{ "text": "2026年AI智能体\n也就是我们常说的AI agent\n一下子就火了\n你随便一搜就能看到各种各样的产品\n比如DeepSeaKarnes\n腾讯的WorkBody\n一个叫HermesAgent的\n还有OpenAI的Codex\n乍一看他们好像\n都差不多都是能自己干活的AI\n我们是不是只要比比谁更聪明就行了\n他们四个其实代表了\n四种完全不同的发展路线\n今天咱们最重要的任务\n就不是给他们排个1234\n而是要把这四家的底细都摸清楚\n他们各自到底在解决什么问题\n最厉害的地方在哪\n又分别适合什么样的人去用\n咱们一个一个来看\n先说deepseek harness\n这个名字里的harness翻译过来有点像\n智能体的运行底座或者框架\n你可以这么理解\n一个大模型\n就像一个聪明的大脑\n它会思考\n会说话\n但是它光有大脑\n是没法帮你完成一个\n现实世界里的具体任务的\n它得有手有脚\n得有工具\n得有记忆\n得有个工作台\n这个harness就是给它提供\n身体操作系统和工作环境的那一整套东西\nDeepSeaKarnes最核心的设计理念叫Everything is a Plugin\n翻译过来就是一切皆插件\n你用哪个大语言模型可以换\n你需要它用什么工具可以换\n它的记忆系统存储方式任务调度方法\n甚至是用户界面全都可以换\n开发者就像搭积木一样\n用配置的方式就能把这些能力重新组合起来\n而不用动框架本身的代码\n这就带来了几个非常突出的优点\n第一 开放性和可组合性超级强\n比如说一家公司想搞一套自己的内部智能体\n又不想被某一家模型公司给绑架\n那就可以把不同公司的模型都接进这个Harness里\n遇到复杂的推理任务就调用能力最强的那个模型\n遇到普通的整理任务\n就用一个成本更低的模型\n甚至处理一些敏感数据\n可以直接用部署在自己服务器上的本地模型\n这就非常灵活了\n第二\n它的运行过程非常透明\n它会详细记录下\n模型每一步都看到了什么\n系统给了它什么指令\n它调用了什么工具\n又派出了哪个小助手去干活\n整个任务过程可以被恢复搜索分叉和重放\n这一点对开发者来说太重要了\n因为AI智能体最让人头疼的一个问题就是\n它要是做错了事\n你根本不知道它为什么错\n有了这份完整的黑匣子飞行记录\n调试评估和审计就都有了依据\n第三\n它不只提供一种固定的工作模式\n它有标准模式适合绝大多数任务\n有代码模式\n让模型直接写程序来组织复杂的工具调用\n还有最小模式只保留最核心的功能\n方便测试模型最原始的智能体能力\n甚至还有一个Creator模式\n专门用来设计新的插件和智能体模板\n当然缺点也很明显\nDeepSeek Harness更像是一个基础设施\n一个给专业人士用的工具包\n而不是咱们普通人下载下来\n就能直接当办公助手用的\n模型怎么选 工具怎么接 权限怎么配\n数据怎么存 这些都得你自己来负责\n而且它目前还处在开发者预览阶段\n接口和插件生态都还在快速变化\n所以它最大的价值\n不是让你马上扔掉现在的AI助手\n而是为那些有开发能力的企业或者团队\n提供了一套可控可扩展的\nagent底盘\n好\n说完了需要自己动手组装的底盘\n我们再来看一个已经组装好加满油\n随时能开的整车\n腾讯的workbody\n如果说deep-sea carless\n是给开发者准备的乐高积木\n那workbody就是一辆已经组装好的\n设计好所有细节的汽车\n它的核心目标非常明确\n就是让普通的职场人用一句大白话\n就能把一个相对完整的工作任务\n交给AI去完成\nWorkBuddy强调的是\n全场景办公\n多专家协作\n和开箱即用\n你可以用它来搜集资料\n做市场调研\n分析表格\n生成PPT\n写内容整理文件\n甚至写点代码\n你只需要告诉它你的目标\n它会自己先去理解\n然后把任务拆解\n再调用不同领域的专家\n并行处理\n最后把结果交到你手上\n你还可以检查和修改\nWorkBuddy最大的优点\n首先就是使用门槛极低\n你完全不需要懂什么\nPython编程\nDocker容器\nAPM要像量数据库这些听起来就头大的词\n翻装好之后直接就能用\n这对大多数人来说太重要了\n因为真正决定生产力的\n往往不是理论上限有多高\n而是一个工具能不能在10分钟内\n就开始帮你创造价值\n第二个优点是它对中文办公\n和国内生态的适配特别好\n它覆盖了Windows和macOS系统\n还打通了桌面端主流的\n即时通讯工具和微信小程序\n无论是写中文报告做PPT处理Excel\n还是跟国内企业的沟通场景\nWorkbody的产品设计\n都更贴近我们的日常办公习惯\n第三个优点\n是它把复杂的多智能体协作\n包装成了一个用户能听懂的概念\n叫专家团队\n你不需要自己去设计什么研究agent\n数据agent 协作agent\n你只需要描述你的最终目标\n系统会自动帮你组织\n不同角色的专家来协作\n这种把复杂技术藏在产品界面背后的设计\n大大降低了用户的理解成本\n当然 成品化也意味着\n你失去了一部分底层的控制权\n你很难像用开源框架那样\n完全掌握它内部的每一个运行细节\n它的透明度\n可定制程度和模型选择的自由度\n通常是不如DeepSeek Harness\n或者我们接下来要说的Herms Agent的\n而且对于一些特别复杂的代码仓库\n严格的软件测试任务\nWorkBuddy也未必比那些更专业的编码Agent有优势\n所以WorkBuddy最适合的是谁呢\n就是那些希望马上提升办公效率\n但又不愿意投入大量时间去研究\n部署和配置系统的人\n它的竞争力不在于某一项底层技术有多么领先\n而在于它把模型、工具、专家、文件和办公入口\n整合成了一个可以直接上手用的产品\n接下来我们聊聊一个很有意思的agent\n叫做Hermes Agent\n它是由一个叫Nunz Research的团队开发的\n这个agent最核心的理念跟前面两个都不太一样\n它认为一个真正有价值的智能体\n不应该每次跟你对话都像第一次见面\n而应该在长期的使用中慢慢形成记忆\n掌握你的工作习惯\n并且把成功的经验\n沉淀成可以反复使用的技能\nHermes最大的特点就是闭环学习\n它有一套长期记忆系统\n可以保存你的个人偏好\n项目背景和过去的经验\n它还有一个技能系统\n可以把一个多步骤的复杂流程\n保存成一个可以一键调用的skill\n也就是技能\n以后再遇到类似的任务\n它就不必重新摸索了\n而是可以直接调用\n已经形成的工作方法\n打个比方\n这就有点像一个实习生\n刚来的时候什么都不会\n但你在带他的过程中\n他会犯错\n你会纠正他\n慢慢的他学会了怎么做报告\n怎么分析数据\n下次再让他做类似的工作\n他就能直接上手\n甚至做的又快又好\nHermes就是这样\n如果他在执行任务时走了弯路\n最后找到了正确的方法\n或者你纠正了他的做法\n他可以把这些经验\n写进他的技能库里\n所以Hermes的价值\n是会随着你使用它的时间\n增长而增长的\n他不是简单的记住聊天内容\n而是在尝试把过去的经验\n转化为程序化\n可附用的工作流程\nHermes的第二个优点\n是模型自由度很高\n它可以连接OpenAI\nCloud\nGemini\nDeepSeek\nQianwen等各种云端模型\n甚至你本地的模型也行\n你的长期记忆技能和工作入口\n是属于HIMES这个员工本身的\n而不是绑定在某一个特定的大模型上\n也就是说\n你可以随时给这个员工换个大脑\n但他已经学会的工作方法还在\n第三个优点是它适合\n长期在线运行\n你可以把它部署在自己的\n电脑上或者云端的服务器里\n甚至可以通过Telegram\nDiscord\nSlack\nWhatsApp\n非输 企业微信\n这些渠道跟他交互\n这意味着就算你人不在电脑前\n也可以通过手机给他发个任务\n让云端的agent继续帮你工作\n他还内置了定时任务和子agent的能力\n特别适合用来做周期性的报告\n资料搜集 这类长期自动化任务\n不过 关于Hermes的自进化\n这里有一个非常重要的澄清\n他通常不是自己重新训练的一个大语言模型\n也不是说用了一段时间之后\n模型本身的神经网络权重就自动改变了\n他所谓的成长主要发生在\n在技能和系统提示这些层面上\n它有一个叫Self Evolution\n也就是自我进化的项目\n会利用执行轨迹和评估数据\n来生成和优化技能和提示词\n但并不是在修改模型本身\n这更接近于是工作流程\n和提示系统的迭代\n而不是模型的重新训练\n这种机制\n既是它的优势\n也带来了风险\n你想想\n如果一个agent总结了错误的经验\n他也可能把这个错误\n写进他的记忆里\n如果某一次偶然的特殊情况\n被他误认为是普遍规律\n那这个错误的流程就可能在以后被反复调用\n所以Hermes提供了记忆写入和技能写入的审批机制\n比较稳妥的做法是让agent先提出修改建议\n然后由人来查看差异并批准\n而不是让他完全不受控制的修改自己\n这也给我们一个很重要的启发\n一个AI的自我改进不能只看他有没有学习能力\n还得看他有没有平衡能力\n回滚\n审批和安全边界\n没有验证机制的自我学习\n可能只是让他在\n更有信心的重复错误\n最后我们来看看OpenAI的Codex\n很多人对Codex的印象\n可能还停留在他是一个编码助手\n但现在它已经发展成一个由桌面端\n命令行 IDE插件和云端环境组成的完整体系了\n严格来说 OpenAI 把那些通用的研究\n报告 表格 PPT 和文件交付等功能\n更多的放在了 ChatGPT Work 里\n而 CodeX 仍然以代码 项目 工具和工程执行为核心\n当然 这两者的能力正在越来越多地共享\nCodeX 最大的优点是它特别强调做完并验证\n而不仅仅是给你一些建议\n它可以读取你整个项目的文件\n搜索代码 修改多个文件\n运行命令 执行测试 查看报错信息\n然后根据测试结果继续修复问题\n最后交付给你的不只是一几段示例代码\n而是一个完整的修改摘要\n文件差异对比 测试结果\n以及可以审查的最终成果\n第二个优点是它的工程环境比较完整\n你在本地的Codex\n可以直接使用你电脑上已经装好的编译器\n测试工具和项目依赖\n而云端的Codex\n可以为不同的任务建立隔离的环境\n让多个任务在后台并行运行\n任务完成后\n你可以清楚地看到他都改了些什么\n可以要求他继续调整\n或者直接把修改提交到你的代码仓库\n第三个优点是\n它的权限和安全边界做的比较成熟\nCodex通过沙箱来限制它对文件和网络的访问\n在执行一些高风险命令\n或者跨越权限边界的时候\n会向你请求批准\n这种设计可能会让操作显得有点谨慎\n但对于一个能够真实修改文件\n运行程序和连接外部工具的智能体来说\n这种谨慎本身就是可靠性的重要组成部分\n第四个优点是它把编码研究和文件交付连接了起来\n在ChatGPT Work这个更大的环境里\n同一套能力还可以用来生成和检查文档\n表格PPT PDF网站和数据分析结果\n这意味着一个复杂的任务可以从资料搜集开始\n经过分析和程序处理\n最后形成可以直接使用的文件\n整个流程都可以在一个环境里完成\n当然Codex的局限也很清楚\n它主要围绕OpenAI自家的模型和ChatGPT服务来运行\n模型的自由度不如Hermes或者DeepSeek Harness\n他虽然也支持技能记忆和项目规则\n但不会把自动修改自己作为最核心的卖点\n他的取向更加保守\n宁可让流程可审查可验证\n也不轻易让agent自行改变长期规则\n好了\n现在我们把这四个智能体放在一起\n就能得到一个非常清晰的结论了\n如果你的目标是开发一个自己的agent的产品\n那DeepSeek Harness最有吸引力\n因为它开放模块化\n运行过程可追溯\n如果你的目标是普通办公人员想马上提高效率\n那Workbody最方便\n因为它把复杂的技术包装成了\n如果你的目标是培养一个能长期在线\n逐渐熟悉你个人工作方式的智能体\n那Hermes最有特色\n如果你的目标是完成复杂的代码\n工具执行和需要严格验证的专业任务\n那Codex的综合完成度更高\n你看他们代表的其实是4条完全不同的路线\nDeepSeek Harness追求的是架构上的自由\nWorkBuddy追求的是使用的便利\nHermes追求的是长期的成长\n而Codex追求的是可靠的交付\n聊到这里\n还有三点特别值得我们进一步思考\n第一 未来agent的竞争绝对不会只发生在模型层面\n模型当然很重要\n它决定了智力的上限\n但真正决定这个智能体能不能稳定落地完成实际工作的\n是模型之外的那个Harness也就是它的运行底座\n它能用什么工具\n能访问哪些文件\n怎么记忆\n怎么调度任务\n出错了能不能恢复\n以及我们作为使用者能不能看清它到底都做了什么\n这些才是关键\n第二 我们刚才聊到的所谓自进化\n绝对不能脱离一个有效的评估系统\n一个agent能够修改自己的技能并不等于他一定会变得更好\n真正有效的自我改进至少需要完整的执行记录\n客观的测试数据 效果比较机智必要的人类审批\n以及出错了能回滚的方案\n否则他可能只是在更有信心的把你最初交给他的那个错误\n一遍又一遍的固化下来\n第三 最好的agent不一定是最自主的agent\n对于很多低风险的重复性的任务\n高程度的自动化确实能节省大量时间\n但是当涉及到付款 删除文件 发布公开内容\n修改正式代码或者处理敏感数据这些关键操作时\n一个能够暂停下来向你解释情况\n并等待你确认的agent\n它的价值其实远远高于一个完全自主\n从不请示的agent\n一个成熟的智能体系统不是要取消人的判断\n而是要把人的判断\n集中到那些最关键的节点上\n最后我们来总结一下今天的内容\nDeepSake Harness\n最适合作为构建自由agent系统的开放底座\n腾讯的WorkBuddy\n最省事\n适合想直接进入日常办公场景的用户\nHermes Agent最像一个会积累记忆和技能的长期AI员工\n而OpenAI的Codex则最擅长把复杂的专业工作\n真正做完并进行严格的验证\n所以下次你在选择的时候不要只问谁最强\n而要问自己三个更具体的问题\n第一 我需要的是一个成品工具\n还是一个可以自己开发的框架\n第二 我更看重的是即时的效率提升\n还是长期的成长和陪伴\n第三 我是否需要掌握对模型\n数据 权限和运行环境的控制权\n当你把这三个问题想清楚之后\n这死者之间的选择就一点都不模糊了\n他们不是简单的\n谁替代谁的关系\n而是分别服务于不同层级\n不同需求的\n未来更常见的情况\n可能也不是某一个超级agent\n包揽所有工作\n而是像今天聊到的成品办公agent\n专业工程agent\n长期个人agent和企业资金底座\n他们相互配合\n共同组成一个真正好用的\nAI工作系统", "language": "multilingual", "duration": 