{"version":"1.0","generated_at":"2026-10-09T08:56:28.254217","id":148,"slug":"socialgrid-a-benchmark-for-planning-and-social-reasoning-in-embodied-multi-agent-systems","title":"论文速读：SocialGrid，提升开发者接入体验","summary":"SocialGrid是一个受《Among Us》启发的多智能体环境，评估LLM代理的社会推理能力，发现即使最强模型也表现不佳，社会推理是瓶颈。","abstract":"论文速读：SocialGrid，提升开发者接入体验 SocialGrid是一个受《Among Us》启发的多智能体环境，评估LLM代理的社会推理能力，发现即使最强模型也表现不佳，社会推理是瓶颈。 SocialGrid是受《Among Us》启发的多智能体环境 最强开放模型任务完成准确率低于60% 代理依赖浅层启发式，无法检测欺骗 提供可选的规划预言以隔离社会推理 自动失败分析和排行榜帮助开发者改进 随着大型语言模型从文本处理器转变为自主代理，评估它们在具身多智能体环境中的社会推理能力变得至关重要。我们提出了SocialGrid，一个受《Among Us》启发的具身多智能体环境，用于评估LLM代理在规划、任务执行和社会推理方面的表现。我们的评估显示，即使是最强的开放模型（GPT-OSS-120B）在任务完成和规划方面的准确率也低于60%，代理会陷入重复行为或无法通过基本障碍。由于导航能力差会混淆对社会智能的评估，SocialGrid提供了一个可选的规划预言（Planning Oracle）来将社会推理与规划缺陷分离开。虽然规划辅助提高了任务完成率，但社会推理仍然是瓶颈：代理无法检测欺骗，几乎与随机猜测无异，无论规模大小，它们依赖浅层启发式而非累积行为证据。SocialGrid提供自动失败分析和细粒度指标，使开发者能够诊断和改进他们的代理。我们还通过对抗联赛的Elo评分建立了一个竞争性排行榜。 这是什么信号？ SocialGrid的出现标志着AI研究进入一个新的阶段：从关注单智能体能力转向多智能体环境下的社会交互。论文揭示了一个关键发现：即使是最强开源模型，在需要社会推理（如识别…","access_level":"member","access_label":"免费阅读","access_mode":"full","is_preview":false,"canonical_url":"https://opc.beizhux.com/content/148/socialgrid-a-benchmark-for-planning-and-social-reasoning-in-embodied-multi-agent-systems","html_url":"https://opc.beizhux.com/content/148/socialgrid-a-benchmark-for-planning-and-social-reasoning-in-embodied-multi-agent-systems","json_url":"https://opc.beizhux.com/content/148/socialgrid-a-benchmark-for-planning-and-social-reasoning-in-embodied-multi-agent-systems.json","published_at":"2026-04-20T12:00:06","updated_at":"2026-10-09T07:59:07","category":{"slug":"ai-skills","name":"AI技能杠杆"},"source":{"site":"arxiv.org","author":"arXiv cs.AI","url":"https://arxiv.org/abs/2604.16022"},"tags":["AI","arXiv cs.AI","LLM Agent","SocialGrid","多智能体","社会推理","自动化"],"topics":[{"slug":"agent-workflow","name":"Agent工作流","url":"https://opc.beizhux.com/topics/agent-workflow","reason":"这篇内容命中「Agent、智能体、工作流」等主题信号。"},{"slug":"ai-tools","name":"AI工具","url":"https://opc.beizhux.com/topics/ai-tools","reason":"这篇内容命中「自动化、模型」等主题信号。"},{"slug":"ai-super-individual","name":"AI超级个体","url":"https://opc.beizhux.com/topics/ai-super-individual","reason":"这篇内容命中「技能」等主题信号。"}],"keywords":["AI技能杠杆","Agent工作流","AI工具","AI超级个体","工具","自动化","模型","Cursor","Claude","Agent","智能体","工作流"],"questions":[{"question":"论文速读：SocialGrid，提升开发者接入体验主要讲什么？","answer":"SocialGrid是一个受《Among Us》启发的多智能体环境，评估LLM代理的社会推理能力，发现即使最强模型也表现不佳，社会推理是瓶颈。"},{"question":"这篇文章最值得关注的要点是什么？","answer":"SocialGrid是一个受《Among Us》启发的多智能体环境，评估LLM代理的社会推理能力，发现即使最强模型也表现不佳，社会推理是瓶颈。