{
  "text": "如果你持续关注硅谷AI产业动态,一定听过这场堪称传奇的人才出走事件。手握20亿美元顶级风头,集结一批从OpenAI出走的核心研发人才。秘密打磨接近两年,所有人都默认他们会拿出一款能够横扫行业的顶尖大模型。可最终的结果却让整个科技圈大跌眼镜。在前不久的发布会结束之后,他们推出的全新模型在主流权威基准评测榜单当中甚至无法冲进前五。\n不少开源模型的综合表现\n都能够实现对它的压制\n回望2024年年底那场震动\n全球AI行业的\nOpenAI高管离职风波\n大家才慢慢意识到\n这根本不是简单的人员流动\n而是一场精心布局\n持续多年的人才收割行动\n米拉姆拉蒂\n也就是曾经短暂执掌OpenAI\n全程主导ChatGPT\n落地研发的核心关键人物\n在2024年9月正式离开老东家\n并且不是孤身一人\n他踏出OpenAI大门之后\n联合创始人约翰·舒尔曼\n前研究副总裁巴雷特·佐夫\n还有一大批资深核心研究员紧随其后递交辞呈\n这群人的集体出走\n几乎直接抽走了OpenAI半数核心研发力量\n在硅谷的产业生态里面\n顶尖人工智能研究员的稀缺程度\n甚至超过英伟达H100系列高端算力芯片\n米拉带走的远远不止数十名资深工程师\n更是一套经过GPT3到GPT4完整项目验证\n成熟可行的大模型训练方法论\n还有整套工程落地的实战经验\n这支团队亲眼见证过\n全球规模顶尖算力集群的运行逻辑\n也踩遍了训练过程里\n各类会引发模型崩溃的隐蔽技术陷阱\n因此当他们对外宣布独立创业的时候\n整个风险投资圈子陷入狂热\n硅谷风投机构最擅长算账\n也最容易陷入害怕错失机遇的情绪\n安德森霍洛维茨\n直接敲定20亿美元现金投入\n投资落地的时候\n这家连正式名称都没有完全敲定\n一行商用产品代码都还没有产出的初创团队\n最终定名Thinking Machines Lab\n投后估值直接冲到120亿美元\n没有任何成熟产品就能拥有120亿美元估值\n放在其他行业很容易被贴上庞氏骗局的标签\n但是在近几年火热的AI赛道\n资本把这称作抢占通往通用人工智能时代最后的传票\n风投机构押注的核心目标\n就是米拉带领的这支团队\n能够打造出一套\n打破OpenAI避原模式垄断的\n全新产业生态\n所有人静静观望\n期待着20亿美元资金\n两年时间能够孵化出\n怎样惊世骇俗的人工智能产品\n整个行业秉析宁神\n等待新一代顶级大模型的\n技术降维打击\n时间来到2026年7月中旬\n也就是近期\n沉寂一年半之久的\nThinking Machines Lab\n终于对外公开酝酿许久的重磅成果\n开源基础模型代号Inklin\n伴随着发布会正式落幕\n整个科技行业陷入巨大的错愕\n这款投入重金闭关研发两年打造的模型\n在各类权威基准测试当中成绩只能用平庸形容\n前方有克劳德·菲波5\nOpenAI主流避原模型两座顶尖避原模型\n稳定占据高位\n同期发布的月之暗面KIMI K3\n这款拥有2.8万亿参数的模型\n同样能够和它正面抗衡\n多项指标实现反超\nInkling在开源模型梯队当中\n综合水平最多只能算作中油水准\n更加尴尬的还有发布时间窗口\nInkling正式上线仅仅一天之后\nKimi K3热度全面爆发\n大量开发者涌入调用\n直接把算力资源耗尽\n几乎抢走了Inkling全部公众关注度\n普通观众看到这里\n大概率都会认定这是一场彻头彻尾的失败\n顶级团队\n顶级资金\n顶级资源\n最后打出一张平平无奇的牌\n甚至被后发竞品全面碾压\n但如果你只看到表面的跑分输赢\n完全看不懂当下AI行业真正的内卷困局\n更看不懂硅谷顶级资本真正的下注逻辑\n现在整个AI行业已经陷入一种\n极度畸形的内卷怪圈\n绝大多数团队、媒体、开发者\n都在疯狂迷信基准测试分数\n把榜单排名当成唯一的评判标准\n为了零点几分的微弱提升\n无数公司动辄烧数亿美金算力\n堆砌海量参数\n清洗冗余数据\n反复微调权重\n最后做出来的模型跑分看似好看\n但是商用成本极高\n落地兼容性极差\n企业部署难度巨大\n完全是为了跑分而生\n而非为了商用而生的产品\n这也是最近一年来\n无数顶级AI模型高开低走\n热度爆红却无法变现的核心原因\n行业早已不再缺聪明的模型\n企业真正缺的\n是稳定便宜可控能落地\n零合规风险的工业化AI基础设施\n而这恰恰就是Inkling精准卡位的赛道\n大家一定要点赞订阅锁住频道\n接下来我逐层拆解这场瞒天过海的AI商业布局\n看懂了你就明白\n这根本不是翻车\n而是近两年最顶级的AI创业降位布局\n我们不妨深入拆解\n这笔20亿美元资金孵化出来的技术底子究竟具备哪些特点\nInkling属于总参数量达到9750亿的混合专家模型\n也就是业内常说的MOE架构\n单单看总参数规模接近万亿\n听起来极具冲击力\n但是熟悉稀疏架构的从业者\n都清楚MOE底层运行逻辑\n每一次处理词源信息的时候\n仅仅只有410亿参数被激活参与运算\n模型原生支持100万长度上下文窗口\n依托45万亿词源规模的文本\n图像 音频混合数据\n完成从零开始的预训链\n整套体系当中最具备革新意义的\n是它重构了多模态信息处理逻辑\n目前市面上绝大多数宣称原生多模态的大模型\n本质只是在基础语言模型外部附加一套处理组件\n遇到图片、音频信息\n需要独立编码器\n先把视觉、听觉信号转化为模型能够识别的次元格式\n再送入主体网络运算\n而Inkling跳过中间转化流程\n可以直接读取原始像素与音频波形信号\n这种底层架构重构的工程难度极高\n也直观体现出这支前OpenAI团队扎实的技术积累\n对比Meta很多开源模型饱受争议的授权条款\nThinking Machines Lab\n这次选择十分开放\n直接采用Apache 2.0开源协议\n没有附加繁杂限制\n完整模型权重直接上传至哈格面部平台\n开发者可以自由修改\n部署 商用\n即便拥有底层架构创新\n搭配宽松友好的开源协议\n依旧无法掩盖一处无法回避的短板\n模型纯粹推理上线并不突出\n当下AI行业早已进入白热化竞争阶段\n各家机构哪怕基准测试出现小数点级别的分数差距\n都会专门撰写论文大肆宣传\n投入20亿美元资金\n最终产出一款激活参数410亿\n综合表现中等的模型\n在外人眼里无疑是一笔严重亏损的投资\n不少业内研究员研读论文\n实测模型之后给出的评价高度统一\n评语大多只有很酷两个字\n站在投资人视角\n花费巨额资金换来一句很酷\n和你邀请他人享用高端米其林晚餐\n最后只收到一句\n你人很好高度相似\n属于十分委婉的负面评价\n很多人会顺势产生疑问\n难道米拉团队已经将狼才进\n脱离大厂充足算力扶持之后\n这群天才工程师再也无法完成前沿模型研发\n答案显然是否定的\n这里藏着一个极异被大众忽略的反常识逻辑\n也是绝大多数行业观察者出现判断偏差的关键\n外界普遍默认研发大模型\n必须无休止开展算力军备竞赛\n追求更高参数 更强跑分\n谁的榜单成绩领先 谁就是最终赢家\n可这套竞争思路已经是两年之前的老旧玩法\n现如今想要依靠基准测试提升1%的准确率\n需要额外投入数亿美元电费 海量算力资源\n还要耗费大量人力筛选稀缺高质量训练数据\n边际成本已经膨胀到难以承受的地步。米拉在OpenAI长期生耕,比任何人都清楚算力成本核算逻辑。\n深刻明白B原大模型无休止堆参数换取微小性能提升,本身就是一条难以持续的死胡同。\n如果运营一家估值120亿美元的企业\n核心目标绝对不是\n打赢一场没有实际商业价值的跑分竞赛\n而是搭建一套能持续自我造血的商业体系\n所以当大家看见一群掌握顶尖AI底层认知的技术人才\n在关键节点推出一款看起来平平无奇的产品\n千万不要简单判定他们决策失误\n他们正在执行一套普通人难以看懂的长线布局\nInkling跑分表现相对平庸\n并不是研发能力不足\n而是精准的商业策略规划\nTai Mankey & G. McKean ES Lab\n从一开始就没有争夺\n全球最强通用大模型的想法\n把一款推理能力中等\n算力消耗低廉\n响应速度出众\n并且完全开源的模型\n免费对外开放\n仅仅是整套战略布局的第一步\n20亿美元的资金投入\n绝大部分运用在\nInkling反直觉的底层机制设计\n以及他们真正用来实现\n商业化变现的秘密武器之上\n倘若大家单纯认为\n这款无法冲进榜单前五的模型\n就是故事终点\n那就严重低估企业及弊端市场竞争的残酷程度\n为了更好落地商业化\nInkling甚至主动削减一部分自由联想能力\n只为持续降低单次调用产生的词源开销\n模型内部运行逻辑\n加上Thinking Machines Lab\n后续依靠估值变现的整套方案\n才是整个科技行业真正需要警惕的地方\n当我们拆解Inkling内部调节推理强度的核心机制\n就能意识到未来人工智能赛道的主战场\n根本不会集中在评测排行榜\n这款模型看似平庸的跑分\n是刻意控制之下形成的结果\n米拉团队在Inkling内部设置三处反直觉\n甚至有些激进的技术设定\n每一项都精准击中\n日军千万次AI调用企业客户的核心痛点\n第一项设定可以称之为智能调节推理力度旋钮\n官方命名为思考力度控制系统\n这项设计充满巧思\n市面上绝大多数主流大模型\n处理简单问答1加1等于计\n和撰写后端微服务架构代码\n后台占用的算力资源差距极小\n好比驾驶重型重载卡车\n仅仅只为前往楼下便利店采购物品\n在企业大规模批量调用场景之下\n这种资源浪费很难长期承受\n依托这套调节旋钮\n运行规则彻底改变\n当模型处理发票分类\n标准化客服问答这类重复性基础任务\n可以下调思考力度\n此时Inkling反馈速度极快\n调用成本压缩到极低水平\n基础答案完全能够满足业务需求\n一旦遭遇核心业务逻辑推演\n复杂代码修复\n高阶数学推理难题\n直接拉高思考力度档位\nInkling会调动更多算力进入深度推理模式\n综合推理效率能够对标行业第一梯队大模型\n同时算力消耗远低于同级产品\n在企业商用场景具备极强优势\n这意味着企业客户终于可以动态管控算力支出\n当后台每日运行数百万次智能体任务\n可以按需调整资源投入\n省下的每一笔开支都会直接转化为企业净利润\n第二项设定更加打破大众固有认知\n研发团队刻意训练模型形成诚实输出特性\n当下绝大多数大模型普遍存在幻觉问题\n无论面对何种提问都会强行生成完整答案\n哪怕内容完全不符合事实\n依旧会以笃定语气输出\n在文学创作\n日常闲聊场景\n这种自主生成能力\n可以称之为创造力\n放到金融预测\n医疗诊断这类企业场景\n毫无依据的编造内容\n会带来致命风险\n为此\nThinking Machines Lab\n引入认知学导向训练方案\n简单来讲\n利用强化学习奖励模型\n主动承认认知边界的行为\n团队持续迭代训练\n引导Inkling在信息不足\n无法得出可靠结论的时候\n不要强行猜测\n直接告知使用者无法解答\n最终训练成果十分突出\nInkling在事件预测\n数据分析类客观评测中\n准确率超越多款主流通用大模型\n达到行业顶尖水准\n对于华尔街对冲基金\n医疗行业大型机构而言\n一台关键时刻敢于坦诚信息不足的模型\n远比一位强行输出答案的通用模型更加可靠\n第三项设定带着一丝黑色幽默\n为极致压缩运行成本\n经过3000万轮强化学习迭代之后\nInkling自主演化出高度精简的内部思考语言\n我们可以简单对比\n差异主流模型处理复杂问题时\n内部推理链条会使用规范\n完整英文语句\n需要消耗大量词源资源\nInkling内部推演过程抛弃完整语法\n只保留核心关键词\n推理效果几乎没有衰减\n但是极大削减词源占用\n听起来十分有趣\n背后体现的是对推理效率极致压榨的工程思路\n读到这里大家就能明白,这从来不是一款用来冲击榜单排名的展示型模型,\n而是面向大规模企业商用量身打造,精打细算,稳定可靠的底层基础设施。\n即便Inclin拥有诸多精妙设计,仅仅依靠开源模型本身无法创造收入,\n能够支撑120亿估值完成变现的核心载体,是配套同步推出的Tinker自动化微调平台。\n这才是SY NTNG MacAINES Lab真正的商业核武器。\n前面我们提到\nInkling完全开源\n任何人都可以免费下载部署\n企业不会为此支付一分钱\n可是真实落地场景当中\n大型企业技术负责人拿到模型权重之后\n很快会遭遇现实阻碍\n想要让模型理解自身行业业务\n读懂内部财务数据\n适配客户服务场景\n必须使用企业私有数据开展微调\n模型微调门槛极高\n研发人员需要精通整套训练架构\n持续维护GPU算力集群\n还要应对训练进程当中频繁出现的程序崩溃问题\n绝大多数传统企业并不具备对应的技术团队\n这个时候Thinking Machines Lab就可以顺势推出解决方案\n向企业推荐Tinker平台\n模型免费交付\n但是微调操作门槛过高\n借助Tinker平台\n使用者不用理解底层框架\n不用管理算力节点\n直接上传私有业务数据\n简单操作就能全自动训练专属垂直领域专家模型\n在官方演示当中还有极具冲击力的功能\n平台下达指令之后\nInklin能够自主编写训练脚本\n自动生成配套训练数据集\n独立完成全部微调流程\n加在全新全重实现自我迭代\n达成全自动自主微调\n这正是米拉真正的野心\n她根本无意打造第二个OpenAI\n依靠售卖标准化按磁源计费的通用API盈利\n现如今基础大模型赛道早已红海厮杀\n通用模型数量众多\n不少开源产品在垂直场景\n性能超越必源方案\nThinking Machines Lab\n真实商业模式十分清晰\n依靠可控性强\n运行稳定\n完全免费的开源模型\n作为引流载体\n吸引大量想要搭建私有人工智能系统的企业\n随后依托简易自动化微调平台锁定客户\n持续收取平台服务费用\n他们不对外售卖通用智能大脑\n而是对外提供批量定制行业专家模型的生产工具\n未来倘若数万企业借助Tinker搭建专属财物\n客服务医疗智能模型\n没有人再会持续纠结\nInkling在评测榜单排名高低\n这笔20亿美元投入真的存在浪费吗\n答案是否定的\n米拉带领顶尖团队\n跳出OpenAI必源竞争的内卷赛道\n在企业定制人工智能赛道\n筑器极高护城河\n他们没有研发失败\n而是在下一盘足以重塑全球AI商业格局的大棋\n我们接着聊硅谷行业内部很少公开讨论的细节\nInkling这套高效混合专家\nMOE架构背后存在不少值得深究的线索\n资深AI架构师研读底层代码之后\n能够发现256个路由专业\n資深AI架構師研讀底層代碼之後,能夠發現256個路由專家單元,Sigmoid路由機制,滑動窗口注意力配比,整套工程設計思路和國內開源大模型深度求索,Deep-Seek V3高度相似。\n除此之外\n模型早期预训练阶段\nThinking Machines Lab大量\n运用成熟开源模型生成的\n合成数据辅助训练\n大幅降低训练成本与失措周期\n一家估值120亿美元\n纯正美国背景顶尖AI实验室\n由前OpenAI核心负责人戴队\n手握英伟达高端算力资源\n最终推出旗舰模型\n大量参考国内开源模型成熟的\n工程优化方案\n很多外行看到消息第一反应\n会质疑团队创新能力\n认为只是照搬现有成果\n站在工程与商业客观视角分析\n这是非常理性务实的选择\n最近几年海外高端算力芯片供给受限\n国内AI研发团队无法像OpenAI一样\n不计成本堆叠算力暴力训练模型\n只能持续在算法效率\n内存优化\n参数调度层面深耕\n倒逼开源模型运行效率突破全新高度\n米拉团队十分清楚这一点\n工程技术不存在面子之争\n既然同行已经验证高效率优化路线可行\n开源协议允许借鉴参考\n站在前人成果之上继续迭代\n就是最高效的选择\n也正是依托成熟的优化思路\n团队大幅压缩研发周期\n快速完成9750亿参数巨型模型的\n完整训练与迭代\n讲到这里\n大家会产生一个直击本质的疑问\n既然国内开源模型效率出众\n多款国产开源模型在代码\n逻辑推理层面\n能够对标国际顶级币源产品\n使用成本极其低廉\n美国本土企业为什么不直接选用现成国产开源模型\n何必耗费资金使用Inkling\n额外采购Tinker平台服务\n这个问题刚好戳破Thinking Machines Lab整套布局最核心的商业命脉\n事实上美国资本方并不缺乏理性\n在Inkling正式发布之前\n美国企业界曾经掀起一波使用海外开源模型的热炒\n查看OPE NRO UTR平台词源路由统计数据\n能够发现高峰期有大量企业调用流量\n转向海外开源模型\nCoinbase这类头部企业都悄悄迁移智能体业务\n直接把运营成本削减一半\n好用\n低价\n资本自然会用脚投票\n全球规模顶尖对冲基金桥水\n就曾利用自身业务数据微调国内开源模型\n在金融推理评测中跑出高分\n全面超越多款高价通用必源模型\n使用成本大幅降低\n低价定制垂直模型碾压昂贵通用大模型\n这笔账所有企业决策者都能够看懂\n但是繁荣周期并没有持续太久\n地缘政策带来的监管压力快速到来\n由于海外开源模型普及速度超出预期\n美国监管机构开始高度警惕\n持续向使用海外模型的企业发送警示信函\n国会相关委员会启动调研\n甚至有部门草拟采购限制条款\n手握海量隐私业务数据\n流水规模动辄百亿的华尔街投行\n医疗集团与科技大厂根本无法承受这类合规风险\n即便模型性能出色 价格低廉\n一旦被贴上数据安全隐患标签\n对于企业而言就是随时可能引爆的经营风险\n与此同时 曾经支撑西方开源生态的Meta\n近期持续收紧授权政策\n后续Lama系列模型逐步转向更加封闭的商业授权模式\n彻底终结了西方企业免费商用开源大模型的红利时代\n就在这个行业窗口期\n西方企业市场出现巨大需求真空\n企业迫切需要开源模型降低算力开销\n希望依靠微调保护私有数据\n但是不敢选用海外免费模型\n本土老牌开源路线又不断抬高准入门槛\n此刻再回看Inkling的发布时机\n一切都豁然开朗\nInkling本身不需要争夺全球最强模型的名号\n它的真实定位\n是当下西方企业市场为数不多\n合规安全\n纯正美国本土血统\n且综合性能达标的开源备选方案\n米拉团队不仅交付一套安全可控的底层模型\n还同步推出轻量化版本Inclean Small\n这款缩小版本总参数规模压缩至2760亿\n单次推理仅激活120亿参数\n轻量化架构适配中小企业硬件设备\n在专项垂直推理场景中表现亮眼\n专门面向有本地私有化部署需求\n服务器硬件资源有限的中小企业\n硬件配置不足可以选用轻量化版本\n缺少技术人员直接采购Tinker平台\n担忧跨境数据合规风险\n它又是本土正规团队开发\n代码完整公开托管在哈格面部\n全部流程透明可查\n把所有线索串联起来\nThinking Machines Lab完整商业版图清晰浮现\n离开OpenAI的MamalaD\n从一开始就没有打算在早已竞争白热化的\n通用大模型赛道\n复刻第二个ChatGPT\n那条赛道只能无休止消耗投资人资金持续内卷\n他依靠20亿美元资金\n在OpenAI币源商业帝国旁边\n开辟全新赛道\n精准预判人工智能未来形态\n不会是无所不能的通用超级大脑\n而是千千万万扎根细分行业\n低成本运行数据归属企业自身的领域专用模型\n借鉴经过市场验证的高效开源架构\n搭配团队成熟的成本控制方案打造Inkling\n作为引流载体\n再借助地缘环境催生的合规焦虑\n吸引急需解决方案的企业客户\n进入Tinker微调服务生态\n花费20亿美元打造一款榜单\n无法跻身前五的模型\n表面看上去像是一场失败投资\n但是倘若这20亿美元换来未来10年\n全球企业及AI定制市场核心入口\n这就是教科书级别的商业布局\n当所有人紧盯评测排行榜相互竞争的时候\n已经有人提前锁定人工智能淘金浪潮当中\n合规产子的独家供应权\n这也是120亿美元估值背后\n最值得细细思考的行业真相\n其实我们在延伸一层\n你就能看懂硅谷资本更深层的赌具逻辑\n近两年无数AI创业公司扎堆入场\n大多都在跟风做通用大模型\n做对话机器人\n做花哨的多模态演示\n真正能落地赚钱的寥寥无几\n绝大多数AI初创企业\n盈利模式极度单一\n要么卖API调用\n要么卖会员套餐\n陷入无尽的价格战和算力内卷\n稍万融资就倒闭\n存活率极低\n而SY NTNG Mac King ES Lab\n彻底避开了这片红海\n它不跟任何人抢C端用户\n不卷跑分\n不卷花式功能\n直接切入最赚钱\n最稳定\n刚需最强的弊端企业服务市场\n对于大型企业来说\nAI成本\n数据安全\n私有化部署\n定制化适配\n是比模型聪不聪明\n更重要的刚需\n哪怕通用能力平庸\n只要足够稳\n足够便宜\n足够安全\n足够好微调\n就是企业愿意\n长期付费的核心资产\n这也是为什么A16Z\n敢于在没有成熟产品\n没有实际营收的情况下\n豪至20亿美金\n资本赌的不是一款模型的输赢\n而是赌未来企业AI私有化\n定制化\n合规化的万亿级风口\n反观行业内很多盲目堆算力\n卷参数\n卷跑分的团队\n看似热度爆棚\n数据亮眼\n实则完全偏离了真实商业需求\n最终只能沦为\n昙花一现的行业过客\n除此之外\n米拉团队的布局\n还有一个极具远见的细节\n那就是彻底放弃\n避原锁库的传统玩法\n坚持全开源 全透明\n在当下AI舆论环境中\n避原模型随时面临授权变动\n服务关停 接口涨价\n功能阉割的风险\n企业用户极其没有安全感\n而Inkling的Apache 2.0协议\n给予了企业最大的自由度\n可商用 可二次开发\n可私有化部署 无板权捆绑\n这种极致开放的姿态\n刚好击中了当下企业最核心的诉求\n免费拿模型\n免费改模型\n免费部署\n最后只为精细化微调服务\n自动化运为工具付费\n这种先让利\n后所刻的商业模式\n远比一次性售卖API要长久\n稳固得多\n很多人始終想不通一個核心問題,為什麼無數大廠耗費巨資打造的頂級通用大模型,落地企業服務時頻頻翻車,反而是Inclin這種跑分中庸的模型,精準踩中了行業痛點。\n答案其实藏在企业真实的AI落地困境里\n绝大多数传统企业\n金融机构\n医疗集团\n制造业工厂\n根本不需要模型创作诗歌\n撰写文案\n完成趣味问答\n也不需要极致的顶尖推力跑分\n他们的核心需求极其朴素\n就是稳定低成本\n可管控\n不幻觉\n能对接自身业务数据\n同时全程保障数据\n不外泄\n过去几年\n大量企业跟风接入各类顶尖臂源大模型API\n看似快速完成了AI数字化升级\n实则埋下了无数隐患\n首先是持续走高的调用成本\n企业日均数万数十万次的重复性基础调用\n累积下来的算力开支\n每年动辄数百万甚至上千万\n长期投入完全没有上线\n其次是数据泄露风险\n企业内部的客户资料\n财务报表\n生产数据\n并立信息\n一旦上传至第三方B元模型云端服务器\n就彻底脱离了自身管控\n随时可能面临数据抓取\n缓存留存\n合规溯源的风险\n更关键的是\n通用大模型的过度聪明\n反而成为企业落地的累赘\n很多企业业务只需要标准化\n流程化的固定输出\n不需要模型自由发挥\n发散创作\n通用模型自带的高自由度推理能力\n不仅会增加算力损耗\n还容易输出不符合企业规章制度\n行业规范的内容\n需要人工二次审核修正\n反而降低了工作效率\n而Inclin的全套设计\n恰好针对性解决了这些行业顽疾\n可调节的思考力度杜绝算力浪费\n极致诚实的输出特性\n规避业务幻觉风险\n轻量化部署适配各类企业硬件条件\n完全开源的属性\n满足私有化部署需求\n再搭配Tinker全自动微调平台\n彻底抹平了企业AI落地的技术鸿沟\n哪怕是没有专业算法团队的传统企业\n也能轻松搭建专属行业AI系统\n这种精准贴合商业落地的产品逻辑\n也是它能够碾压一众跑分更强模型的核心底气\n除此之外\n我们还能从行业迭代周期看懂这套布局的前瞻性\n如今通用大模型的技术迭代已经进入瓶颈期\n性能提升的边际效益持续递减\n单纯比拼参数和算力已经没有实质商业意义\n未来AI行业的竞争早已从技术参数竞赛转向商业落地竞赛\n谁能更低成本更合规\n更便捷地帮助各行各业\n完成AI私有化\n定制化改造\n谁就能掌握行业话语权\n米拉团队放弃无意义的跑分内卷\n深耕企业服务赛道\n本质是提前预判了AI行业的终局形态\n放弃短期的流量热度\n锁定了长期的行业垄断地位\n看完这期视频\n你如何看待大模型一味比拼跑分的内卷现状\n又觉得未来企业AI定制服务\n会走向怎样的格局\n欢迎在评论区理性交流\n如果你觉得本期深度行业解读干货充足\n记得点赞并且订阅频道\n后续持续跟踪全球AI巨头的战略博弈\n不要错过更多前沿产业分析\n我们下期再见",
  "language": "zh",
  "duration": 1632.455688,
  "segments": [
    {
      "start": 0.0,
      "end": 28.54,
      "text": "如果你持续关注硅谷AI产业动态,一定听过这场堪称传奇的人才出走事件。手握20亿美元顶级风头,集结一批从OpenAI出走的核心研发人才。秘密打磨接近两年,所有人都默认他们会拿出一款能够横扫行业的顶尖大模型。可最终的结果却让整个科技圈大跌眼镜。在前不久的发布会结束之后,他们推出的全新模型在主流权威基准评测榜单当中甚至无法冲进前五。",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.03962820768356323,
      "no_speech_prob": 7.608256039071115e-12,
      "compression_ratio": 1.2574525745257452
    },
    {
      "start": 28.54,
      "end": 30.64,
      "text": "不少开源模型的综合表现",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0823630534956608,
      "no_speech_prob": 1.3161305378872612e-11,
      "compression_ratio": 1.2166666666666666
    },
    {
      "start": 30.64,
      "end": 32.24,
      "text": "都能够实现对它的压制",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0823630534956608,
      "no_speech_prob": 1.3161305378872612e-11,
      "compression_ratio": 1.2166666666666666
    },
    {
      "start": 32.24,
      "end": 34.5,
      "text": "回望2024年年底那场震动",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0823630534956608,
      "no_speech_prob": 1.3161305378872612e-11,
      "compression_ratio": 1.2166666666666666
    },
    {
      "start": 34.5,
      "end": 35.68,
      "text": "全球AI行业的",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0823630534956608,
      "no_speech_prob": 1.3161305378872612e-11,
      "compression_ratio": 1.2166666666666666
    },
    {
      "start": 35.68,
      "end": 37.24,
      "text": "OpenAI高管离职风波",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0823630534956608,
      "no_speech_prob": 1.3161305378872612e-11,
      "compression_ratio": 1.2166666666666666
    },
    {
      "start": 37.24,
      "end": 38.78,
      "text": "大家才慢慢意识到",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0823630534956608,
      "no_speech_prob": 1.3161305378872612e-11,
      "compression_ratio": 1.2166666666666666
    },
    {
      "start": 38.78,
      "end": 41.2,
      "text": "这根本不是简单的人员流动",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0823630534956608,
      "no_speech_prob": 1.3161305378872612e-11,
      "compression_ratio": 1.2166666666666666
    },
    {
      "start": 41.2,
      "end": 42.56,
      "text": "而是一场精心布局",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0823630534956608,
      "no_speech_prob": 1.3161305378872612e-11,
      "compression_ratio": 1.2166666666666666
    },
    {
      "start": 42.56,
      "end": 44.82,
      "text": "持续多年的人才收割行动",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0823630534956608,
      "no_speech_prob": 1.3161305378872612e-11,
      "compression_ratio": 1.2166666666666666
    },
    {
      "start": 44.82,
      "end": 45.78,
      "text": "米拉姆拉蒂",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0823630534956608,
      "no_speech_prob": 1.3161305378872612e-11,
      "compression_ratio": 1.2166666666666666
    },
    {
      "start": 45.78,
      "end": 48.12,
      "text": "也就是曾经短暂执掌OpenAI",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0823630534956608,
      "no_speech_prob": 1.3161305378872612e-11,
      "compression_ratio": 1.2166666666666666
    },
    {
      "start": 48.12,
      "end": 49.94,
      "text": "全程主导ChatGPT",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0823630534956608,
      "no_speech_prob": 1.3161305378872612e-11,
      "compression_ratio": 1.2166666666666666
    },
    {
      "start": 49.94,
      "end": 51.74,
      "text": "落地研发的核心关键人物",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0823630534956608,
      "no_speech_prob": 1.3161305378872612e-11,
      "compression_ratio": 1.2166666666666666
    },
    {
      "start": 51.74,
      "end": 54.46,
      "text": "在2024年9月正式离开老东家",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0823630534956608,
      "no_speech_prob": 1.3161305378872612e-11,
      "compression_ratio": 1.2166666666666666
    },
    {
      "start": 54.46,
      "end": 56.3,
      "text": "并且不是孤身一人",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0823630534956608,
      "no_speech_prob": 1.3161305378872612e-11,
      "compression_ratio": 1.2166666666666666
    },
    {
      "start": 56.3,
      "end": 58.32,
      "text": "他踏出OpenAI大门之后",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0823630534956608,
      "no_speech_prob": 1.3161305378872612e-11,
      "compression_ratio": 1.2166666666666666
    },
    {
      "start": 58.32,
      "end": 60.16,
      "text": "联合创始人约翰·舒尔曼",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06519466666288154,
      "no_speech_prob": 1.4563111133680806e-11,
      "compression_ratio": 1.1994301994301995
    },
    {
      "start": 60.16,
      "end": 62.38,
      "text": "前研究副总裁巴雷特·佐夫",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06519466666288154,
      "no_speech_prob": 1.4563111133680806e-11,
      "compression_ratio": 1.1994301994301995
    },
    {
      "start": 62.38,
      "end": 66.16,
      "text": "还有一大批资深核心研究员紧随其后递交辞呈",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06519466666288154,
      "no_speech_prob": 1.4563111133680806e-11,
      "compression_ratio": 1.1994301994301995
    },
    {
      "start": 66.16,
      "end": 67.8,
      "text": "这群人的集体出走",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06519466666288154,
      "no_speech_prob": 1.4563111133680806e-11,
      "compression_ratio": 1.1994301994301995
    },
    {
      "start": 67.8,
      "end": 71.12,
      "text": "几乎直接抽走了OpenAI半数核心研发力量",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06519466666288154,
      "no_speech_prob": 1.4563111133680806e-11,
      "compression_ratio": 1.1994301994301995
    },
    {
      "start": 71.12,
      "end": 73.18,
      "text": "在硅谷的产业生态里面",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06519466666288154,
      "no_speech_prob": 1.4563111133680806e-11,
      "compression_ratio": 1.1994301994301995
    },
    {
      "start": 73.18,
      "end": 75.5,
      "text": "顶尖人工智能研究员的稀缺程度",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06519466666288154,
      "no_speech_prob": 1.4563111133680806e-11,
      "compression_ratio": 1.1994301994301995
    },
    {
      "start": 75.5,
      "end": 78.84,
      "text": "甚至超过英伟达H100系列高端算力芯片",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06519466666288154,
      "no_speech_prob": 1.4563111133680806e-11,
      "compression_ratio": 1.1994301994301995
    },
    {
      "start": 78.84,
      "end": 81.96,
      "text": "米拉带走的远远不止数十名资深工程师",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06519466666288154,
      "no_speech_prob": 1.4563111133680806e-11,
      "compression_ratio": 1.1994301994301995
    },
    {
      "start": 81.96,
      "end": 86.06,
      "text": "更是一套经过GPT3到GPT4完整项目验证",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06519466666288154,
      "no_speech_prob": 1.4563111133680806e-11,
      "compression_ratio": 1.1994301994301995
    },
    {
      "start": 86.06,
      "end": 88.76,
      "text": "成熟可行的大模型训练方法论",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07410321553548177,
      "no_speech_prob": 1.4514257851150347e-11,
      "compression_ratio": 1.2768361581920904
    },
    {
      "start": 88.76,
      "end": 91.22,
      "text": "还有整套工程落地的实战经验",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07410321553548177,
      "no_speech_prob": 1.4514257851150347e-11,
      "compression_ratio": 1.2768361581920904
    },
    {
      "start": 91.22,
      "end": 93.06,
      "text": "这支团队亲眼见证过",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07410321553548177,
      "no_speech_prob": 1.4514257851150347e-11,
      "compression_ratio": 1.2768361581920904
    },
    {
      "start": 93.06,
      "end": 95.66,
      "text": "全球规模顶尖算力集群的运行逻辑",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07410321553548177,
      "no_speech_prob": 1.4514257851150347e-11,
      "compression_ratio": 1.2768361581920904
    },
    {
      "start": 95.66,
      "end": 97.42,
      "text": "也踩遍了训练过程里",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07410321553548177,
      "no_speech_prob": 1.4514257851150347e-11,