800.832, "segments": [ { "start": 0.0, "end": 1.36, "text": "2026年AI智能体", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 1.36, "end": 2.96, "text": "也就是我们常说的AI agent", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 2.96, "end": 3.84, "text": "一下子就火了", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 3.84, "end": 5.76, "text": "你随便一搜就能看到各种各样的产品", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 5.76, "end": 6.96, "text": "比如DeepSeaKarnes", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 7.32, "end": 8.32, "text": "腾讯的WorkBody", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 8.32, "end": 10.0, "text": "一个叫HermesAgent的", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 10.0, "end": 11.52, "text": "还有OpenAI的Codex", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 11.52, "end": 12.64, "text": "乍一看他们好像", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 13.2, "end": 14.88, "text": "都差不多都是能自己干活的AI", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 15.12, "end": 16.96, "text": "我们是不是只要比比谁更聪明就行了", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 16.96, "end": 18.16, "text": "他们四个其实代表了", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 18.16, "end": 20.24, "text": "四种完全不同的发展路线", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 20.24, "end": 22.08, "text": "今天咱们最重要的任务", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 22.08, "end": 23.84, "text": "就不是给他们排个1234", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 23.84, "end": 25.52, "text": "而是要把这四家的底细都摸清楚", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 25.52, "end": 27.76, "text": "他们各自到底在解决什么问题", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 27.76, "end": 28.72, "text": "最厉害的地方在哪", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 28.72, "end": 30.38, "text": "又分别适合什么样的人去用", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 30.38, "end": 31.46, "text": "咱们一个一个来看", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 31.46, "end": 33.22, "text": "先说deepseek harness", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 33.22, "end": 35.56, "text": "这个名字里的harness翻译过来有点像", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 35.56, "end": 38.32, "text": "智能体的运行底座或者框架", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 38.32, "end": 39.12, "text": "你可以这么理解", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 39.12, "end": 40.32, "text": "一个大模型", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 40.32, "end": 41.76, "text": "就像一个聪明的大脑", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 41.76, "end": 42.8, "text": "它会思考", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 42.8, "end": 43.7, "text": "会说话", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 43.7, "end": 45.4, "text": "但是它光有大脑", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 45.4, "end": 46.9, "text": "是没法帮你完成一个", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 46.9, "end": 48.4, "text": "现实世界里的具体任务的", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 48.4, "end": 49.9, "text": "它得有手有脚", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 49.9, "end": 51.1, "text": "得有工具", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 51.1, "end": 52.1, "text": "得有记忆", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 52.1, "end": 53.3, "text": "得有个工作台", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 53.3, "end": 55.0, "text": "这个harness就是给它提供", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 55.0, "end": 57.94, "text": "身体操作系统和工作环境的那一整套东西", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 57.94, "end": 60.74, "text": "DeepSeaKarnes最核心的设计理念叫Everything is a Plugin", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 60.74, "end": 62.74, "text": "翻译过来就是一切皆插件", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 62.74, "end": 64.74, "text": "你用哪个大语言模型可以换", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 64.74, "end": 66.84, "text": "你需要它用什么工具可以换", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 66.84, "end": 69.64, "text": "它的记忆系统存储方式任务调度方法", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 69.64, "end": 71.74, "text": "甚至是用户界面全都可以换", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 71.74, "end": 73.24, "text": "开发者就像搭积木一样", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 73.24, "end": 75.84, "text": "用配置的方式就能把这些能力重新组合起来", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 75.84, "end": 77.34, "text": "而不用动框架本身的代码", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 77.34, "end": 79.54, "text": "这就带来了几个非常突出的优点", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 79.54, "end": 81.84, "text": "第一 开放性和可组合性超级强", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 81.84, "end": 84.84, "text": "比如说一家公司想搞一套自己的内部智能体", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 84.84, "end": 87.54, "text": "又不想被某一家模型公司给绑架", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 87.54, "end": 90.04, "text": "那就可以把不同公司的模型都接进这个Harness里", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 90.04, "end": 92.64, "text": "遇到复杂的推理任务就调用能力最强的那个模型", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 92.64, "end": 93.94, "text": "遇到普通的整理任务", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 93.94, "end": 95.84, "text": "就用一个成本更低的模型", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 95.84, "end": 97.44, "text": "甚至处理一些敏感数据", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 97.44, "end": 100.34, "text": "可以直接用部署在自己服务器上的本地模型", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 100.34, "end": 101.64, "text": "这就非常灵活了", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 101.64, "end": 102.24, "text": "第二", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 102.24, "end": 103.84, "text": "它的运行过程非常透明", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 103.84, "end": 105.04, "text": "它会详细记录下", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 105.04, "end": 106.64, "text": "模型每一步都看到了什么", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 106.64, "end": 108.04, "text": "系统给了它什么指令", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 108.04, "end": 109.14, "text": "它调用了什么工具", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 109.14, "end": 111.34, "text": "又派出了哪个小助手去干活", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 111.34, "end": 114.74, "text": "整个任务过程可以被恢复搜索分叉和重放", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 