；SocialGrid是受《Among Us》启发的多智能体环境；最强开放模型任务完成准确率低于60%；代理依赖浅层启发式，无法检测欺骗"},{"question":"这篇文章和哪些AI专题相关？","answer":"它适合放在Agent工作流、AI工具、AI超级个体专题里阅读。 关联原因：这篇内容命中「Agent、智能体、工作流」等主题信号。；这篇内容命中「自动化、模型」等主题信号。；这篇内容命中「技能」等主题信号。"},{"question":"阅读这篇文章建议先理解哪些关键词？","answer":"建议先理解AI工具、工具、自动化、模型、Cursor这些关键词，再结合正文判断工具、机会或风险是否值得进入自己的工作流。"}],"terms":[{"slug":"ai-tools-term","name":"AI工具","definition":"在AI觉醒星球里，「AI工具」属于「AI工具」方向。围绕AI工具、模型能力、自动化流程和真实使用场景做整理，优先关注能提升效率、降低成本和创造新交付的工具。 不只看工具热度，更看它能不能进入真实流程。","topic_slug":"ai-tools","topic_name":"AI工具","topic_title":"AI工具：模型、插件、自动化与实战场景","topic_url":"https://opc.beizhux.com/topics/ai-tools","topic_path":"/topics/ai-tools","url":"https://opc.beizhux.com/glossary/ai-tools-term","path":"/glossary/ai-tools-term","json_url":"https://opc.beizhux.com/glossary/ai-tools-term.json"},{"slug":"tools","name":"工具","definition":"在AI觉醒星球里，「工具」属于「AI工具」方向。围绕AI工具、模型能力、自动化流程和真实使用场景做整理，优先关注能提升效率、降低成本和创造新交付的工具。 不只看工具热度，更看它能不能进入真实流程。","topic_slug":"ai-tools","topic_name":"AI工具","topic_title":"AI工具：模型、插件、自动化与实战场景","topic_url":"https://opc.beizhux.com/topics/ai-tools","topic_path":"/topics/ai-tools","url":"https://opc.beizhux.com/glossary/tools","path":"/glossary/tools","json_url":"https://opc.beizhux.com/glossary/tools.json"},{"slug":"automation","name":"自动化","definition":"在AI觉醒星球里，「自动化」属于「AI工具」方向。围绕AI工具、模型能力、自动化流程和真实使用场景做整理，优先关注能提升效率、降低成本和创造新交付的工具。 不只看工具热度，更看它能不能进入真实流程。","topic_slug":"ai-tools","topic_name":"AI工具","topic_title":"AI工具：模型、插件、自动化与实战场景","topic_url":"https://opc.beizhux.com/topics/ai-tools","topic_path":"/topics/ai-tools","url":"https://opc.beizhux.com/glossary/automation","path":"/glossary/automation","json_url":"https://opc.beizhux.com/glossary/automation.json"},{"slug":"ai-model","name":"模型","definition":"在AI觉醒星球里，「模型」属于「AI工具」方向。围绕AI工具、模型能力、自动化流程和真实使用场景做整理，优先关注能提升效率、降低成本和创造新交付的工具。 不只看工具热度，更看它能不能进入真实流程。","topic_slug":"ai-tools","topic_name":"AI工具","topic_title":"AI工具：模型、插件、自动化与实战场景","topic_url":"https://opc.beizhux.com/topics/ai-tools","topic_path":"/topics/ai-tools","url":"https://opc.beizhux.com/glossary/ai-model","path":"/glossary/ai-model","json_url":"https://opc.beizhux.com/glossary/ai-model.json"},{"slug":"cursor","name":"Cursor","definition":"在AI觉醒星球里，「Cursor」属于「AI工具」方向。围绕AI工具、模型能力、自动化流程和真实使用场景做整理，优先关注能提升效率、降低成本和创造新交付的工具。 不只看工具热度，更看它能不能进入真实流程。","topic_slug":"ai-tools","topic_name":"AI工具","topic_title":"AI工具：模型、插件、自动化与实战场景","topic_url":"https://opc.beizhux.com/topics/ai-tools","topic_path":"/topics/ai-tools","url":"https://opc.beizhux.com/glossary/cursor","path":"/glossary/cursor","json_url":"https://opc.beizhux.com/glossary/cursor.json"},{"slug":"claude","name":"Claude","definition":"在AI觉醒星球里，「Claude」属于「AI工具」方向。