      "compression_ratio": 1.2768361581920904
    },
    {
      "start": 97.42,
      "end": 100.24,
      "text": "各类会引发模型崩溃的隐蔽技术陷阱",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07410321553548177,
      "no_speech_prob": 1.4514257851150347e-11,
      "compression_ratio": 1.2768361581920904
    },
    {
      "start": 100.24,
      "end": 102.86,
      "text": "因此当他们对外宣布独立创业的时候",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07410321553548177,
      "no_speech_prob": 1.4514257851150347e-11,
      "compression_ratio": 1.2768361581920904
    },
    {
      "start": 102.86,
      "end": 105.24,
      "text": "整个风险投资圈子陷入狂热",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07410321553548177,
      "no_speech_prob": 1.4514257851150347e-11,
      "compression_ratio": 1.2768361581920904
    },
    {
      "start": 105.24,
      "end": 107.46,
      "text": "硅谷风投机构最擅长算账",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07410321553548177,
      "no_speech_prob": 1.4514257851150347e-11,
      "compression_ratio": 1.2768361581920904
    },
    {
      "start": 107.46,
      "end": 110.3,
      "text": "也最容易陷入害怕错失机遇的情绪",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07410321553548177,
      "no_speech_prob": 1.4514257851150347e-11,
      "compression_ratio": 1.2768361581920904
    },
    {
      "start": 110.3,
      "end": 111.6,
      "text": "安德森霍洛维茨",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07410321553548177,
      "no_speech_prob": 1.4514257851150347e-11,
      "compression_ratio": 1.2768361581920904
    },
    {
      "start": 111.6,
      "end": 113.9,
      "text": "直接敲定20亿美元现金投入",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07410321553548177,
      "no_speech_prob": 1.4514257851150347e-11,
      "compression_ratio": 1.2768361581920904
    },
    {
      "start": 113.9,
      "end": 115.32,
      "text": "投资落地的时候",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08297526364946096,
      "no_speech_prob": 1.305051032518545e-11,
      "compression_ratio": 1.2095808383233533
    },
    {
      "start": 115.32,
      "end": 117.96,
      "text": "这家连正式名称都没有完全敲定",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08297526364946096,
      "no_speech_prob": 1.305051032518545e-11,
      "compression_ratio": 1.2095808383233533
    },
    {
      "start": 117.96,
      "end": 121.54,
      "text": "一行商用产品代码都还没有产出的初创团队",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08297526364946096,
      "no_speech_prob": 1.305051032518545e-11,
      "compression_ratio": 1.2095808383233533
    },
    {
      "start": 121.54,
      "end": 123.7,
      "text": "最终定名Thinking Machines Lab",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08297526364946096,
      "no_speech_prob": 1.305051032518545e-11,
      "compression_ratio": 1.2095808383233533
    },
    {
      "start": 123.7,
      "end": 126.18,
      "text": "投后估值直接冲到120亿美元",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08297526364946096,
      "no_speech_prob": 1.305051032518545e-11,
      "compression_ratio": 1.2095808383233533
    },
    {
      "start": 126.18,
      "end": 130.02,
      "text": "没有任何成熟产品就能拥有120亿美元估值",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08297526364946096,
      "no_speech_prob": 1.305051032518545e-11,
      "compression_ratio": 1.2095808383233533
    },
    {
      "start": 130.02,
      "end": 133.76,
      "text": "放在其他行业很容易被贴上庞氏骗局的标签",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08297526364946096,
      "no_speech_prob": 1.305051032518545e-11,
      "compression_ratio": 1.2095808383233533
    },
    {
      "start": 133.76,
      "end": 136.14,
      "text": "但是在近几年火热的AI赛道",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08297526364946096,
      "no_speech_prob": 1.305051032518545e-11,
      "compression_ratio": 1.2095808383233533
    },
    {
      "start": 136.14,
      "end": 140.36,
      "text": "资本把这称作抢占通往通用人工智能时代最后的传票",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08297526364946096,
      "no_speech_prob": 1.305051032518545e-11,
      "compression_ratio": 1.2095808383233533
    },
    {
      "start": 140.36,
      "end": 142.46,
      "text": "风投机构押注的核心目标",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09427797105577257,
      "no_speech_prob": 1.3255478312212965e-11,
      "compression_ratio": 1.1886304909560723
    },
    {
      "start": 142.46,
      "end": 144.4,
      "text": "就是米拉带领的这支团队",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09427797105577257,
      "no_speech_prob": 1.3255478312212965e-11,
      "compression_ratio": 1.1886304909560723
    },
    {
      "start": 144.4,
      "end": 145.72,
      "text": "能够打造出一套",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09427797105577257,
      "no_speech_prob": 1.3255478312212965e-11,
      "compression_ratio": 1.1886304909560723
    },
    {
      "start": 145.72,
      "end": 147.98,
      "text": "打破OpenAI避原模式垄断的",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09427797105577257,
      "no_speech_prob": 1.3255478312212965e-11,
      "compression_ratio": 1.1886304909560723
    },
    {
      "start": 147.98,
      "end": 148.96,
      "text": "全新产业生态",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09427797105577257,
      "no_speech_prob": 1.3255478312212965e-11,
      "compression_ratio": 1.1886304909560723
    },
    {
      "start": 148.96,
      "end": 150.5,
      "text": "所有人静静观望",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09427797105577257,
      "no_speech_prob": 1.3255478312212965e-11,
      "compression_ratio": 1.1886304909560723
    },
    {
      "start": 150.5,
      "end": 152.38,
      "text": "期待着20亿美元资金",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09427797105577257,
      "no_speech_prob": 1.3255478312212965e-11,
      "compression_ratio": 1.1886304909560723
    },
    {
      "start": 152.38,
      "end": 154.28,
      "text": "两年时间能够孵化出",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09427797105577257,
      "no_speech_prob": 1.3255478312212965e-11,
      "compression_ratio": 1.1886304909560723
    },
    {
      "start": 154.28,
      "end": 156.64,
      "text": "怎样惊世骇俗的人工智能产品",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09427797105577257,
      "no_speech_prob": 1.3255478312212965e-11,
      "compression_ratio": 1.1886304909560723
    },
    {
      "start": 156.64,
      "end": 158.44,
      "text": "整个行业秉析宁神",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09427797105577257,
      "no_speech_prob": 1.3255478312212965e-11,
      "compression_ratio": 1.1886304909560723
    },
    {
      "start": 158.44,
      "end": 160.58,
      "text": "等待新一代顶级大模型的",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09427797105577257,
      "no_speech_prob": 1.3255478312212965e-11,
      "compression_ratio": 1.1886304909560723
    },
    {
      "start": 160.58,
      "end": 161.58,
      "text": "技术降维打击",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09427797105577257,
      "no_speech_prob": 1.3255478312212965e-11,
      "compression_ratio": 1.1886304909560723
    },
    {
      "start": 161.58,
      "end": 163.98,
      "text": "时间来到2026年7月中旬",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09427797105577257,
      "no_speech_prob": 1.3255478312212965e-11,
      "compression_ratio": 1.1886304909560723
    },
    {
      "start": 163.98,
      "end": 165.06,
      "text": "也就是近期",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09427797105577257,
      "no_speech_prob": 1.3255478312212965e-11,
      "compression_ratio": 1.1886304909560723
    },
    {
      "start": 165.06,
      "end": 166.36,
      "text": "沉寂一年半之久的",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09427797105577257,
      "no_speech_prob": 1.3255478312212965e-11,
      "compression_ratio": 1.1886304909560723
    },
    {
      "start": 166.36,
      "end": 167.42,
      "text": "Thinking Machines Lab",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09427797105577257,
      "no_speech_prob": 1.3255478312212965e-11,
      "compression_ratio": 1.1886304909560723
    },
    {
      "start": 167.42,
      "end": 170.14,
      "text": "终于对外公开酝酿许久的重磅成果",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0921341391170726,
      "no_speech_prob": 1.428146229748295e-11,
      "compression_ratio": 1.2075471698113207
    },
    {
      "start": 170.14,
      "end": 172.26,
      "text": "开源基础模型代号Inklin",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0921341391170726,
      "no_speech_prob": 1.428146229748295e-11,
      "compression_ratio": 1.2075471698113207
    },
    {
      "start": 172.26,
      "end": 174.02,
      "text": "伴随着发布会正式落幕",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0921341391170726,
      "no_speech_prob": 1.428146229748295e-11,
      "compression_ratio": 1.2075471698113207
    },
    {
      "start": 174.02,
      "end": 176.76,
      "text": "整个科技行业陷入巨大的错愕",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0921341391170726,
      "no_speech_prob": 1.428146229748295e-11,
      "compression_ratio": 1.2075471698113207
    },
    {
      "start": 176.76,
      "end": 179.86,
      "text": "这款投入重金闭关研发两年打造的模型",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0921341391170726,
      "no_speech_prob": 1.428146229748295e-11,
      "compression_ratio": 1.2075471698113207
    },
    {
      "start": 179.86,
      "end": 183.64,
      "text": "在各类权威基准测试当中成绩只能用平庸形容",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0921341391170726,
      "no_speech_prob": 1.428146229748295e-11,
      "compression_ratio": 1.2075471698113207
    },
    {
      "start": 183.64,
      "end": 185.36,
      "text": "前方有克劳德·菲波5",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0921341391170726,
      "no_speech_prob": 1.428146229748295e-11,
      "compression_ratio": 1.2075471698113207
    },
    {
      "start": 185.36,
      "end": 188.88,
      "text": "OpenAI主流避原模型两座顶尖避原模型",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0921341391170726,
      "no_speech_prob": 1.428146229748295e-11,
      "compression_ratio": 1.2075471698113207
    },
    {
      "start": 188.88,
      "end": 190.06,
      "text": "稳定占据高位",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0921341391170726,
      "no_speech_prob": 1.428146229748295e-11,
      "compression_ratio": 1.2075471698113207
    },
    {
      "start": 190.06,
      "end": 192.42,
      "text": "同期发布的月之暗面KIMI K3",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0921341391170726,
      "no_speech_prob": 1.428146229748295e-11,
      "compression_ratio": 1.2075471698113207
    },
    {
      "start": 192.42,
      "end": 194.74,
      "text": "这款拥有2.8万亿参数的模型",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0921341391170726,
      "no_speech_prob": 1.428146229748295e-11,
      "compression_ratio": 1.2075471698113207
    },
    {
      "start": 194.74,
      "end": 196.64,
      "text": "同样能够和它正面抗衡",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0921341391170726,
      "no_speech_prob": 1.428146229748295e-11,
      "compression_ratio": 1.2075471698113207
    },
    {
      "start": 196.64,
      "end": 198.48,
      "text": "多项指标实现反超",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08005795146143714,
      "no_speech_prob": 1.4286044569544742e-11,
      "compression_ratio": 1.2727272727272727
    },
    {
      "start": 198.48,
      "end": 200.72,
      "text": "Inkling在开源模型梯队当中",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08005795146143714,
      "no_speech_prob": 1.4286044569544742e-11,
      "compression_ratio": 1.2727272727272727
    },
    {
      "start": 200.72,
      "end": 203.26,
      "text": "综合水平最多只能算作中油水准",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08005795146143714,
      "no_speech_prob": 1.4286044569544742e-11,
      "compression_ratio": 1.2727272727272727
    },
    {
      "start": 203.26,
      "end": 205.86,
      "text": "更加尴尬的还有发布时间窗口",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08005795146143714,
      "no_speech_prob": 1.4286044569544742e-11,
      "compression_ratio": 1.2727272727272727
    },
    {
      "start": 205.86,
      "end": 208.46,
      "text": "Inkling正式上线仅仅一天之后",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08005795146143714,
      "no_speech_prob": 1.4286044569544742e-11,
      "compression_ratio": 1.2727272727272727
    },
    {
      "start": 208.46,
      "end": 210.44,
      "text": "Kimi K3热度全面爆发",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08005795146143714,
      "no_speech_prob": 1.4286044569544742e-11,
      "compression_ratio": 1.2727272727272727
    },
    {
      "start": 210.44,
      "end": 212.12,
      "text": "大量开发者涌入调用",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08005795146143714,
      "no_speech_prob": 1.4286044569544742e-11,
      "compression_ratio": 1.2727272727272727
    },
    {
      "start": 212.12,
      "end": 213.8,
      "text": "直接把算力资源耗尽",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08005795146143714,
      "no_speech_prob": 1.4286044569544742e-11,
      "compression_ratio": 1.2727272727272727
    },
    {
      "start": 213.8,
      "end": 216.56,
      "text": "几乎抢走了Inkling全部公众关注度",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08005795146143714,
      "no_speech_prob": 1.4286044569544742e-11,
      "compression_ratio": 1.2727272727272727
    },
    {
      "start": 216.56,
      "end": 218.0,
      "text": "普通观众看到这里",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08005795146143714,
      "no_speech_prob": 1.4286044569544742e-11,
      "compression_ratio": 1.2727272727272727
    },
    {
      "start": 218.0,
      "end": 220.98,
      "text": "大概率都会认定这是一场彻头彻尾的失败",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08005795146143714,
      "no_speech_prob": 1.4286044569544742e-11,
      "compression_ratio": 1.2727272727272727
    },
    {
      "start": 220.98,
      "end": 221.96,
      "text": "顶级团队",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08005795146143714,
      "no_speech_prob": 1.4286044569544742e-11,
      "compression_ratio": 1.2727272727272727
    },
    {
      "start": 221.96,
      "end": 222.78,
      "text": "顶级资金",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08005795146143714,
      "no_speech_prob": 1.4286044569544742e-11,
      "compression_ratio": 1.2727272727272727
    },
    {
      "start": 222.78,
      "end": 223.62,
      "text": "顶级资源",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08005795146143714,
      "no_speech_prob": 1.4286044569544742e-11,
      "compression_ratio": 1.2727272727272727
    },
    {
      "start": 223.62,
      "end": 225.74,
      "text": "最后打出一张平平无奇的牌",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08005795146143714,
      "no_speech_prob": 1.4286044569544742e-11,
      "compression_ratio": 1.2727272727272727
    },
    {
      "start": 225.74,
      "end": 228.1,
      "text": "甚至被后发竞品全面碾压",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06426919389654089,
      "no_speech_prob": 1.1981481778944314e-11,
      "compression_ratio": 1.302325581395349
    },
    {
      "start": 228.1,
      "end": 230.7,
      "text": "但如果你只看到表面的跑分输赢",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06426919389654089,
      "no_speech_prob": 1.1981481778944314e-11,
      "compression_ratio": 1.302325581395349
    },
    {
      "start": 230.7,
      "end": 234.18,
      "text": "完全看不懂当下AI行业真正的内卷困局",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06426919389654089,
      "no_speech_prob": 1.1981481778944314e-11,
      "compression_ratio": 1.302325581395349
    },
    {
      "start": 234.18,
      "end": 237.36,
      "text": "更看不懂硅谷顶级资本真正的下注逻辑",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06426919389654089,
      "no_speech_prob": 1.1981481778944314e-11,
      "compression_ratio": 1.302325581395349
    },
    {
      "start": 237.36,
      "end": 239.68,
      "text": "现在整个AI行业已经陷入一种",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06426919389654089,
      "no_speech_prob": 1.1981481778944314e-11,
      "compression_ratio": 1.302325581395349
    },
    {
      "start": 239.68,
      "end": 241.34,
      "text": "极度畸形的内卷怪圈",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06426919389654089,
      "no_speech_prob": 1.1981481778944314e-11,
      "compression_ratio": 1.302325581395349
    },
    {
      "start": 241.34,
      "end": 243.9,
      "text": "绝大多数团队、媒体、开发者",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06426919389654089,
      "no_speech_prob": 1.1981481778944314e-11,
      "compression_ratio": 1.302325581395349
    },
    {
      "start": 243.9,
      "end": 246.38,
      "text": "都在疯狂迷信基准测试分数",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06426919389654089,
      "no_speech_prob": 1.1981481778944314e-11,
      "compression_ratio": 1.302325581395349
    },
    {
      "start": 246.38,
      "end": 249.26,
      "text": "把榜单排名当成唯一的评判标准",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06426919389654089,
      "no_speech_prob": 1.1981481778944314e-11,
      "compression_ratio": 1.302325581395349
    },
    {
      "start": 249.26,
      "end": 251.34,
      "text": "为了零点几分的微弱提升",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06426919389654089,
      "no_speech_prob": 1.1981481778944314e-11,
      "compression_ratio": 1.302325581395349
    },
    {
      "start": 251.34,
      "end": 254.04,
      "text": "无数公司动辄烧数亿美金算力",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06426919389654089,
      "no_speech_prob": 1.1981481778944314e-11,
      "compression_ratio": 1.302325581395349
    },
    {
      "start": 254.04,
      "end": 255.42,
      "text": "堆砌海量参数",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06426919389654089,
      "no_speech_prob": 1.1981481778944314e-11,
      "compression_ratio": 1.302325581395349
    },
    {
      "start": 255.42,
      "end": 256.8,
      "text": "清洗冗余数据",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06418798511154986,
      "no_speech_prob": 1.3892189482112016e-11,
      "compression_ratio": 1.3190184049079754
    },
    {
      "start": 256.8,
      "end": 258.22,
      "text": "反复微调权重",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06418798511154986,
      "no_speech_prob": 1.3892189482112016e-11,
      "compression_ratio": 1.3190184049079754
    },
    {
      "start": 258.22,
      "end": 260.82,