114.74, "end": 116.74, "text": "这一点对开发者来说太重要了", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 116.74, "end": 119.14, "text": "因为AI智能体最让人头疼的一个问题就是", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 119.14, "end": 119.94, "text": "它要是做错了事", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 119.94, "end": 121.24, "text": "你根本不知道它为什么错", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 121.24, "end": 123.44, "text": "有了这份完整的黑匣子飞行记录", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 123.44, "end": 125.74, "text": "调试评估和审计就都有了依据", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 125.74, "end": 126.24, "text": "第三", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 126.24, "end": 128.24, "text": "它不只提供一种固定的工作模式", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 128.24, "end": 130.84, "text": "它有标准模式适合绝大多数任务", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 130.84, "end": 132.34, "text": "有代码模式", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 132.34, "end": 135.44, "text": "让模型直接写程序来组织复杂的工具调用", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 135.44, "end": 137.94, "text": "还有最小模式只保留最核心的功能", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 137.94, "end": 140.42, "text": "方便测试模型最原始的智能体能力", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 140.42, "end": 141.86, "text": "甚至还有一个Creator模式", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 141.86, "end": 144.5, "text": "专门用来设计新的插件和智能体模板", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 144.5, "end": 145.86, "text": "当然缺点也很明显", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 145.86, "end": 147.82, "text": "DeepSeek Harness更像是一个基础设施", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 147.82, "end": 149.34, "text": "一个给专业人士用的工具包", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 149.34, "end": 150.7, "text": "而不是咱们普通人下载下来", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 150.7, "end": 152.22, "text": "就能直接当办公助手用的", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 152.22, "end": 154.5, "text": "模型怎么选 工具怎么接 权限怎么配", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 154.5, "end": 156.7, "text": "数据怎么存 这些都得你自己来负责", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 156.7, "end": 159.26, "text": "而且它目前还处在开发者预览阶段", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 159.26, "end": 161.22, "text": "接口和插件生态都还在快速变化", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 161.22, "end": 162.86, "text": "所以它最大的价值", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 162.86, "end": 165.46, "text": "不是让你马上扔掉现在的AI助手", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 165.46, "end": 168.02, "text": "而是为那些有开发能力的企业或者团队", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 168.02, "end": 170.14, "text": "提供了一套可控可扩展的", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 170.42, "end": 171.26, "text": "agent底盘", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 171.26, "end": 171.74, "text": "好", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 171.74, "end": 174.06, "text": "说完了需要自己动手组装的底盘", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 174.06, "end": 176.3, "text": "我们再来看一个已经组装好加满油", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 176.3, "end": 177.7, "text": "随时能开的整车", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 177.7, "end": 178.9, "text": "腾讯的workbody", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 178.9, "end": 180.18, "text": "如果说deep-sea carless", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 180.18, "end": 181.82, "text": "是给开发者准备的乐高积木", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 181.82, "end": 183.42, "text": "那workbody就是一辆已经组装好的", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 183.42, "end": 184.86, "text": "设计好所有细节的汽车", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 184.86, "end": 186.34, "text": "它的核心目标非常明确", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 186.34, "end": 189.26, "text": "就是让普通的职场人用一句大白话", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 189.26, "end": 191.62, "text": "就能把一个相对完整的工作任务", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 191.62, "end": 192.92, "text": "交给AI去完成", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 192.92, "end": 194.08, "text": "WorkBuddy强调的是", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 194.08, "end": 195.02, "text": "全场景办公", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 195.02, "end": 195.88, "text": "多专家协作", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 195.88, "end": 196.72, "text": "和开箱即用", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 196.72, "end": 198.42, "text": "你可以用它来搜集资料", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 198.42, "end": 199.48, "text": "做市场调研", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 199.48, "end": 200.48, "text": "分析表格", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 200.48, "end": 201.58, "text": "生成PPT", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 201.58, "end": 202.88, "text": "写内容整理文件", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 202.88, "end": 203.88, "text": "甚至写点代码", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 203.88, "end": 205.42, "text": "你只需要告诉它你的目标", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 205.42, "end": 206.62, "text": "它会自己先去理解", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 206.62, "end": 207.58, "text": "然后把任务拆解", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 207.58, "end": 209.18, "text": "再调用不同领域的专家", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 209.18, "end": 209.98, "text": "并行处理", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 209.98, "end": 211.42, "text": "最后把结果交到你手上", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 211.42, "end": 212.98, "text": "你还可以检查和修改", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 212.98, "end": 214.02, "text": "WorkBuddy最大的优点", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 214.02, "end": 215.58, "text": "首先就是使用门槛极低", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 215.58, "end": 216.62, "text": "你完全不需要懂什么", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 216.62, "end": 217.22, "text": "Python编程", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 217.22, "end": 217.82, "text": "Docker容器", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 217.82, "end": 