{
  "text": "谷歌内部又出现了\n一个叫carbon的模型\n有人用完之后把它的编程\n表现拿来跟Claudeopus 5.5比较\n但这轮竞争最值得追问的地方是\n模型越来越会写代码\n为什么企业离放心\n交出工作还是差那么远\ncarbon的消息加上微软刚推出的decision\n恰好把这个缺口摆到了台面上\n先把消息的分量放准\ncarbon目前受到关注\n主要来自business Insider对\n谷歌内部材料的报道\n员工正在内部编程\n环境Jetbrains里测试它\n有人给出积极评价\n也有人提醒还需要更多测试\n这属于接触过产品的人的使用反馈\n它能说明谷歌正在试验什么\n暂时还回答不了\n公开产品最终有多强\n报道涉及Argon Barium和carbon几个名字\n公开亮相的Argon与内部\n候选版本的选择有关\n这些代号容易制造一种错觉\n好像元素名称排到后面\n性能就自动升一级\n开发流程里却经常\n同时存在多个候选版本\n名字可以帮助团队区分实验\n不能拿来替代完整的测评\n其中最容易被传播的\n是员工把carbon的编程\n感觉与Opus5.5放在一起谈\n这个比较有新闻价值\n因为编程已经成为前沿\n模型竞争的重要场景\n但员工熟悉的仓库使用的工具以及\n愿意交给模型的任务都会影响体验\n一个团队觉得顺手\n距离跨行业稳定领先\n还需要经过公开检验\n谷歌并没有在这篇报道\n中披露carbon的参数规模\n外界也没有因此拿到\n可以复现的完整测试\n所以庞然大物适合\n表达市场对它的期待\n不能直接写成参数记录\n真正有用的问题是\n它能不能把工作做完\n而且做完以后\n别人敢不敢接手\n这个区别落到编程上\n很容易看出来\n让模型生成一个独立的小程序\n通常可以很快看到成果\n让它接手一套已经运行多年的系统\n情况就复杂了\n某个函数看起来多余\n可能正好处理一个\n老客户的特殊情况\n模型删掉它\n测试暂时照样通过\n客户第二天才会发现问题\n企业付钱买的是经过验证的修改\n代码只是修改过程\n留下的一部分结果\n模型需要找到相关模块\n理解接口约束\n再检查改动有没有影响别处\n某些任务还得等待构建完成\n读错误日志\n重新定位问题\n每个环节都可能让第一次\n生成时的漂亮答案失去价值因此\n内部工程师愿意继续使用一个模型\n往往比一张孤立的\n成绩表更值得关注\n他们已经知道哪些建议会浪费时间\n也知道哪些回答看似\n复杂却解决不了问题\n如果carbon能在日常\n任务里减少反复解释\n减少审查后的返工\n这种体验就可能转化成真实需求\n现在的报道提供了一个观察入口\n还没有给出最终结论\n接下来要看的是连续工作能力\n单次回答做对和一连串操作做对\n难度差别很大\n一个代理可能正确理解需求\n却选错文件\n也可能改对文件\n却漏跑相关测试\n最后汇报完成的时候\n他甚至可能没有\n注意到工具已经报错\n用户看到的一个结果背后\n其实经过了很多次判断\n用一个纯粹释义的\n算式就能理解风险\n假设10个步骤彼此独立\n每一步成功率都是95%\n10步全部成功的概率也只有大约6成\n这并非carbon的测试数据\n实际任务也未必满足独立假设\n这个例子只是提醒\n长任务会放大局部失误\n单步表现不错\n不能保证整体可靠\n真正成熟的系统必须允许纠错\n工具执行失败\n代理要发现失败\n发现之后还得知道应该换方法\n还是停下来询问\n更麻烦的情况是\n工具没有报错\n但结果不符合业务要求\n比如系统成功发出了一封邮件\n收件人却选错了\n技术执行成功\n业务已经出事\n这也是编程模型值得研究的原因\n软件开发能提供比较明确的反馈\n编译器会报错\n测试会失败\n运行结果可以检查\n模型有机会利用这些反馈修正操作\n相比完全凭主观判断的任务\n这种环境更适合测量长流程能力\n不过测试覆盖不到的业务\n规则仍然需要其他检查\n评估carbon时\n可以先观察它遇到失败的表现\n第一次没有解决问题\n会不会继续重复同一套方案\n面对相互矛盾的文件\n会不会主动查找依据\n已经花费很多时间之后\n会不会为了尽快结束而宣称成功\n这些行为比生成答案\n时的语气有多自信\n更接近企业真正担心的事\n内部测试还有一层价值\n谷歌的工程环境能够提供真实任务\n让模型遇到教科书上没有的麻烦\n但这种环境也会引入条件差异\n如果gensky提供了更好的代码检索\n或者自动准备好了上下文\n最终体验就包含工具的贡献\n模型和工程系统应该一起评估\n不能把所有提升都归到模型权重上\n外界经常把新模型的进步\n想成一次训练带来的跳跃\n实际产品可能同时改了\n提示方式和工具调用\n也可能换了检索方法\n调整了计算预算\n哪怕底层模型相同\n任务完成率也可能变化\n看到内部版本好用\n更稳妥的追问是哪些改进\n能够保留到正式服务里\n这就牵涉到成本\n测试时允许模型思考很久\n调用很多次工具\n容易换来更好的结果\n正式上线之后\n服务需要承受大量用户同时请求\n等待时间过长\n普通用户会放弃\n每个任务的计算费用太高\n企业即使喜欢效果\n也未必愿意大规模采购\n所以一次内部体验要成为商业突破\n需要经过几道现实关卡\n同样的任务\n在可接受的时间里能不能做完\n用户增加之后\n服务质量能不能维持\n团队升级工具之后\n旧流程会不会失效\n这些问题没有一项能靠\n感觉像顶级模型直接解决\n却会决定产品最后有没有人续费\n再看已经公开的Argon\n谷歌的官方介绍\n采用分阶段开放安排\n涉及受信任的网路防御使用者\n随后逐步扩展访问\n公开产品和内部候选\n版本之间存在发布距离\n部分用户能接触到某项能力\n并不代表所有账户都\n已经拥有相同条件\n更不代表carbon已经成为\n公开可购买的服务\n这种距离也会让截图变得危险\n测试界面里出现某个名称\n可能只是灰度实验\n显示较长的上下文\n也可能对应特定入口\n用户最需要确认的是\n正式文档中的可用范围\n以及实际账户能否调用\n单张界面截图无法证明永久额度\n更无法承诺服务上线的时间\n上下文长度尤其容易引发误判\n能放进去很多资料\n只说明输入容量扩大\n模型能否找到关键证据\n能否区分新版文件和旧版文件\n还要另行测试\n一个系统把所有历史记录都塞进去\n可能增加信息\n也可能增加冲突\n资料越多\n选择依据的能力越重要\n企业真正需要的是可追溯\n模型说某条规则要求这样处理\n工作人员能否查到具体出处\n规则刚刚更新\n系统会不会仍然引用旧版\n同一段资料被不同\n部门解释成不同意思\n代理会怎样处理\n这些细节没有发布会上那么好看\n却直接决定自动化能走多远\n围绕carbon\n还有一个更容易制造轰动的词\nRSI也就是递归自我改进\n相关讨论把模型变强与模型\n帮助改进下一代模型联系起来\n但目前提到的内部测试\n报道并没有提供足够证据\n让外界宣布谷歌已经实现\n了完整的自主递归升级\n这里必须把几种过程拆开\n第一种过程是模型修改\n自己刚刚生成的答案\n编译失败后补一个导入\n测试失败后调整函数\n这属于当前任务里的纠错\n第二种过程是\n研究人员让模型参与开发\n帮助写实验代码或者分析结果\n它能够节约研究时间\n但研究方向和验收\n标准仍然可能由人决定\n更强的说法是系统\n自主提出改进方案\n完成训练和评估\n再确认新系统确实更好\n并继续下一轮\n要支撑这种说法\n至少需要看到比较完整的过程\n模型改了什么\n投入多少资源\n在哪些没见过的任务上进步\n出现退步又怎样处理\n少了这些信息\n自我进化很容易只剩一个标签\n尤其要防止把熟悉\n考试当成能力增长\n模型反复接触同一批测试\n可能越来越擅长那批测试\n研究人员就算没有刻意作弊\n选择实验的习惯也可能\n让结果偏向既有指标\n检验新版本需要留出独立的任务\n还要观察其他能力有没有被牺牲\n局部得分提高\n不能自动推出全面提升\n研究循环也受现实资源限制\n提出一个训练方案很快\n完成训练需要计算\n某些实验必须等待结果\n失败以后还得重新安排\n涉及真实世界数据或者硬件时\n等待会更长\n模型帮助研究并不\n意味着这些资源限制消失\n递归过程有没有发生\n与它能不能无限\n加速是两个不同问题\n这并不减损carbon可能具有的价值\n即使它只是把研究人员\n处理复杂代码的时间缩短\n也已经值得关注\n一个工具每天节约\n少量高技能人员的时间\n积累起来就可能改变开发节奏\n这样的进步可以测量\n也可以讨论商业意义\n无需提前附送智能\n爆炸已经开始的结论\n谷歌还有一个独特条件\n它既研发模型\n也经营开发工具和办公产品\n模型如果能够稳定处理工作\n有机会通过这些入口触达用户\n不过入口优势不会\n自动变成使用习惯\n企业已经建立的\n审批流程和权限规则\n往往比换一个聊天窗口难改的多\n想象一个普通的企业需求\n要求代理汇总某个项目的进展\n资料分散在邮件和文档中\n部分内容只有特定成员能看\n系统要先确认提问者有权访问什么\n再判断哪些资料仍然有效\n最终输出还得说明依据\n回答文字写的顺\n不代表整个流程已经合格\n如果任务升级成直接更新项目状态\n要求又高了一层\n更新哪条记录\n是否会通知客户\n谁有权批准\n都需要被明确设置\n能力增强会扩大可执行的动作\n也会扩大错误的影响\n模型越强\n企业对边界的要求往往越具体\n不能靠一句更智能略过去\n这里就轮到微软decision一登场了\n它关注的任务更窄\n给定问题和备选项\n输出各个选项的概率\n微软把它定位在\n决策与分类工作负载\n基于千问3.59B做后训练\n它选择处理大量重复的判断\n不要求所有问题都交给最大模型\n微软公布的测试覆盖36个基准\n约15万道问题\n并报告了相对GPT6索尔的延迟优势\n这些是厂商提供的评估试用条件\n需要保留\n官方也介绍了对输入脑洞的测试\n他们值得关注的地方是把\n一致性和速度当成产品指标\n不能据此断言任何\n业务场景都能照搬结果\n这种模型能放在哪里\n客服系统收到一条请求\n可以判断应当交给哪个业务组\n代理生成一份答案之后\n也可以根据预设标准分类检查\n是否需要人工复核\n可以被设计成一个明确选项\n这些都是工作流的应用释义\n不等于相关企业已经\n部署了同样的流程\n最有意思的是\n这类系统未必需要写出一大段解释\n业务有时只需要一个\n格式稳定的选择结果\n输出越容易被程序读取\n后面的动作就越容易安排\n大模型擅长展开思路\n专用判断模型则可以把\n计算集中在有限选择上\n两种能力有机会组合起来\n服务同一个任务\n但概率必须谨慎解释\n某个选项显示90%不等于当前\n答案拿到了真实性担保\n校准讨论的是一批预测\n与实际结果之间的关系\n企业需要在自己的数据上\n检查这种关系是否成立\n业务变化之后\n原来可靠的概率\n也可能需要重新评估比如\n一个系统在普通咨询里表现稳定\n遇到新的诈骗话术\n判断可能失准\n输入很像熟悉的类别\n实际风险却不同\n如果流程把高分直接当成放行许可\n评分错误就会沿着自动化继续传播\n可靠的设计需要给陌生情况留出口\n包括拒绝判断或者交给人处理\n这也解释了为什么判断\n模型的选项设置很重要\n如果只给执行和不执行\n一些模糊情况会被硬塞进去\n增加信息不足可能更符合业务\n但新增选项之后还要\n测试它是不是过度回避\n企业需要按照错误代价设计流程\n不能只追求一个漂亮的平均准确率\n不同错误的代价可以相差很大\n把普通咨询错分到另一个客服组\n可能只是多等几分钟\n把需要人工审批的支付\n操作判断成可以自动执行\n后果就严重得多\n模型在两类任务上\n得到相似的正确率\n也不代表承担了相似的风险\n评估必须看错在哪里\n微软官方给decision\n一列出的输入价格很低\n输出不另行收费\n并区分了当前可用入口和后续计划\n这个定价值得讨论\n但它只是模型调用的一部分成本\n系统还需要连接业务数据\n维护规则\n保存日志\n出了问题\n团队还得有人处理\n低价分类可以节约预算\n不能让整个工程免费\n延迟同样如此\n一个判断步骤明显变快\n如果整条流程大部分\n时间都在等待数据库\n用户感受到的改善就有限\n业务负责人需要测端到端时间\n每个模型单独跑得快\n系统之间来回传输很慢\n最终体验还是可能卡住\n局部性能只有放进\n流程里才知道实际价值\n把carbon和decision e放到一起\n竞争路线就更清楚了\ncarbon受到关注\n因为它可能承担更\n复杂的生成和编程工作\ndecision e则试图把高频\n选择变得便宜而稳定\n一个工作流完全可以\n同时需要这两类能力\n再让人保留重要审批\n企业采购的对象正在\n变成一整套任务完成方案\n模型之间怎样分工\n也会带来新的问题\n分类系统把任务分错\n大模型可能认真解决了错误的问题\n大模型输出的信息不完整\n后续评分也可能建立在缺失材料上\n多个组件串联之后\n责任必须能追溯\n系统应该记录每个阶段看到了什么\n做出了什么选择\n这类记录还有一个实际用途\n帮助团队判断应该升级哪一部分\n失败发生在检索\n就不该盲目换更贵的模型\n规则写的模糊\n就应该先修规则\n只有确实属于能力不足的任务\n升级模型才可能解决\n没有分阶段记录\n所有问题都会被归结\n成一句AI还不够强\n企业可以先选一个范围有限的任务\n保留原流程作为对照\n评价标准提前确定\n既记录完成时间\n也记录返工和错误\n这样做看起来没有\n全面上线那么热闹\n却更容易判断投入有没有回报\n任务边界清楚\n失败也更容易回滚\n团队能够积累真正有用的经验\n对照测试还要防止选择偏差\n员工往往把容易的任务交给AI\n把复杂任务留给自己\n最后看使用记录\nAI完成率很高\n人类耗时很长\n比较就已经不公平\n不同任务的难度必须被考虑\n否则一套未经校正的报表就\n可能推动错误的用工决定\n模型更新后也要重测\n企业可能喜欢新版本的总体表现\n却发现某个关键分类变差了\n如果自动切换版本\n没有复核就继续运行\n原先建立的流程可能悄悄失效\n稳定的模型服务需要版本管理\n重要任务还要允许暂时保留旧版本\n更新速度和部署速度可以不同\n成本评估也应覆盖失败的任务\n只算成功那一次的调用费用\n会漏掉之前的重试\n只算生成速度\n会漏掉人工检查\n团队需要知道\n完成一个合格任务总共花了多少钱\n发生严重错误的频率是多少\n这两个数字\n比每秒生成多少\n文字更接近商业决策\n这也是卡本后续最值得观察的方向\n公开测试如果出现\n应该看任务是否覆盖真实仓库\n测试条件是否一致\n失败记录是否完整\n产品如果上线\n应该看普通付费用户\n能够获得什么条件\n内部员工的惊喜很有吸引力\n可持续的服务体验\n才会决定市场位置\n还有一个会影响采购的因素\n任务到底由谁定义完成\n编程测试通常有预设目标\n企业请求却可能很模糊\n客户说希望页面更快\n可能指加载时间\n也可能指操作步骤太多\n模型直接优化代码\n未必解决客户最在意的问题\n复杂任务需要先确认目标\n随后才适合自动执行\n需求澄清也可以被评估\n系统能否识别缺失条件\n能否只问必要的问题\n都会影响合作体验\n每一步都要求人确认\n效率上不去\n什么都不问\n又容易走偏\n企业需要寻找合适的询问频率\n并按任务风险设置不同阈值\n这个问题通常比聊天\n回答是否自然更难处理\n长任务还会跨越时间\n代理今天拿到的资料\n明天可能被更新\n用户中途改变要求\n他得知道哪些工作可以保留\n哪些已经作废\n如果系统继续沿着旧目标执行\n能力再强也只是更快的做错事\n保存工作状态的同时\n还要允许目标发生变化\n这是长期运行的实际要求\n任务中断之后能否\n恢复也值得单独测试\n网路暂时断开\n已经完成的步骤是否会重复\n一项操作已经提交\n但确认消息没有返回\n代理会不会再提交一次\n这些情况在演示中很少\n出现在真实服务里却躲不开\n恢复机制做不好\n自动化就可能制造\n重复订单或者重复修改\n在这些问题上\n模型不一定需要独自解决所有细节\n应用可以记录操作状态\n阻止重复提交\n再让模型决定下一步\n确定性的程序负责确定性的约束\n模型处理需要理解和选择的部分\n企业能够把两者结合的多好\n会直接影响carbon这类\n强模型能发挥多少价值\n数据保密则决定哪些\n任务根本不能随意交出去\n代码仓库里可能包含商业逻辑\n文档里可能有客户信息\n企业选择模型需要查清\n数据保存方式和服务条款\n模型路由经过不同供应商时\n也要确认资料流向\n使用效果相似的两个服务\n合规条件可能完全不同\n由此还会出现迁移成本\n一个团队围绕某个模型\n调好了提示和评分规则\n换模型之后未必照样有效\n输出格式稍有变化\n下游程序就可能出错\n企业可以保留共同的任务接口\n再为不同模型做适配\n这样以后选择carbon或者其他候选\n服务才不至于重新搭建整个流程\n比较时也应该避免只\n挑自己擅长的任务\n一家模型擅长从零搭建页面\n另一家擅长修复旧项目\n两者都可以展示出经验结果\n采购方需要按照真实\n任务占比构建测试集\n很少遇到的高难问题可以单独关注\n不能让它完全代表日常使用成本\n还有用户经常忽略的等待分布\n平均耗时不错\n偶尔卡住很久\n同样会破坏工作安排\n客服和实时业务尤其需要看慢请求\n研究型任务能够容忍更长等待\n标准又不同\n用一个平均延迟比较所有产品\n很容易选出对业务并不合适的模型\ndecision一这类模型的机会\n就来自任务需求的差别\n高频而简单的分类可以单独处理\n把昂贵计算留给复杂部分\n但路由系统自身也需要观察\n不能为了省一笔调用费\n让后面产生更多返工\n模型组合的价值应当\n体现为合格交付成本下降\n不能只体现在某个组件便宜\n对个人开发者同样\n可以用较小的方式验证\n选一批已经知道解决方法的旧问题\n隐藏答案\n让新工具独立尝试\n比较它使用的时间和修改范围\n再检查有没有新增问题\n这样的测试无法替代专业基准\n却能帮助判断新工具\n是否适合手上的项目\n减少仅凭宣传切换工具的成本\n网络安全相关能力\n还需要额外的边界\n模型能够发现复杂软件问题\n对防御工作可能很有帮助\n但使用者身份和\n操作权限必须被检查\n授权研究与未经许可的操作不能\n因为同样调用一个模型就混在一起\n安全发布的难点也包括让\n有益能力进入正确的工作流程\n把所有能力封成一个聊天窗口\n用户可能很难察觉这些差别\n企业却必须知道哪些动作只是建议\n哪些动作真的改变系统\n一次代码审查可以人工确认\n一次生产配置修改则\n需要更严格的审批\n产品越接近实际执行\n界面上越应该清楚的展示动作范围\n未来的竞争也就不会\n只停留在模型排行榜\n模型厂商需要证明复杂任务的能力\n工具平台需要证明连接可靠\n企业则要证明\n自动化确实节约总成本\n任何一环跟不上\n都可能让看起来领先\n的技术停在演示阶段\n这个过程没有统一的捷径\n只能靠连续的部署和复合\ncarbon目前最重要的信号是谷歌内部\n的编程候选模型正在获得积极反馈\n这个信号值得追踪\n却不足以写成全面领先的事实\n微软decision亦提供了\n另一个观察角度\nAI产品还在向具体判断任务细分\n规模和专用化都可能带来价值\n最终要接受同一套业务验收\n开头的问题现在可以\n落到一个明确标准上\n企业离放心交出工作还有距离\n因为交付涉及连续判断\n而且错误有真实代价\ncarbon如果确实是个强大的新模型\n最有说服力的证明就是帮助用户\n在可承受的成本下完成合格任务\n能被复核\n能被回滚\n下一次还能稳定做对\n这样的进步才会留在账上",
  "language": "zh",
  "duration": 1510.0,
  "segments": [
    {
      "id": 0,
      "start": 0.2,
      "end": 1.533,
      "text": "谷歌内部又出现了",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 1,
      "start": 1.533,
      "end": 3.166,
      "text": "一个叫carbon的模型",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 2,
      "start": 3.3,
      "end": 5.7,
      "text": "有人用完之后把它的编程",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 3,
      "start": 5.7,
      "end": 9.333,
      "text": "表现拿来跟Claudeopus 5.5比较",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 4,
      "start": 9.7,
      "end": 12.3,
      "text": "但这轮竞争最值得追问的地方是",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 5,
      "start": 12.4,
      "end": 14.166,
      "text": "模型越来越会写代码",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 6,
      "start": 14.3,
      "end": 15.766,
      "text": "为什么企业离放心",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 7,
      "start": 15.766,
      "end": 18.0,
      "text": "交出工作还是差那么远",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 8,
      "start": 18.266,
      "end": 21.566,
      "text": "carbon的消息加上微软刚推出的decision",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 9,
      "start": 22.2,
      "end": 24.733,
      "text": "恰好把这个缺口摆到了台面上",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 10,
      "start": 24.8,
      "end": 26.8,
      "text": "先把消息的分量放准",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 11,
      "start": 27.466,
      "end": 29.166,
      "text": "carbon目前受到关注",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 12,
      "start": 29.5,
      "end": 31.533,
      "text": "主要来自business Insider对",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 13,
      "start": 31.533,
      "end": 33.2,
      "text": "谷歌内部材料的报道",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 14,
      "start": 33.333,
      "end": 34.733,
      "text": "员工正在内部编程",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 15,
      "start": 34.733,
      "end": 36.6,
      "text": "环境Jetbrains里测试它",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 16,
      "start": 36.8,
      "end": 38.466,
      "text": "有人给出积极评价",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 17,
      "start": 38.466,
      "end": 40.8,
      "text": "也有人提醒还需要更多测试",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 18,
      "start": 40.933,
      "end": 44.166,
      "text": "这属于接触过产品的人的使用反馈",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 19,
      "start": 44.7,
      "end": 46.8,
      "text": "它能说明谷歌正在试验什么",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 20,
      "start": 46.9,
      "end": 48.1,
      "text": "暂时还回答不了",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 21,
      "start": 48.1,
      "end": 50.0,
      "text": "公开产品最终有多强",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 22,
      "start": 50.133,
      "end": 53.6,
      "text": "报道涉及Argon Barium和carbon几个名字",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 23,
      "start": 54.166,
      "end": 55.966,
      "text": "公开亮相的Argon与内部",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 24,
      "start": 55.966,
      "end": 57.866,
      "text": "候选版本的选择有关",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 25,
      "start": 58.166,
      "end": 60.533,
      "text": "这些代号容易制造一种错觉",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 26,
      "start": 60.7,
      "end": 62.9,
      "text": "好像元素名称排到后面",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 27,
      "start": 63.0,
      "end": 64.7,
      "text": "性能就自动升一级",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 28,
      "start": 65.3,
      "end": 66.8,
      "text": "开发流程里却经常",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 29,
      "start": 66.8,
      "end": 68.866,
      "text": "同时存在多个候选版本",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 30,
      "start": 69.0,
      "end": 71.666,