      "text": "最后做出来的模型跑分看似好看",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06418798511154986,
      "no_speech_prob": 1.3892189482112016e-11,
      "compression_ratio": 1.3190184049079754
    },
    {
      "start": 260.82,
      "end": 262.3,
      "text": "但是商用成本极高",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06418798511154986,
      "no_speech_prob": 1.3892189482112016e-11,
      "compression_ratio": 1.3190184049079754
    },
    {
      "start": 262.3,
      "end": 263.88,
      "text": "落地兼容性极差",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06418798511154986,
      "no_speech_prob": 1.3892189482112016e-11,
      "compression_ratio": 1.3190184049079754
    },
    {
      "start": 263.88,
      "end": 265.58,
      "text": "企业部署难度巨大",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06418798511154986,
      "no_speech_prob": 1.3892189482112016e-11,
      "compression_ratio": 1.3190184049079754
    },
    {
      "start": 265.58,
      "end": 267.4,
      "text": "完全是为了跑分而生",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06418798511154986,
      "no_speech_prob": 1.3892189482112016e-11,
      "compression_ratio": 1.3190184049079754
    },
    {
      "start": 267.4,
      "end": 269.48,
      "text": "而非为了商用而生的产品",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06418798511154986,
      "no_speech_prob": 1.3892189482112016e-11,
      "compression_ratio": 1.3190184049079754
    },
    {
      "start": 269.48,
      "end": 271.04,
      "text": "这也是最近一年来",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06418798511154986,
      "no_speech_prob": 1.3892189482112016e-11,
      "compression_ratio": 1.3190184049079754
    },
    {
      "start": 271.04,
      "end": 273.24,
      "text": "无数顶级AI模型高开低走",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06418798511154986,
      "no_speech_prob": 1.3892189482112016e-11,
      "compression_ratio": 1.3190184049079754
    },
    {
      "start": 273.24,
      "end": 275.92,
      "text": "热度爆红却无法变现的核心原因",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06418798511154986,
      "no_speech_prob": 1.3892189482112016e-11,
      "compression_ratio": 1.3190184049079754
    },
    {
      "start": 275.92,
      "end": 278.26,
      "text": "行业早已不再缺聪明的模型",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06418798511154986,
      "no_speech_prob": 1.3892189482112016e-11,
      "compression_ratio": 1.3190184049079754
    },
    {
      "start": 278.26,
      "end": 279.64,
      "text": "企业真正缺的",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06418798511154986,
      "no_speech_prob": 1.3892189482112016e-11,
      "compression_ratio": 1.3190184049079754
    },
    {
      "start": 279.64,
      "end": 281.66,
      "text": "是稳定便宜可控能落地",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06418798511154986,
      "no_speech_prob": 1.3892189482112016e-11,
      "compression_ratio": 1.3190184049079754
    },
    {
      "start": 281.66,
      "end": 284.06,
      "text": "零合规风险的工业化AI基础设施",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06418798511154986,
      "no_speech_prob": 1.3892189482112016e-11,
      "compression_ratio": 1.3190184049079754
    },
    {
      "start": 284.06,
      "end": 286.96,
      "text": "而这恰恰就是Inkling精准卡位的赛道",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06629844985083136,
      "no_speech_prob": 1.152986907337894e-11,
      "compression_ratio": 1.2155172413793103
    },
    {
      "start": 286.96,
      "end": 289.66,
      "text": "大家一定要点赞订阅锁住频道",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06629844985083136,
      "no_speech_prob": 1.152986907337894e-11,
      "compression_ratio": 1.2155172413793103
    },
    {
      "start": 289.66,
      "end": 293.48,
      "text": "接下来我逐层拆解这场瞒天过海的AI商业布局",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06629844985083136,
      "no_speech_prob": 1.152986907337894e-11,
      "compression_ratio": 1.2155172413793103
    },
    {
      "start": 293.48,
      "end": 294.84,
      "text": "看懂了你就明白",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06629844985083136,
      "no_speech_prob": 1.152986907337894e-11,
      "compression_ratio": 1.2155172413793103
    },
    {
      "start": 294.84,
      "end": 296.24,
      "text": "这根本不是翻车",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06629844985083136,
      "no_speech_prob": 1.152986907337894e-11,
      "compression_ratio": 1.2155172413793103
    },
    {
      "start": 296.24,
      "end": 299.24,
      "text": "而是近两年最顶级的AI创业降位布局",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06629844985083136,
      "no_speech_prob": 1.152986907337894e-11,
      "compression_ratio": 1.2155172413793103
    },
    {
      "start": 299.24,
      "end": 300.88,
      "text": "我们不妨深入拆解",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06629844985083136,
      "no_speech_prob": 1.152986907337894e-11,
      "compression_ratio": 1.2155172413793103
    },
    {
      "start": 300.88,
      "end": 305.52,
      "text": "这笔20亿美元资金孵化出来的技术底子究竟具备哪些特点",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06629844985083136,
      "no_speech_prob": 1.152986907337894e-11,
      "compression_ratio": 1.2155172413793103
    },
    {
      "start": 305.52,
      "end": 310.24,
      "text": "Inkling属于总参数量达到9750亿的混合专家模型",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06629844985083136,
      "no_speech_prob": 1.152986907337894e-11,
      "compression_ratio": 1.2155172413793103
    },
    {
      "start": 310.24,
      "end": 312.46,
      "text": "也就是业内常说的MOE架构",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06629844985083136,
      "no_speech_prob": 1.152986907337894e-11,
      "compression_ratio": 1.2155172413793103
    },
    {
      "start": 312.46,
      "end": 314.86,
      "text": "单单看总参数规模接近万亿",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06451689914504499,
      "no_speech_prob": 1.1894935557366093e-11,
      "compression_ratio": 1.2566371681415929
    },
    {
      "start": 314.86,
      "end": 316.44,
      "text": "听起来极具冲击力",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06451689914504499,
      "no_speech_prob": 1.1894935557366093e-11,
      "compression_ratio": 1.2566371681415929
    },
    {
      "start": 316.44,
      "end": 318.92,
      "text": "但是熟悉稀疏架构的从业者",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06451689914504499,
      "no_speech_prob": 1.1894935557366093e-11,
      "compression_ratio": 1.2566371681415929
    },
    {
      "start": 318.92,
      "end": 321.18,
      "text": "都清楚MOE底层运行逻辑",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06451689914504499,
      "no_speech_prob": 1.1894935557366093e-11,
      "compression_ratio": 1.2566371681415929
    },
    {
      "start": 321.18,
      "end": 323.3,
      "text": "每一次处理词源信息的时候",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06451689914504499,
      "no_speech_prob": 1.1894935557366093e-11,
      "compression_ratio": 1.2566371681415929
    },
    {
      "start": 323.3,
      "end": 326.5,
      "text": "仅仅只有410亿参数被激活参与运算",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06451689914504499,
      "no_speech_prob": 1.1894935557366093e-11,
      "compression_ratio": 1.2566371681415929
    },
    {
      "start": 326.5,
      "end": 329.46,
      "text": "模型原生支持100万长度上下文窗口",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06451689914504499,
      "no_speech_prob": 1.1894935557366093e-11,
      "compression_ratio": 1.2566371681415929
    },
    {
      "start": 329.46,
      "end": 331.76,
      "text": "依托45万亿词源规模的文本",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06451689914504499,
      "no_speech_prob": 1.1894935557366093e-11,
      "compression_ratio": 1.2566371681415929
    },
    {
      "start": 331.76,
      "end": 333.68,
      "text": "图像 音频混合数据",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06451689914504499,
      "no_speech_prob": 1.1894935557366093e-11,
      "compression_ratio": 1.2566371681415929
    },
    {
      "start": 333.68,
      "end": 335.52,
      "text": "完成从零开始的预训链",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06451689914504499,
      "no_speech_prob": 1.1894935557366093e-11,
      "compression_ratio": 1.2566371681415929
    },
    {
      "start": 335.52,
      "end": 338.16,
      "text": "整套体系当中最具备革新意义的",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06451689914504499,
      "no_speech_prob": 1.1894935557366093e-11,
      "compression_ratio": 1.2566371681415929
    },
    {
      "start": 338.16,
      "end": 340.8,
      "text": "是它重构了多模态信息处理逻辑",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06451689914504499,
      "no_speech_prob": 1.1894935557366093e-11,
      "compression_ratio": 1.2566371681415929
    },
    {
      "start": 340.8,
      "end": 344.76,
      "text": "目前市面上绝大多数宣称原生多模态的大模型",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06734864631395661,
      "no_speech_prob": 1.2364038612378003e-11,
      "compression_ratio": 1.236760124610592
    },
    {
      "start": 344.76,
      "end": 348.82,
      "text": "本质只是在基础语言模型外部附加一套处理组件",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06734864631395661,
      "no_speech_prob": 1.2364038612378003e-11,
      "compression_ratio": 1.236760124610592
    },
    {
      "start": 348.82,
      "end": 350.78,
      "text": "遇到图片、音频信息",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06734864631395661,
      "no_speech_prob": 1.2364038612378003e-11,
      "compression_ratio": 1.236760124610592
    },
    {
      "start": 350.78,
      "end": 352.18,
      "text": "需要独立编码器",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06734864631395661,
      "no_speech_prob": 1.2364038612378003e-11,
      "compression_ratio": 1.236760124610592
    },
    {
      "start": 352.18,
      "end": 356.58,
      "text": "先把视觉、听觉信号转化为模型能够识别的次元格式",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06734864631395661,
      "no_speech_prob": 1.2364038612378003e-11,
      "compression_ratio": 1.236760124610592
    },
    {
      "start": 356.58,
      "end": 358.46,
      "text": "再送入主体网络运算",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06734864631395661,
      "no_speech_prob": 1.2364038612378003e-11,
      "compression_ratio": 1.236760124610592
    },
    {
      "start": 358.46,
      "end": 360.78,
      "text": "而Inkling跳过中间转化流程",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06734864631395661,
      "no_speech_prob": 1.2364038612378003e-11,
      "compression_ratio": 1.236760124610592
    },
    {
      "start": 360.78,
      "end": 364.26,
      "text": "可以直接读取原始像素与音频波形信号",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06734864631395661,
      "no_speech_prob": 1.2364038612378003e-11,
      "compression_ratio": 1.236760124610592
    },
    {
      "start": 364.26,
      "end": 367.16,
      "text": "这种底层架构重构的工程难度极高",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06734864631395661,
      "no_speech_prob": 1.2364038612378003e-11,
      "compression_ratio": 1.236760124610592
    },
    {
      "start": 367.16,
      "end": 371.32,
      "text": "也直观体现出这支前OpenAI团队扎实的技术积累",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06643996575866083,
      "no_speech_prob": 1.1687622224199057e-11,
      "compression_ratio": 1.1779661016949152
    },
    {
      "start": 371.32,
      "end": 374.76,
      "text": "对比Meta很多开源模型饱受争议的授权条款",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06643996575866083,
      "no_speech_prob": 1.1687622224199057e-11,
      "compression_ratio": 1.1779661016949152
    },
    {
      "start": 374.76,
      "end": 376.16,
      "text": "Thinking Machines Lab",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06643996575866083,
      "no_speech_prob": 1.1687622224199057e-11,
      "compression_ratio": 1.1779661016949152
    },
    {
      "start": 376.16,
      "end": 377.76,
      "text": "这次选择十分开放",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06643996575866083,
      "no_speech_prob": 1.1687622224199057e-11,
      "compression_ratio": 1.1779661016949152
    },
    {
      "start": 377.76,
      "end": 380.36,
      "text": "直接采用Apache 2.0开源协议",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06643996575866083,
      "no_speech_prob": 1.1687622224199057e-11,
      "compression_ratio": 1.1779661016949152
    },
    {
      "start": 380.36,
      "end": 381.98,
      "text": "没有附加繁杂限制",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06643996575866083,
      "no_speech_prob": 1.1687622224199057e-11,
      "compression_ratio": 1.1779661016949152
    },
    {
      "start": 381.98,
      "end": 385.22,
      "text": "完整模型权重直接上传至哈格面部平台",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06643996575866083,
      "no_speech_prob": 1.1687622224199057e-11,
      "compression_ratio": 1.1779661016949152
    },
    {
      "start": 385.22,
      "end": 387.08,
      "text": "开发者可以自由修改",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06643996575866083,
      "no_speech_prob": 1.1687622224199057e-11,
      "compression_ratio": 1.1779661016949152
    },
    {
      "start": 387.08,
      "end": 388.48,
      "text": "部署 商用",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06643996575866083,
      "no_speech_prob": 1.1687622224199057e-11,
      "compression_ratio": 1.1779661016949152
    },
    {
      "start": 388.48,
      "end": 390.5,
      "text": "即便拥有底层架构创新",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06643996575866083,
      "no_speech_prob": 1.1687622224199057e-11,
      "compression_ratio": 1.1779661016949152
    },
    {
      "start": 390.5,
      "end": 392.64,
      "text": "搭配宽松友好的开源协议",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06643996575866083,
      "no_speech_prob": 1.1687622224199057e-11,
      "compression_ratio": 1.1779661016949152
    },
    {
      "start": 392.64,
      "end": 395.5,
      "text": "依旧无法掩盖一处无法回避的短板",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06643996575866083,
      "no_speech_prob": 1.1687622224199057e-11,
      "compression_ratio": 1.1779661016949152
    },
    {
      "start": 395.5,
      "end": 398.04,
      "text": "模型纯粹推理上线并不突出",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06267819710827749,
      "no_speech_prob": 1.2032709030551647e-11,
      "compression_ratio": 1.237142857142857
    },
    {
      "start": 398.04,
      "end": 401.06,
      "text": "当下AI行业早已进入白热化竞争阶段",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06267819710827749,
      "no_speech_prob": 1.2032709030551647e-11,
      "compression_ratio": 1.237142857142857
    },
    {
      "start": 401.06,
      "end": 405.36,
      "text": "各家机构哪怕基准测试出现小数点级别的分数差距",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06267819710827749,
      "no_speech_prob": 1.2032709030551647e-11,
      "compression_ratio": 1.237142857142857
    },
    {
      "start": 405.36,
      "end": 407.98,
      "text": "都会专门撰写论文大肆宣传",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06267819710827749,
      "no_speech_prob": 1.2032709030551647e-11,
      "compression_ratio": 1.237142857142857
    },
    {
      "start": 407.98,
      "end": 409.52,
      "text": "投入20亿美元资金",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06267819710827749,
      "no_speech_prob": 1.2032709030551647e-11,
      "compression_ratio": 1.237142857142857
    },
    {
      "start": 409.52,
      "end": 412.34,
      "text": "最终产出一款激活参数410亿",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06267819710827749,
      "no_speech_prob": 1.2032709030551647e-11,
      "compression_ratio": 1.237142857142857
    },
    {
      "start": 412.34,
      "end": 414.3,
      "text": "综合表现中等的模型",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06267819710827749,
      "no_speech_prob": 1.2032709030551647e-11,
      "compression_ratio": 1.237142857142857
    },
    {
      "start": 414.3,
      "end": 417.46,
      "text": "在外人眼里无疑是一笔严重亏损的投资",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06267819710827749,
      "no_speech_prob": 1.2032709030551647e-11,
      "compression_ratio": 1.237142857142857
    },
    {
      "start": 417.46,
      "end": 419.74,
      "text": "不少业内研究员研读论文",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06267819710827749,
      "no_speech_prob": 1.2032709030551647e-11,
      "compression_ratio": 1.237142857142857
    },
    {
      "start": 419.74,
      "end": 422.54,
      "text": "实测模型之后给出的评价高度统一",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06267819710827749,
      "no_speech_prob": 1.2032709030551647e-11,
      "compression_ratio": 1.237142857142857
    },
    {
      "start": 422.54,
      "end": 424.76,
      "text": "评语大多只有很酷两个字",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06267819710827749,
      "no_speech_prob": 1.2032709030551647e-11,
      "compression_ratio": 1.237142857142857
    },
    {
      "start": 424.76,
      "end": 426.26,
      "text": "站在投资人视角",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07595833261214678,
      "no_speech_prob": 1.5999000735344993e-11,
      "compression_ratio": 1.238938053097345
    },
    {
      "start": 426.26,
      "end": 428.52,
      "text": "花费巨额资金换来一句很酷",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07595833261214678,
      "no_speech_prob": 1.5999000735344993e-11,
      "compression_ratio": 1.238938053097345
    },
    {
      "start": 428.52,
      "end": 431.16,
      "text": "和你邀请他人享用高端米其林晚餐",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07595833261214678,
      "no_speech_prob": 1.5999000735344993e-11,
      "compression_ratio": 1.238938053097345
    },
    {
      "start": 431.16,
      "end": 432.58,
      "text": "最后只收到一句",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07595833261214678,
      "no_speech_prob": 1.5999000735344993e-11,
      "compression_ratio": 1.238938053097345
    },
    {
      "start": 432.58,
      "end": 434.12,
      "text": "你人很好高度相似",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07595833261214678,
      "no_speech_prob": 1.5999000735344993e-11,
      "compression_ratio": 1.238938053097345
    },
    {
      "start": 434.12,
      "end": 436.28,
      "text": "属于十分委婉的负面评价",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07595833261214678,
      "no_speech_prob": 1.5999000735344993e-11,
      "compression_ratio": 1.238938053097345
    },
    {
      "start": 436.28,
      "end": 438.14,
      "text": "很多人会顺势产生疑问",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07595833261214678,
      "no_speech_prob": 1.5999000735344993e-11,