220.58, "text": "APM要像量数据库这些听起来就头大的词", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 220.58, "end": 221.9, "text": "翻装好之后直接就能用", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 221.9, "end": 223.66, "text": "这对大多数人来说太重要了", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 223.66, "end": 225.14, "text": "因为真正决定生产力的", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 225.14, "end": 226.74, "text": "往往不是理论上限有多高", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 226.74, "end": 228.54, "text": "而是一个工具能不能在10分钟内", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 228.54, "end": 229.98, "text": "就开始帮你创造价值", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 229.98, "end": 231.86, "text": "第二个优点是它对中文办公", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 231.86, "end": 233.54, "text": "和国内生态的适配特别好", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 233.54, "end": 235.74, "text": "它覆盖了Windows和macOS系统", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 235.74, "end": 237.7, "text": "还打通了桌面端主流的", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 237.7, "end": 239.86, "text": "即时通讯工具和微信小程序", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 239.86, "end": 242.58, "text": "无论是写中文报告做PPT处理Excel", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 242.58, "end": 244.42, "text": "还是跟国内企业的沟通场景", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 244.42, "end": 245.86, "text": "Workbody的产品设计", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 245.86, "end": 247.86, "text": "都更贴近我们的日常办公习惯", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 247.86, "end": 248.86, "text": "第三个优点", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 248.86, "end": 251.46, "text": "是它把复杂的多智能体协作", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 251.46, "end": 253.76, "text": "包装成了一个用户能听懂的概念", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 253.76, "end": 254.96, "text": "叫专家团队", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 254.96, "end": 257.06, "text": "你不需要自己去设计什么研究agent", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 257.06, "end": 258.66, "text": "数据agent 协作agent", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 258.66, "end": 260.36, "text": "你只需要描述你的最终目标", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 260.36, "end": 262.06, "text": "系统会自动帮你组织", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 262.06, "end": 263.86, "text": "不同角色的专家来协作", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 263.86, "end": 266.66, "text": "这种把复杂技术藏在产品界面背后的设计", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 266.66, "end": 268.26, "text": "大大降低了用户的理解成本", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 268.26, "end": 269.56, "text": "当然 成品化也意味着", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 269.56, "end": 271.66, "text": "你失去了一部分底层的控制权", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 271.66, "end": 273.26, "text": "你很难像用开源框架那样", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 273.26, "end": 275.82, "text": "完全掌握它内部的每一个运行细节", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 275.82, "end": 277.1, "text": "它的透明度", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 277.1, "end": 279.14, "text": "可定制程度和模型选择的自由度", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 279.14, "end": 280.74, "text": "通常是不如DeepSeek Harness", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 280.74, "end": 282.7, "text": "或者我们接下来要说的Herms Agent的", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 282.7, "end": 285.06, "text": "而且对于一些特别复杂的代码仓库", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 285.06, "end": 286.38, "text": "严格的软件测试任务", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 286.38, "end": 289.46, "text": "WorkBuddy也未必比那些更专业的编码Agent有优势", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 289.46, "end": 292.02, "text": "所以WorkBuddy最适合的是谁呢", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 292.02, "end": 294.18, "text": "就是那些希望马上提升办公效率", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 294.18, "end": 296.3, "text": "但又不愿意投入大量时间去研究", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 296.3, "end": 297.66, "text": "部署和配置系统的人", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 297.66, "end": 301.1, "text": "它的竞争力不在于某一项底层技术有多么领先", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 301.1, "end": 305.22, "text": "而在于它把模型、工具、专家、文件和办公入口", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 305.22, "end": 307.38, "text": "整合成了一个可以直接上手用的产品", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 307.38, "end": 309.46, "text": "接下来我们聊聊一个很有意思的agent", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 309.46, "end": 310.62, "text": "叫做Hermes Agent", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 310.62, "end": 313.66, "text": "它是由一个叫Nunz Research的团队开发的", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 313.66, "end": 316.34, "text": "这个agent最核心的理念跟前面两个都不太一样", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 316.34, "end": 318.86, "text": "它认为一个真正有价值的智能体", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 318.86, "end": 320.7, "text": "不应该每次跟你对话都像第一次见面", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 320.7, "end": 323.38, "text": "而应该在长期的使用中慢慢形成记忆", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 323.38, "end": 324.62, "text": "掌握你的工作习惯", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 324.62, "end": 325.9, "text": "并且把成功的经验", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 325.9, "end": 327.78, "text": "沉淀成可以反复使用的技能", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 327.78, "end": 329.9, "text": "Hermes最大的特点就是闭环学习", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 329.9, "end": 331.58, "text": "它有一套长期记忆系统", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 331.58, "end": 332.86, "text": "可以保存你的个人偏好", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 332.86, "end": 334.26, "text": "项目背景和过去的经验", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 334.26, "end": 335.78, "text": "它还有一个技能系统", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 335.78, "end": 338.54, "text": "可以把一个多步骤的复杂流程", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 338.54, "end": 340.74, "text": "保存成一个可以一键调用的skill", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 340.74, "end": 341.62, "text": "也就是技能", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 341.62, "end": 