      "text": "名字可以帮助团队区分实验",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 31,
      "start": 71.7,
      "end": 73.933,
      "text": "不能拿来替代完整的测评",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 32,
      "start": 74.166,
      "end": 76.066,
      "text": "其中最容易被传播的",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 33,
      "start": 76.066,
      "end": 77.933,
      "text": "是员工把carbon的编程",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 34,
      "start": 77.933,
      "end": 80.533,
      "text": "感觉与Opus5.5放在一起谈",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 35,
      "start": 81.133,
      "end": 82.766,
      "text": "这个比较有新闻价值",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 36,
      "start": 82.766,
      "end": 84.733,
      "text": "因为编程已经成为前沿",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 37,
      "start": 84.733,
      "end": 86.733,
      "text": "模型竞争的重要场景",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 38,
      "start": 86.766,
      "end": 90.5,
      "text": "但员工熟悉的仓库使用的工具以及",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 39,
      "start": 90.5,
      "end": 93.733,
      "text": "愿意交给模型的任务都会影响体验",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 40,
      "start": 94.1,
      "end": 95.566,
      "text": "一个团队觉得顺手",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 41,
      "start": 95.666,
      "end": 97.666,
      "text": "距离跨行业稳定领先",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 42,
      "start": 97.733,
      "end": 99.666,
      "text": "还需要经过公开检验",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 43,
      "start": 99.866,
      "end": 101.366,
      "text": "谷歌并没有在这篇报道",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 44,
      "start": 101.366,
      "end": 103.5,
      "text": "中披露carbon的参数规模",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 45,
      "start": 104.1,
      "end": 105.466,
      "text": "外界也没有因此拿到",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 46,
      "start": 105.466,
      "end": 107.133,
      "text": "可以复现的完整测试",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 47,
      "start": 107.366,
      "end": 108.9,
      "text": "所以庞然大物适合",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 48,
      "start": 108.9,
      "end": 110.6,
      "text": "表达市场对它的期待",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 49,
      "start": 110.666,
      "end": 112.733,
      "text": "不能直接写成参数记录",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 50,
      "start": 112.933,
      "end": 114.866,
      "text": "真正有用的问题是",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 51,
      "start": 114.866,
      "end": 116.6,
      "text": "它能不能把工作做完",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 52,
      "start": 116.733,
      "end": 118.0,
      "text": "而且做完以后",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 53,
      "start": 118.0,
      "end": 119.466,
      "text": "别人敢不敢接手",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 54,
      "start": 119.533,
      "end": 121.533,
      "text": "这个区别落到编程上",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 55,
      "start": 121.566,
      "end": 122.8,
      "text": "很容易看出来",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 56,
      "start": 123.4,
      "end": 125.9,
      "text": "让模型生成一个独立的小程序",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 57,
      "start": 126.066,
      "end": 128.066,
      "text": "通常可以很快看到成果",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 58,
      "start": 128.333,
      "end": 131.466,
      "text": "让它接手一套已经运行多年的系统",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 59,
      "start": 131.533,
      "end": 133.0,
      "text": "情况就复杂了",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 60,
      "start": 133.0,
      "end": 134.6,
      "text": "某个函数看起来多余",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 61,
      "start": 135.0,
      "end": 136.1,
      "text": "可能正好处理一个",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 62,
      "start": 136.1,
      "end": 137.733,
      "text": "老客户的特殊情况",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 63,
      "start": 138.0,
      "end": 139.2,
      "text": "模型删掉它",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 64,
      "start": 139.466,
      "end": 141.266,
      "text": "测试暂时照样通过",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 65,
      "start": 141.4,
      "end": 143.366,
      "text": "客户第二天才会发现问题",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 66,
      "start": 143.5,
      "end": 146.966,
      "text": "企业付钱买的是经过验证的修改",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 67,
      "start": 147.666,
      "end": 149.166,
      "text": "代码只是修改过程",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 68,
      "start": 149.166,
      "end": 150.7,
      "text": "留下的一部分结果",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 69,
      "start": 150.966,
      "end": 152.966,
      "text": "模型需要找到相关模块",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 70,
      "start": 153.266,
      "end": 154.666,
      "text": "理解接口约束",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 71,
      "start": 154.966,
      "end": 157.533,
      "text": "再检查改动有没有影响别处",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 72,
      "start": 157.766,
      "end": 160.1,
      "text": "某些任务还得等待构建完成",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 73,
      "start": 160.666,
      "end": 161.7,
      "text": "读错误日志",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 74,
      "start": 161.866,
      "end": 163.2,
      "text": "重新定位问题",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 75,
      "start": 163.366,
      "end": 165.2,
      "text": "每个环节都可能让第一次",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 76,
      "start": 165.2,
      "end": 168.466,
      "text": "生成时的漂亮答案失去价值因此",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 77,
      "start": 168.466,
      "end": 171.166,
      "text": "内部工程师愿意继续使用一个模型",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 78,
      "start": 171.4,
      "end": 172.7,
      "text": "往往比一张孤立的",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 79,
      "start": 172.7,
      "end": 174.266,
      "text": "成绩表更值得关注",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 80,
      "start": 174.7,
      "end": 177.166,
      "text": "他们已经知道哪些建议会浪费时间",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 81,
      "start": 177.333,
      "end": 179.0,
      "text": "也知道哪些回答看似",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 82,
      "start": 179.0,
      "end": 180.733,
      "text": "复杂却解决不了问题",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 83,
      "start": 180.866,
      "end": 182.266,
      "text": "如果carbon能在日常",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 84,
      "start": 182.266,
      "end": 184.0,
      "text": "任务里减少反复解释",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 85,
      "start": 184.1,
      "end": 185.6,
      "text": "减少审查后的返工",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 86,
      "start": 185.766,
      "end": 188.766,
      "text": "这种体验就可能转化成真实需求",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 87,
      "start": 189.666,
      "end": 192.1,
      "text": "现在的报道提供了一个观察入口",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 88,
      "start": 192.166,
      "end": 193.9,
      "text": "还没有给出最终结论",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 89,
      "start": 194.1,
      "end": 196.7,
      "text": "接下来要看的是连续工作能力",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 90,
      "start": 196.766,
      "end": 199.966,
      "text": "单次回答做对和一连串操作做对",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 91,
      "start": 199.966,
      "end": 201.333,
      "text": "难度差别很大",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 92,
      "start": 201.6,
      "end": 203.733,
      "text": "一个代理可能正确理解需求",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 93,
      "start": 203.9,
      "end": 205.166,
      "text": "却选错文件",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 94,
      "start": 205.333,
      "end": 206.9,
      "text": "也可能改对文件",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 95,
      "start": 206.933,
      "end": 208.566,
      "text": "却漏跑相关测试",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 96,
      "start": 208.733,
      "end": 210.4,
      "text": "最后汇报完成的时候",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 97,
      "start": 210.466,
      "end": 211.366,
      "text": "他甚至可能没有",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 98,
      "start": 211.366,
      "end": 213.066,
      "text": "注意到工具已经报错",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 99,
      "start": 213.5,
      "end": 215.6,
      "text": "用户看到的一个结果背后",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 100,
      "start": 215.6,
      "end": 217.566,
      "text": "其实经过了很多次判断",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 101,
      "start": 217.7,
      "end": 219.0,
      "text": "用一个纯粹释义的",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 102,
      "start": 219.0,
      "end": 220.733,
      "text": "算式就能理解风险",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 103,
      "start": 221.0,
      "end": 223.1,
      "text": "假设10个步骤彼此独立",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 104,
      "start": 223.4,
      "end": 225.766,
      "text": "每一步成功率都是95%",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 105,
      "start": 225.9,
      "end": 228.966,
      "text": "10步全部成功的概率也只有大约6成",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 106,
      "start": 229.166,
      "end": 231.333,
      "text": "这并非carbon的测试数据",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 107,
      "start": 231.6,
      "end": 234.2,
      "text": "实际任务也未必满足独立假设",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 108,
      "start": 234.366,
      "end": 235.9,
      "text": "这个例子只是提醒",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 109,
      "start": 236.266,
      "end": 238.366,
      "text": "长任务会放大局部失误",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 110,
      "start": 238.9,
      "end": 240.1,
      "text": "单步表现不错",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 111,
      "start": 240.166,
      "end": 241.933,
      "text": "不能保证整体可靠",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 112,
      "start": 242.1,
      "end": 244.933,
      "text": "真正成熟的系统必须允许纠错",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 113,
      "start": 245.2,
      "end": 246.666,
      "text": "工具执行失败",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 114,
      "start": 246.866,
      "end": 248.366,
      "text": "代理要发现失败",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 115,
      "start": 248.766,
      "end": 251.666,
      "text": "发现之后还得知道应该换方法",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 116,
      "start": 251.733,
      "end": 253.366,
      "text": "还是停下来询问",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 117,
      "start": 253.466,
      "end": 254.933,
      "text": "更麻烦的情况是",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 118,
      "start": 254.933,
      "end": 256.166,
      "text": "工具没有报错",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 119,
      "start": 256.166,
      "end": 258.166,
      "text": "但结果不符合业务要求",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 120,
      "start": 258.333,
      "end": 261.133,
      "text": "比如系统成功发出了一封邮件",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 121,
      "start": 261.566,
      "end": 263.133,
      "text": "收件人却选错了",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 122,
      "start": 263.3,
      "end": 264.733,
      "text": "技术执行成功",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 123,
      "start": 264.8,
      "end": 266.1,
      "text": "业务已经出事",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 124,
      "start": 266.266,
      "end": 269.166,
      "text": "这也是编程模型值得研究的原因",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 125,
      "start": 269.866,
      "end": 272.5,