      "compression_ratio": 1.238938053097345
    },
    {
      "start": 438.14,
      "end": 440.64,
      "text": "难道米拉团队已经将狼才进",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07595833261214678,
      "no_speech_prob": 1.5999000735344993e-11,
      "compression_ratio": 1.238938053097345
    },
    {
      "start": 440.64,
      "end": 443.04,
      "text": "脱离大厂充足算力扶持之后",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07595833261214678,
      "no_speech_prob": 1.5999000735344993e-11,
      "compression_ratio": 1.238938053097345
    },
    {
      "start": 443.04,
      "end": 446.54,
      "text": "这群天才工程师再也无法完成前沿模型研发",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07595833261214678,
      "no_speech_prob": 1.5999000735344993e-11,
      "compression_ratio": 1.238938053097345
    },
    {
      "start": 446.54,
      "end": 448.12,
      "text": "答案显然是否定的",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07595833261214678,
      "no_speech_prob": 1.5999000735344993e-11,
      "compression_ratio": 1.238938053097345
    },
    {
      "start": 448.12,
      "end": 451.42,
      "text": "这里藏着一个极异被大众忽略的反常识逻辑",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07595833261214678,
      "no_speech_prob": 1.5999000735344993e-11,
      "compression_ratio": 1.238938053097345
    },
    {
      "start": 451.42,
      "end": 455.26,
      "text": "也是绝大多数行业观察者出现判断偏差的关键",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.054232386442331165,
      "no_speech_prob": 1.1976314905071117e-11,
      "compression_ratio": 1.2349570200573066
    },
    {
      "start": 455.26,
      "end": 457.42,
      "text": "外界普遍默认研发大模型",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.054232386442331165,
      "no_speech_prob": 1.1976314905071117e-11,
      "compression_ratio": 1.2349570200573066
    },
    {
      "start": 457.42,
      "end": 459.76,
      "text": "必须无休止开展算力军备竞赛",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.054232386442331165,
      "no_speech_prob": 1.1976314905071117e-11,
      "compression_ratio": 1.2349570200573066
    },
    {
      "start": 459.76,
      "end": 461.98,
      "text": "追求更高参数 更强跑分",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.054232386442331165,
      "no_speech_prob": 1.1976314905071117e-11,
      "compression_ratio": 1.2349570200573066
    },
    {
      "start": 461.98,
      "end": 465.06,
      "text": "谁的榜单成绩领先 谁就是最终赢家",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.054232386442331165,
      "no_speech_prob": 1.1976314905071117e-11,
      "compression_ratio": 1.2349570200573066
    },
    {
      "start": 465.06,
      "end": 468.56,
      "text": "可这套竞争思路已经是两年之前的老旧玩法",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.054232386442331165,
      "no_speech_prob": 1.1976314905071117e-11,
      "compression_ratio": 1.2349570200573066
    },
    {
      "start": 468.56,
      "end": 472.16,
      "text": "现如今想要依靠基准测试提升1%的准确率",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.054232386442331165,
      "no_speech_prob": 1.1976314905071117e-11,
      "compression_ratio": 1.2349570200573066
    },
    {
      "start": 472.16,
      "end": 475.98,
      "text": "需要额外投入数亿美元电费 海量算力资源",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.054232386442331165,
      "no_speech_prob": 1.1976314905071117e-11,
      "compression_ratio": 1.2349570200573066
    },
    {
      "start": 476.06,
      "end": 479.76,
      "text": "还要耗费大量人力筛选稀缺高质量训练数据",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.054232386442331165,
      "no_speech_prob": 1.1976314905071117e-11,
      "compression_ratio": 1.2349570200573066
    },
    {
      "start": 480.0,
      "end": 487.48,
      "text": "边际成本已经膨胀到难以承受的地步。米拉在OpenAI长期生耕,比任何人都清楚算力成本核算逻辑。",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.095818865951614,
      "no_speech_prob": 7.295012684205293e-12,
      "compression_ratio": 1.2323759791122715
    },
    {
      "start": 487.48,
      "end": 494.96,
      "text": "深刻明白B原大模型无休止堆参数换取微小性能提升,本身就是一条难以持续的死胡同。",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.095818865951614,
      "no_speech_prob": 7.295012684205293e-12,
      "compression_ratio": 1.2323759791122715
    },
    {
      "start": 494.96,
      "end": 498.04,
      "text": "如果运营一家估值120亿美元的企业",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.058957826744005515,
      "no_speech_prob": 1.1175027049559194e-11,
      "compression_ratio": 1.2644628099173554
    },
    {
      "start": 498.04,
      "end": 499.8,
      "text": "核心目标绝对不是",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.058957826744005515,
      "no_speech_prob": 1.1175027049559194e-11,
      "compression_ratio": 1.2644628099173554
    },
    {
      "start": 499.8,
      "end": 502.96,
      "text": "打赢一场没有实际商业价值的跑分竞赛",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.058957826744005515,
      "no_speech_prob": 1.1175027049559194e-11,
      "compression_ratio": 1.2644628099173554
    },
    {
      "start": 502.96,
      "end": 506.6,
      "text": "而是搭建一套能持续自我造血的商业体系",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.058957826744005515,
      "no_speech_prob": 1.1175027049559194e-11,
      "compression_ratio": 1.2644628099173554
    },
    {
      "start": 506.6,
      "end": 510.86,
      "text": "所以当大家看见一群掌握顶尖AI底层认知的技术人才",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.058957826744005515,
      "no_speech_prob": 1.1175027049559194e-11,
      "compression_ratio": 1.2644628099173554
    },
    {
      "start": 510.86,
      "end": 514.36,
      "text": "在关键节点推出一款看起来平平无奇的产品",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.058957826744005515,
      "no_speech_prob": 1.1175027049559194e-11,
      "compression_ratio": 1.2644628099173554
    },
    {
      "start": 514.36,
      "end": 516.92,
      "text": "千万不要简单判定他们决策失误",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.058957826744005515,
      "no_speech_prob": 1.1175027049559194e-11,
      "compression_ratio": 1.2644628099173554
    },
    {
      "start": 516.92,
      "end": 520.32,
      "text": "他们正在执行一套普通人难以看懂的长线布局",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.058957826744005515,
      "no_speech_prob": 1.1175027049559194e-11,
      "compression_ratio": 1.2644628099173554
    },
    {
      "start": 520.32,
      "end": 522.44,
      "text": "Inkling跑分表现相对平庸",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.058957826744005515,
      "no_speech_prob": 1.1175027049559194e-11,
      "compression_ratio": 1.2644628099173554
    },
    {
      "start": 522.44,
      "end": 523.96,
      "text": "并不是研发能力不足",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.058957826744005515,
      "no_speech_prob": 1.1175027049559194e-11,
      "compression_ratio": 1.2644628099173554
    },
    {
      "start": 523.96,
      "end": 526.04,
      "text": "而是精准的商业策略规划",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10113045854388543,
      "no_speech_prob": 1.798607270064867e-11,
      "compression_ratio": 1.1675392670157068
    },
    {
      "start": 526.04,
      "end": 528.44,
      "text": "Tai Mankey & G. McKean ES Lab",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10113045854388543,
      "no_speech_prob": 1.798607270064867e-11,
      "compression_ratio": 1.1675392670157068
    },
    {
      "start": 528.44,
      "end": 530.16,
      "text": "从一开始就没有争夺",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10113045854388543,
      "no_speech_prob": 1.798607270064867e-11,
      "compression_ratio": 1.1675392670157068
    },
    {
      "start": 530.16,
      "end": 532.72,
      "text": "全球最强通用大模型的想法",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10113045854388543,
      "no_speech_prob": 1.798607270064867e-11,
      "compression_ratio": 1.1675392670157068
    },
    {
      "start": 532.72,
      "end": 534.44,
      "text": "把一款推理能力中等",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10113045854388543,
      "no_speech_prob": 1.798607270064867e-11,
      "compression_ratio": 1.1675392670157068
    },
    {
      "start": 534.44,
      "end": 535.66,
      "text": "算力消耗低廉",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10113045854388543,
      "no_speech_prob": 1.798607270064867e-11,
      "compression_ratio": 1.1675392670157068
    },
    {
      "start": 535.66,
      "end": 536.82,
      "text": "响应速度出众",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10113045854388543,
      "no_speech_prob": 1.798607270064867e-11,
      "compression_ratio": 1.1675392670157068
    },
    {
      "start": 536.82,
      "end": 538.36,
      "text": "并且完全开源的模型",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10113045854388543,
      "no_speech_prob": 1.798607270064867e-11,
      "compression_ratio": 1.1675392670157068
    },
    {
      "start": 538.36,
      "end": 539.56,
      "text": "免费对外开放",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10113045854388543,
      "no_speech_prob": 1.798607270064867e-11,
      "compression_ratio": 1.1675392670157068
    },
    {
      "start": 539.56,
      "end": 541.98,
      "text": "仅仅是整套战略布局的第一步",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10113045854388543,
      "no_speech_prob": 1.798607270064867e-11,
      "compression_ratio": 1.1675392670157068
    },
    {
      "start": 541.98,
      "end": 543.76,
      "text": "20亿美元的资金投入",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10113045854388543,
      "no_speech_prob": 1.798607270064867e-11,
      "compression_ratio": 1.1675392670157068
    },
    {
      "start": 543.76,
      "end": 545.06,
      "text": "绝大部分运用在",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10113045854388543,
      "no_speech_prob": 1.798607270064867e-11,
      "compression_ratio": 1.1675392670157068
    },
    {
      "start": 545.06,
      "end": 547.28,
      "text": "Inkling反直觉的底层机制设计",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10113045854388543,
      "no_speech_prob": 1.798607270064867e-11,
      "compression_ratio": 1.1675392670157068
    },
    {
      "start": 547.28,
      "end": 549.18,
      "text": "以及他们真正用来实现",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10113045854388543,
      "no_speech_prob": 1.798607270064867e-11,
      "compression_ratio": 1.1675392670157068
    },
    {
      "start": 549.18,
      "end": 551.24,
      "text": "商业化变现的秘密武器之上",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10113045854388543,
      "no_speech_prob": 1.798607270064867e-11,
      "compression_ratio": 1.1675392670157068
    },
    {
      "start": 551.24,
      "end": 552.88,
      "text": "倘若大家单纯认为",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10113045854388543,
      "no_speech_prob": 1.798607270064867e-11,
      "compression_ratio": 1.1675392670157068
    },
    {
      "start": 552.88,
      "end": 555.18,
      "text": "这款无法冲进榜单前五的模型",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07026442166032462,
      "no_speech_prob": 1.3778332641489754e-11,
      "compression_ratio": 1.2168021680216803
    },
    {
      "start": 555.18,
      "end": 556.14,
      "text": "就是故事终点",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07026442166032462,
      "no_speech_prob": 1.3778332641489754e-11,
      "compression_ratio": 1.2168021680216803
    },
    {
      "start": 556.14,
      "end": 560.08,
      "text": "那就严重低估企业及弊端市场竞争的残酷程度",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07026442166032462,
      "no_speech_prob": 1.3778332641489754e-11,
      "compression_ratio": 1.2168021680216803
    },
    {
      "start": 560.08,
      "end": 561.78,
      "text": "为了更好落地商业化",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07026442166032462,
      "no_speech_prob": 1.3778332641489754e-11,
      "compression_ratio": 1.2168021680216803
    },
    {
      "start": 561.78,
      "end": 564.94,
      "text": "Inkling甚至主动削减一部分自由联想能力",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07026442166032462,
      "no_speech_prob": 1.3778332641489754e-11,
      "compression_ratio": 1.2168021680216803
    },
    {
      "start": 564.94,
      "end": 568.34,
      "text": "只为持续降低单次调用产生的词源开销",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07026442166032462,
      "no_speech_prob": 1.3778332641489754e-11,
      "compression_ratio": 1.2168021680216803
    },
    {
      "start": 568.34,
      "end": 570.02,
      "text": "模型内部运行逻辑",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07026442166032462,
      "no_speech_prob": 1.3778332641489754e-11,
      "compression_ratio": 1.2168021680216803
    },
    {
      "start": 570.02,
      "end": 571.64,
      "text": "加上Thinking Machines Lab",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07026442166032462,
      "no_speech_prob": 1.3778332641489754e-11,
      "compression_ratio": 1.2168021680216803
    },
    {
      "start": 571.64,
      "end": 573.98,
      "text": "后续依靠估值变现的整套方案",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07026442166032462,
      "no_speech_prob": 1.3778332641489754e-11,
      "compression_ratio": 1.2168021680216803
    },
    {
      "start": 573.98,
      "end": 577.2,
      "text": "才是整个科技行业真正需要警惕的地方",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07026442166032462,
      "no_speech_prob": 1.3778332641489754e-11,
      "compression_ratio": 1.2168021680216803
    },
    {
      "start": 577.2,
      "end": 580.9,
      "text": "当我们拆解Inkling内部调节推理强度的核心机制",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07026442166032462,
      "no_speech_prob": 1.3778332641489754e-11,
      "compression_ratio": 1.2168021680216803
    },
    {
      "start": 580.9,
      "end": 583.9,
      "text": "就能意识到未来人工智能赛道的主战场",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06621444981272627,
      "no_speech_prob": 1.2018286539572376e-11,
      "compression_ratio": 1.2682215743440233
    },
    {
      "start": 583.9,
      "end": 586.34,
      "text": "根本不会集中在评测排行榜",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06621444981272627,
      "no_speech_prob": 1.2018286539572376e-11,
      "compression_ratio": 1.2682215743440233
    },
    {
      "start": 586.34,
      "end": 588.34,
      "text": "这款模型看似平庸的跑分",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06621444981272627,
      "no_speech_prob": 1.2018286539572376e-11,
      "compression_ratio": 1.2682215743440233
    },
    {
      "start": 588.34,
      "end": 590.74,
      "text": "是刻意控制之下形成的结果",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06621444981272627,
      "no_speech_prob": 1.2018286539572376e-11,
      "compression_ratio": 1.2682215743440233
    },
    {
      "start": 590.74,
      "end": 593.92,
      "text": "米拉团队在Inkling内部设置三处反直觉",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06621444981272627,
      "no_speech_prob": 1.2018286539572376e-11,
      "compression_ratio": 1.2682215743440233
    },
    {
      "start": 593.92,
      "end": 595.92,
      "text": "甚至有些激进的技术设定",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06621444981272627,
      "no_speech_prob": 1.2018286539572376e-11,
      "compression_ratio": 1.2682215743440233
    },
    {
      "start": 595.92,
      "end": 597.64,
      "text": "每一项都精准击中",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06621444981272627,
      "no_speech_prob": 1.2018286539572376e-11,
      "compression_ratio": 1.2682215743440233
    },
    {
      "start": 597.64,
      "end": 600.76,
      "text": "日军千万次AI调用企业客户的核心痛点",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06621444981272627,
      "no_speech_prob": 1.2018286539572376e-11,
      "compression_ratio": 1.2682215743440233
    },
    {
      "start": 600.76,
      "end": 604.34,
      "text": "第一项设定可以称之为智能调节推理力度旋钮",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06621444981272627,
      "no_speech_prob": 1.2018286539572376e-11,
      "compression_ratio": 1.2682215743440233
    },
    {
      "start": 604.34,
      "end": 607.36,
      "text": "官方命名为思考力度控制系统",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06621444981272627,
      "no_speech_prob": 1.2018286539572376e-11,
      "compression_ratio": 1.2682215743440233
    },
    {
      "start": 607.36,
      "end": 609.18,
      "text": "这项设计充满巧思",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06621444981272627,
      "no_speech_prob": 1.2018286539572376e-11,
      "compression_ratio": 1.2682215743440233
    },
    {
      "start": 609.18,
      "end": 611.44,
      "text": "市面上绝大多数主流大模型",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06842770046657987,
      "no_speech_prob": 1.3264442495775075e-11,
      "compression_ratio": 1.2740524781341107
    },
    {
      "start": 611.44,
      "end": 613.56,
      "text": "处理简单问答1加1等于计",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06842770046657987,
      "no_speech_prob": 1.3264442495775075e-11,
      "compression_ratio": 1.2740524781341107
    },
    {
      "start": 613.56,
      "end": 616.12,
      "text": "和撰写后端微服务架构代码",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06842770046657987,
      "no_speech_prob": 1.3264442495775075e-11,
      "compression_ratio": 1.2740524781341107
    },
    {
      "start": 616.12,
      "end": 618.26,
      "text": "后台占用的算力资源差距极小",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06842770046657987,
      "no_speech_prob": 1.3264442495775075e-11,
      "compression_ratio": 1.2740524781341107
    },
    {
      "start": 618.26,
      "end": 620.26,
      "text": "好比驾驶重型重载卡车",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06842770046657987,
      "no_speech_prob": 1.3264442495775075e-11,
      "compression_ratio": 1.2740524781341107
    },
    {
      "start": 620.26,
      "end": 623.18,
      "text": "仅仅只为前往楼下便利店采购物品",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06842770046657987,
      "no_speech_prob": 1.3264442495775075e-11,
      "compression_ratio": 1.2740524781341107
    },
    {
      "start": 623.18,
      "end": 625.98,
      "text": "在企业大规模批量调用场景之下",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06842770046657987,
      "no_speech_prob": 1.3264442495775075e-11,
      "compression_ratio": 1.2740524781341107
    },
    {
      "start": 625.98,
      "end": 628.62,
      "text": "这种资源浪费很难长期承受",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06842770046657987,
      "no_speech_prob": 1.3264442495775075e-11,
      "compression_ratio": 1.2740524781341107
    },
    {
      "start": 628.62,
      "end": 630.24,