342.9, "text": "以后再遇到类似的任务", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 342.9, "end": 344.18, "text": "它就不必重新摸索了", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 344.18, "end": 345.34, "text": "而是可以直接调用", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 345.34, "end": 346.62, "text": "已经形成的工作方法", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 346.62, "end": 347.14, "text": "打个比方", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 347.14, "end": 348.66, "text": "这就有点像一个实习生", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 348.66, "end": 349.94, "text": "刚来的时候什么都不会", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 349.94, "end": 351.02, "text": "但你在带他的过程中", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 351.02, "end": 351.74, "text": "他会犯错", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 351.74, "end": 352.58, "text": "你会纠正他", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 352.58, "end": 354.3, "text": "慢慢的他学会了怎么做报告", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 354.3, "end": 355.14, "text": "怎么分析数据", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 355.14, "end": 356.78, "text": "下次再让他做类似的工作", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 356.78, "end": 357.7, "text": "他就能直接上手", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 357.7, "end": 358.74, "text": "甚至做的又快又好", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 358.74, "end": 359.5, "text": "Hermes就是这样", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 359.5, "end": 361.38, "text": "如果他在执行任务时走了弯路", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 361.38, "end": 362.62, "text": "最后找到了正确的方法", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 362.62, "end": 363.9, "text": "或者你纠正了他的做法", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 363.9, "end": 364.82, "text": "他可以把这些经验", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 364.82, "end": 365.86, "text": "写进他的技能库里", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 365.86, "end": 367.34, "text": "所以Hermes的价值", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 367.34, "end": 368.98, "text": "是会随着你使用它的时间", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 368.98, "end": 369.82, "text": "增长而增长的", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 369.82, "end": 371.66, "text": "他不是简单的记住聊天内容", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 371.66, "end": 373.46, "text": "而是在尝试把过去的经验", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 373.46, "end": 374.66, "text": "转化为程序化", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 374.66, "end": 376.22, "text": "可附用的工作流程", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 376.22, "end": 377.42, "text": "Hermes的第二个优点", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 377.42, "end": 378.9, "text": "是模型自由度很高", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 378.9, "end": 380.7, "text": "它可以连接OpenAI", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 380.7, "end": 381.5, "text": "Cloud", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 381.5, "end": 382.5, "text": "Gemini", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 382.5, "end": 383.5, "text": "DeepSeek", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 383.5, "end": 385.34, "text": "Qianwen等各种云端模型", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 385.34, "end": 386.7, "text": "甚至你本地的模型也行", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 386.7, "end": 389.02, "text": "你的长期记忆技能和工作入口", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 389.02, "end": 391.18, "text": "是属于HIMES这个员工本身的", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 391.18, "end": 393.46, "text": "而不是绑定在某一个特定的大模型上", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 393.46, "end": 394.14, "text": "也就是说", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 394.14, "end": 396.42, "text": "你可以随时给这个员工换个大脑", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 396.42, "end": 398.18, "text": "但他已经学会的工作方法还在", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 398.18, "end": 399.86, "text": "第三个优点是它适合", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 399.86, "end": 401.06, "text": "长期在线运行", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 401.06, "end": 402.62, "text": "你可以把它部署在自己的", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 402.62, "end": 404.62, "text": "电脑上或者云端的服务器里", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 404.62, "end": 405.86, "text": "甚至可以通过Telegram", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 405.86, "end": 406.3, "text": "Discord", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 406.3, "end": 406.66, "text": "Slack", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 406.66, "end": 407.1, "text": "WhatsApp", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 407.1, "end": 408.76, "text": "非输 企业微信", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 408.76, "end": 410.1, "text": "这些渠道跟他交互", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 410.1, "end": 412.34, "text": "这意味着就算你人不在电脑前", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 412.34, "end": 413.9, "text": "也可以通过手机给他发个任务", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 413.9, "end": 415.44, "text": "让云端的agent继续帮你工作", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 415.44, "end": 417.76, "text": "他还内置了定时任务和子agent的能力", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 417.76, "end": 419.5, "text": "特别适合用来做周期性的报告", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 419.5, "end": 421.64, "text": "资料搜集 这类长期自动化任务", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 421.64, "end": 423.64, "text": "不过 关于Hermes的自进化", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 423.64, "end": 425.2, "text": "这里有一个非常重要的澄清", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 425.2, "end": 427.54, "text": "他通常不是自己重新训练的一个大语言模型", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 427.54, "end": 428.94, "text": "也不是说用了一段时间之后", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 428.94, "end": 431.44, "text": "模型本身的神经网络权重就自动改变了", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 431.44, "end": 432.9, "text": "他所谓的成长主要发生在", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 432.9, "end": 435.5, "text": "在技能和系统提示这些层面上", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 435.5, "end": 