---
title: "Small LLMs Are Weak Tool Learners: A Multi-LLM Agent (2024)"
description: "The authors propose α-UMi, which splits these duties across three LLMs: a planner reasons about the… 35 references, cited by 35 papers here."
authors: ["Weizhou Shen", "Chenliang Li", "Hongzhan Chen", "Ming Yan", "Xiaojun Quan", "Hehong Chen", "Ji Zhang", "Fei Huang"]
year: 2024
venue: "Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing"
source: "https://aclanthology.org/2024.emnlp-main.929/"
doi: "10.18653/v1/2024.emnlp-main.929"
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/shen-2024-small-llms-are-weak-tool/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/shen-2024-small-llms-are-weak-tool/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2024](https://md.research.elamur.ai/free/en/papers/2024/page/5/) / Small LLMs Are Weak Tool Learners: A Multi-LLM Agent

# Small LLMs Are Weak Tool Learners: A Multi-LLM Agent

**Authors:** [Weizhou Shen](https://md.research.elamur.ai/free/en/authors/weizhou-shen/), [Chenliang Li](https://md.research.elamur.ai/free/en/authors/chenliang-li/), [Hongzhan Chen](https://md.research.elamur.ai/free/en/authors/hongzhan-chen/), [Ming Yan](https://md.research.elamur.ai/free/en/authors/ming-yan/), [Xiaojun Quan](https://md.research.elamur.ai/free/en/authors/xiaojun-quan/), Hehong Chen, [Ji Zhang](https://md.research.elamur.ai/free/en/authors/ji-zhang/) and [Fei Huang](https://md.research.elamur.ai/free/en/authors/fei-huang/) · **Year:** 2024 · **Venue:** [Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing](https://md.research.elamur.ai/free/en/venues/emnlp/) ([EMNLP 2024](https://md.research.elamur.ai/free/en/venues/emnlp/2024/)) · **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/2024.emnlp-main.929/) · [PDF](https://aclanthology.org/2024.emnlp-main.929.pdf) · [DOI 10.18653/v1/2024.emnlp-main.929](https://doi.org/10.18653/v1/2024.emnlp-main.929) · [OpenAlex](https://openalex.org/W4404782859)

**Citations:** 30 citations (OpenAlex, 2026-10-14) · [35](https://md.research.elamur.ai/free/en/papers/shen-2024-small-llms-are-weak-tool/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
The paper tackles a weakness of compact open-source LLMs used as tool-using agents. A single small model struggles to plan, call tools correctly and summarize results all at once. The authors propose α-UMi, which splits these duties across three LLMs:
- a **planner** reasons about the current state and decides whether to call a tool, write the final answer or give up;
- a **caller** turns the planner's rationale into tool invocations;
- a **summarizer** composes the user-facing answer from the execution trajectory.

Training uses a two-phase scheme called global-to-local progressive fine-tuning (GLPFT). A LLaMA-2 backbone is first tuned on the complete tool-use data. It is then cloned into the three roles, and each clone continues training on data restructured for its own sub-task. On tool-learning benchmarks built from ToolBench and ToolAlpaca, α-UMi beats single-LLM baselines across model and data sizes. Loss curves suggest it keeps improving where a lone model plateaus. The work matters because it shows modular, separately updatable small models can rival or exceed agents that rely on one larger model.

## Key points
- α-UMi is named after Polaris, a triple star system. The bright star stands for the planner, and the two fainter companions stand for the caller and summarizer.
- In GLPFT, the role-specific models branch off from the shared model after two epochs of global fine-tuning. From the third epoch, rationale and action losses keep falling for α-UMi but stagnate for Single-LLM.
- In the data-scaling study, Single-LLM peaks on different metrics at different data sizes: Plan ACC at 31.3k instances, Arg. F1 and R-L at 62.7k. α-UMi improves steadily on all metrics as data grows.
- The evaluation covers four tool-learning benchmarks, using LLaMA-2(-chat) backbones, with ToolBench and ToolAlpaca as training data. Code is available at github.com/X-PLUG/Multi-LLM-Agent.
- The appendices report additional analyses:
  - training and deployment cost;
  - Single-LLM gaining little from extra compute;
  - the caller implicitly correcting flawed rationales;
  - pairing planners and callers of different sizes;
  - ablations that merge components.

*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

- [Alibaba Group](https://md.research.elamur.ai/free/en/companies/alibaba-group/) — Company, Hangzhou, China · 168 works · 15,386 citations
- [Sun Yat-sen University](https://md.research.elamur.ai/free/en/universities/sun-yat-sen-university/) — University, Guangzhou, China · 53 works · 9,185 citations

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works

## Cite

Cite as: Weizhou Shen, Chenliang Li, Hongzhan Chen, Ming Yan, Xiaojun Quan, Hehong Chen, Ji Zhang and Fei Huang. 2024. Small LLMs Are Weak Tool Learners: A Multi-LLM Agent. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing.