      "text": "依托这套调节旋钮",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06842770046657987,
      "no_speech_prob": 1.3264442495775075e-11,
      "compression_ratio": 1.2740524781341107
    },
    {
      "start": 630.24,
      "end": 632.04,
      "text": "运行规则彻底改变",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06842770046657987,
      "no_speech_prob": 1.3264442495775075e-11,
      "compression_ratio": 1.2740524781341107
    },
    {
      "start": 632.04,
      "end": 633.88,
      "text": "当模型处理发票分类",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06842770046657987,
      "no_speech_prob": 1.3264442495775075e-11,
      "compression_ratio": 1.2740524781341107
    },
    {
      "start": 633.88,
      "end": 637.16,
      "text": "标准化客服问答这类重复性基础任务",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06842770046657987,
      "no_speech_prob": 1.3264442495775075e-11,
      "compression_ratio": 1.2740524781341107
    },
    {
      "start": 637.16,
      "end": 638.42,
      "text": "可以下调思考力度",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.058158924466087704,
      "no_speech_prob": 1.8969497847520778e-11,
      "compression_ratio": 1.2996941896024465
    },
    {
      "start": 638.42,
      "end": 640.66,
      "text": "此时Inkling反馈速度极快",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.058158924466087704,
      "no_speech_prob": 1.8969497847520778e-11,
      "compression_ratio": 1.2996941896024465
    },
    {
      "start": 640.66,
      "end": 642.72,
      "text": "调用成本压缩到极低水平",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.058158924466087704,
      "no_speech_prob": 1.8969497847520778e-11,
      "compression_ratio": 1.2996941896024465
    },
    {
      "start": 642.72,
      "end": 645.1,
      "text": "基础答案完全能够满足业务需求",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.058158924466087704,
      "no_speech_prob": 1.8969497847520778e-11,
      "compression_ratio": 1.2996941896024465
    },
    {
      "start": 645.1,
      "end": 647.16,
      "text": "一旦遭遇核心业务逻辑推演",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.058158924466087704,
      "no_speech_prob": 1.8969497847520778e-11,
      "compression_ratio": 1.2996941896024465
    },
    {
      "start": 647.16,
      "end": 648.4,
      "text": "复杂代码修复",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.058158924466087704,
      "no_speech_prob": 1.8969497847520778e-11,
      "compression_ratio": 1.2996941896024465
    },
    {
      "start": 648.4,
      "end": 650.04,
      "text": "高阶数学推理难题",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.058158924466087704,
      "no_speech_prob": 1.8969497847520778e-11,
      "compression_ratio": 1.2996941896024465
    },
    {
      "start": 650.04,
      "end": 651.92,
      "text": "直接拉高思考力度档位",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.058158924466087704,
      "no_speech_prob": 1.8969497847520778e-11,
      "compression_ratio": 1.2996941896024465
    },
    {
      "start": 651.92,
      "end": 655.34,
      "text": "Inkling会调动更多算力进入深度推理模式",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.058158924466087704,
      "no_speech_prob": 1.8969497847520778e-11,
      "compression_ratio": 1.2996941896024465
    },
    {
      "start": 655.34,
      "end": 658.6,
      "text": "综合推理效率能够对标行业第一梯队大模型",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.058158924466087704,
      "no_speech_prob": 1.8969497847520778e-11,
      "compression_ratio": 1.2996941896024465
    },
    {
      "start": 658.6,
      "end": 661.02,
      "text": "同时算力消耗远低于同级产品",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.058158924466087704,
      "no_speech_prob": 1.8969497847520778e-11,
      "compression_ratio": 1.2996941896024465
    },
    {
      "start": 661.02,
      "end": 663.6,
      "text": "在企业商用场景具备极强优势",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.058158924466087704,
      "no_speech_prob": 1.8969497847520778e-11,
      "compression_ratio": 1.2996941896024465
    },
    {
      "start": 663.6,
      "end": 667.5,
      "text": "这意味着企业客户终于可以动态管控算力支出",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06302283499478163,
      "no_speech_prob": 1.2366740444191837e-11,
      "compression_ratio": 1.3061797752808988
    },
    {
      "start": 667.5,
      "end": 670.78,
      "text": "当后台每日运行数百万次智能体任务",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06302283499478163,
      "no_speech_prob": 1.2366740444191837e-11,
      "compression_ratio": 1.3061797752808988
    },
    {
      "start": 670.78,
      "end": 672.58,
      "text": "可以按需调整资源投入",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06302283499478163,
      "no_speech_prob": 1.2366740444191837e-11,
      "compression_ratio": 1.3061797752808988
    },
    {
      "start": 672.58,
      "end": 676.06,
      "text": "省下的每一笔开支都会直接转化为企业净利润",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06302283499478163,
      "no_speech_prob": 1.2366740444191837e-11,
      "compression_ratio": 1.3061797752808988
    },
    {
      "start": 676.06,
      "end": 678.84,
      "text": "第二项设定更加打破大众固有认知",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06302283499478163,
      "no_speech_prob": 1.2366740444191837e-11,
      "compression_ratio": 1.3061797752808988
    },
    {
      "start": 678.84,
      "end": 682.46,
      "text": "研发团队刻意训练模型形成诚实输出特性",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06302283499478163,
      "no_speech_prob": 1.2366740444191837e-11,
      "compression_ratio": 1.3061797752808988
    },
    {
      "start": 682.46,
      "end": 685.78,
      "text": "当下绝大多数大模型普遍存在幻觉问题",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06302283499478163,
      "no_speech_prob": 1.2366740444191837e-11,
      "compression_ratio": 1.3061797752808988
    },
    {
      "start": 685.78,
      "end": 689.66,
      "text": "无论面对何种提问都会强行生成完整答案",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06302283499478163,
      "no_speech_prob": 1.2366740444191837e-11,
      "compression_ratio": 1.3061797752808988
    },
    {
      "start": 689.66,
      "end": 691.52,
      "text": "哪怕内容完全不符合事实",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06302283499478163,
      "no_speech_prob": 1.2366740444191837e-11,
      "compression_ratio": 1.3061797752808988
    },
    {
      "start": 691.52,
      "end": 693.54,
      "text": "依旧会以笃定语气输出",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06302283499478163,
      "no_speech_prob": 1.2366740444191837e-11,
      "compression_ratio": 1.3061797752808988
    },
    {
      "start": 693.54,
      "end": 694.68,
      "text": "在文学创作",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06822508917914497,
      "no_speech_prob": 1.4766448480640904e-11,
      "compression_ratio": 1.2225274725274726
    },
    {
      "start": 694.68,
      "end": 696.06,
      "text": "日常闲聊场景",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06822508917914497,
      "no_speech_prob": 1.4766448480640904e-11,
      "compression_ratio": 1.2225274725274726
    },
    {
      "start": 696.06,
      "end": 697.6,
      "text": "这种自主生成能力",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06822508917914497,
      "no_speech_prob": 1.4766448480640904e-11,
      "compression_ratio": 1.2225274725274726
    },
    {
      "start": 697.6,
      "end": 699.0,
      "text": "可以称之为创造力",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06822508917914497,
      "no_speech_prob": 1.4766448480640904e-11,
      "compression_ratio": 1.2225274725274726
    },
    {
      "start": 699.0,
      "end": 700.22,
      "text": "放到金融预测",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06822508917914497,
      "no_speech_prob": 1.4766448480640904e-11,
      "compression_ratio": 1.2225274725274726
    },
    {
      "start": 700.22,
      "end": 702.16,
      "text": "医疗诊断这类企业场景",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06822508917914497,
      "no_speech_prob": 1.4766448480640904e-11,
      "compression_ratio": 1.2225274725274726
    },
    {
      "start": 702.16,
      "end": 703.72,
      "text": "毫无依据的编造内容",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06822508917914497,
      "no_speech_prob": 1.4766448480640904e-11,
      "compression_ratio": 1.2225274725274726
    },
    {
      "start": 703.72,
      "end": 704.96,
      "text": "会带来致命风险",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06822508917914497,
      "no_speech_prob": 1.4766448480640904e-11,
      "compression_ratio": 1.2225274725274726
    },
    {
      "start": 704.96,
      "end": 705.68,
      "text": "为此",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06822508917914497,
      "no_speech_prob": 1.4766448480640904e-11,
      "compression_ratio": 1.2225274725274726
    },
    {
      "start": 705.68,
      "end": 707.08,
      "text": "Thinking Machines Lab",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06822508917914497,
      "no_speech_prob": 1.4766448480640904e-11,
      "compression_ratio": 1.2225274725274726
    },
    {
      "start": 707.08,
      "end": 709.18,
      "text": "引入认知学导向训练方案",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06822508917914497,
      "no_speech_prob": 1.4766448480640904e-11,
      "compression_ratio": 1.2225274725274726
    },
    {
      "start": 709.18,
      "end": 710.2,
      "text": "简单来讲",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06822508917914497,
      "no_speech_prob": 1.4766448480640904e-11,
      "compression_ratio": 1.2225274725274726
    },
    {
      "start": 710.2,
      "end": 712.08,
      "text": "利用强化学习奖励模型",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06822508917914497,
      "no_speech_prob": 1.4766448480640904e-11,
      "compression_ratio": 1.2225274725274726
    },
    {
      "start": 712.08,
      "end": 714.1,
      "text": "主动承认认知边界的行为",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06822508917914497,
      "no_speech_prob": 1.4766448480640904e-11,
      "compression_ratio": 1.2225274725274726
    },
    {
      "start": 714.1,
      "end": 715.76,
      "text": "团队持续迭代训练",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06822508917914497,
      "no_speech_prob": 1.4766448480640904e-11,
      "compression_ratio": 1.2225274725274726
    },
    {
      "start": 715.76,
      "end": 717.6,
      "text": "引导Inkling在信息不足",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06822508917914497,
      "no_speech_prob": 1.4766448480640904e-11,
      "compression_ratio": 1.2225274725274726
    },
    {
      "start": 717.6,
      "end": 719.54,
      "text": "无法得出可靠结论的时候",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06822508917914497,
      "no_speech_prob": 1.4766448480640904e-11,
      "compression_ratio": 1.2225274725274726
    },
    {
      "start": 719.54,
      "end": 720.98,
      "text": "不要强行猜测",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06822508917914497,
      "no_speech_prob": 1.4766448480640904e-11,
      "compression_ratio": 1.2225274725274726
    },
    {
      "start": 720.98,
      "end": 723.1,
      "text": "直接告知使用者无法解答",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06917105469049192,
      "no_speech_prob": 1.3900616768758312e-11,
      "compression_ratio": 1.244712990936556
    },
    {
      "start": 723.1,
      "end": 725.12,
      "text": "最终训练成果十分突出",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06917105469049192,
      "no_speech_prob": 1.3900616768758312e-11,
      "compression_ratio": 1.244712990936556
    },
    {
      "start": 725.12,
      "end": 726.62,
      "text": "Inkling在事件预测",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06917105469049192,
      "no_speech_prob": 1.3900616768758312e-11,
      "compression_ratio": 1.244712990936556
    },
    {
      "start": 726.62,
      "end": 728.68,
      "text": "数据分析类客观评测中",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06917105469049192,
      "no_speech_prob": 1.3900616768758312e-11,
      "compression_ratio": 1.244712990936556
    },
    {
      "start": 728.68,
      "end": 731.54,
      "text": "准确率超越多款主流通用大模型",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06917105469049192,
      "no_speech_prob": 1.3900616768758312e-11,
      "compression_ratio": 1.244712990936556
    },
    {
      "start": 731.54,
      "end": 733.28,
      "text": "达到行业顶尖水准",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06917105469049192,
      "no_speech_prob": 1.3900616768758312e-11,
      "compression_ratio": 1.244712990936556
    },
    {
      "start": 733.28,
      "end": 735.04,
      "text": "对于华尔街对冲基金",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06917105469049192,
      "no_speech_prob": 1.3900616768758312e-11,
      "compression_ratio": 1.244712990936556
    },
    {
      "start": 735.04,
      "end": 736.96,
      "text": "医疗行业大型机构而言",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06917105469049192,
      "no_speech_prob": 1.3900616768758312e-11,
      "compression_ratio": 1.244712990936556
    },
    {
      "start": 736.96,
      "end": 740.22,
      "text": "一台关键时刻敢于坦诚信息不足的模型",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06917105469049192,
      "no_speech_prob": 1.3900616768758312e-11,
      "compression_ratio": 1.244712990936556
    },
    {
      "start": 740.22,
      "end": 743.88,
      "text": "远比一位强行输出答案的通用模型更加可靠",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06917105469049192,
      "no_speech_prob": 1.3900616768758312e-11,
      "compression_ratio": 1.244712990936556
    },
    {
      "start": 743.88,
      "end": 746.48,
      "text": "第三项设定带着一丝黑色幽默",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06917105469049192,
      "no_speech_prob": 1.3900616768758312e-11,
      "compression_ratio": 1.244712990936556
    },
    {
      "start": 746.48,
      "end": 748.46,
      "text": "为极致压缩运行成本",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06917105469049192,
      "no_speech_prob": 1.3900616768758312e-11,
      "compression_ratio": 1.244712990936556
    },
    {
      "start": 748.46,
      "end": 751.26,
      "text": "经过3000万轮强化学习迭代之后",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07148519387611976,
      "no_speech_prob": 1.2411561362002388e-11,
      "compression_ratio": 1.2956521739130435
    },
    {
      "start": 751.26,
      "end": 754.78,
      "text": "Inkling自主演化出高度精简的内部思考语言",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07148519387611976,
      "no_speech_prob": 1.2411561362002388e-11,
      "compression_ratio": 1.2956521739130435
    },
    {
      "start": 754.78,
      "end": 756.22,
      "text": "我们可以简单对比",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07148519387611976,
      "no_speech_prob": 1.2411561362002388e-11,
      "compression_ratio": 1.2956521739130435
    },
    {
      "start": 756.22,
      "end": 758.86,
      "text": "差异主流模型处理复杂问题时",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07148519387611976,
      "no_speech_prob": 1.2411561362002388e-11,
      "compression_ratio": 1.2956521739130435
    },
    {
      "start": 758.86,
      "end": 761.02,
      "text": "内部推理链条会使用规范",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07148519387611976,
      "no_speech_prob": 1.2411561362002388e-11,
      "compression_ratio": 1.2956521739130435
    },
    {
      "start": 761.02,
      "end": 762.36,
      "text": "完整英文语句",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07148519387611976,
      "no_speech_prob": 1.2411561362002388e-11,
      "compression_ratio": 1.2956521739130435
    },
    {
      "start": 762.36,
      "end": 764.54,
      "text": "需要消耗大量词源资源",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07148519387611976,
      "no_speech_prob": 1.2411561362002388e-11,
      "compression_ratio": 1.2956521739130435
    },
    {
      "start": 764.54,
      "end": 767.32,
      "text": "Inkling内部推演过程抛弃完整语法",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07148519387611976,
      "no_speech_prob": 1.2411561362002388e-11,
      "compression_ratio": 1.2956521739130435
    },
    {
      "start": 767.32,
      "end": 769.0,
      "text": "只保留核心关键词",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07148519387611976,
      "no_speech_prob": 1.2411561362002388e-11,
      "compression_ratio": 1.2956521739130435
    },
    {
      "start": 769.0,
      "end": 770.8,
      "text": "推理效果几乎没有衰减",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07148519387611976,
      "no_speech_prob": 1.2411561362002388e-11,
      "compression_ratio": 1.2956521739130435
    },
    {
      "start": 770.8,
      "end": 772.92,
      "text": "但是极大削减词源占用",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07148519387611976,
      "no_speech_prob": 1.2411561362002388e-11,
      "compression_ratio": 1.2956521739130435
    },
    {
      "start": 772.92,
      "end": 774.32,
      "text": "听起来十分有趣",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07148519387611976,
      "no_speech_prob": 1.2411561362002388e-11,
      "compression_ratio": 1.2956521739130435
    },
    {
      "start": 774.32,
      "end": 778.16,
      "text": "背后体现的是对推理效率极致压榨的工程思路",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07148519387611976,
      "no_speech_prob": 1.2411561362002388e-11,
      "compression_ratio": 1.2956521739130435
    },
    {
      "start": 778.46,
      "end": 783.4,
      "text": "读到这里大家就能明白,这从来不是一款用来冲击榜单排名的展示型模型,",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10035434061167192,
      "no_speech_prob": 1.7426724993607756e-11,
      "compression_ratio": 1.1436031331592689
    },
    {
      "start": 783.8,
      "end": 789.48,
      "text": "而是面向大规模企业商用量身打造,精打细算,稳定可靠的底层基础设施。",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10035434061167192,
      "no_speech_prob": 1.7426724993607756e-11,
      "compression_ratio": 1.1436031331592689
    },
    {
      "start": 789.84,
      "end": 794.98,
      "text": "即便Inclin拥有诸多精妙设计,仅仅依靠开源模型本身无法创造收入,",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10035434061167192,
      "no_speech_prob": 1.7426724993607756e-11,
      "compression_ratio": 1.1436031331592689
    },
    {
      "start": 795.08,
      "end": 802.06,
      "text": "能够支撑120亿估值完成变现的核心载体,是配套同步推出的Tinker自动化微调平台。",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10035434061167192,
      "no_speech_prob": 1.7426724993607756e-11,
      "compression_ratio": 1.1436031331592689
    },
    {
      "start": 802.42,
      "end": 806.74,
      "text": "这才是SY NTNG MacAINES Lab真正的商业核武器。",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10035434061167192,