437.06, "text": "它有一个叫Self Evolution", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 437.06, "end": 438.7, "text": "也就是自我进化的项目", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 438.7, "end": 440.98, "text": "会利用执行轨迹和评估数据", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 440.98, "end": 442.86, "text": "来生成和优化技能和提示词", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 442.86, "end": 444.66, "text": "但并不是在修改模型本身", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 444.66, "end": 446.9, "text": "这更接近于是工作流程", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 446.9, "end": 448.54, "text": "和提示系统的迭代", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 448.54, "end": 450.42, "text": "而不是模型的重新训练", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 450.42, "end": 451.26, "text": "这种机制", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 451.26, "end": 452.3, "text": "既是它的优势", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 452.3, "end": 453.26, "text": "也带来了风险", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 453.26, "end": 453.82, "text": "你想想", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 453.82, "end": 455.86, "text": "如果一个agent总结了错误的经验", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 455.86, "end": 456.78, "text": "他也可能把这个错误", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 456.78, "end": 457.62, "text": "写进他的记忆里", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 457.62, "end": 459.14, "text": "如果某一次偶然的特殊情况", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 459.14, "end": 460.46, "text": "被他误认为是普遍规律", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 460.46, "end": 462.78, "text": "那这个错误的流程就可能在以后被反复调用", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 462.78, "end": 466.5, "text": "所以Hermes提供了记忆写入和技能写入的审批机制", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 466.5, "end": 469.58, "text": "比较稳妥的做法是让agent先提出修改建议", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 469.58, "end": 471.62, "text": "然后由人来查看差异并批准", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 471.62, "end": 474.22, "text": "而不是让他完全不受控制的修改自己", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 474.22, "end": 475.86, "text": "这也给我们一个很重要的启发", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 475.86, "end": 478.74, "text": "一个AI的自我改进不能只看他有没有学习能力", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 478.74, "end": 480.0, "text": "还得看他有没有平衡能力", "language": "zh", "chunk": 0, "local_language_run": 0, "language_run": 0 }, { "start": 480.3, "end": 481.1, "text": "回滚", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 481.1, "end": 482.4, "text": "审批和安全边界", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 482.4, "end": 483.9, "text": "没有验证机制的自我学习", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 483.9, "end": 485.1, "text": "可能只是让他在", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 485.1, "end": 486.6, "text": "更有信心的重复错误", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 486.6, "end": 489.1, "text": "最后我们来看看OpenAI的Codex", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 489.1, "end": 490.3, "text": "很多人对Codex的印象", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 490.3, "end": 492.4, "text": "可能还停留在他是一个编码助手", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 492.4, "end": 494.64, "text": "但现在它已经发展成一个由桌面端", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 494.64, "end": 497.8, "text": "命令行 IDE插件和云端环境组成的完整体系了", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 497.8, "end": 500.28, "text": "严格来说 OpenAI 把那些通用的研究", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 500.28, "end": 503.08, "text": "报告 表格 PPT 和文件交付等功能", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 503.08, "end": 504.92, "text": "更多的放在了 ChatGPT Work 里", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 504.92, "end": 508.52, "text": "而 CodeX 仍然以代码 项目 工具和工程执行为核心", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 508.52, "end": 511.0, "text": "当然 这两者的能力正在越来越多地共享", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 511.0, "end": 514.16, "text": "CodeX 最大的优点是它特别强调做完并验证", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 514.16, "end": 516.16, "text": "而不仅仅是给你一些建议", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 516.16, "end": 518.2, "text": "它可以读取你整个项目的文件", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 518.2, "end": 519.84, "text": "搜索代码 修改多个文件", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 519.84, "end": 522.44, "text": "运行命令 执行测试 查看报错信息", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 522.44, "end": 524.24, "text": "然后根据测试结果继续修复问题", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 524.24, "end": 526.64, "text": "最后交付给你的不只是一几段示例代码", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 526.64, "end": 528.64, "text": "而是一个完整的修改摘要", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 528.64, "end": 531.04, "text": "文件差异对比 测试结果", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 531.04, "end": 533.14, "text": "以及可以审查的最终成果", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 533.14, "end": 536.34, "text": "第二个优点是它的工程环境比较完整", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 536.34, "end": 537.74, "text": "你在本地的Codex", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 537.74, "end": 540.64, "text": "可以直接使用你电脑上已经装好的编译器", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 540.64, "end": 542.14, "text": "测试工具和项目依赖", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 542.14, "end": 543.14, "text": "而云端的Codex", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 543.14, "end": 545.14, "text": "可以为不同的任务建立隔离的环境", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 545.14, "end": 547.54, "text": "让多个任务在后台并行运行", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 547.54, "end": 548.14, "text": "任务完成后", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 548.14, "end": 549.86, "text": "你可以清楚地看到他都改了些什么", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 549.86, "end": 551.22, "text": "可以要求他继续调整", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 551.22, "end": 554.46, "text": "或者直接把修改提交到你的代码仓库", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 554.46, "end": 555.38, "text": "第三个优点是", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 555.38, "end": 557.78, "text": "它的权限和安全边界做的比较成熟", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 557.78, "end": 