv:2410.01707. — [arXiv:2410.01707](https://arxiv.org/abs/2410.01707) find
  9. Kye Gomez. 2023. Tree of thoughts. https://github.com/kyegomez/tree-of-thoughts. ↗
  10. Shibo Hao, Yi Gu, Haodi Ma, Joshua Jiahua Hong, Zhen Wang, Daisy Zhe Wang, and Zhiting Hu. 2023. Reasoning with language model is planning with world model. arXiv preprint arXiv:2305.14992. find
  11. Yang He, Xiao Ding, Bibo Cai, Yufei Zhang, Kai Xiong, Zhouhao Sun, Bing Qin, and Ting Liu. 2025. Self-route: Automatic mode switching via capability estimation for efficient reasoning. arXiv preprint arXiv:2505.20664. — [arXiv:2505.20664](https://arxiv.org/abs/2505.20664) find
  12. Guiyang Hou, Wenqi Zhang, Yongliang Shen, Linjuan Wu, and Weiming Lu. 2024. Timetom: Temporal space is the key to unlocking the door of large language models’ theory-of-mind. arXiv preprint arXiv:2407.01455. find
  13. Jie Huang and Kevin Chen-Chuan Chang. 2022. Towards reasoning in large language models: A survey. arXiv preprint arXiv:2212.10403.
  14. Hugging Face. 2024. Llama-3.1-8B-Instruct. find
  15. Aaron Hurst, Adam Lerer, Adam P Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Ostrow, Akila Welihinda, Alan Hayes, Alec Radford, and 1 others. 2024. Gpt-4o system card. arXiv preprint arXiv:2410.21276.
  16. Ziqi Jin and Wei Lu. 2024. Self-harmonized chain of thought. arXiv preprint arXiv:2409.04057. find
  17. Jaehun Jung, Lianhui Qin, Sean Welleck, Faeze Brahman, Chandra Bhagavatula, Ronan Le Bras, and Yejin Choi. 2022. Maieutic prompting: Logically consistent reasoning with recursive explanations. arXiv preprint arXiv:2205.11822.
  18. Takeshi Kojima, Shixiang Shane Gu, Machel Reid, Yutaka Matsuo, and Yusuke Iwasawa. 2022. Large language models are zero-shot reasoners. Advances in neural information processing systems, 35:22199– 22213.
  19. Suphakit Niwattanakul, Jatsada Singthongchai, Ekkachai Naenudorn, and Supachanun Wanapu. 2013. Using of jaccard coefficient for keywords similarity. In Proceedings of the international multiconference of engineers and computer scientists, volume 1, pages 380–384. find
  20. OpenAI. 2023. Gpt-4. https://openai.com/index/ gpt-4-research/. ↗
  21. OpenAI. 2024. GPT-4o-mini: Advancing Cost-Efficient Intelligence. Accessed: 2024-10-05. find
  22. Zhiyuan Li, Hong Liu, Denny Zhou, and Tengyu Ma. 2024. Chain of thought empowers transformers to solve inherently serial problems. arXiv preprint arXiv:2402.12875. find
  23. Qianjun Pan, Wenkai Ji, Yuyang Ding, Junsong Li, Shilian Chen, Junyi Wang, Jie Zhou, Qin Chen, Min Zhang, Yulan Wu, and 1 others. 2025. A survey of slow thinking-based reasoning llms using reinforced learning and inference-time scaling law. arXiv preprint arXiv:2505.02665. find
  24. Xun Liang, Shichao Song, Zifan Zheng, Hanyu Wang, Qingchen Yu, Xunkai Li, Rong-Hua Li, Feiyu Xiong, and Zhiyu Li. 2024. Internal consistency and self-feedback in large language models: A survey. arXiv preprint arXiv:2407.14507. — [arXiv:2407.14507](https://arxiv.org/abs/2407.14507)
  25. Deepak Pandita, Tharindu Cyril Weerasooriya, Ankit Parag Shah, Christopher M Homan, and Wei Wei. 2025. Prorefine: Inference-time prompt refinement with textual feedback. arXiv preprint arXiv:2506.05305. — [arXiv:2506.05305](https://arxiv.org/abs/2506.05305) find
  26. Hunter Lightman, Vineet Kosaraju, Yuri Burda, Harrison Edwards, Bowen Baker, Teddy Lee, Jan Leike, John Schulman, Ilya Sutskever, and Karl Cobbe. 2023. Let’s verify step by step. In The Twelfth International Conference on Learning Representations.
  27. David Patterson, Joseph Gonzalez, Urs Hölzle, Quoc Le, Chen Liang, Lluis-Miquel Munguia, Daniel Rothchild, David R So, Maud Texier, and Jeff Dean. 2022. The carbon footprint of machine learning training will plateau, then shrink. Computer, 55(7):18– 28. find
  28. Ryan Liu, Jiayi Geng, Addison J. Wu, Ilia Sucholutsky, Tania Lombrozo, and Thomas L. Griffiths. 2025. Mind your step (by step): Chain-of-thought can reduce performance on tasks where thinking makes humans worse. Preprint, arXiv:2410.21333. — [arXiv:2410.21333](https://arxiv.org/abs/2410.21333) find
  29. Tongxuan Liu, Wenjiang Xu, Weizhe Huang, Xingyu Wang, Jiaxing Wang, Hailong Yang, and Jing Li. 2024. Logic-of-thought: Injecting logic into contexts for full reasoning in large language models. arXiv preprint arXiv:2409.17539. find
  30. Jieyi Long. 2023. Large language model guided tree-of-thought. arXiv preprint arXiv:2305.08291. find
  31. Kelong Mao, Zhicheng Dou, Fengran Mo, Jiewen Hou, Haonan Chen, and Hongjin Qian. 2023. Large language models know your contextual search intent: A prompting framework for conversational search. Preprint, arXiv:2303.06573. find
  32. Math-AI. 2024. Aime 2024. https://huggingface.co/datasets/math-ai/aime24. Accessed: 2025-10-06. ↗
  33. Math-AI. 2025. Aime 2025. https://huggingface.co/datasets/math-ai/aime25. Accessed: 2025-10-06. ↗
  34. Shentong Mo and Miao Xin. 2024. Tree of uncertain thoughts reasoning for large language models. In ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 12742–12746. IEEE. find
  35. David Rein, Betty Li Hou, Asa Cooper Stickland, Jackson Petty, Richard Yuanzhe Pang, Julien Dirani, Julian Michael, and Samuel R Bowman. 2024. Gpqa: A graduate-level google-proof q&a benchmark. In First Conference on Language Modeling.