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

```bibtex
@inproceedings{shen2024small,
  title = {{Small LLMs Are Weak Tool Learners: A Multi-LLM Agent}},
  author = {Weizhou Shen, Chenliang Li, Hongzhan Chen, Ming Yan, Xiaojun Quan, Hehong Chen, Ji Zhang and Fei Huang},
  booktitle = {Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing},
  year = {2024},
  doi = {10.18653/v1/2024.emnlp-main.929},
  url = {https://doi.org/10.18653/v1/2024.emnlp-main.929}
}
```

- [← previous On the Multilingual Ability of Decoder-based Pre-trained Language Models: Finding and Controlling Language-Specific Neurons](https://md.research.elamur.ai/free/en/papers/kojima-2024-multilingual-ability-decoder-based-pre/)
- [next → A Peek into Token Bias: Large Language Models Are Not Yet Genuine Reasoners](https://md.research.elamur.ai/free/en/papers/jiang-2024-peek-into-token-bias-large/)
- [↑ back to papers of 2024](https://md.research.elamur.ai/free/en/papers/2024/page/5/)

## References

35 references, 19 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Chenliang Li, Hehong Chen, Ming Yan, Weizhou Shen, Haiyang Xu, Zhikai Wu, Zhicheng Zhang, Wenmeng Zhou, Yingda Chen, Chen Cheng, Hongzhu Shi, Ji Zhang, Fei Huang, and Jingren Zhou. 2023. Modelscope-agent: Building your customizable agent system with open-source large language models. [find](https://scholar.google.com/scholar?q=Chenliang+Li%2C+Hehong+Chen%2C+Ming+Yan%2C+Weizhou+Shen%2C+Haiyang+Xu%2C+Zhikai+Wu%2C+Zhicheng+Zhang%2C+Wenmeng+Zhou%2C+Yingda+Chen%2C+Chen+Cheng%2C+Hongzhu+Shi%2C+Ji+Zhang%2C+Fei+Huang%2C+and+Jingren+Zhou%2E+2023%2E+Modelscope%2Dag)
2. Anthropic. 2023. Claude-2. Website. [https://www.anthropic.com/news/claude-2](https://www.anthropic.com/news/claude-2). [↗](https://www.anthropic.com/news/claude-2)
3. Yohei Nakajima. 2023. Babyagi. Github repository. [https://github.com/yoheinakajima/babyagi](https://github.com/yoheinakajima/babyagi). [↗](https://github.com/yoheinakajima/babyagi)
4. Baian Chen, Chang Shu, Ehsan Shareghi, Nigel Collier, Karthik Narasimhan, and Shunyu Yao. 2023. [Fireact: Toward language agent fine-tuning](https://md.research.elamur.ai/free/en/papers/chen-2023-fireact-toward-language-agent-fine/). arXiv preprint arXiv:2310.05915.
5. Reiichiro Nakano, Jacob Hilton, Suchir Balaji, Jeff Wu, Long Ouyang, Christina Kim, Christopher Hesse, Shantanu Jain, Vineet Kosaraju, William Saunders, et al. 2021. [Webgpt: Browser-assisted question-answering with human feedback](https://md.research.elamur.ai/free/en/papers/nakano-2021-webgpt-browser-assisted-question-answering/). arXiv preprint arXiv:2112.09332.
6. Hyung Won Chung, Le Hou, Shayne Longpre, Barret Zoph, Yi Tay, William Fedus, Yunxuan Li, Xuezhi Wang, Mostafa Dehghani, Siddhartha Brahma, et al. 2022. Scaling instruction-finetuned language models. arXiv preprint arXiv:2210.11416. [find](https://scholar.google.com/scholar?q=Hyung+Won+Chung%2C+Le+Hou%2C+Shayne+Longpre%2C+Barret+Zoph%2C+Yi+Tay%2C+William+Fedus%2C+Yunxuan+Li%2C+Xuezhi+Wang%2C+Mostafa+Dehghani%2C+Siddhartha+Brahma%2C+et+al%2E+2022%2E+Scaling+instruction%2Dfinetuned+language+models%2E+a)
7. Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, et al. 2021. [Training verifiers to solve math word problems](https://md.research.elamur.ai/free/en/papers/cobbe-2021-training-verifiers-solve-math-word/). arXiv preprint arXiv:2110.14168.