560.9, "text": "Codex通过沙箱来限制它对文件和网络的访问", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 560.9, "end": 562.26, "text": "在执行一些高风险命令", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 562.26, "end": 563.94, "text": "或者跨越权限边界的时候", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 563.94, "end": 565.14, "text": "会向你请求批准", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 565.14, "end": 567.34, "text": "这种设计可能会让操作显得有点谨慎", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 567.34, "end": 569.62, "text": "但对于一个能够真实修改文件", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 569.62, "end": 572.38, "text": "运行程序和连接外部工具的智能体来说", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 572.38, "end": 575.02, "text": "这种谨慎本身就是可靠性的重要组成部分", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 575.02, "end": 578.62, "text": "第四个优点是它把编码研究和文件交付连接了起来", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 578.62, "end": 580.62, "text": "在ChatGPT Work这个更大的环境里", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 580.62, "end": 583.22, "text": "同一套能力还可以用来生成和检查文档", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 583.22, "end": 587.02, "text": "表格PPT PDF网站和数据分析结果", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 587.02, "end": 590.62, "text": "这意味着一个复杂的任务可以从资料搜集开始", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 590.62, "end": 592.02, "text": "经过分析和程序处理", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 592.02, "end": 594.12, "text": "最后形成可以直接使用的文件", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 594.12, "end": 596.02, "text": "整个流程都可以在一个环境里完成", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 596.02, "end": 597.92, "text": "当然Codex的局限也很清楚", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 597.92, "end": 601.72, "text": "它主要围绕OpenAI自家的模型和ChatGPT服务来运行", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 601.72, "end": 604.92, "text": "模型的自由度不如Hermes或者DeepSeek Harness", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 604.92, "end": 607.48, "text": "他虽然也支持技能记忆和项目规则", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 607.48, "end": 610.32, "text": "但不会把自动修改自己作为最核心的卖点", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 610.32, "end": 611.68, "text": "他的取向更加保守", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 611.68, "end": 614.2, "text": "宁可让流程可审查可验证", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 614.2, "end": 616.52, "text": "也不轻易让agent自行改变长期规则", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 616.52, "end": 616.88, "text": "好了", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 616.88, "end": 618.84, "text": "现在我们把这四个智能体放在一起", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 618.84, "end": 620.6, "text": "就能得到一个非常清晰的结论了", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 620.6, "end": 623.04, "text": "如果你的目标是开发一个自己的agent的产品", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 623.04, "end": 625.48, "text": "那DeepSeek Harness最有吸引力", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 625.48, "end": 627.44, "text": "因为它开放模块化", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 627.44, "end": 628.8, "text": "运行过程可追溯", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 628.8, "end": 631.52, "text": "如果你的目标是普通办公人员想马上提高效率", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 631.52, "end": 632.6, "text": "那Workbody最方便", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 632.6, "end": 634.28, "text": "因为它把复杂的技术包装成了", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 634.92, "end": 637.8, "text": "如果你的目标是培养一个能长期在线", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 637.8, "end": 640.0, "text": "逐渐熟悉你个人工作方式的智能体", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 640.0, "end": 641.64, "text": "那Hermes最有特色", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 641.64, "end": 643.48, "text": "如果你的目标是完成复杂的代码", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 643.48, "end": 645.96, "text": "工具执行和需要严格验证的专业任务", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 645.96, "end": 647.92, "text": "那Codex的综合完成度更高", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 647.92, "end": 651.56, "text": "你看他们代表的其实是4条完全不同的路线", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 651.56, "end": 653.96, "text": "DeepSeek Harness追求的是架构上的自由", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 653.96, "end": 655.88, "text": "WorkBuddy追求的是使用的便利", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 655.88, "end": 657.96, "text": "Hermes追求的是长期的成长", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 657.96, "end": 660.2, "text": "而Codex追求的是可靠的交付", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 660.2, "end": 660.84, "text": "聊到这里", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 660.84, "end": 663.08, "text": "还有三点特别值得我们进一步思考", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 663.08, "end": 666.72, "text": "第一 未来agent的竞争绝对不会只发生在模型层面", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 666.72, "end": 668.08, "text": "模型当然很重要", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 668.08, "end": 669.48, "text": "它决定了智力的上限", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 669.48, "end": 672.92, "text": "但真正决定这个智能体能不能稳定落地完成实际工作的", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 672.92, "end": 676.12, "text": "是模型之外的那个Harness也就是它的运行底座", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 676.12, "end": 677.08, "text": "它能用什么工具", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 677.08, "end": 678.48, "text": "能访问哪些文件", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 678.48, "end": 679.16, "text": "怎么记忆", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 679.16, "end": 680.28, "text": "怎么调度任务", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 680.28, "end": 681.4, "text": "出错了能不能恢复", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 681.4, "end": 685.08, "text": "以及我们作为使用者能不能看清它到底都做了什么", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 685.08, "end": 686.36, "text": "这些才是关键", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 686.36, "end": 688.48, "text": "第二 我们刚才聊到的所谓自进化", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 688.48, "end": 690.6, "text": "绝对不能脱离一个有效的评估系统", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 690.6, "end": 694.12, "text": "一个agent能够修改自己的技能并不等于他一定会变得更好", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 694.12, "end": 