{"paper":{"title":"Ghost in the Minecraft: Generally Capable Agents for Open-World Environments via Large Language Models with Text-based Knowledge and Memory","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Large language models with text memory and knowledge let agents complete Minecraft's full Overworld item tree for the first time.","cross_cats":["cs.CL","cs.CV","cs.LG"],"primary_cat":"cs.AI","authors_text":"Bin Li, Chenxin Tao, Chenyu Yang, Gao Huang, Hao Tian, Jifeng Dai, Lewei Lu, Weijie Su, Xiaogang Wang, Xizhou Zhu, Yuntao Chen, Yu Qiao, Zhaoxiang Zhang","submitted_at":"2023-05-25T17:59:49Z","abstract_excerpt":"The captivating realm of Minecraft has attracted substantial research interest in recent years, serving as a rich platform for developing intelligent agents capable of functioning in open-world environments. However, the current research landscape predominantly focuses on specific objectives, such as the popular \"ObtainDiamond\" task, and has not yet shown effective generalization to a broader spectrum of tasks. Furthermore, the current leading success rate for the \"ObtainDiamond\" task stands at around 20%, highlighting the limitations of Reinforcement Learning (RL) based controllers used in ex"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"The resulting LLM-based agent markedly surpasses previous methods, achieving a remarkable improvement of +47.5% in success rate on the 'ObtainDiamond' task... Notably, our agent is the first to procure all items in the Minecraft Overworld technology tree.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That large language models already contain sufficient logic and common sense to generate reliable long-horizon action plans for sparse-reward open-world tasks when given only text-based state and memory.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"GITM uses LLMs to generate action plans from text knowledge and memory, enabling agents to complete long-horizon Minecraft tasks at much higher success rates than prior RL methods.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Large language models with text memory and knowledge let agents complete Minecraft's full Overworld item tree for the first time.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"49a8c854c3822a50e5a9b48045d7012b4bf1d57af374abb8d23fbb5a96ed10a1"},"source":{"id":"2305.17144","kind":"arxiv","version":2},"verdict":{"id":"70236ee0-85e4-4b3d-9b37-38c40548c3dc","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T18:32:52.638408Z","strongest_claim":"The resulting LLM-based agent markedly surpasses previous methods, achieving a remarkable improvement of +47.5% in success rate on the 'ObtainDiamond' task... Notably, our agent is the first to procure all items in the Minecraft Overworld technology tree.","one_line_summary":"GITM uses LLMs to generate action plans from text knowledge and memory, enabling agents to complete long-horizon Minecraft tasks at much higher success rates than prior RL methods.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That large language models already contain sufficient logic and common sense to generate reliable long-horizon action plans for sparse-reward open-world tasks when given only text-based state and memory.","pith_extraction_headline":"Large language models with text memory and knowledge let agents complete Minecraft's full Overworld item tree for the first time."},"references":{"count":45,"sample":[{"doi":"","year":2007,"title":"A. Amiranashvili, N. Dorka, W. Burgard, V . Koltun, and T. Brox. Scaling imitation learning in minecraft. arXiv preprint arXiv:2007.02701, 2020","work_id":"e7b16df6-9e5c-4fac-8a21-737b11f1e8b0","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"B. Baker, I. Akkaya, P. Zhokov, J. Huizinga, J. Tang, A. Ecoffet, B. Houghton, R. Sampedro, and J. Clune. Video pretraining (vpt): Learning to act by watching unlabeled online videos. Advances in Neur","work_id":"75b9b01d-e05c-4cc5-aa7d-f927f9399a9e","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Open-world multi-task control through goal-aware representation learning and adaptive horizon prediction","work_id":"af3d3319-28aa-4659-b04a-04dd057709b2","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"PaLM-E: An Embodied Multimodal Language Model","work_id":"5b99811a-1d93-47e2-9d59-f4045a0b74a2","ref_index":4,"cited_arxiv_id":"2303.03378","is_internal_anchor":true},{"doi":"","year":2022,"title":"doi:10.48550/arXiv.2206.08853 , author =","work_id":"ded038df-ed98-4558-8bab-af5a5338aad5","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":45,"snapshot_sha256":"1057049b21a3a2b89d97c09f654924fa318a81f36fcdd8ad013c4404d244ab20","internal_anchors":10},"formal_canon":{"evidence_count":3,"snapshot_sha256":"c8edf64d1837e8b0cd2d509dfdec619ef87a05a68e713467314e60c349ee11e3"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}