      "text": "软件开发能提供比较明确的反馈",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 126,
      "start": 273.1,
      "end": 274.6,
      "text": "编译器会报错",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 127,
      "start": 274.7,
      "end": 275.966,
      "text": "测试会失败",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 128,
      "start": 276.2,
      "end": 277.933,
      "text": "运行结果可以检查",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 129,
      "start": 278.533,
      "end": 281.6,
      "text": "模型有机会利用这些反馈修正操作",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 130,
      "start": 281.733,
      "end": 284.133,
      "text": "相比完全凭主观判断的任务",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 131,
      "start": 284.3,
      "end": 287.266,
      "text": "这种环境更适合测量长流程能力",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 132,
      "start": 287.933,
      "end": 289.8,
      "text": "不过测试覆盖不到的业务",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 133,
      "start": 289.8,
      "end": 292.2,
      "text": "规则仍然需要其他检查",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 134,
      "start": 292.366,
      "end": 293.766,
      "text": "评估carbon时",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 135,
      "start": 293.766,
      "end": 296.7,
      "text": "可以先观察它遇到失败的表现",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 136,
      "start": 296.766,
      "end": 298.4,
      "text": "第一次没有解决问题",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 137,
      "start": 298.6,
      "end": 300.9,
      "text": "会不会继续重复同一套方案",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 138,
      "start": 301.066,
      "end": 302.9,
      "text": "面对相互矛盾的文件",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 139,
      "start": 303.166,
      "end": 304.966,
      "text": "会不会主动查找依据",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 140,
      "start": 305.333,
      "end": 307.466,
      "text": "已经花费很多时间之后",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 141,
      "start": 307.9,
      "end": 310.566,
      "text": "会不会为了尽快结束而宣称成功",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 142,
      "start": 310.7,
      "end": 312.533,
      "text": "这些行为比生成答案",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 143,
      "start": 312.533,
      "end": 314.3,
      "text": "时的语气有多自信",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 144,
      "start": 314.3,
      "end": 316.733,
      "text": "更接近企业真正担心的事",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 145,
      "start": 316.866,
      "end": 318.8,
      "text": "内部测试还有一层价值",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 146,
      "start": 319.366,
      "end": 322.166,
      "text": "谷歌的工程环境能够提供真实任务",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 147,
      "start": 322.466,
      "end": 325.166,
      "text": "让模型遇到教科书上没有的麻烦",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 148,
      "start": 325.166,
      "end": 327.766,
      "text": "但这种环境也会引入条件差异",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 149,
      "start": 327.9,
      "end": 330.6,
      "text": "如果gensky提供了更好的代码检索",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 150,
      "start": 330.7,
      "end": 332.866,
      "text": "或者自动准备好了上下文",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 151,
      "start": 332.933,
      "end": 335.2,
      "text": "最终体验就包含工具的贡献",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 152,
      "start": 335.733,
      "end": 338.266,
      "text": "模型和工程系统应该一起评估",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 153,
      "start": 338.333,
      "end": 341.533,
      "text": "不能把所有提升都归到模型权重上",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 154,
      "start": 341.6,
      "end": 343.766,
      "text": "外界经常把新模型的进步",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 155,
      "start": 343.766,
      "end": 345.933,
      "text": "想成一次训练带来的跳跃",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 156,
      "start": 346.3,
      "end": 348.2,
      "text": "实际产品可能同时改了",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 157,
      "start": 348.2,
      "end": 350.166,
      "text": "提示方式和工具调用",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 158,
      "start": 350.566,
      "end": 352.366,
      "text": "也可能换了检索方法",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 159,
      "start": 352.533,
      "end": 354.1,
      "text": "调整了计算预算",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 160,
      "start": 354.3,
      "end": 355.966,
      "text": "哪怕底层模型相同",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 161,
      "start": 355.966,
      "end": 357.866,
      "text": "任务完成率也可能变化",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 162,
      "start": 357.966,
      "end": 359.9,
      "text": "看到内部版本好用",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 163,
      "start": 360.0,
      "end": 362.4,
      "text": "更稳妥的追问是哪些改进",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 164,
      "start": 362.4,
      "end": 364.333,
      "text": "能够保留到正式服务里",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 165,
      "start": 364.466,
      "end": 365.8,
      "text": "这就牵涉到成本",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 166,
      "start": 366.0,
      "end": 368.4,
      "text": "测试时允许模型思考很久",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 167,
      "start": 368.5,
      "end": 369.966,
      "text": "调用很多次工具",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 168,
      "start": 370.066,
      "end": 371.9,
      "text": "容易换来更好的结果",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 169,
      "start": 372.1,
      "end": 373.4,
      "text": "正式上线之后",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 170,
      "start": 373.4,
      "end": 376.533,
      "text": "服务需要承受大量用户同时请求",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 171,
      "start": 376.733,
      "end": 378.133,
      "text": "等待时间过长",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 172,
      "start": 378.266,
      "end": 379.733,
      "text": "普通用户会放弃",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 173,
      "start": 379.866,
      "end": 381.9,
      "text": "每个任务的计算费用太高",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 174,
      "start": 382.366,
      "end": 383.933,
      "text": "企业即使喜欢效果",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 175,
      "start": 383.933,
      "end": 385.966,
      "text": "也未必愿意大规模采购",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 176,
      "start": 386.0,
      "end": 389.266,
      "text": "所以一次内部体验要成为商业突破",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 177,
      "start": 389.4,
      "end": 391.266,
      "text": "需要经过几道现实关卡",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 178,
      "start": 391.466,
      "end": 392.4,
      "text": "同样的任务",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 179,
      "start": 392.466,
      "end": 395.0,
      "text": "在可接受的时间里能不能做完",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 180,
      "start": 395.333,
      "end": 396.566,
      "text": "用户增加之后",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 181,
      "start": 396.7,
      "end": 398.333,
      "text": "服务质量能不能维持",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 182,
      "start": 398.566,
      "end": 400.333,
      "text": "团队升级工具之后",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 183,
      "start": 400.366,
      "end": 402.1,
      "text": "旧流程会不会失效",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 184,
      "start": 402.166,
      "end": 404.366,
      "text": "这些问题没有一项能靠",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 185,
      "start": 404.366,
      "end": 406.933,
      "text": "感觉像顶级模型直接解决",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 186,
      "start": 407.2,
      "end": 410.0,
      "text": "却会决定产品最后有没有人续费",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 187,
      "start": 410.1,
      "end": 412.1,
      "text": "再看已经公开的Argon",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 188,
      "start": 412.666,
      "end": 414.1,
      "text": "谷歌的官方介绍",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 189,
      "start": 414.1,
      "end": 416.166,
      "text": "采用分阶段开放安排",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 190,
      "start": 416.266,
      "end": 419.066,
      "text": "涉及受信任的网路防御使用者",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 191,
      "start": 419.333,
      "end": 421.266,
      "text": "随后逐步扩展访问",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 192,
      "start": 421.566,
      "end": 423.3,
      "text": "公开产品和内部候选",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 193,
      "start": 423.3,
      "end": 425.566,
      "text": "版本之间存在发布距离",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 194,
      "start": 425.733,
      "end": 428.066,
      "text": "部分用户能接触到某项能力",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 195,
      "start": 428.2,
      "end": 429.933,
      "text": "并不代表所有账户都",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 196,
      "start": 429.933,
      "end": 431.566,
      "text": "已经拥有相同条件",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 197,
      "start": 431.9,
      "end": 433.733,
      "text": "更不代表carbon已经成为",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 198,
      "start": 433.733,
      "end": 435.466,
      "text": "公开可购买的服务",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 199,
      "start": 435.466,
      "end": 438.066,
      "text": "这种距离也会让截图变得危险",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 200,
      "start": 438.6,
      "end": 440.766,
      "text": "测试界面里出现某个名称",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 201,
      "start": 440.8,
      "end": 442.466,
      "text": "可能只是灰度实验",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 202,
      "start": 442.733,
      "end": 444.733,
      "text": "显示较长的上下文",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 203,
      "start": 444.733,
      "end": 446.7,
      "text": "也可能对应特定入口",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 204,
      "start": 446.866,
      "end": 449.066,
      "text": "用户最需要确认的是",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 205,
      "start": 449.066,
      "end": 451.1,
      "text": "正式文档中的可用范围",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 206,
      "start": 451.9,
      "end": 454.2,
      "text": "以及实际账户能否调用",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 207,
      "start": 454.3,
      "end": 457.266,
      "text": "单张界面截图无法证明永久额度",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 208,
      "start": 457.333,
      "end": 459.6,
      "text": "更无法承诺服务上线的时间",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 209,
      "start": 459.733,
      "end": 462.3,
      "text": "上下文长度尤其容易引发误判",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 210,
      "start": 462.666,
      "end": 464.366,
      "text": "能放进去很多资料",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 211,
      "start": 464.5,
      "end": 466.6,
      "text": "只说明输入容量扩大",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 212,
      "start": 467.266,
      "end": 469.2,
      "text": "模型能否找到关键证据",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 213,
      "start": 469.4,
      "end": 472.133,
      "text": "能否区分新版文件和旧版文件",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 214,
      "start": 472.266,
      "end": 473.6,
      "text": "还要另行测试",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 215,
      "start": 473.733,
      "end": 476.533,
      "text": "一个系统把所有历史记录都塞进去",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 216,
      "start": 476.666,
      "end": 477.966,
      "text": "可能增加信息",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 217,
      "start": 478.0,
      "end": 479.4,
      "text": "也可能增加冲突",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 218,
      "start": 480.066,
      "end": 481.0,
      "text": "资料越多",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 219,
      "start": 481.0,
      "end": 482.933,
      "text": "选择依据的能力越重要",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 220,