8. Guanting Dong, Hongyi Yuan, Keming Lu, Chengpeng Li, Mingfeng Xue, Dayiheng Liu, Wei Wang, OpenAI. 2022. Chatgpt: Conversational ai language model. Website. [https://openai.com/chatgpt](https://openai.com/chatgpt). [↗](https://openai.com/chatgpt)
9. OpenAI. 2023a. Gpt-4 code interpreter. Website. [https://chat.openai.com/?model=](https://chat.openai.com/?model=) gpt-4-code-interpreter. [↗](https://chat.openai.com/?model=)
10. OpenAI. 2023b. [Gpt-4 technical report](https://md.research.elamur.ai/free/en/papers/openai-2023-gpt-4-technical-report/).
11. Joon Sung Park, Joseph O’Brien, Carrie Jun Cai, Meredith Ringel Morris, Percy Liang, and Michael S Bernstein. 2023. Generative agents: Interactive simulacra of human behavior. In Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology, pages 1–22. [find](https://scholar.google.com/scholar?q=Joon+Sung+Park%2C+Joseph+O%E2%80%99Brien%2C+Carrie+Jun+Cai%2C+Meredith+Ringel+Morris%2C+Percy+Liang%2C+and+Michael+S+Bernstein%2E+2023%2E+Generative+agents%3A+Interactive+simulacra+of+human+behavior%2E+In+Proceedings+of+the+36)
12. Shishir G. Patil, Tianjun Zhang, Xin Wang, and Joseph E. Gonzalez. 2023. [Gorilla: Large language model connected with massive apis](https://md.research.elamur.ai/free/en/papers/patil-2023-gorilla-large-language-model-connected/). arXiv preprint arXiv:2305.15334.
13. Chen Qian, Xin Cong, Cheng Yang, Weize Chen, Yusheng Su, Juyuan Xu, Zhiyuan Liu, and Maosong Sun. 2023a. [Communicative agents for software development](https://md.research.elamur.ai/free/en/papers/qian-2023-chatdev-communicative-agents-software/). arXiv preprint arXiv:2307.07924.
14. Chen Qian, Yufan Dang, Jiahao Li, Wei Liu, Weize Chen, Cheng Yang, Zhiyuan Liu, and Maosong Sun. 2023b. Experiential co-learning of software-developing agents. [find](https://scholar.google.com/scholar?q=Chen+Qian%2C+Yufan+Dang%2C+Jiahao+Li%2C+Wei+Liu%2C+Weize+Chen%2C+Cheng+Yang%2C+Zhiyuan+Liu%2C+and+Maosong+Sun%2E+2023b%2E+Experiential+co%2Dlearning+of+software%2Ddeveloping+agents%2E)
15. Shuofei Qiao, Ningyu Zhang, Runnan Fang, Yujie Luo, Wangchunshu Zhou, Yuchen Eleanor Jiang, Chengfei Lv, and Huajun Chen. 2024. Autoact: Automatic agent learning from scratch via self-planning. [find](https://scholar.google.com/scholar?q=Shuofei+Qiao%2C+Ningyu+Zhang%2C+Runnan+Fang%2C+Yujie+Luo%2C+Wangchunshu+Zhou%2C+Yuchen+Eleanor+Jiang%2C+Chengfei+Lv%2C+and+Huajun+Chen%2E+2024%2E+Autoact%3A+Automatic+agent+learning+from+scratch+via+self%2Dplanning%2E)
16. Yujia Qin, Shengding Hu, Yankai Lin, Weize Chen, Ning Ding, Ganqu Cui, Zheni Zeng, Yufei Huang, Chaojun Xiao, Chi Han, et al. 2023a. [Tool learning with foundation models](https://md.research.elamur.ai/free/en/papers/qin-2023-tool-learning-foundation-models/). arXiv preprint arXiv:2304.08354.
17. Yujia Qin, Shihao Liang, Yining Ye, Kunlun Zhu, Lan Yan, Yaxi Lu, Yankai Lin, Xin Cong, Xiangru Tang, Bill Qian, Sihan Zhao, Runchu Tian, Ruobing Xie, Jie Zhou, Mark Gerstein, Dahai Li, Zhiyuan Liu, and Maosong Sun. 2023b. [Toolllm: Facilitating large language models to master 16000+ real-world apis](https://md.research.elamur.ai/free/en/papers/qin-2023-toolllm-facilitating-large-language-models/).