围绕AI工具、模型能力、自动化流程和真实使用场景做整理，优先关注能提升效率、降低成本和创造新交付的工具。 不只看工具热度，更看它能不能进入真实流程。","topic_slug":"ai-tools","topic_name":"AI工具","topic_title":"AI工具：模型、插件、自动化与实战场景","topic_url":"https://opc.beizhux.com/topics/ai-tools","topic_path":"/topics/ai-tools","url":"https://opc.beizhux.com/glossary/claude","path":"/glossary/claude","json_url":"https://opc.beizhux.com/glossary/claude.json"},{"slug":"agent","name":"Agent","definition":"在AI觉醒星球里，「Agent」属于「Agent工作流」方向。聚合Agent、AI编程、Claude Code、Cursor、工作流自动化等内容，帮助你把智能体能力落到真实任务链路。 把智能体当成任务系统，而不是聊天窗口。","topic_slug":"agent-workflow","topic_name":"Agent工作流","topic_title":"Agent工作流：智能体、AI编程与自动化协作","topic_url":"https://opc.beizhux.com/topics/agent-workflow","topic_path":"/topics/agent-workflow","url":"https://opc.beizhux.com/glossary/agent","path":"/glossary/agent","json_url":"https://opc.beizhux.com/glossary/agent.json"},{"slug":"ai-agent","name":"智能体","definition":"在AI觉醒星球里，「智能体」属于「Agent工作流」方向。聚合Agent、AI编程、Claude Code、Cursor、工作流自动化等内容，帮助你把智能体能力落到真实任务链路。 把智能体当成任务系统，而不是聊天窗口。","topic_slug":"agent-workflow","topic_name":"Agent工作流","topic_title":"Agent工作流：智能体、AI编程与自动化协作","topic_url":"https://opc.beizhux.com/topics/agent-workflow","topic_path":"/topics/agent-workflow","url":"https://opc.beizhux.com/glossary/ai-agent","path":"/glossary/ai-agent","json_url":"https://opc.beizhux.com/glossary/ai-agent.json"}],"preview_hidden_sections":[],"related_items":[{"id":3312,"title":"如何在单张 RTX 5090 上用 MiniMax H3 生成实时视频","url":"https://opc.beizhux.com/content/3312/rtx-5090-minimax-h3","summary":"ComfyUI 作者用智能体汇总优化，把 MiniMax H3 实时视频生成压缩到单张 RTX 5090（32 GB 显存）上运行；生成 30 秒视频耗时 23.51 秒，达到 1.28 倍实时吞吐。","access_level":"member"},{"id":3288,"title":"更好的工作是否总意味着更好的工作者？","url":"https://opc.beizhux.com/content/3288/does-better-work-always-mean-better-workers","summary":"Google Research Blog 分享一项为期三个月的随机对照实验：133 名专利律师使用 AI 专利撰写助手后，工作质量普遍提升，10 天提高 0.34 个标准差，90 天提高 0.38 个标准差；但 AI 对在职学习的影响因资历而异——资深律师判断力增强，初级律师平均技能无增益，得分反而分化为更多高分和更…","access_level":"member"},{"id":3292,"title":"LangChain 重构 Deep Agents 的 Skills 支持，新增工具绑定、固定技能与线程内重载","url":"https://opc.beizhux.com/content/3292/langchain-deep-agents-skills","summary":"LangChain 对 Deep Agents 的 Skills 支持做了重构，面向企业技能库膨胀到数千个技能的场景推出三项更新：工具可绑定到技能、仅在该技能被读取时才加载；用户可用 /meeting-prep 这类显式请求固定技能，使其在首次模型调用前加载；长线程中可通过把 skills_metadata 设为 N…","access_level":"member"}],"usage_policy":{"summarizable":true,"preferred_url":"https://opc.beizhux.com/content/148/socialgrid-a-benchmark-for-planning-and-social-reasoning-in-embodied-multi-agent-systems","private_fields_excluded":true}}