      "start": 483.133,
      "end": 485.133,
      "text": "企业真正需要的是可追溯",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 221,
      "start": 485.4,
      "end": 487.8,
      "text": "模型说某条规则要求这样处理",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 222,
      "start": 487.933,
      "end": 490.066,
      "text": "工作人员能否查到具体出处",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 223,
      "start": 490.3,
      "end": 491.666,
      "text": "规则刚刚更新",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 224,
      "start": 491.966,
      "end": 494.1,
      "text": "系统会不会仍然引用旧版",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 225,
      "start": 494.3,
      "end": 495.466,
      "text": "同一段资料被不同",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 226,
      "start": 495.466,
      "end": 497.2,
      "text": "部门解释成不同意思",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 227,
      "start": 497.533,
      "end": 499.0,
      "text": "代理会怎样处理",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 228,
      "start": 499.1,
      "end": 501.666,
      "text": "这些细节没有发布会上那么好看",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 229,
      "start": 501.933,
      "end": 504.533,
      "text": "却直接决定自动化能走多远",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 230,
      "start": 504.7,
      "end": 505.933,
      "text": "围绕carbon",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 231,
      "start": 505.933,
      "end": 508.3,
      "text": "还有一个更容易制造轰动的词",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 232,
      "start": 508.933,
      "end": 511.666,
      "text": "RSI也就是递归自我改进",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 233,
      "start": 511.933,
      "end": 514.8,
      "text": "相关讨论把模型变强与模型",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 234,
      "start": 514.8,
      "end": 517.333,
      "text": "帮助改进下一代模型联系起来",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 235,
      "start": 517.466,
      "end": 519.166,
      "text": "但目前提到的内部测试",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 236,
      "start": 519.166,
      "end": 521.266,
      "text": "报道并没有提供足够证据",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 237,
      "start": 521.7,
      "end": 523.8,
      "text": "让外界宣布谷歌已经实现",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 238,
      "start": 523.8,
      "end": 525.766,
      "text": "了完整的自主递归升级",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 239,
      "start": 526.066,
      "end": 528.3,
      "text": "这里必须把几种过程拆开",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 240,
      "start": 528.4,
      "end": 530.666,
      "text": "第一种过程是模型修改",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 241,
      "start": 530.666,
      "end": 532.3,
      "text": "自己刚刚生成的答案",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 242,
      "start": 532.466,
      "end": 534.466,
      "text": "编译失败后补一个导入",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 243,
      "start": 534.6,
      "end": 536.6,
      "text": "测试失败后调整函数",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 244,
      "start": 536.666,
      "end": 538.9,
      "text": "这属于当前任务里的纠错",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 245,
      "start": 539.2,
      "end": 540.766,
      "text": "第二种过程是",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 246,
      "start": 540.766,
      "end": 543.1,
      "text": "研究人员让模型参与开发",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 247,
      "start": 543.266,
      "end": 545.866,
      "text": "帮助写实验代码或者分析结果",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 248,
      "start": 546.0,
      "end": 547.8,
      "text": "它能够节约研究时间",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 249,
      "start": 548.5,
      "end": 550.1,
      "text": "但研究方向和验收",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 250,
      "start": 550.1,
      "end": 552.133,
      "text": "标准仍然可能由人决定",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 251,
      "start": 552.266,
      "end": 554.133,
      "text": "更强的说法是系统",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 252,
      "start": 554.133,
      "end": 555.8,
      "text": "自主提出改进方案",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 253,
      "start": 555.933,
      "end": 557.533,
      "text": "完成训练和评估",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 254,
      "start": 557.666,
      "end": 559.866,
      "text": "再确认新系统确实更好",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 255,
      "start": 559.866,
      "end": 561.133,
      "text": "并继续下一轮",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 256,
      "start": 561.666,
      "end": 563.166,
      "text": "要支撑这种说法",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 257,
      "start": 563.266,
      "end": 565.5,
      "text": "至少需要看到比较完整的过程",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 258,
      "start": 565.7,
      "end": 566.966,
      "text": "模型改了什么",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 259,
      "start": 567.0,
      "end": 568.4,
      "text": "投入多少资源",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 260,
      "start": 568.5,
      "end": 570.8,
      "text": "在哪些没见过的任务上进步",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 261,
      "start": 570.866,
      "end": 572.933,
      "text": "出现退步又怎样处理",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 262,
      "start": 573.533,
      "end": 574.733,
      "text": "少了这些信息",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 263,
      "start": 574.866,
      "end": 577.5,
      "text": "自我进化很容易只剩一个标签",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 264,
      "start": 577.666,
      "end": 579.333,
      "text": "尤其要防止把熟悉",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 265,
      "start": 579.333,
      "end": 581.1,
      "text": "考试当成能力增长",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 266,
      "start": 581.5,
      "end": 583.933,
      "text": "模型反复接触同一批测试",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 267,
      "start": 584.2,
      "end": 586.5,
      "text": "可能越来越擅长那批测试",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 268,
      "start": 586.8,
      "end": 589.133,
      "text": "研究人员就算没有刻意作弊",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 269,
      "start": 589.366,
      "end": 591.266,
      "text": "选择实验的习惯也可能",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 270,
      "start": 591.266,
      "end": 593.333,
      "text": "让结果偏向既有指标",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 271,
      "start": 593.7,
      "end": 596.566,
      "text": "检验新版本需要留出独立的任务",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 272,
      "start": 596.666,
      "end": 599.1,
      "text": "还要观察其他能力有没有被牺牲",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 273,
      "start": 599.466,
      "end": 600.733,
      "text": "局部得分提高",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 274,
      "start": 600.8,
      "end": 602.866,
      "text": "不能自动推出全面提升",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 275,
      "start": 603.1,
      "end": 605.766,
      "text": "研究循环也受现实资源限制",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 276,
      "start": 606.166,
      "end": 608.266,
      "text": "提出一个训练方案很快",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 277,
      "start": 608.6,
      "end": 610.4,
      "text": "完成训练需要计算",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 278,
      "start": 610.533,
      "end": 612.6,
      "text": "某些实验必须等待结果",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 279,
      "start": 613.133,
      "end": 615.166,
      "text": "失败以后还得重新安排",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 280,
      "start": 615.366,
      "end": 618.066,
      "text": "涉及真实世界数据或者硬件时",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 281,
      "start": 618.133,
      "end": 619.4,
      "text": "等待会更长",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 282,
      "start": 619.566,
      "end": 621.266,
      "text": "模型帮助研究并不",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 283,
      "start": 621.266,
      "end": 623.333,
      "text": "意味着这些资源限制消失",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 284,
      "start": 623.533,
      "end": 625.266,
      "text": "递归过程有没有发生",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 285,
      "start": 625.266,
      "end": 626.533,
      "text": "与它能不能无限",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 286,
      "start": 626.533,
      "end": 628.6,
      "text": "加速是两个不同问题",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 287,
      "start": 628.7,
      "end": 631.566,
      "text": "这并不减损carbon可能具有的价值",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 288,
      "start": 632.066,
      "end": 633.733,
      "text": "即使它只是把研究人员",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 289,
      "start": 633.733,
      "end": 635.966,
      "text": "处理复杂代码的时间缩短",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 290,
      "start": 636.1,
      "end": 637.566,
      "text": "也已经值得关注",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 291,
      "start": 637.7,
      "end": 639.266,
      "text": "一个工具每天节约",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 292,
      "start": 639.266,
      "end": 641.166,
      "text": "少量高技能人员的时间",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 293,
      "start": 641.566,
      "end": 643.733,
      "text": "积累起来就可能改变开发节奏",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 294,
      "start": 643.8,
      "end": 645.4,
      "text": "这样的进步可以测量",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 295,
      "start": 645.5,
      "end": 647.466,
      "text": "也可以讨论商业意义",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 296,
      "start": 647.7,
      "end": 649.5,
      "text": "无需提前附送智能",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 297,
      "start": 649.5,
      "end": 651.366,
      "text": "爆炸已经开始的结论",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 298,
      "start": 651.5,
      "end": 653.333,
      "text": "谷歌还有一个独特条件",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 299,
      "start": 653.6,
      "end": 655.066,
      "text": "它既研发模型",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 300,
      "start": 655.066,
      "end": 657.6,
      "text": "也经营开发工具和办公产品",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 301,
      "start": 658.266,
      "end": 660.5,
      "text": "模型如果能够稳定处理工作",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 302,
      "start": 660.533,
      "end": 663.166,
      "text": "有机会通过这些入口触达用户",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 303,
      "start": 663.3,
      "end": 665.1,
      "text": "不过入口优势不会",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 304,
      "start": 665.1,
      "end": 666.766,
      "text": "自动变成使用习惯",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 305,
      "start": 667.0,
      "end": 668.3,
      "text": "企业已经建立的",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 306,
      "start": 668.3,
      "end": 670.466,
      "text": "审批流程和权限规则",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 307,
      "start": 670.666,
      "end": 673.566,
      "text": "往往比换一个聊天窗口难改的多",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 308,
      "start": 673.6,
      "end": 675.733,
      "text": "想象一个普通的企业需求",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 309,
      "start": 676.1,
      "end": 678.533,
      "text": "要求代理汇总某个项目的进展",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 310,
      "start": 678.766,
      "end": 681.133,
      "text": "资料分散在邮件和文档中",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 311,
      "start": 681.466,
      "end": 683.9,
      "text": "部分内容只有特定成员能看",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 312,
      "start": 684.066,
      "end": 687.2,
      "text": "系统要先确认提问者有权访问什么",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 313,
      "start": 687.3,
      "end": 689.7,
      "text": "再判断哪些资料仍然有效",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 314,
      "start": 689.766,