      "no_speech_prob": 1.7426724993607756e-11,
      "compression_ratio": 1.1436031331592689
    },
    {
      "start": 806.74,
      "end": 808.06,
      "text": "前面我们提到",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08864939817085793,
      "no_speech_prob": 1.2359994971955501e-11,
      "compression_ratio": 1.2528735632183907
    },
    {
      "start": 808.06,
      "end": 809.38,
      "text": "Inkling完全开源",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08864939817085793,
      "no_speech_prob": 1.2359994971955501e-11,
      "compression_ratio": 1.2528735632183907
    },
    {
      "start": 809.38,
      "end": 811.54,
      "text": "任何人都可以免费下载部署",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08864939817085793,
      "no_speech_prob": 1.2359994971955501e-11,
      "compression_ratio": 1.2528735632183907
    },
    {
      "start": 811.54,
      "end": 813.66,
      "text": "企业不会为此支付一分钱",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08864939817085793,
      "no_speech_prob": 1.2359994971955501e-11,
      "compression_ratio": 1.2528735632183907
    },
    {
      "start": 813.66,
      "end": 815.6,
      "text": "可是真实落地场景当中",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08864939817085793,
      "no_speech_prob": 1.2359994971955501e-11,
      "compression_ratio": 1.2528735632183907
    },
    {
      "start": 815.6,
      "end": 818.9,
      "text": "大型企业技术负责人拿到模型权重之后",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08864939817085793,
      "no_speech_prob": 1.2359994971955501e-11,
      "compression_ratio": 1.2528735632183907
    },
    {
      "start": 818.9,
      "end": 820.7,
      "text": "很快会遭遇现实阻碍",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08864939817085793,
      "no_speech_prob": 1.2359994971955501e-11,
      "compression_ratio": 1.2528735632183907
    },
    {
      "start": 820.7,
      "end": 823.16,
      "text": "想要让模型理解自身行业业务",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08864939817085793,
      "no_speech_prob": 1.2359994971955501e-11,
      "compression_ratio": 1.2528735632183907
    },
    {
      "start": 823.16,
      "end": 824.6,
      "text": "读懂内部财务数据",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08864939817085793,
      "no_speech_prob": 1.2359994971955501e-11,
      "compression_ratio": 1.2528735632183907
    },
    {
      "start": 824.6,
      "end": 826.46,
      "text": "适配客户服务场景",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08864939817085793,
      "no_speech_prob": 1.2359994971955501e-11,
      "compression_ratio": 1.2528735632183907
    },
    {
      "start": 826.46,
      "end": 829.22,
      "text": "必须使用企业私有数据开展微调",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08864939817085793,
      "no_speech_prob": 1.2359994971955501e-11,
      "compression_ratio": 1.2528735632183907
    },
    {
      "start": 829.22,
      "end": 830.88,
      "text": "模型微调门槛极高",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08864939817085793,
      "no_speech_prob": 1.2359994971955501e-11,
      "compression_ratio": 1.2528735632183907
    },
    {
      "start": 830.88,
      "end": 833.44,
      "text": "研发人员需要精通整套训练架构",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08864939817085793,
      "no_speech_prob": 1.2359994971955501e-11,
      "compression_ratio": 1.2528735632183907
    },
    {
      "start": 833.44,
      "end": 835.66,
      "text": "持续维护GPU算力集群",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08864939817085793,
      "no_speech_prob": 1.2359994971955501e-11,
      "compression_ratio": 1.2528735632183907
    },
    {
      "start": 835.66,
      "end": 839.44,
      "text": "还要应对训练进程当中频繁出现的程序崩溃问题",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06692768319305283,
      "no_speech_prob": 1.28103205127017e-11,
      "compression_ratio": 1.2532637075718016
    },
    {
      "start": 839.44,
      "end": 842.62,
      "text": "绝大多数传统企业并不具备对应的技术团队",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06692768319305283,
      "no_speech_prob": 1.28103205127017e-11,
      "compression_ratio": 1.2532637075718016
    },
    {
      "start": 842.62,
      "end": 846.42,
      "text": "这个时候Thinking Machines Lab就可以顺势推出解决方案",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06692768319305283,
      "no_speech_prob": 1.28103205127017e-11,
      "compression_ratio": 1.2532637075718016
    },
    {
      "start": 846.42,
      "end": 848.32,
      "text": "向企业推荐Tinker平台",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06692768319305283,
      "no_speech_prob": 1.28103205127017e-11,
      "compression_ratio": 1.2532637075718016
    },
    {
      "start": 848.32,
      "end": 849.6,
      "text": "模型免费交付",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06692768319305283,
      "no_speech_prob": 1.28103205127017e-11,
      "compression_ratio": 1.2532637075718016
    },
    {
      "start": 849.6,
      "end": 851.54,
      "text": "但是微调操作门槛过高",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06692768319305283,
      "no_speech_prob": 1.28103205127017e-11,
      "compression_ratio": 1.2532637075718016
    },
    {
      "start": 851.54,
      "end": 852.94,
      "text": "借助Tinker平台",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06692768319305283,
      "no_speech_prob": 1.28103205127017e-11,
      "compression_ratio": 1.2532637075718016
    },
    {
      "start": 852.94,
      "end": 854.98,
      "text": "使用者不用理解底层框架",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06692768319305283,
      "no_speech_prob": 1.28103205127017e-11,
      "compression_ratio": 1.2532637075718016
    },
    {
      "start": 854.98,
      "end": 856.56,
      "text": "不用管理算力节点",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06692768319305283,
      "no_speech_prob": 1.28103205127017e-11,
      "compression_ratio": 1.2532637075718016
    },
    {
      "start": 856.56,
      "end": 858.44,
      "text": "直接上传私有业务数据",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06692768319305283,
      "no_speech_prob": 1.28103205127017e-11,
      "compression_ratio": 1.2532637075718016
    },
    {
      "start": 858.44,
      "end": 862.38,
      "text": "简单操作就能全自动训练专属垂直领域专家模型",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06692768319305283,
      "no_speech_prob": 1.28103205127017e-11,
      "compression_ratio": 1.2532637075718016
    },
    {
      "start": 862.38,
      "end": 865.3,
      "text": "在官方演示当中还有极具冲击力的功能",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06692768319305283,
      "no_speech_prob": 1.28103205127017e-11,
      "compression_ratio": 1.2532637075718016
    },
    {
      "start": 865.3,
      "end": 867.02,
      "text": "平台下达指令之后",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10901414253861431,
      "no_speech_prob": 1.3498553837199001e-11,
      "compression_ratio": 1.25
    },
    {
      "start": 867.02,
      "end": 869.52,
      "text": "Inklin能够自主编写训练脚本",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10901414253861431,
      "no_speech_prob": 1.3498553837199001e-11,
      "compression_ratio": 1.25
    },
    {
      "start": 869.52,
      "end": 871.68,
      "text": "自动生成配套训练数据集",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10901414253861431,
      "no_speech_prob": 1.3498553837199001e-11,
      "compression_ratio": 1.25
    },
    {
      "start": 871.68,
      "end": 873.98,
      "text": "独立完成全部微调流程",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10901414253861431,
      "no_speech_prob": 1.3498553837199001e-11,
      "compression_ratio": 1.25
    },
    {
      "start": 873.98,
      "end": 876.14,
      "text": "加在全新全重实现自我迭代",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10901414253861431,
      "no_speech_prob": 1.3498553837199001e-11,
      "compression_ratio": 1.25
    },
    {
      "start": 876.14,
      "end": 878.02,
      "text": "达成全自动自主微调",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10901414253861431,
      "no_speech_prob": 1.3498553837199001e-11,
      "compression_ratio": 1.25
    },
    {
      "start": 878.02,
      "end": 880.0,
      "text": "这正是米拉真正的野心",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10901414253861431,
      "no_speech_prob": 1.3498553837199001e-11,
      "compression_ratio": 1.25
    },
    {
      "start": 880.0,
      "end": 882.24,
      "text": "她根本无意打造第二个OpenAI",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10901414253861431,
      "no_speech_prob": 1.3498553837199001e-11,
      "compression_ratio": 1.25
    },
    {
      "start": 882.24,
      "end": 885.74,
      "text": "依靠售卖标准化按磁源计费的通用API盈利",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10901414253861431,
      "no_speech_prob": 1.3498553837199001e-11,
      "compression_ratio": 1.25
    },
    {
      "start": 885.74,
      "end": 888.94,
      "text": "现如今基础大模型赛道早已红海厮杀",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10901414253861431,
      "no_speech_prob": 1.3498553837199001e-11,
      "compression_ratio": 1.25
    },
    {
      "start": 888.94,
      "end": 890.86,
      "text": "通用模型数量众多",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10901414253861431,
      "no_speech_prob": 1.3498553837199001e-11,
      "compression_ratio": 1.25
    },
    {
      "start": 890.86,
      "end": 893.22,
      "text": "不少开源产品在垂直场景",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10901414253861431,
      "no_speech_prob": 1.3498553837199001e-11,
      "compression_ratio": 1.25
    },
    {
      "start": 893.22,
      "end": 894.74,
      "text": "性能超越必源方案",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.10901414253861431,
      "no_speech_prob": 1.3498553837199001e-11,
      "compression_ratio": 1.25
    },
    {
      "start": 894.74,
      "end": 896.16,
      "text": "Thinking Machines Lab",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07586494060831332,
      "no_speech_prob": 1.4003137191465065e-11,
      "compression_ratio": 1.2591549295774649
    },
    {
      "start": 896.16,
      "end": 898.18,
      "text": "真实商业模式十分清晰",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07586494060831332,
      "no_speech_prob": 1.4003137191465065e-11,
      "compression_ratio": 1.2591549295774649
    },
    {
      "start": 898.18,
      "end": 899.78,
      "text": "依靠可控性强",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07586494060831332,
      "no_speech_prob": 1.4003137191465065e-11,
      "compression_ratio": 1.2591549295774649
    },
    {
      "start": 899.78,
      "end": 900.72,
      "text": "运行稳定",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07586494060831332,
      "no_speech_prob": 1.4003137191465065e-11,
      "compression_ratio": 1.2591549295774649
    },
    {
      "start": 900.72,
      "end": 902.58,
      "text": "完全免费的开源模型",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07586494060831332,
      "no_speech_prob": 1.4003137191465065e-11,
      "compression_ratio": 1.2591549295774649
    },
    {
      "start": 902.58,
      "end": 903.78,
      "text": "作为引流载体",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07586494060831332,
      "no_speech_prob": 1.4003137191465065e-11,
      "compression_ratio": 1.2591549295774649
    },
    {
      "start": 903.78,
      "end": 907.22,
      "text": "吸引大量想要搭建私有人工智能系统的企业",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07586494060831332,
      "no_speech_prob": 1.4003137191465065e-11,
      "compression_ratio": 1.2591549295774649
    },
    {
      "start": 907.22,
      "end": 910.68,
      "text": "随后依托简易自动化微调平台锁定客户",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07586494060831332,
      "no_speech_prob": 1.4003137191465065e-11,
      "compression_ratio": 1.2591549295774649
    },
    {
      "start": 910.68,
      "end": 912.82,
      "text": "持续收取平台服务费用",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07586494060831332,
      "no_speech_prob": 1.4003137191465065e-11,
      "compression_ratio": 1.2591549295774649
    },
    {
      "start": 912.82,
      "end": 915.42,
      "text": "他们不对外售卖通用智能大脑",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07586494060831332,
      "no_speech_prob": 1.4003137191465065e-11,
      "compression_ratio": 1.2591549295774649
    },
    {
      "start": 915.42,
      "end": 919.02,
      "text": "而是对外提供批量定制行业专家模型的生产工具",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07586494060831332,
      "no_speech_prob": 1.4003137191465065e-11,
      "compression_ratio": 1.2591549295774649
    },
    {
      "start": 919.02,
      "end": 922.54,
      "text": "未来倘若数万企业借助Tinker搭建专属财物",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07586494060831332,
      "no_speech_prob": 1.4003137191465065e-11,
      "compression_ratio": 1.2591549295774649
    },
    {
      "start": 922.54,
      "end": 924.36,
      "text": "客服务医疗智能模型",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07586494060831332,
      "no_speech_prob": 1.4003137191465065e-11,
      "compression_ratio": 1.2591549295774649
    },
    {
      "start": 924.36,
      "end": 926.18,
      "text": "没有人再会持续纠结",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08678077697753907,
      "no_speech_prob": 1.2318574112157865e-11,
      "compression_ratio": 1.2034883720930232
    },
    {
      "start": 926.18,
      "end": 928.38,
      "text": "Inkling在评测榜单排名高低",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08678077697753907,
      "no_speech_prob": 1.2318574112157865e-11,
      "compression_ratio": 1.2034883720930232
    },
    {
      "start": 928.38,
      "end": 931.18,
      "text": "这笔20亿美元投入真的存在浪费吗",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08678077697753907,
      "no_speech_prob": 1.2318574112157865e-11,
      "compression_ratio": 1.2034883720930232
    },
    {
      "start": 931.18,
      "end": 932.38,
      "text": "答案是否定的",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08678077697753907,
      "no_speech_prob": 1.2318574112157865e-11,
      "compression_ratio": 1.2034883720930232
    },
    {
      "start": 932.38,
      "end": 933.86,
      "text": "米拉带领顶尖团队",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08678077697753907,
      "no_speech_prob": 1.2318574112157865e-11,
      "compression_ratio": 1.2034883720930232
    },
    {
      "start": 933.86,
      "end": 936.36,
      "text": "跳出OpenAI必源竞争的内卷赛道",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08678077697753907,
      "no_speech_prob": 1.2318574112157865e-11,
      "compression_ratio": 1.2034883720930232
    },
    {
      "start": 936.36,
      "end": 938.52,
      "text": "在企业定制人工智能赛道",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08678077697753907,
      "no_speech_prob": 1.2318574112157865e-11,
      "compression_ratio": 1.2034883720930232
    },
    {
      "start": 938.52,
      "end": 940.0,
      "text": "筑器极高护城河",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08678077697753907,
      "no_speech_prob": 1.2318574112157865e-11,
      "compression_ratio": 1.2034883720930232
    },
    {
      "start": 940.0,
      "end": 941.54,
      "text": "他们没有研发失败",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08678077697753907,
      "no_speech_prob": 1.2318574112157865e-11,
      "compression_ratio": 1.2034883720930232
    },
    {
      "start": 941.54,
      "end": 945.66,
      "text": "而是在下一盘足以重塑全球AI商业格局的大棋",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08678077697753907,
      "no_speech_prob": 1.2318574112157865e-11,
      "compression_ratio": 1.2034883720930232
    },
    {
      "start": 945.66,
      "end": 949.18,
      "text": "我们接着聊硅谷行业内部很少公开讨论的细节",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08678077697753907,
      "no_speech_prob": 1.2318574112157865e-11,
      "compression_ratio": 1.2034883720930232
    },
    {
      "start": 949.18,
      "end": 951.52,
      "text": "Inkling这套高效混合专家",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08678077697753907,
      "no_speech_prob": 1.2318574112157865e-11,
      "compression_ratio": 1.2034883720930232
    },
    {
      "start": 951.52,
      "end": 954.8,
      "text": "MOE架构背后存在不少值得深究的线索",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0995050763326978,
      "no_speech_prob": 8.852936612957496e-12,
      "compression_ratio": 0.9153846153846154
    },
    {
      "start": 954.8,
      "end": 957.76,
      "text": "资深AI架构师研读底层代码之后",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0995050763326978,
      "no_speech_prob": 8.852936612957496e-12,
      "compression_ratio": 0.9153846153846154
    },
    {
      "start": 957.76,
      "end": 960.0,
      "text": "能够发现256个路由专业",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0995050763326978,
      "no_speech_prob": 8.852936612957496e-12,
      "compression_ratio": 0.9153846153846154
    },
    {
      "start": 960.0,
      "end": 970.0,
      "text": "資深AI架構師研讀底層代碼之後,能夠發現256個路由專家單元,Sigmoid路由機制,滑動窗口注意力配比,整套工程設計思路和國內開源大模型深度求索,Deep-Seek V3高度相似。",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.15287464543392784,
      "no_speech_prob": 7.861310560852708e-12,
      "compression_ratio": 1.1211180124223603
    },
    {
      "start": 970.0,
      "end": 970.72,
      "text": "除此之外",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07213649391568323,
      "no_speech_prob": 1.4613850060629652e-11,
      "compression_ratio": 1.2106741573033708
    },
    {
      "start": 970.72,
      "end": 972.48,
      "text": "模型早期预训练阶段",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07213649391568323,
      "no_speech_prob": 1.4613850060629652e-11,
      "compression_ratio": 1.2106741573033708
    },
    {
      "start": 972.48,
      "end": 974.2,
      "text": "Thinking Machines Lab大量",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07213649391568323,
      "no_speech_prob": 1.4613850060629652e-11,
      "compression_ratio": 1.2106741573033708
    },
    {
      "start": 974.2,
      "end": 976.32,
      "text": "运用成熟开源模型生成的",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07213649391568323,
      "no_speech_prob": 1.4613850060629652e-11,
      "compression_ratio": 1.2106741573033708
    },
    {
      "start": 976.32,
      "end": 977.78,
      "text": "合成数据辅助训练",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07213649391568323,
      "no_speech_prob": 1.4613850060629652e-11,
      "compression_ratio": 1.2106741573033708
    },
    {
      "start": 977.78,
      "end": 980.68,