18. Samyam Rajbhandari, Olatunji Ruwase, Jeff Rasley, Shaden Smith, and Yuxiong He. 2021. Zero-infinity: Breaking the gpu memory wall for extreme scale deep learning. In SC21: International Conference for High Performance Computing, Networking, Storage and Analysis, pages 1–15. IEEE Computer Society. [find](https://scholar.google.com/scholar?q=Samyam+Rajbhandari%2C+Olatunji+Ruwase%2C+Jeff+Rasley%2C+Shaden+Smith%2C+and+Yuxiong+He%2E+2021%2E+Zero%2Dinfinity%3A+Breaking+the+gpu+memory+wall+for+extreme+scale+deep+learning%2E+In+SC21%3A+International+Conference+for)
19. Timo Schick, Jane Dwivedi-Yu, Roberto Dessì, Roberta Raileanu, Maria Lomeli, Luke Zettlemoyer, Nicola Cancedda, and Thomas Scialom. 2023. [Toolformer: Language models can teach themselves to use tools](https://md.research.elamur.ai/free/en/papers/schick-2023-toolformer-language-models-can-teach/). arXiv preprint arXiv:2302.04761.
20. Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurelien Rodriguez, Armand Joulin, Edouard Grave, and Guillaume Lample. 2023a. [Llama: Open and efficient foundation language models](https://md.research.elamur.ai/free/en/papers/touvron-2023-llama-open-efficient-foundation-language/). arXiv p…
21. Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, et al. 2023b. [Llama 2: Open foundation and fine-tuned chat models](https://md.research.elamur.ai/free/en/papers/touvron-2023-llama-2-open-foundation-fine/). arXiv preprint arXiv:2307.09288.
22. Guanzhi Wang, Yuqi Xie, Yunfan Jiang, Ajay Mandlekar, Chaowei Xiao, Yuke Zhu, Linxi Fan, and Anima Anandkumar. 2023. Voyager: An open-ended embodied agent with large language models. arXiv preprint arXiv:2305.16291. [find](https://scholar.google.com/scholar?q=Guanzhi+Wang%2C+Yuqi+Xie%2C+Yunfan+Jiang%2C+Ajay+Mandlekar%2C+Chaowei+Xiao%2C+Yuke+Zhu%2C+Linxi+Fan%2C+and+Anima+Anandkumar%2E+2023%2E+Voyager%3A+An+open%2Dended+embodied+agent+with+large+language+models%2E+arXiv+preprint+ar)
23. Jimmy Wei, Kurt Shuster, Arthur Szlam, Jason Weston, Jack Urbanek, and Mojtaba Komeili. 2023. Multi-party chat: Conversational agents in group settings with humans and models. arXiv preprint arXiv:2304.13835. — [arXiv:2304.13835](https://arxiv.org/abs/2304.13835) [find](https://scholar.google.com/scholar?q=Jimmy+Wei%2C+Kurt+Shuster%2C+Arthur+Szlam%2C+Jason+Weston%2C+Jack+Urbanek%2C+and+Mojtaba+Komeili%2E+2023%2E+Multi%2Dparty+chat%3A+Conversational+agents+in+group+settings+with+humans+and+models%2E+arXiv+preprint+arXiv%3A230)
24. Chenfei Wu, Shengming Yin, Weizhen Qi, Xiaodong Wang, Zecheng Tang, and Nan Duan. 2023a. [Visual chatgpt: Talking, drawing and editing with visual foundation models](https://md.research.elamur.ai/free/en/papers/wu-2023-visual-chatgpt-talking-drawing-editing/). arXiv preprint arXiv:2303.04671.
25. Qingyun Wu, Gagan Bansal, Jieyu Zhang, Yiran Wu, Shaokun Zhang, Erkang Zhu, Beibin Li, Li Jiang, Xiaoyun Zhang, and Chi Wang. 2023b. [Autogen: Enabling next-gen llm applications via multi-agent conversation](https://md.research.elamur.ai/free/en/papers/wu-2023-autogen-enabling-next-gen-llm/) framework. arXiv preprint arXiv:2308.08155.
26. Linyao Yang, Hongyang Chen, Zhao Li, Xiao Ding, and Xindong Wu. 2023a. Chatgpt is not enough: Enhancing large language models with knowledge graphs for fact-aware language modeling. arXiv preprint arXiv:2306.11489. [find](https://scholar.google.com/scholar?q=Linyao+Yang%2C+Hongyang+Chen%2C+Zhao+Li%2C+Xiao+Ding%2C+and+Xindong+Wu%2E+2023a%2E+Chatgpt+is+not+enough%3A+Enhancing+large+language+models+with+knowledge+graphs+for+fact%2Daware+language+modeling%2E+arXiv+preprint+arX)