      "end": 691.9,
      "text": "最终输出还得说明依据",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 315,
      "start": 692.066,
      "end": 693.533,
      "text": "回答文字写的顺",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 316,
      "start": 694.1,
      "end": 696.0,
      "text": "不代表整个流程已经合格",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 317,
      "start": 696.133,
      "end": 699.1,
      "text": "如果任务升级成直接更新项目状态",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 318,
      "start": 699.1,
      "end": 700.5,
      "text": "要求又高了一层",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 319,
      "start": 701.066,
      "end": 702.366,
      "text": "更新哪条记录",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 320,
      "start": 702.5,
      "end": 704.1,
      "text": "是否会通知客户",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 321,
      "start": 704.466,
      "end": 705.8,
      "text": "谁有权批准",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 322,
      "start": 705.866,
      "end": 707.566,
      "text": "都需要被明确设置",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 323,
      "start": 707.8,
      "end": 710.6,
      "text": "能力增强会扩大可执行的动作",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 324,
      "start": 711.166,
      "end": 712.933,
      "text": "也会扩大错误的影响",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 325,
      "start": 713.133,
      "end": 714.166,
      "text": "模型越强",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 326,
      "start": 714.366,
      "end": 717.2,
      "text": "企业对边界的要求往往越具体",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 327,
      "start": 717.6,
      "end": 720.666,
      "text": "不能靠一句更智能略过去",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 328,
      "start": 720.666,
      "end": 723.366,
      "text": "这里就轮到微软decision一登场了",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 329,
      "start": 723.466,
      "end": 725.066,
      "text": "它关注的任务更窄",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 330,
      "start": 725.166,
      "end": 726.966,
      "text": "给定问题和备选项",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 331,
      "start": 727.0,
      "end": 728.8,
      "text": "输出各个选项的概率",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 332,
      "start": 729.2,
      "end": 730.5,
      "text": "微软把它定位在",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 333,
      "start": 730.5,
      "end": 732.333,
      "text": "决策与分类工作负载",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 334,
      "start": 732.933,
      "end": 735.366,
      "text": "基于千问3.59B做后训练",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 335,
      "start": 735.666,
      "end": 737.8,
      "text": "它选择处理大量重复的判断",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 336,
      "start": 738.2,
      "end": 740.766,
      "text": "不要求所有问题都交给最大模型",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 337,
      "start": 740.966,
      "end": 743.966,
      "text": "微软公布的测试覆盖36个基准",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 338,
      "start": 743.966,
      "end": 745.533,
      "text": "约15万道问题",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 339,
      "start": 745.566,
      "end": 749.1,
      "text": "并报告了相对GPT6索尔的延迟优势",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 340,
      "start": 749.266,
      "end": 752.133,
      "text": "这些是厂商提供的评估试用条件",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 341,
      "start": 752.166,
      "end": 753.133,
      "text": "需要保留",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 342,
      "start": 753.5,
      "end": 756.4,
      "text": "官方也介绍了对输入脑洞的测试",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 343,
      "start": 756.6,
      "end": 758.7,
      "text": "他们值得关注的地方是把",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 344,
      "start": 758.7,
      "end": 761.3,
      "text": "一致性和速度当成产品指标",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 345,
      "start": 761.3,
      "end": 762.966,
      "text": "不能据此断言任何",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 346,
      "start": 762.966,
      "end": 765.166,
      "text": "业务场景都能照搬结果",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 347,
      "start": 765.266,
      "end": 767.133,
      "text": "这种模型能放在哪里",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 348,
      "start": 767.2,
      "end": 769.266,
      "text": "客服系统收到一条请求",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 349,
      "start": 769.366,
      "end": 771.9,
      "text": "可以判断应当交给哪个业务组",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 350,
      "start": 772.566,
      "end": 774.466,
      "text": "代理生成一份答案之后",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 351,
      "start": 774.533,
      "end": 777.333,
      "text": "也可以根据预设标准分类检查",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 352,
      "start": 777.5,
      "end": 779.466,
      "text": "是否需要人工复核",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 353,
      "start": 779.866,
      "end": 782.366,
      "text": "可以被设计成一个明确选项",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 354,
      "start": 782.533,
      "end": 784.8,
      "text": "这些都是工作流的应用释义",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 355,
      "start": 784.966,
      "end": 786.933,
      "text": "不等于相关企业已经",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 356,
      "start": 786.933,
      "end": 788.5,
      "text": "部署了同样的流程",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 357,
      "start": 788.866,
      "end": 789.766,
      "text": "最有意思的是",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 358,
      "start": 789.766,
      "end": 792.8,
      "text": "这类系统未必需要写出一大段解释",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 359,
      "start": 793.2,
      "end": 794.8,
      "text": "业务有时只需要一个",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 360,
      "start": 794.8,
      "end": 796.6,
      "text": "格式稳定的选择结果",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 361,
      "start": 797.266,
      "end": 799.366,
      "text": "输出越容易被程序读取",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 362,
      "start": 799.5,
      "end": 801.7,
      "text": "后面的动作就越容易安排",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 363,
      "start": 801.933,
      "end": 804.0,
      "text": "大模型擅长展开思路",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 364,
      "start": 804.066,
      "end": 805.7,
      "text": "专用判断模型则可以把",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 365,
      "start": 805.7,
      "end": 807.966,
      "text": "计算集中在有限选择上",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 366,
      "start": 808.2,
      "end": 810.333,
      "text": "两种能力有机会组合起来",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 367,
      "start": 810.466,
      "end": 811.8,
      "text": "服务同一个任务",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 368,
      "start": 811.8,
      "end": 813.8,
      "text": "但概率必须谨慎解释",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 369,
      "start": 814.3,
      "end": 817.5,
      "text": "某个选项显示90%不等于当前",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 370,
      "start": 817.5,
      "end": 819.6,
      "text": "答案拿到了真实性担保",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 371,
      "start": 819.933,
      "end": 821.766,
      "text": "校准讨论的是一批预测",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 372,
      "start": 821.766,
      "end": 823.866,
      "text": "与实际结果之间的关系",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 373,
      "start": 824.066,
      "end": 826.1,
      "text": "企业需要在自己的数据上",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 374,
      "start": 826.1,
      "end": 828.266,
      "text": "检查这种关系是否成立",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 375,
      "start": 828.4,
      "end": 829.866,
      "text": "业务变化之后",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 376,
      "start": 830.1,
      "end": 831.766,
      "text": "原来可靠的概率",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 377,
      "start": 832.133,
      "end": 834.766,
      "text": "也可能需要重新评估比如",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 378,
      "start": 835.0,
      "end": 837.966,
      "text": "一个系统在普通咨询里表现稳定",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 379,
      "start": 838.266,
      "end": 840.066,
      "text": "遇到新的诈骗话术",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 380,
      "start": 840.1,
      "end": 841.566,
      "text": "判断可能失准",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 381,
      "start": 841.9,
      "end": 843.866,
      "text": "输入很像熟悉的类别",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 382,
      "start": 844.1,
      "end": 845.7,
      "text": "实际风险却不同",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 383,
      "start": 846.0,
      "end": 848.9,
      "text": "如果流程把高分直接当成放行许可",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 384,
      "start": 849.1,
      "end": 851.8,
      "text": "评分错误就会沿着自动化继续传播",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 385,
      "start": 851.966,
      "end": 855.1,
      "text": "可靠的设计需要给陌生情况留出口",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 386,
      "start": 855.7,
      "end": 858.333,
      "text": "包括拒绝判断或者交给人处理",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 387,
      "start": 858.333,
      "end": 860.0,
      "text": "这也解释了为什么判断",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 388,
      "start": 860.0,
      "end": 862.1,
      "text": "模型的选项设置很重要",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 389,
      "start": 862.366,
      "end": 864.466,
      "text": "如果只给执行和不执行",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 390,
      "start": 864.6,
      "end": 867.2,
      "text": "一些模糊情况会被硬塞进去",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 391,
      "start": 867.566,
      "end": 870.066,
      "text": "增加信息不足可能更符合业务",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 392,
      "start": 870.066,
      "end": 872.266,
      "text": "但新增选项之后还要",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 393,
      "start": 872.266,
      "end": 874.266,
      "text": "测试它是不是过度回避",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 394,
      "start": 874.5,
      "end": 877.333,
      "text": "企业需要按照错误代价设计流程",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 395,
      "start": 877.766,
      "end": 880.7,
      "text": "不能只追求一个漂亮的平均准确率",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 396,
      "start": 880.733,
      "end": 883.333,
      "text": "不同错误的代价可以相差很大",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 397,
      "start": 883.8,
      "end": 886.533,
      "text": "把普通咨询错分到另一个客服组",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 398,
      "start": 886.666,
      "end": 888.533,
      "text": "可能只是多等几分钟",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 399,
      "start": 889.0,
      "end": 890.933,
      "text": "把需要人工审批的支付",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 400,
      "start": 890.933,
      "end": 893.2,
      "text": "操作判断成可以自动执行",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 401,
      "start": 893.933,
      "end": 895.5,
      "text": "后果就严重得多",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 402,
      "start": 895.666,
      "end": 897.266,
      "text": "模型在两类任务上",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 403,
      "start": 897.266,
      "end": 899.0,
      "text": "得到相似的正确率",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 404,
      "start": 899.133,
      "end": 901.566,
      "text": "也不代表承担了相似的风险",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 405,
      "start": 902.266,
      "end": 904.066,
      "text": "评估必须看错在哪里",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 406,
      "start": 904.2,
      "end": 905.7,
      "text": "微软官方给decision",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 407,
      "start": 906.0,
      "end": 908.0,
      "text": "一列出的输入价格很低",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 408,
      "start": 908.166,
      "end": 909.7,