      "text": "大幅降低训练成本与失措周期",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07213649391568323,
      "no_speech_prob": 1.4613850060629652e-11,
      "compression_ratio": 1.2106741573033708
    },
    {
      "start": 980.68,
      "end": 982.84,
      "text": "一家估值120亿美元",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07213649391568323,
      "no_speech_prob": 1.4613850060629652e-11,
      "compression_ratio": 1.2106741573033708
    },
    {
      "start": 982.84,
      "end": 985.86,
      "text": "纯正美国背景顶尖AI实验室",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07213649391568323,
      "no_speech_prob": 1.4613850060629652e-11,
      "compression_ratio": 1.2106741573033708
    },
    {
      "start": 985.86,
      "end": 987.9,
      "text": "由前OpenAI核心负责人戴队",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07213649391568323,
      "no_speech_prob": 1.4613850060629652e-11,
      "compression_ratio": 1.2106741573033708
    },
    {
      "start": 987.9,
      "end": 990.14,
      "text": "手握英伟达高端算力资源",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07213649391568323,
      "no_speech_prob": 1.4613850060629652e-11,
      "compression_ratio": 1.2106741573033708
    },
    {
      "start": 990.14,
      "end": 991.86,
      "text": "最终推出旗舰模型",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07213649391568323,
      "no_speech_prob": 1.4613850060629652e-11,
      "compression_ratio": 1.2106741573033708
    },
    {
      "start": 991.86,
      "end": 994.38,
      "text": "大量参考国内开源模型成熟的",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07213649391568323,
      "no_speech_prob": 1.4613850060629652e-11,
      "compression_ratio": 1.2106741573033708
    },
    {
      "start": 994.38,
      "end": 995.36,
      "text": "工程优化方案",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07213649391568323,
      "no_speech_prob": 1.4613850060629652e-11,
      "compression_ratio": 1.2106741573033708
    },
    {
      "start": 995.36,
      "end": 997.42,
      "text": "很多外行看到消息第一反应",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07213649391568323,
      "no_speech_prob": 1.4613850060629652e-11,
      "compression_ratio": 1.2106741573033708
    },
    {
      "start": 997.42,
      "end": 999.02,
      "text": "会质疑团队创新能力",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07213649391568323,
      "no_speech_prob": 1.4613850060629652e-11,
      "compression_ratio": 1.2106741573033708
    },
    {
      "start": 999.02,
      "end": 1001.08,
      "text": "认为只是照搬现有成果",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07646247964156301,
      "no_speech_prob": 1.2698776925834654e-11,
      "compression_ratio": 1.2462006079027355
    },
    {
      "start": 1001.08,
      "end": 1003.36,
      "text": "站在工程与商业客观视角分析",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07646247964156301,
      "no_speech_prob": 1.2698776925834654e-11,
      "compression_ratio": 1.2462006079027355
    },
    {
      "start": 1003.36,
      "end": 1005.5,
      "text": "这是非常理性务实的选择",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07646247964156301,
      "no_speech_prob": 1.2698776925834654e-11,
      "compression_ratio": 1.2462006079027355
    },
    {
      "start": 1005.5,
      "end": 1008.58,
      "text": "最近几年海外高端算力芯片供给受限",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07646247964156301,
      "no_speech_prob": 1.2698776925834654e-11,
      "compression_ratio": 1.2462006079027355
    },
    {
      "start": 1008.58,
      "end": 1011.72,
      "text": "国内AI研发团队无法像OpenAI一样",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07646247964156301,
      "no_speech_prob": 1.2698776925834654e-11,
      "compression_ratio": 1.2462006079027355
    },
    {
      "start": 1011.72,
      "end": 1014.38,
      "text": "不计成本堆叠算力暴力训练模型",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07646247964156301,
      "no_speech_prob": 1.2698776925834654e-11,
      "compression_ratio": 1.2462006079027355
    },
    {
      "start": 1014.38,
      "end": 1016.3,
      "text": "只能持续在算法效率",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07646247964156301,
      "no_speech_prob": 1.2698776925834654e-11,
      "compression_ratio": 1.2462006079027355
    },
    {
      "start": 1016.3,
      "end": 1017.32,
      "text": "内存优化",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07646247964156301,
      "no_speech_prob": 1.2698776925834654e-11,
      "compression_ratio": 1.2462006079027355
    },
    {
      "start": 1017.32,
      "end": 1018.92,
      "text": "参数调度层面深耕",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07646247964156301,
      "no_speech_prob": 1.2698776925834654e-11,
      "compression_ratio": 1.2462006079027355
    },
    {
      "start": 1018.92,
      "end": 1022.22,
      "text": "倒逼开源模型运行效率突破全新高度",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07646247964156301,
      "no_speech_prob": 1.2698776925834654e-11,
      "compression_ratio": 1.2462006079027355
    },
    {
      "start": 1022.22,
      "end": 1024.12,
      "text": "米拉团队十分清楚这一点",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07646247964156301,
      "no_speech_prob": 1.2698776925834654e-11,
      "compression_ratio": 1.2462006079027355
    },
    {
      "start": 1024.12,
      "end": 1026.42,
      "text": "工程技术不存在面子之争",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07646247964156301,
      "no_speech_prob": 1.2698776925834654e-11,
      "compression_ratio": 1.2462006079027355
    },
    {
      "start": 1026.42,
      "end": 1029.56,
      "text": "既然同行已经验证高效率优化路线可行",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06285964965820312,
      "no_speech_prob": 1.1887392112330808e-11,
      "compression_ratio": 1.3126843657817109
    },
    {
      "start": 1029.56,
      "end": 1031.5,
      "text": "开源协议允许借鉴参考",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06285964965820312,
      "no_speech_prob": 1.1887392112330808e-11,
      "compression_ratio": 1.3126843657817109
    },
    {
      "start": 1031.5,
      "end": 1033.66,
      "text": "站在前人成果之上继续迭代",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06285964965820312,
      "no_speech_prob": 1.1887392112330808e-11,
      "compression_ratio": 1.3126843657817109
    },
    {
      "start": 1033.66,
      "end": 1035.58,
      "text": "就是最高效的选择",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06285964965820312,
      "no_speech_prob": 1.1887392112330808e-11,
      "compression_ratio": 1.3126843657817109
    },
    {
      "start": 1035.58,
      "end": 1037.56,
      "text": "也正是依托成熟的优化思路",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06285964965820312,
      "no_speech_prob": 1.1887392112330808e-11,
      "compression_ratio": 1.3126843657817109
    },
    {
      "start": 1037.56,
      "end": 1039.7,
      "text": "团队大幅压缩研发周期",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06285964965820312,
      "no_speech_prob": 1.1887392112330808e-11,
      "compression_ratio": 1.3126843657817109
    },
    {
      "start": 1039.7,
      "end": 1043.3,
      "text": "快速完成9750亿参数巨型模型的",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06285964965820312,
      "no_speech_prob": 1.1887392112330808e-11,
      "compression_ratio": 1.3126843657817109
    },
    {
      "start": 1043.3,
      "end": 1044.66,
      "text": "完整训练与迭代",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06285964965820312,
      "no_speech_prob": 1.1887392112330808e-11,
      "compression_ratio": 1.3126843657817109
    },
    {
      "start": 1044.66,
      "end": 1045.52,
      "text": "讲到这里",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06285964965820312,
      "no_speech_prob": 1.1887392112330808e-11,
      "compression_ratio": 1.3126843657817109
    },
    {
      "start": 1045.52,
      "end": 1047.72,
      "text": "大家会产生一个直击本质的疑问",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06285964965820312,
      "no_speech_prob": 1.1887392112330808e-11,
      "compression_ratio": 1.3126843657817109
    },
    {
      "start": 1047.72,
      "end": 1049.92,
      "text": "既然国内开源模型效率出众",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06285964965820312,
      "no_speech_prob": 1.1887392112330808e-11,
      "compression_ratio": 1.3126843657817109
    },
    {
      "start": 1049.92,
      "end": 1052.36,
      "text": "多款国产开源模型在代码",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06285964965820312,
      "no_speech_prob": 1.1887392112330808e-11,
      "compression_ratio": 1.3126843657817109
    },
    {
      "start": 1052.36,
      "end": 1053.3,
      "text": "逻辑推理层面",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06285964965820312,
      "no_speech_prob": 1.1887392112330808e-11,
      "compression_ratio": 1.3126843657817109
    },
    {
      "start": 1053.3,
      "end": 1055.6,
      "text": "能够对标国际顶级币源产品",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06285964965820312,
      "no_speech_prob": 1.1887392112330808e-11,
      "compression_ratio": 1.3126843657817109
    },
    {
      "start": 1055.6,
      "end": 1057.44,
      "text": "使用成本极其低廉",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08972459741540857,
      "no_speech_prob": 1.1606944572140065e-11,
      "compression_ratio": 1.217008797653959
    },
    {
      "start": 1057.44,
      "end": 1061.32,
      "text": "美国本土企业为什么不直接选用现成国产开源模型",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08972459741540857,
      "no_speech_prob": 1.1606944572140065e-11,
      "compression_ratio": 1.217008797653959
    },
    {
      "start": 1061.32,
      "end": 1063.48,
      "text": "何必耗费资金使用Inkling",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08972459741540857,
      "no_speech_prob": 1.1606944572140065e-11,
      "compression_ratio": 1.217008797653959
    },
    {
      "start": 1063.48,
      "end": 1065.9,
      "text": "额外采购Tinker平台服务",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08972459741540857,
      "no_speech_prob": 1.1606944572140065e-11,
      "compression_ratio": 1.217008797653959
    },
    {
      "start": 1065.9,
      "end": 1070.3,
      "text": "这个问题刚好戳破Thinking Machines Lab整套布局最核心的商业命脉",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08972459741540857,
      "no_speech_prob": 1.1606944572140065e-11,
      "compression_ratio": 1.217008797653959
    },
    {
      "start": 1070.3,
      "end": 1072.96,
      "text": "事实上美国资本方并不缺乏理性",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08972459741540857,
      "no_speech_prob": 1.1606944572140065e-11,
      "compression_ratio": 1.217008797653959
    },
    {
      "start": 1072.96,
      "end": 1074.86,
      "text": "在Inkling正式发布之前",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08972459741540857,
      "no_speech_prob": 1.1606944572140065e-11,
      "compression_ratio": 1.217008797653959
    },
    {
      "start": 1074.86,
      "end": 1079.0,
      "text": "美国企业界曾经掀起一波使用海外开源模型的热炒",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08972459741540857,
      "no_speech_prob": 1.1606944572140065e-11,
      "compression_ratio": 1.217008797653959
    },
    {
      "start": 1079.0,
      "end": 1083.16,
      "text": "查看OPE NRO UTR平台词源路由统计数据",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08972459741540857,
      "no_speech_prob": 1.1606944572140065e-11,
      "compression_ratio": 1.217008797653959
    },
    {
      "start": 1083.16,
      "end": 1086.16,
      "text": "能够发现高峰期有大量企业调用流量",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0660912137512767,
      "no_speech_prob": 1.5150842386235652e-11,
      "compression_ratio": 1.314540059347181
    },
    {
      "start": 1086.16,
      "end": 1087.72,
      "text": "转向海外开源模型",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0660912137512767,
      "no_speech_prob": 1.5150842386235652e-11,
      "compression_ratio": 1.314540059347181
    },
    {
      "start": 1087.72,
      "end": 1091.72,
      "text": "Coinbase这类头部企业都悄悄迁移智能体业务",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0660912137512767,
      "no_speech_prob": 1.5150842386235652e-11,
      "compression_ratio": 1.314540059347181
    },
    {
      "start": 1091.72,
      "end": 1093.78,
      "text": "直接把运营成本削减一半",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0660912137512767,
      "no_speech_prob": 1.5150842386235652e-11,
      "compression_ratio": 1.314540059347181
    },
    {
      "start": 1093.78,
      "end": 1094.52,
      "text": "好用",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0660912137512767,
      "no_speech_prob": 1.5150842386235652e-11,
      "compression_ratio": 1.314540059347181
    },
    {
      "start": 1094.52,
      "end": 1095.06,
      "text": "低价",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0660912137512767,
      "no_speech_prob": 1.5150842386235652e-11,
      "compression_ratio": 1.314540059347181
    },
    {
      "start": 1095.06,
      "end": 1096.94,
      "text": "资本自然会用脚投票",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0660912137512767,
      "no_speech_prob": 1.5150842386235652e-11,
      "compression_ratio": 1.314540059347181
    },
    {
      "start": 1096.94,
      "end": 1099.56,
      "text": "全球规模顶尖对冲基金桥水",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0660912137512767,
      "no_speech_prob": 1.5150842386235652e-11,
      "compression_ratio": 1.314540059347181
    },
    {
      "start": 1099.56,
      "end": 1102.84,
      "text": "就曾利用自身业务数据微调国内开源模型",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0660912137512767,
      "no_speech_prob": 1.5150842386235652e-11,
      "compression_ratio": 1.314540059347181
    },
    {
      "start": 1102.84,
      "end": 1105.06,
      "text": "在金融推理评测中跑出高分",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0660912137512767,
      "no_speech_prob": 1.5150842386235652e-11,
      "compression_ratio": 1.314540059347181
    },
    {
      "start": 1105.06,
      "end": 1107.9,
      "text": "全面超越多款高价通用必源模型",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0660912137512767,
      "no_speech_prob": 1.5150842386235652e-11,
      "compression_ratio": 1.314540059347181
    },
    {
      "start": 1107.9,
      "end": 1109.54,
      "text": "使用成本大幅降低",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0660912137512767,
      "no_speech_prob": 1.5150842386235652e-11,
      "compression_ratio": 1.314540059347181
    },
    {
      "start": 1109.54,
      "end": 1112.92,
      "text": "低价定制垂直模型碾压昂贵通用大模型",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0660912137512767,
      "no_speech_prob": 1.5150842386235652e-11,
      "compression_ratio": 1.314540059347181
    },
    {
      "start": 1112.92,
      "end": 1115.84,
      "text": "这笔账所有企业决策者都能够看懂",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.05124220530192057,
      "no_speech_prob": 1.2415172188917634e-11,
      "compression_ratio": 1.3072625698324023
    },
    {
      "start": 1115.84,
      "end": 1118.08,
      "text": "但是繁荣周期并没有持续太久",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.05124220530192057,
      "no_speech_prob": 1.2415172188917634e-11,
      "compression_ratio": 1.3072625698324023
    },
    {
      "start": 1118.08,
      "end": 1120.98,
      "text": "地缘政策带来的监管压力快速到来",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.05124220530192057,
      "no_speech_prob": 1.2415172188917634e-11,
      "compression_ratio": 1.3072625698324023
    },
    {
      "start": 1120.98,
      "end": 1123.92,
      "text": "由于海外开源模型普及速度超出预期",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.05124220530192057,
      "no_speech_prob": 1.2415172188917634e-11,
      "compression_ratio": 1.3072625698324023
    },
    {
      "start": 1123.92,
      "end": 1126.16,
      "text": "美国监管机构开始高度警惕",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.05124220530192057,
      "no_speech_prob": 1.2415172188917634e-11,
      "compression_ratio": 1.3072625698324023
    },
    {
      "start": 1126.16,
      "end": 1129.68,
      "text": "持续向使用海外模型的企业发送警示信函",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.05124220530192057,
      "no_speech_prob": 1.2415172188917634e-11,
      "compression_ratio": 1.3072625698324023
    },
    {
      "start": 1129.68,
      "end": 1131.8,
      "text": "国会相关委员会启动调研",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.05124220530192057,
      "no_speech_prob": 1.2415172188917634e-11,
      "compression_ratio": 1.3072625698324023
    },
    {
      "start": 1131.8,
      "end": 1134.4,
      "text": "甚至有部门草拟采购限制条款",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.05124220530192057,
      "no_speech_prob": 1.2415172188917634e-11,
      "compression_ratio": 1.3072625698324023
    },
    {
      "start": 1134.4,
      "end": 1136.28,
      "text": "手握海量隐私业务数据",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.05124220530192057,
      "no_speech_prob": 1.2415172188917634e-11,
      "compression_ratio": 1.3072625698324023
    },
    {
      "start": 1136.28,
      "end": 1138.84,
      "text": "流水规模动辄百亿的华尔街投行",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.05124220530192057,
      "no_speech_prob": 1.2415172188917634e-11,
      "compression_ratio": 1.3072625698324023
    },
    {
      "start": 1138.84,
      "end": 1142.76,
      "text": "医疗集团与科技大厂根本无法承受这类合规风险",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06513902726867389,
      "no_speech_prob": 1.2607242506901262e-11,
      "compression_ratio": 1.2614942528735633
    },
    {
      "start": 1142.76,
      "end": 1145.3,
      "text": "即便模型性能出色 价格低廉",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06513902726867389,
      "no_speech_prob": 1.2607242506901262e-11,
      "compression_ratio": 1.2614942528735633
    },
    {
      "start": 1145.3,
      "end": 1147.92,
      "text": "一旦被贴上数据安全隐患标签",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06513902726867389,
      "no_speech_prob": 1.2607242506901262e-11,
      "compression_ratio": 1.2614942528735633
    },
    {
      "start": 1147.92,
      "end": 1151.46,
      "text": "对于企业而言就是随时可能引爆的经营风险",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06513902726867389,
      "no_speech_prob": 1.2607242506901262e-11,
      "compression_ratio": 1.2614942528735633
    },
    {
      "start": 1151.46,
      "end": 1154.66,
      "text": "与此同时 曾经支撑西方开源生态的Meta",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06513902726867389,
      "no_speech_prob": 1.2607242506901262e-11,
      "compression_ratio": 1.2614942528735633
    },
    {
      "start": 1154.66,
      "end": 1156.86,