27. Rui Yang, Lin Song, Yanwei Li, Sijie Zhao, Yixiao Ge, Xiu Li, and Ying Shan. 2023b. Gpt4tools: Teaching large language model to use tools via self-instruction. arXiv preprint arXiv:2305.18752. [find](https://scholar.google.com/scholar?q=Rui+Yang%2C+Lin+Song%2C+Yanwei+Li%2C+Sijie+Zhao%2C+Yixiao+Ge%2C+Xiu+Li%2C+and+Ying+Shan%2E+2023b%2E+Gpt4tools%3A+Teaching+large+language+model+to+use+tools+via+self%2Dinstruction%2E+arXiv+preprint+arXiv%3A2305%2E18752%2E)
28. Yongliang Shen, Kaitao Song, Xu Tan, Dongsheng Li, Weiming Lu, and Yueting Zhuang. 2023. [Hugginggpt: Solving ai tasks with chatgpt and its friends in hugging face](https://md.research.elamur.ai/free/en/papers/shen-2023-hugginggpt-solving-ai-tasks-chatgpt/). arXiv preprint arXiv:2303.17580.
29. Zhengyuan Yang, Linjie Li, Jianfeng Wang, Kevin Lin, Ehsan Azarnasab, Faisal Ahmed, Zicheng Liu, Ce Liu, Michael Zeng, and Lijuan Wang. 2023c. Mm-react: Prompting chatgpt for multimodal reasoning and action. arXiv preprint arXiv:2303.11381. [find](https://scholar.google.com/scholar?q=Zhengyuan+Yang%2C+Linjie+Li%2C+Jianfeng+Wang%2C+Kevin+Lin%2C+Ehsan+Azarnasab%2C+Faisal+Ahmed%2C+Zicheng+Liu%2C+Ce+Liu%2C+Michael+Zeng%2C+and+Lijuan+Wang%2E+2023c%2E+Mm%2Dreact%3A+Prompting+chatgpt+for+multimodal+reasoning+and+)
30. Yifan Song, Weimin Xiong, Dawei Zhu, Wenhao Wu, Han Qian, Mingbo Song, Hailiang Huang, Cheng Li, Ke Wang, Rong Yao, Ye Tian, and Sujian Li. 2023. [Restgpt: Connecting large language models with real-world restful apis](https://md.research.elamur.ai/free/en/papers/song-2023-restgpt-connecting-large-language-models/).
31. Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik R Narasimhan, and Yuan Cao. 2022. [React: Synergizing reasoning and acting in language models](https://md.research.elamur.ai/free/en/papers/yao-2022-react-synergizing-reasoning-acting-language/). In The Eleventh International Conference on Learning Representations.
32. Qiaoyu Tang, Ziliang Deng, Hongyu Lin, Xianpei Han, Qiao Liang, and Le Sun. 2023. [Toolalpaca: Generalized tool learning for language models with 3000 simulated cases](https://md.research.elamur.ai/free/en/papers/tang-2023-toolalpaca-generalized-tool-learning-language/). arXiv preprint arXiv:2306.05301.
33. Qin Yujia, Yankai Lin Shengding Hu, Weize Chen, Ning Ding, Ganqu Cui, Zheni Zeng, et al. 2023. [Tool learning with foundation models](https://md.research.elamur.ai/free/en/papers/qin-2023-tool-learning-foundation-models/). arXiv preprint arXiv:2304.08354.
34. Aohan Zeng, Mingdao Liu, Rui Lu, Bowen Wang, Xiao Liu, Yuxiao Dong, and Jie Tang. 2023. [Agenttuning: Enabling generalized agent abilities for llms](https://md.research.elamur.ai/free/en/papers/zeng-2023-agenttuning-enabling-generalized-agent/). arXiv preprint arXiv:2310.12823.
35. Xizhou Zhu, Yuntao Chen, Hao Tian, Chenxin Tao, Weijie Su, Chenyu Yang, Gao Huang, Bin Li, Lewei Lu, Xiaogang Wang, et al. 2023. Ghost in the minecraft: Generally capable agents for open-world enviroments via large language models with text-based knowledge and memory. arXiv preprint arXiv:2305.17144… [find](https://scholar.google.com/scholar?q=Xizhou+Zhu%2C+Yuntao+Chen%2C+Hao+Tian%2C+Chenxin+Tao%2C+Weijie+Su%2C+Chenyu+Yang%2C+Gao+Huang%2C+Bin+Li%2C+Lewei+Lu%2C+Xiaogang+Wang%2C+et+al%2E+2023%2E+Ghost+in+the+minecraft%3A+Generally+capable+agents+for+open%2Dworld+envirom)

*Updated 30 October 2026*