      "text": "输出不另行收费",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 409,
      "start": 909.9,
      "end": 912.9,
      "text": "并区分了当前可用入口和后续计划",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 410,
      "start": 913.466,
      "end": 914.933,
      "text": "这个定价值得讨论",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 411,
      "start": 914.966,
      "end": 917.766,
      "text": "但它只是模型调用的一部分成本",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 412,
      "start": 917.966,
      "end": 920.1,
      "text": "系统还需要连接业务数据",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 413,
      "start": 920.333,
      "end": 921.3,
      "text": "维护规则",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 414,
      "start": 921.5,
      "end": 922.5,
      "text": "保存日志",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 415,
      "start": 922.666,
      "end": 923.666,
      "text": "出了问题",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 416,
      "start": 923.766,
      "end": 925.333,
      "text": "团队还得有人处理",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 417,
      "start": 925.733,
      "end": 927.6,
      "text": "低价分类可以节约预算",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 418,
      "start": 927.766,
      "end": 929.666,
      "text": "不能让整个工程免费",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 419,
      "start": 929.9,
      "end": 931.1,
      "text": "延迟同样如此",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 420,
      "start": 931.3,
      "end": 933.6,
      "text": "一个判断步骤明显变快",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 421,
      "start": 933.666,
      "end": 935.266,
      "text": "如果整条流程大部分",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 422,
      "start": 935.266,
      "end": 936.933,
      "text": "时间都在等待数据库",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 423,
      "start": 937.2,
      "end": 939.2,
      "text": "用户感受到的改善就有限",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 424,
      "start": 939.6,
      "end": 942.366,
      "text": "业务负责人需要测端到端时间",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 425,
      "start": 942.466,
      "end": 944.4,
      "text": "每个模型单独跑得快",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 426,
      "start": 944.533,
      "end": 946.6,
      "text": "系统之间来回传输很慢",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 427,
      "start": 946.6,
      "end": 948.733,
      "text": "最终体验还是可能卡住",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 428,
      "start": 949.133,
      "end": 950.533,
      "text": "局部性能只有放进",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 429,
      "start": 950.533,
      "end": 952.733,
      "text": "流程里才知道实际价值",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 430,
      "start": 952.8,
      "end": 955.2,
      "text": "把carbon和decision e放到一起",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 431,
      "start": 955.5,
      "end": 957.5,
      "text": "竞争路线就更清楚了",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 432,
      "start": 957.933,
      "end": 959.266,
      "text": "carbon受到关注",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 433,
      "start": 959.333,
      "end": 960.7,
      "text": "因为它可能承担更",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 434,
      "start": 960.7,
      "end": 962.8,
      "text": "复杂的生成和编程工作",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 435,
      "start": 963.6,
      "end": 965.333,
      "text": "decision e则试图把高频",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 436,
      "start": 965.333,
      "end": 967.1,
      "text": "选择变得便宜而稳定",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 437,
      "start": 967.166,
      "end": 968.766,
      "text": "一个工作流完全可以",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 438,
      "start": 968.766,
      "end": 970.6,
      "text": "同时需要这两类能力",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 439,
      "start": 970.733,
      "end": 972.6,
      "text": "再让人保留重要审批",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 440,
      "start": 973.166,
      "end": 975.1,
      "text": "企业采购的对象正在",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 441,
      "start": 975.1,
      "end": 977.266,
      "text": "变成一整套任务完成方案",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 442,
      "start": 977.5,
      "end": 979.133,
      "text": "模型之间怎样分工",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 443,
      "start": 979.133,
      "end": 980.666,
      "text": "也会带来新的问题",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 444,
      "start": 980.966,
      "end": 982.566,
      "text": "分类系统把任务分错",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 445,
      "start": 982.6,
      "end": 985.166,
      "text": "大模型可能认真解决了错误的问题",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 446,
      "start": 985.2,
      "end": 987.2,
      "text": "大模型输出的信息不完整",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 447,
      "start": 987.3,
      "end": 990.766,
      "text": "后续评分也可能建立在缺失材料上",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 448,
      "start": 990.766,
      "end": 992.4,
      "text": "多个组件串联之后",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 449,
      "start": 992.4,
      "end": 993.9,
      "text": "责任必须能追溯",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 450,
      "start": 994.133,
      "end": 997.1,
      "text": "系统应该记录每个阶段看到了什么",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 451,
      "start": 997.1,
      "end": 998.4,
      "text": "做出了什么选择",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 452,
      "start": 998.566,
      "end": 1000.766,
      "text": "这类记录还有一个实际用途",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 453,
      "start": 1001.066,
      "end": 1003.5,
      "text": "帮助团队判断应该升级哪一部分",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 454,
      "start": 1003.9,
      "end": 1005.466,
      "text": "失败发生在检索",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 455,
      "start": 1005.533,
      "end": 1007.866,
      "text": "就不该盲目换更贵的模型",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 456,
      "start": 1008.533,
      "end": 1009.7,
      "text": "规则写的模糊",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 457,
      "start": 1009.8,
      "end": 1011.366,
      "text": "就应该先修规则",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 458,
      "start": 1011.6,
      "end": 1014.133,
      "text": "只有确实属于能力不足的任务",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 459,
      "start": 1014.366,
      "end": 1016.2,
      "text": "升级模型才可能解决",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 460,
      "start": 1016.3,
      "end": 1017.733,
      "text": "没有分阶段记录",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 461,
      "start": 1018.166,
      "end": 1019.533,
      "text": "所有问题都会被归结",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 462,
      "start": 1019.533,
      "end": 1021.6,
      "text": "成一句AI还不够强",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 463,
      "start": 1021.666,
      "end": 1024.3,
      "text": "企业可以先选一个范围有限的任务",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 464,
      "start": 1024.666,
      "end": 1026.5,
      "text": "保留原流程作为对照",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 465,
      "start": 1027.2,
      "end": 1028.9,
      "text": "评价标准提前确定",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 466,
      "start": 1028.966,
      "end": 1030.466,
      "text": "既记录完成时间",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 467,
      "start": 1030.466,
      "end": 1032.1,
      "text": "也记录返工和错误",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 468,
      "start": 1032.2,
      "end": 1033.366,
      "text": "这样做看起来没有",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 469,
      "start": 1033.366,
      "end": 1034.933,
      "text": "全面上线那么热闹",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 470,
      "start": 1035.133,
      "end": 1037.566,
      "text": "却更容易判断投入有没有回报",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 471,
      "start": 1037.666,
      "end": 1038.966,
      "text": "任务边界清楚",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 472,
      "start": 1039.166,
      "end": 1040.866,
      "text": "失败也更容易回滚",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 473,
      "start": 1041.166,
      "end": 1043.733,
      "text": "团队能够积累真正有用的经验",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 474,
      "start": 1043.933,
      "end": 1046.5,
      "text": "对照测试还要防止选择偏差",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 475,
      "start": 1047.066,
      "end": 1049.733,
      "text": "员工往往把容易的任务交给AI",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 476,
      "start": 1049.966,
      "end": 1051.9,
      "text": "把复杂任务留给自己",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 477,
      "start": 1052.166,
      "end": 1053.733,
      "text": "最后看使用记录",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 478,
      "start": 1054.0,
      "end": 1055.566,
      "text": "AI完成率很高",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 479,
      "start": 1055.766,
      "end": 1057.266,
      "text": "人类耗时很长",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 480,
      "start": 1057.3,
      "end": 1058.9,
      "text": "比较就已经不公平",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 481,
      "start": 1059.2,
      "end": 1061.5,
      "text": "不同任务的难度必须被考虑",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 482,
      "start": 1061.666,
      "end": 1064.2,
      "text": "否则一套未经校正的报表就",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 483,
      "start": 1064.2,
      "end": 1066.3,
      "text": "可能推动错误的用工决定",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 484,
      "start": 1066.5,
      "end": 1068.4,
      "text": "模型更新后也要重测",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 485,
      "start": 1068.866,
      "end": 1071.7,
      "text": "企业可能喜欢新版本的总体表现",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 486,
      "start": 1071.9,
      "end": 1074.666,
      "text": "却发现某个关键分类变差了",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 487,
      "start": 1074.733,
      "end": 1076.4,
      "text": "如果自动切换版本",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 488,
      "start": 1076.4,
      "end": 1078.533,
      "text": "没有复核就继续运行",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 489,
      "start": 1078.866,
      "end": 1081.533,
      "text": "原先建立的流程可能悄悄失效",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 490,
      "start": 1081.8,
      "end": 1084.333,
      "text": "稳定的模型服务需要版本管理",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 491,
      "start": 1084.766,
      "end": 1088.0,
      "text": "重要任务还要允许暂时保留旧版本",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 492,
      "start": 1088.333,
      "end": 1090.766,
      "text": "更新速度和部署速度可以不同",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 493,
      "start": 1090.866,
      "end": 1093.333,
      "text": "成本评估也应覆盖失败的任务",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 494,
      "start": 1093.866,
      "end": 1096.166,
      "text": "只算成功那一次的调用费用",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 495,
      "start": 1096.266,
      "end": 1097.866,
      "text": "会漏掉之前的重试",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 496,
      "start": 1098.166,
      "end": 1099.6,
      "text": "只算生成速度",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 497,
      "start": 1099.733,
      "end": 1101.333,
      "text": "会漏掉人工检查",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 498,
      "start": 1101.566,
      "end": 1103.066,
      "text": "团队需要知道",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 499,
      "start": 1103.3,
      "end": 1106.2,
      "text": "完成一个合格任务总共花了多少钱",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 500,
      "start": 1106.3,
      "end": 1108.733,
      "text": "发生严重错误的频率是多少",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 501,
      "start": 1108.766,
      "end": 1109.933,
      "text": "这两个数字",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 502,