697.36, "text": "真正有效的自我改进至少需要完整的执行记录", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 697.36, "end": 700.64, "text": "客观的测试数据 效果比较机智必要的人类审批", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 700.64, "end": 702.24, "text": "以及出错了能回滚的方案", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 702.24, "end": 706.52, "text": "否则他可能只是在更有信心的把你最初交给他的那个错误", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 706.52, "end": 708.24, "text": "一遍又一遍的固化下来", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 708.24, "end": 710.96, "text": "第三 最好的agent不一定是最自主的agent", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 710.96, "end": 713.36, "text": "对于很多低风险的重复性的任务", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 713.36, "end": 715.8, "text": "高程度的自动化确实能节省大量时间", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 715.8, "end": 720.16, "text": "但是当涉及到付款 删除文件 发布公开内容", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 720.16, "end": 723.44, "text": "修改正式代码或者处理敏感数据这些关键操作时", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 723.44, "end": 725.76, "text": "一个能够暂停下来向你解释情况", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 725.76, "end": 726.96, "text": "并等待你确认的agent", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 726.96, "end": 729.76, "text": "它的价值其实远远高于一个完全自主", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 729.76, "end": 731.16, "text": "从不请示的agent", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 731.16, "end": 734.24, "text": "一个成熟的智能体系统不是要取消人的判断", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 734.24, "end": 735.76, "text": "而是要把人的判断", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 735.76, "end": 738.16, "text": "集中到那些最关键的节点上", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 738.16, "end": 739.92, "text": "最后我们来总结一下今天的内容", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 739.92, "end": 740.96, "text": "DeepSake Harness", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 740.96, "end": 744.36, "text": "最适合作为构建自由agent系统的开放底座", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 744.36, "end": 745.56, "text": "腾讯的WorkBuddy", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 745.56, "end": 746.32, "text": "最省事", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 746.32, "end": 748.92, "text": "适合想直接进入日常办公场景的用户", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 748.92, "end": 752.56, "text": "Hermes Agent最像一个会积累记忆和技能的长期AI员工", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 752.56, "end": 755.8, "text": "而OpenAI的Codex则最擅长把复杂的专业工作", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 755.8, "end": 758.08, "text": "真正做完并进行严格的验证", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 758.08, "end": 761.48, "text": "所以下次你在选择的时候不要只问谁最强", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 761.48, "end": 763.44, "text": "而要问自己三个更具体的问题", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 763.44, "end": 765.32, "text": "第一 我需要的是一个成品工具", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 765.32, "end": 766.88, "text": "还是一个可以自己开发的框架", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 766.88, "end": 769.84, "text": "第二 我更看重的是即时的效率提升", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 769.84, "end": 771.52, "text": "还是长期的成长和陪伴", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 771.52, "end": 775.16, "text": "第三 我是否需要掌握对模型", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 775.16, "end": 778.36, "text": "数据 权限和运行环境的控制权", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 778.36, "end": 779.96, "text": "当你把这三个问题想清楚之后", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 779.96, "end": 781.96, "text": "这死者之间的选择就一点都不模糊了", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 781.96, "end": 783.16, "text": "他们不是简单的", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 783.16, "end": 784.64, "text": "谁替代谁的关系", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 784.64, "end": 786.72, "text": "而是分别服务于不同层级", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 786.72, "end": 787.6, "text": "不同需求的", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 787.6, "end": 789.12, "text": "未来更常见的情况", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 789.12, "end": 790.76, "text": "可能也不是某一个超级agent", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 790.76, "end": 791.72, "text": "包揽所有工作", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 791.72, "end": 793.8, "text": "而是像今天聊到的成品办公agent", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 793.8, "end": 794.76, "text": "专业工程agent", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 794.76, "end": 796.76, "text": "长期个人agent和企业资金底座", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 796.76, "end": 797.72, "text": "他们相互配合", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 797.72, "end": 799.72, "text": "共同组成一个真正好用的", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 }, { "start": 799.72, "end": 800.72, "text": "AI工作系统", "language": "zh", "chunk": 1, "local_language_run": 0, "language_run": 0 } ], "segmented_multilingual_transcription": { "status": "STABLE_SEGMENTED_MULTILINGUAL_OK", "created_at": "2026-08-19T07:50:51", "manifest": "/Users/sagawa/AI_WORK/video_notes/url/20260819_073449__四大AI_Agent实测_Codex_Hermes_WorkBuddy_DeepSeek谁更能干活/segmented_work/chunk_manifest.json", "audio": "/Users/sagawa/AI_WORK/video_notes/url/20260819_073449__四大AI_Agent实测_Codex_Hermes_WorkBuddy_DeepSeek谁更能干活/audio_16k_mono.wav", "audio_sha256": "226cd962fe3b8a5a4822f44c89677c6d5b9879bd7939f88db5d3018fac5c322d", "audio_duration": 800.832, "model": "/Users/sagawa/AI_WORK/cache/huggingface/hub/models--mlx-community--whisper-large-v2", "language_profile": "/Users/sagawa/AI_WORK/video_notes/url/20260819_073449__四大AI_Agent实测_Codex_Hermes_WorkBuddy_DeepSeek谁更能干活/multilingual_work/multilingual_language_profile.json", "dominant_languages": [ "zh" ], "suppressed_languages": [], "chunk_count": 2, "completed_chunks": 2, "resumed_chunks": 0, "merged_segment_count": 431, "detected_languages": [ "zh" ], "transcript_languages": [ "zh" ], "missing_languages": [], "unexpected_languages": [], "language_sequence": [ "zh" ], "speech_region_seconds": 800.832, "speech_covered_seconds": 798.34, "speech_coverage_ratio": 0.9968882362343163, "tail_repeat_run": 1, "tail_repeat_text": "2026年AI智能体", "chunk_needs_review": [], "qc_ok": true, "qc_errors": [], "complete_review_artifact": true } }