  36. Sina J Semnani, Violet Z Yao, Heidi C Zhang, and Monica S Lam. 2023. Wikichat: Stopping the hallucination of large language model chatbots by few-shot grounding on wikipedia. arXiv preprint arXiv:2305.14292. find
  37. Noah Shinn, Beck Labash, and Ashwin Gopinath. 2023. Reflexion: an autonomous agent with dynamic memory and self-reflection. arXiv preprint arXiv:2303.11366, 2(5):9.
  38. Zayne Sprague, Fangcong Yin, Juan Diego Rodriguez, Dongwei Jiang, Manya Wadhwa, Prasann Singhal, Xinyu Zhao, Xi Ye, Kyle Mahowald, and Greg Durrett. 2025. To cot or not to cot? chain-of-thought helps mainly on math and symbolic reasoning. Preprint, arXiv:2409.12183.
  39. Kaya Stechly, Karthik Valmeekam, and Subbarao Kambhampati. 2024. Chain of thoughtlessness: An analysis of cot in planning. arXiv preprint arXiv:2405.04776. — [arXiv:2405.04776](https://arxiv.org/abs/2405.04776) find
  40. Weikuan Wang and Ao Feng. 2021. Self-information loss compensation learning for machine-generated text detection. Mathematical Problems in Engineering, 2021(1):6669468. find
  41. Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. 2022. Self-consistency improves chain of thought reasoning in language models. arXiv preprint arXiv:2203.11171.
  42. Xuezhi Wang and Denny Zhou. 2024. Chain-of-thought reasoning without prompting. arXiv preprint arXiv:2402.10200. find
  43. Yifan Zhang, Yang Yuan, and Andrew Chi-Chih Yao. 2024d. On the diagram of thought. arXiv preprint arXiv:2409.10038. — [arXiv:2409.10038](https://arxiv.org/abs/2409.10038) find
  44. Xinyu Zhu, Junjie Wang, Lin Zhang, Yuxiang Zhang, Yongfeng Huang, Ruyi Gan, Jiaxing Zhang, and Yujiu Yang. 2022. Solving math word problems via cooperative reasoning induced language models. arXiv preprint arXiv:2210.16257.
  45. Yaoting Wang, Shengqiong Wu, Yuecheng Zhang, Shuicheng Yan, Ziwei Liu, Jiebo Luo, and Hao Fei. 2025. Multimodal chain-of-thought reasoning: A comprehensive survey. arXiv preprint arXiv:2503.12605. — [arXiv:2503.12605](https://arxiv.org/abs/2503.12605) find
  46. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, and 1 others. 2022. Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems, 35:24824– 24837.
  47. Yu Xia, Rui Wang, Xu Liu, Mingyan Li, Tong Yu, Xiang Chen, Julian McAuley, and Shuai Li. 2024. Beyond chain-of-thought: A survey of chain-of-x paradigms for llms. arXiv preprint arXiv:2404.15676.
  48. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, and 1 others. 2025. Qwen3 technical report. arXiv preprint arXiv:2505.09388.
  49. Shunyu Yao, Dian Yu, Jeffrey Zhao, Izhak Shafran, Tom Griffiths, Yuan Cao, and Karthik Narasimhan. 2023. Tree of thoughts: Deliberate problem solving with large language models. In Advances in Neural Information Processing Systems, volume 36, pages 11809–11822. Curran Associates, Inc. find
  50. Shunyu Yao, Dian Yu, Jeffrey Zhao, Izhak Shafran, Tom Griffiths, Yuan Cao, and Karthik Narasimhan. 2024. Tree of thoughts: Deliberate problem solving with large language models. Advances in Neural Information Processing Systems, 36. find
  51. Jinghan Zhang, Xiting Wang, Yiqiao Jin, Changyu Chen, Xinhao Zhang, and Kunpeng Liu. 2024a. Prototypical reward network for data-efficient rlhf. Preprint, arXiv:2406.06606. find
  52. Jinghan Zhang, Xiting Wang, Fengran Mo, Yeyang Zhou, Wanfu Gao, and Kunpeng Liu. 2025. Entropy-based exploration conduction for multi-step reasoning. arXiv preprint arXiv:2503.15848. find
  53. Jinghan Zhang, Xiting Wang, Weijieying Ren, Lu Jiang, Dongjie Wang, and Kunpeng Liu. 2024b. Ratt: Athought structure for coherent and correct llmrea-soning. arXiv preprint arXiv:2406.02746. — [arXiv:2406.02746](https://arxiv.org/abs/2406.02746) find
  54. Xuan Zhang, Chao Du, Tianyu Pang, Qian Liu, Wei Gao, and Min Lin. 2024c. Chain of preference optimization: Improving chain-of-thought reasoning in llms. arXiv preprint arXiv:2406.09136. — [arXiv:2406.09136](https://arxiv.org/abs/2406.09136) find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zhang-2026-blind-spot-navigation-large-language/