      "start": 1110.6,
      "end": 1111.766,
      "text": "比每秒生成多少",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 503,
      "start": 1111.766,
      "end": 1113.7,
      "text": "文字更接近商业决策",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 504,
      "start": 1113.9,
      "end": 1117.1,
      "text": "这也是卡本后续最值得观察的方向",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 505,
      "start": 1117.9,
      "end": 1119.566,
      "text": "公开测试如果出现",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 506,
      "start": 1119.6,
      "end": 1122.4,
      "text": "应该看任务是否覆盖真实仓库",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 507,
      "start": 1123.066,
      "end": 1124.7,
      "text": "测试条件是否一致",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 508,
      "start": 1124.866,
      "end": 1126.766,
      "text": "失败记录是否完整",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 509,
      "start": 1127.1,
      "end": 1128.2,
      "text": "产品如果上线",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 510,
      "start": 1128.2,
      "end": 1130.066,
      "text": "应该看普通付费用户",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 511,
      "start": 1130.1,
      "end": 1131.566,
      "text": "能够获得什么条件",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 512,
      "start": 1131.7,
      "end": 1133.9,
      "text": "内部员工的惊喜很有吸引力",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 513,
      "start": 1134.066,
      "end": 1135.366,
      "text": "可持续的服务体验",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 514,
      "start": 1135.366,
      "end": 1136.8,
      "text": "才会决定市场位置",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 515,
      "start": 1136.966,
      "end": 1139.1,
      "text": "还有一个会影响采购的因素",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 516,
      "start": 1139.333,
      "end": 1141.366,
      "text": "任务到底由谁定义完成",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 517,
      "start": 1142.1,
      "end": 1144.3,
      "text": "编程测试通常有预设目标",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 518,
      "start": 1144.533,
      "end": 1146.566,
      "text": "企业请求却可能很模糊",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 519,
      "start": 1146.766,
      "end": 1148.7,
      "text": "客户说希望页面更快",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 520,
      "start": 1148.733,
      "end": 1150.3,
      "text": "可能指加载时间",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 521,
      "start": 1150.366,
      "end": 1152.5,
      "text": "也可能指操作步骤太多",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 522,
      "start": 1152.666,
      "end": 1154.3,
      "text": "模型直接优化代码",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 523,
      "start": 1154.333,
      "end": 1156.733,
      "text": "未必解决客户最在意的问题",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 524,
      "start": 1157.4,
      "end": 1159.766,
      "text": "复杂任务需要先确认目标",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 525,
      "start": 1159.966,
      "end": 1162.1,
      "text": "随后才适合自动执行",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 526,
      "start": 1162.166,
      "end": 1164.133,
      "text": "需求澄清也可以被评估",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 527,
      "start": 1164.3,
      "end": 1166.4,
      "text": "系统能否识别缺失条件",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 528,
      "start": 1166.466,
      "end": 1168.266,
      "text": "能否只问必要的问题",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 529,
      "start": 1168.333,
      "end": 1170.0,
      "text": "都会影响合作体验",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 530,
      "start": 1170.1,
      "end": 1171.8,
      "text": "每一步都要求人确认",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 531,
      "start": 1171.9,
      "end": 1173.066,
      "text": "效率上不去",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 532,
      "start": 1173.066,
      "end": 1173.933,
      "text": "什么都不问",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 533,
      "start": 1173.933,
      "end": 1175.066,
      "text": "又容易走偏",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 534,
      "start": 1175.333,
      "end": 1177.866,
      "text": "企业需要寻找合适的询问频率",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 535,
      "start": 1177.866,
      "end": 1180.333,
      "text": "并按任务风险设置不同阈值",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 536,
      "start": 1180.733,
      "end": 1182.333,
      "text": "这个问题通常比聊天",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 537,
      "start": 1182.333,
      "end": 1184.366,
      "text": "回答是否自然更难处理",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 538,
      "start": 1184.566,
      "end": 1186.533,
      "text": "长任务还会跨越时间",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 539,
      "start": 1187.1,
      "end": 1188.866,
      "text": "代理今天拿到的资料",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 540,
      "start": 1188.9,
      "end": 1190.4,
      "text": "明天可能被更新",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 541,
      "start": 1191.166,
      "end": 1192.866,
      "text": "用户中途改变要求",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 542,
      "start": 1193.0,
      "end": 1195.5,
      "text": "他得知道哪些工作可以保留",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 543,
      "start": 1195.5,
      "end": 1196.666,
      "text": "哪些已经作废",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 544,
      "start": 1196.933,
      "end": 1199.466,
      "text": "如果系统继续沿着旧目标执行",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 545,
      "start": 1199.866,
      "end": 1202.333,
      "text": "能力再强也只是更快的做错事",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 546,
      "start": 1202.4,
      "end": 1204.2,
      "text": "保存工作状态的同时",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 547,
      "start": 1204.2,
      "end": 1206.133,
      "text": "还要允许目标发生变化",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 548,
      "start": 1206.266,
      "end": 1208.5,
      "text": "这是长期运行的实际要求",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 549,
      "start": 1208.6,
      "end": 1210.0,
      "text": "任务中断之后能否",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 550,
      "start": 1210.0,
      "end": 1212.1,
      "text": "恢复也值得单独测试",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 551,
      "start": 1212.7,
      "end": 1213.966,
      "text": "网路暂时断开",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 552,
      "start": 1213.966,
      "end": 1216.2,
      "text": "已经完成的步骤是否会重复",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 553,
      "start": 1216.4,
      "end": 1218.133,
      "text": "一项操作已经提交",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 554,
      "start": 1218.133,
      "end": 1220.2,
      "text": "但确认消息没有返回",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 555,
      "start": 1220.4,
      "end": 1222.366,
      "text": "代理会不会再提交一次",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 556,
      "start": 1222.4,
      "end": 1224.066,
      "text": "这些情况在演示中很少",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 557,
      "start": 1224.066,
      "end": 1226.766,
      "text": "出现在真实服务里却躲不开",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 558,
      "start": 1227.366,
      "end": 1228.9,
      "text": "恢复机制做不好",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 559,
      "start": 1229.133,
      "end": 1230.566,
      "text": "自动化就可能制造",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 560,
      "start": 1230.566,
      "end": 1232.666,
      "text": "重复订单或者重复修改",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 561,
      "start": 1232.766,
      "end": 1233.966,
      "text": "在这些问题上",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 562,
      "start": 1233.966,
      "end": 1236.966,
      "text": "模型不一定需要独自解决所有细节",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 563,
      "start": 1237.933,
      "end": 1239.9,
      "text": "应用可以记录操作状态",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 564,
      "start": 1240.0,
      "end": 1241.3,
      "text": "阻止重复提交",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 565,
      "start": 1241.3,
      "end": 1243.0,
      "text": "再让模型决定下一步",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 566,
      "start": 1243.3,
      "end": 1245.866,
      "text": "确定性的程序负责确定性的约束",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 567,
      "start": 1246.4,
      "end": 1249.1,
      "text": "模型处理需要理解和选择的部分",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 568,
      "start": 1249.3,
      "end": 1251.533,
      "text": "企业能够把两者结合的多好",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 569,
      "start": 1251.933,
      "end": 1253.466,
      "text": "会直接影响carbon这类",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 570,
      "start": 1253.466,
      "end": 1255.366,
      "text": "强模型能发挥多少价值",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 571,
      "start": 1255.566,
      "end": 1257.166,
      "text": "数据保密则决定哪些",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 572,
      "start": 1257.166,
      "end": 1259.266,
      "text": "任务根本不能随意交出去",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 573,
      "start": 1259.4,
      "end": 1261.766,
      "text": "代码仓库里可能包含商业逻辑",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 574,
      "start": 1262.1,
      "end": 1263.966,
      "text": "文档里可能有客户信息",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 575,
      "start": 1264.2,
      "end": 1266.1,
      "text": "企业选择模型需要查清",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 576,
      "start": 1266.1,
      "end": 1268.333,
      "text": "数据保存方式和服务条款",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 577,
      "start": 1268.733,
      "end": 1271.266,
      "text": "模型路由经过不同供应商时",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 578,
      "start": 1271.3,
      "end": 1273.133,
      "text": "也要确认资料流向",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 579,
      "start": 1273.366,
      "end": 1275.266,
      "text": "使用效果相似的两个服务",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 580,
      "start": 1275.466,
      "end": 1277.333,
      "text": "合规条件可能完全不同",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 581,
      "start": 1277.466,
      "end": 1279.6,
      "text": "由此还会出现迁移成本",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 582,
      "start": 1280.1,
      "end": 1281.966,
      "text": "一个团队围绕某个模型",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 583,
      "start": 1281.966,
      "end": 1284.166,
      "text": "调好了提示和评分规则",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 584,
      "start": 1284.466,
      "end": 1286.766,
      "text": "换模型之后未必照样有效",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 585,
      "start": 1286.966,
      "end": 1288.766,
      "text": "输出格式稍有变化",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 586,
      "start": 1288.766,
      "end": 1290.8,
      "text": "下游程序就可能出错",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 587,
      "start": 1290.966,
      "end": 1293.466,
      "text": "企业可以保留共同的任务接口",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 588,
      "start": 1293.9,
      "end": 1295.8,
      "text": "再为不同模型做适配",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 589,
      "start": 1295.933,
      "end": 1298.333,
      "text": "这样以后选择carbon或者其他候选",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 590,
      "start": 1298.333,
      "end": 1301.266,
      "text": "服务才不至于重新搭建整个流程",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 591,
      "start": 1301.4,
      "end": 1303.3,
      "text": "比较时也应该避免只",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 592,
      "start": 1303.3,
      "end": 1304.766,
      "text": "挑自己擅长的任务",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 593,
      "start": 1305.0,
      "end": 1307.533,
      "text": "一家模型擅长从零搭建页面",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 594,
      "start": 1307.6,
      "end": 1309.8,
      "text": "另一家擅长修复旧项目",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 595,
      "start": 1309.933,
      "end": 1312.333,
      "text": "两者都可以展示出经验结果",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 596,
      "start": 1312.533,
      "end": 1314.166,
      "text": "采购方需要按照真实",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 597,
      "start": 1314.166,
      "end": 1316.1,
      "text": "任务占比构建测试集",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 598,
      "start": 1316.3,
      "end": 1319.1,
      "text": "很少遇到的高难问题可以单独关注",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 599,
      "start": 1319.2,
      "end": 1321.866,
      "text": "不能让它完全代表日常使用成本",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 600,
      "start": 1321.9,
      "end": 1324.6,
      "text": "还有用户经常忽略的等待分布",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 601,
      "start": 1325.4,
      "end": 1326.866,
      "text": "平均耗时不错",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 602,
      "start": 1326.9,
      "end": 1328.2,
      "text": "偶尔卡住很久",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 603,
      "start": 1328.2,
      "end": 1330.066,
      "text": "同样会破坏工作安排",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 604,
      "start": 1330.766,
      "end": 1334.133,
      "text": "客服和实时业务尤其需要看慢请求",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 605,
      "start": 1334.333,
      "end": 1337.0,
      "text": "研究型任务能够容忍更长等待",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 606,
      "start": 1337.1,
      "end": 1338.266,
      "text": "标准又不同",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 607,
      "start": 1338.3,
      "end": 1340.7,
      "text": "用一个平均延迟比较所有产品",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 608,
      "start": 1341.1,
      "end": 1343.9,
      "text": "很容易选出对业务并不合适的模型",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 609,
      "start": 1344.3,
      "end": 1346.166,
      "text": "decision一这类模型的机会",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 610,
      "start": 1346.166,
      "end": 1348.266,
      "text": "就来自任务需求的差别",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 611,
      "start": 1348.666,
      "end": 1351.533,
      "text": "高频而简单的分类可以单独处理",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 612,
      "start": 1351.8,
      "end": 1354.166,
      "text": "把昂贵计算留给复杂部分",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 613,
      "start": 1354.733,
      "end": 1357.0,
      "text": "但路由系统自身也需要观察",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 614,
      "start": 1357.0,
      "end": 1358.9,
      "text": "不能为了省一笔调用费",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 615,
      "start": 1358.933,
      "end": 1360.933,
      "text": "让后面产生更多返工",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 616,
      "start": 1361.2,
      "end": 1363.333,
      "text": "模型组合的价值应当",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 617,
      "start": 1363.333,
      "end": 1366.1,
      "text": "体现为合格交付成本下降",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 618,
      "start": 1366.266,
      "end": 1368.6,
      "text": "不能只体现在某个组件便宜",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 619,
      "start": 1368.733,
      "end": 1370.4,
      "text": "对个人开发者同样",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 620,
      "start": 1370.4,
      "end": 1372.133,
      "text": "可以用较小的方式验证",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 621,
      "start": 1372.9,
      "end": 1375.566,
      "text": "选一批已经知道解决方法的旧问题",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 622,
      "start": 1375.9,
      "end": 1376.8,
      "text": "隐藏答案",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 623,
      "start": 1376.866,
      "end": 1378.6,
      "text": "让新工具独立尝试",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 624,
      "start": 1378.733,
      "end": 1381.366,
      "text": "比较它使用的时间和修改范围",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 625,
      "start": 1381.5,
      "end": 1383.533,
      "text": "再检查有没有新增问题",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 626,
      "start": 1383.733,
      "end": 1386.266,
      "text": "这样的测试无法替代专业基准",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 627,
      "start": 1386.5,
      "end": 1388.466,
      "text": "却能帮助判断新工具",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 628,
      "start": 1388.466,
      "end": 1390.366,
      "text": "是否适合手上的项目",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 629,
      "start": 1390.966,
      "end": 1393.7,
      "text": "减少仅凭宣传切换工具的成本",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 630,
      "start": 1393.9,
      "end": 1395.2,
      "text": "网络安全相关能力",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 631,
      "start": 1395.2,
      "end": 1396.533,
      "text": "还需要额外的边界",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 632,
      "start": 1396.966,
      "end": 1399.166,
      "text": "模型能够发现复杂软件问题",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 633,
      "start": 1399.3,
      "end": 1401.533,
      "text": "对防御工作可能很有帮助",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 634,
      "start": 1401.866,
      "end": 1403.1,
      "text": "但使用者身份和",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 635,
      "start": 1403.1,
      "end": 1404.966,
      "text": "操作权限必须被检查",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 636,
      "start": 1405.266,
      "end": 1408.166,
      "text": "授权研究与未经许可的操作不能",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 637,
      "start": 1408.166,
      "end": 1410.933,
      "text": "因为同样调用一个模型就混在一起",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 638,
      "start": 1411.066,
      "end": 1413.3,
      "text": "安全发布的难点也包括让",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 639,
      "start": 1413.3,
      "end": 1415.666,
      "text": "有益能力进入正确的工作流程",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 640,
      "start": 1415.8,
      "end": 1418.266,
      "text": "把所有能力封成一个聊天窗口",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 641,
      "start": 1418.6,
      "end": 1420.766,
      "text": "用户可能很难察觉这些差别",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 642,
      "start": 1420.966,
      "end": 1423.866,
      "text": "企业却必须知道哪些动作只是建议",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 643,
      "start": 1424.0,
      "end": 1425.933,
      "text": "哪些动作真的改变系统",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 644,
      "start": 1425.966,
      "end": 1428.3,
      "text": "一次代码审查可以人工确认",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 645,
      "start": 1428.533,
      "end": 1430.4,
      "text": "一次生产配置修改则",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 646,
      "start": 1430.4,
      "end": 1431.966,
      "text": "需要更严格的审批",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 647,
      "start": 1432.4,
      "end": 1434.3,
      "text": "产品越接近实际执行",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 648,
      "start": 1434.4,
      "end": 1437.466,
      "text": "界面上越应该清楚的展示动作范围",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 649,
      "start": 1437.666,
      "end": 1439.266,
      "text": "未来的竞争也就不会",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 650,
      "start": 1439.266,
      "end": 1441.133,
      "text": "只停留在模型排行榜",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 651,
      "start": 1441.666,
      "end": 1444.466,
      "text": "模型厂商需要证明复杂任务的能力",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 652,
      "start": 1444.6,
      "end": 1447.2,
      "text": "工具平台需要证明连接可靠",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 653,
      "start": 1447.5,
      "end": 1448.733,
      "text": "企业则要证明",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 654,
      "start": 1448.733,
      "end": 1450.966,
      "text": "自动化确实节约总成本",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 655,
      "start": 1451.166,
      "end": 1452.6,
      "text": "任何一环跟不上",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 656,
      "start": 1452.966,
      "end": 1454.466,
      "text": "都可能让看起来领先",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 657,
      "start": 1454.466,
      "end": 1456.5,
      "text": "的技术停在演示阶段",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 658,
      "start": 1456.666,
      "end": 1458.666,
      "text": "这个过程没有统一的捷径",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 659,
      "start": 1458.8,
      "end": 1461.0,
      "text": "只能靠连续的部署和复合",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 660,
      "start": 1461.3,
      "end": 1464.666,
      "text": "carbon目前最重要的信号是谷歌内部",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 661,
      "start": 1464.666,
      "end": 1467.7,
      "text": "的编程候选模型正在获得积极反馈",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 662,
      "start": 1468.1,
      "end": 1469.6,
      "text": "这个信号值得追踪",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 663,
      "start": 1469.766,
      "end": 1472.533,
      "text": "却不足以写成全面领先的事实",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 664,
      "start": 1472.8,
      "end": 1474.6,
      "text": "微软decision亦提供了",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 665,
      "start": 1474.6,
      "end": 1476.0,
      "text": "另一个观察角度",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 666,
      "start": 1476.333,
      "end": 1479.533,
      "text": "AI产品还在向具体判断任务细分",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 667,
      "start": 1479.7,
      "end": 1482.366,
      "text": "规模和专用化都可能带来价值",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 668,
      "start": 1482.533,
      "end": 1484.966,
      "text": "最终要接受同一套业务验收",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 669,
      "start": 1485.2,
      "end": 1486.566,
      "text": "开头的问题现在可以",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 670,
      "start": 1486.566,
      "end": 1488.333,
      "text": "落到一个明确标准上",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 671,
      "start": 1488.466,
      "end": 1491.066,
      "text": "企业离放心交出工作还有距离",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 672,
      "start": 1491.1,
      "end": 1493.166,
      "text": "因为交付涉及连续判断",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 673,
      "start": 1493.3,
      "end": 1495.133,
      "text": "而且错误有真实代价",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 674,
      "start": 1495.5,
      "end": 1497.933,
      "text": "carbon如果确实是个强大的新模型",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 675,
      "start": 1498.133,
      "end": 1500.966,
      "text": "最有说服力的证明就是帮助用户",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 676,
      "start": 1500.966,
      "end": 1503.9,
      "text": "在可承受的成本下完成合格任务",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 677,
      "start": 1504.166,
      "end": 1505.166,
      "text": "能被复核",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 678,
      "start": 1505.2,
      "end": 1506.166,
      "text": "能被回滚",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 679,
      "start": 1506.5,
      "end": 1508.3,
      "text": "下一次还能稳定做对",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 680,
      "start": 1508.3,
      "end": 1510.0,
      "text": "这样的进步才会留在账上",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    }
  ],
  "transcript_selection": {
    "schema": "videonote.transcript_selection.v1",
    "status": "SELECTED",
    "requested_language": "zh",
    "duration": 1510.0,
    "reference_path": "/Users/sagawa/AI_WORK/video_notes/url/20261013_175016__谷歌內部核彈級洩露_Carbon_模型橫空出世_Claude_3.5_真的要涼了/reference_subtitle.vtt",
    "reference_kind": "manual",
    "reference_language": "zh",
    "selected_source": "uploaded_master_subtitle",
    "source_caption_kind": "manual",
    "source_caption_processing": "verbatim_direct",
    "youtube_overlap_clamps": [],
    "general_high_confidence_repairs": [],
    "decision_reason": "explicit_adopt_manual_no_whisper",
    "source_strategy": "ADOPT_MANUAL",
    "segmentation_policy": "PRESERVE_SOURCE_CUES",
    "whisper_rejected": false,
    "source_profile": {
      "segment_count": 681,
      "duration_coverage": 0.913662
    }
  }
}