      "text": "近期持续收紧授权政策",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06513902726867389,
      "no_speech_prob": 1.2607242506901262e-11,
      "compression_ratio": 1.2614942528735633
    },
    {
      "start": 1156.86,
      "end": 1161.24,
      "text": "后续Lama系列模型逐步转向更加封闭的商业授权模式",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06513902726867389,
      "no_speech_prob": 1.2607242506901262e-11,
      "compression_ratio": 1.2614942528735633
    },
    {
      "start": 1161.24,
      "end": 1165.64,
      "text": "彻底终结了西方企业免费商用开源大模型的红利时代",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06513902726867389,
      "no_speech_prob": 1.2607242506901262e-11,
      "compression_ratio": 1.2614942528735633
    },
    {
      "start": 1165.64,
      "end": 1167.24,
      "text": "就在这个行业窗口期",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06513902726867389,
      "no_speech_prob": 1.2607242506901262e-11,
      "compression_ratio": 1.2614942528735633
    },
    {
      "start": 1167.24,
      "end": 1169.98,
      "text": "西方企业市场出现巨大需求真空",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06970722476641338,
      "no_speech_prob": 1.4264717011769346e-11,
      "compression_ratio": 1.3134796238244515
    },
    {
      "start": 1169.98,
      "end": 1173.1,
      "text": "企业迫切需要开源模型降低算力开销",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06970722476641338,
      "no_speech_prob": 1.4264717011769346e-11,
      "compression_ratio": 1.3134796238244515
    },
    {
      "start": 1173.1,
      "end": 1175.56,
      "text": "希望依靠微调保护私有数据",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06970722476641338,
      "no_speech_prob": 1.4264717011769346e-11,
      "compression_ratio": 1.3134796238244515
    },
    {
      "start": 1175.56,
      "end": 1177.74,
      "text": "但是不敢选用海外免费模型",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06970722476641338,
      "no_speech_prob": 1.4264717011769346e-11,
      "compression_ratio": 1.3134796238244515
    },
    {
      "start": 1177.74,
      "end": 1180.82,
      "text": "本土老牌开源路线又不断抬高准入门槛",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06970722476641338,
      "no_speech_prob": 1.4264717011769346e-11,
      "compression_ratio": 1.3134796238244515
    },
    {
      "start": 1180.82,
      "end": 1183.28,
      "text": "此刻再回看Inkling的发布时机",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06970722476641338,
      "no_speech_prob": 1.4264717011769346e-11,
      "compression_ratio": 1.3134796238244515
    },
    {
      "start": 1183.28,
      "end": 1184.74,
      "text": "一切都豁然开朗",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06970722476641338,
      "no_speech_prob": 1.4264717011769346e-11,
      "compression_ratio": 1.3134796238244515
    },
    {
      "start": 1184.74,
      "end": 1188.2,
      "text": "Inkling本身不需要争夺全球最强模型的名号",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06970722476641338,
      "no_speech_prob": 1.4264717011769346e-11,
      "compression_ratio": 1.3134796238244515
    },
    {
      "start": 1188.2,
      "end": 1189.38,
      "text": "它的真实定位",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06970722476641338,
      "no_speech_prob": 1.4264717011769346e-11,
      "compression_ratio": 1.3134796238244515
    },
    {
      "start": 1189.38,
      "end": 1191.8,
      "text": "是当下西方企业市场为数不多",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06970722476641338,
      "no_speech_prob": 1.4264717011769346e-11,
      "compression_ratio": 1.3134796238244515
    },
    {
      "start": 1191.8,
      "end": 1192.9,
      "text": "合规安全",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06970722476641338,
      "no_speech_prob": 1.4264717011769346e-11,
      "compression_ratio": 1.3134796238244515
    },
    {
      "start": 1192.9,
      "end": 1194.58,
      "text": "纯正美国本土血统",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06970722476641338,
      "no_speech_prob": 1.4264717011769346e-11,
      "compression_ratio": 1.3134796238244515
    },
    {
      "start": 1194.58,
      "end": 1197.38,
      "text": "且综合性能达标的开源备选方案",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07549253798209095,
      "no_speech_prob": 1.351170130642343e-11,
      "compression_ratio": 1.2903225806451613
    },
    {
      "start": 1197.38,
      "end": 1200.72,
      "text": "米拉团队不仅交付一套安全可控的底层模型",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07549253798209095,
      "no_speech_prob": 1.351170130642343e-11,
      "compression_ratio": 1.2903225806451613
    },
    {
      "start": 1200.72,
      "end": 1203.52,
      "text": "还同步推出轻量化版本Inclean Small",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07549253798209095,
      "no_speech_prob": 1.351170130642343e-11,
      "compression_ratio": 1.2903225806451613
    },
    {
      "start": 1203.52,
      "end": 1207.5,
      "text": "这款缩小版本总参数规模压缩至2760亿",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07549253798209095,
      "no_speech_prob": 1.351170130642343e-11,
      "compression_ratio": 1.2903225806451613
    },
    {
      "start": 1207.5,
      "end": 1209.84,
      "text": "单次推理仅激活120亿参数",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07549253798209095,
      "no_speech_prob": 1.351170130642343e-11,
      "compression_ratio": 1.2903225806451613
    },
    {
      "start": 1209.84,
      "end": 1212.88,
      "text": "轻量化架构适配中小企业硬件设备",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07549253798209095,
      "no_speech_prob": 1.351170130642343e-11,
      "compression_ratio": 1.2903225806451613
    },
    {
      "start": 1212.88,
      "end": 1215.6,
      "text": "在专项垂直推理场景中表现亮眼",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07549253798209095,
      "no_speech_prob": 1.351170130642343e-11,
      "compression_ratio": 1.2903225806451613
    },
    {
      "start": 1215.6,
      "end": 1218.42,
      "text": "专门面向有本地私有化部署需求",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07549253798209095,
      "no_speech_prob": 1.351170130642343e-11,
      "compression_ratio": 1.2903225806451613
    },
    {
      "start": 1218.42,
      "end": 1221.04,
      "text": "服务器硬件资源有限的中小企业",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07549253798209095,
      "no_speech_prob": 1.351170130642343e-11,
      "compression_ratio": 1.2903225806451613
    },
    {
      "start": 1221.04,
      "end": 1223.92,
      "text": "硬件配置不足可以选用轻量化版本",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07549253798209095,
      "no_speech_prob": 1.351170130642343e-11,
      "compression_ratio": 1.2903225806451613
    },
    {
      "start": 1223.92,
      "end": 1226.74,
      "text": "缺少技术人员直接采购Tinker平台",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0940118722580186,
      "no_speech_prob": 1.422031242759303e-11,
      "compression_ratio": 1.121372031662269
    },
    {
      "start": 1226.74,
      "end": 1228.66,
      "text": "担忧跨境数据合规风险",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0940118722580186,
      "no_speech_prob": 1.422031242759303e-11,
      "compression_ratio": 1.121372031662269
    },
    {
      "start": 1228.66,
      "end": 1230.92,
      "text": "它又是本土正规团队开发",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0940118722580186,
      "no_speech_prob": 1.422031242759303e-11,
      "compression_ratio": 1.121372031662269
    },
    {
      "start": 1230.92,
      "end": 1233.52,
      "text": "代码完整公开托管在哈格面部",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0940118722580186,
      "no_speech_prob": 1.422031242759303e-11,
      "compression_ratio": 1.121372031662269
    },
    {
      "start": 1233.52,
      "end": 1235.24,
      "text": "全部流程透明可查",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0940118722580186,
      "no_speech_prob": 1.422031242759303e-11,
      "compression_ratio": 1.121372031662269
    },
    {
      "start": 1235.24,
      "end": 1236.98,
      "text": "把所有线索串联起来",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0940118722580186,
      "no_speech_prob": 1.422031242759303e-11,
      "compression_ratio": 1.121372031662269
    },
    {
      "start": 1236.98,
      "end": 1240.14,
      "text": "Thinking Machines Lab完整商业版图清晰浮现",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0940118722580186,
      "no_speech_prob": 1.422031242759303e-11,
      "compression_ratio": 1.121372031662269
    },
    {
      "start": 1240.14,
      "end": 1241.96,
      "text": "离开OpenAI的MamalaD",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0940118722580186,
      "no_speech_prob": 1.422031242759303e-11,
      "compression_ratio": 1.121372031662269
    },
    {
      "start": 1241.96,
      "end": 1244.88,
      "text": "从一开始就没有打算在早已竞争白热化的",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0940118722580186,
      "no_speech_prob": 1.422031242759303e-11,
      "compression_ratio": 1.121372031662269
    },
    {
      "start": 1244.88,
      "end": 1246.18,
      "text": "通用大模型赛道",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0940118722580186,
      "no_speech_prob": 1.422031242759303e-11,
      "compression_ratio": 1.121372031662269
    },
    {
      "start": 1246.18,
      "end": 1248.04,
      "text": "复刻第二个ChatGPT",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0940118722580186,
      "no_speech_prob": 1.422031242759303e-11,
      "compression_ratio": 1.121372031662269
    },
    {
      "start": 1248.04,
      "end": 1251.78,
      "text": "那条赛道只能无休止消耗投资人资金持续内卷",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0940118722580186,
      "no_speech_prob": 1.422031242759303e-11,
      "compression_ratio": 1.121372031662269
    },
    {
      "start": 1251.78,
      "end": 1253.54,
      "text": "他依靠20亿美元资金",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0739037309374128,
      "no_speech_prob": 1.3004006725603201e-11,
      "compression_ratio": 1.2113821138211383
    },
    {
      "start": 1253.54,
      "end": 1255.76,
      "text": "在OpenAI币源商业帝国旁边",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0739037309374128,
      "no_speech_prob": 1.3004006725603201e-11,
      "compression_ratio": 1.2113821138211383
    },
    {
      "start": 1255.76,
      "end": 1256.96,
      "text": "开辟全新赛道",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0739037309374128,
      "no_speech_prob": 1.3004006725603201e-11,
      "compression_ratio": 1.2113821138211383
    },
    {
      "start": 1256.96,
      "end": 1259.0,
      "text": "精准预判人工智能未来形态",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0739037309374128,
      "no_speech_prob": 1.3004006725603201e-11,
      "compression_ratio": 1.2113821138211383
    },
    {
      "start": 1259.0,
      "end": 1261.68,
      "text": "不会是无所不能的通用超级大脑",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0739037309374128,
      "no_speech_prob": 1.3004006725603201e-11,
      "compression_ratio": 1.2113821138211383
    },
    {
      "start": 1261.68,
      "end": 1263.96,
      "text": "而是千千万万扎根细分行业",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0739037309374128,
      "no_speech_prob": 1.3004006725603201e-11,
      "compression_ratio": 1.2113821138211383
    },
    {
      "start": 1263.96,
      "end": 1267.9,
      "text": "低成本运行数据归属企业自身的领域专用模型",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0739037309374128,
      "no_speech_prob": 1.3004006725603201e-11,
      "compression_ratio": 1.2113821138211383
    },
    {
      "start": 1267.9,
      "end": 1270.64,
      "text": "借鉴经过市场验证的高效开源架构",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0739037309374128,
      "no_speech_prob": 1.3004006725603201e-11,
      "compression_ratio": 1.2113821138211383
    },
    {
      "start": 1270.64,
      "end": 1273.84,
      "text": "搭配团队成熟的成本控制方案打造Inkling",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0739037309374128,
      "no_speech_prob": 1.3004006725603201e-11,
      "compression_ratio": 1.2113821138211383
    },
    {
      "start": 1273.84,
      "end": 1275.1,
      "text": "作为引流载体",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0739037309374128,
      "no_speech_prob": 1.3004006725603201e-11,
      "compression_ratio": 1.2113821138211383
    },
    {
      "start": 1275.1,
      "end": 1277.72,
      "text": "再借助地缘环境催生的合规焦虑",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0739037309374128,
      "no_speech_prob": 1.3004006725603201e-11,
      "compression_ratio": 1.2113821138211383
    },
    {
      "start": 1277.72,
      "end": 1280.04,
      "text": "吸引急需解决方案的企业客户",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0739037309374128,
      "no_speech_prob": 1.3004006725603201e-11,
      "compression_ratio": 1.2113821138211383
    },
    {
      "start": 1280.04,
      "end": 1282.4,
      "text": "进入Tinker微调服务生态",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07179425068390675,
      "no_speech_prob": 1.2452913700222723e-11,
      "compression_ratio": 1.2283236994219653
    },
    {
      "start": 1282.4,
      "end": 1284.46,
      "text": "花费20亿美元打造一款榜单",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07179425068390675,
      "no_speech_prob": 1.2452913700222723e-11,
      "compression_ratio": 1.2283236994219653
    },
    {
      "start": 1284.46,
      "end": 1286.16,
      "text": "无法跻身前五的模型",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07179425068390675,
      "no_speech_prob": 1.2452913700222723e-11,
      "compression_ratio": 1.2283236994219653
    },
    {
      "start": 1286.16,
      "end": 1288.5,
      "text": "表面看上去像是一场失败投资",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07179425068390675,
      "no_speech_prob": 1.2452913700222723e-11,
      "compression_ratio": 1.2283236994219653
    },
    {
      "start": 1288.5,
      "end": 1291.3,
      "text": "但是倘若这20亿美元换来未来10年",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07179425068390675,
      "no_speech_prob": 1.2452913700222723e-11,
      "compression_ratio": 1.2283236994219653
    },
    {
      "start": 1291.3,
      "end": 1294.12,
      "text": "全球企业及AI定制市场核心入口",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07179425068390675,
      "no_speech_prob": 1.2452913700222723e-11,
      "compression_ratio": 1.2283236994219653
    },
    {
      "start": 1294.12,
      "end": 1296.62,
      "text": "这就是教科书级别的商业布局",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07179425068390675,
      "no_speech_prob": 1.2452913700222723e-11,
      "compression_ratio": 1.2283236994219653
    },
    {
      "start": 1296.62,
      "end": 1299.96,
      "text": "当所有人紧盯评测排行榜相互竞争的时候",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07179425068390675,
      "no_speech_prob": 1.2452913700222723e-11,
      "compression_ratio": 1.2283236994219653
    },
    {
      "start": 1299.96,
      "end": 1303.62,
      "text": "已经有人提前锁定人工智能淘金浪潮当中",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07179425068390675,
      "no_speech_prob": 1.2452913700222723e-11,
      "compression_ratio": 1.2283236994219653
    },
    {
      "start": 1303.62,
      "end": 1305.48,
      "text": "合规产子的独家供应权",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07179425068390675,
      "no_speech_prob": 1.2452913700222723e-11,
      "compression_ratio": 1.2283236994219653
    },
    {
      "start": 1305.48,
      "end": 1307.8,
      "text": "这也是120亿美元估值背后",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07179425068390675,
      "no_speech_prob": 1.2452913700222723e-11,
      "compression_ratio": 1.2283236994219653
    },
    {
      "start": 1307.8,
      "end": 1310.24,
      "text": "最值得细细思考的行业真相",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08202179128473455,
      "no_speech_prob": 1.4553388008597956e-11,
      "compression_ratio": 1.2899408284023668
    },
    {
      "start": 1310.24,
      "end": 1311.84,
      "text": "其实我们在延伸一层",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08202179128473455,
      "no_speech_prob": 1.4553388008597956e-11,
      "compression_ratio": 1.2899408284023668
    },
    {
      "start": 1311.84,
      "end": 1315.08,
      "text": "你就能看懂硅谷资本更深层的赌具逻辑",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08202179128473455,
      "no_speech_prob": 1.4553388008597956e-11,
      "compression_ratio": 1.2899408284023668
    },
    {
      "start": 1315.08,
      "end": 1318.0,
      "text": "近两年无数AI创业公司扎堆入场",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08202179128473455,
      "no_speech_prob": 1.4553388008597956e-11,
      "compression_ratio": 1.2899408284023668
    },
    {
      "start": 1318.0,
      "end": 1320.22,
      "text": "大多都在跟风做通用大模型",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08202179128473455,
      "no_speech_prob": 1.4553388008597956e-11,
      "compression_ratio": 1.2899408284023668
    },
    {
      "start": 1320.22,
      "end": 1321.5,
      "text": "做对话机器人",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08202179128473455,
      "no_speech_prob": 1.4553388008597956e-11,
      "compression_ratio": 1.2899408284023668
    },
    {
      "start": 1321.5,
      "end": 1323.14,
      "text": "做花哨的多模态演示",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08202179128473455,
      "no_speech_prob": 1.4553388008597956e-11,
      "compression_ratio": 1.2899408284023668
    },
    {
      "start": 1323.14,
      "end": 1325.84,
      "text": "真正能落地赚钱的寥寥无几",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08202179128473455,
      "no_speech_prob": 1.4553388008597956e-11,
      "compression_ratio": 1.2899408284023668
    },
    {
      "start": 1325.84,
      "end": 1327.82,
      "text": "绝大多数AI初创企业",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08202179128473455,
      "no_speech_prob": 1.4553388008597956e-11,
      "compression_ratio": 1.2899408284023668
    },
    {
      "start": 1327.82,
      "end": 1329.48,
      "text": "盈利模式极度单一",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08202179128473455,
      "no_speech_prob": 1.4553388008597956e-11,
      "compression_ratio": 1.2899408284023668
    },
    {
      "start": 1329.48,
      "end": 1330.98,
      "text": "要么卖API调用",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08202179128473455,
      "no_speech_prob": 1.4553388008597956e-11,
      "compression_ratio": 1.2899408284023668
    },
    {
      "start": 1330.98,
      "end": 1332.26,
      "text": "要么卖会员套餐",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08202179128473455,
      "no_speech_prob": 1.4553388008597956e-11,
      "compression_ratio": 1.2899408284023668
    },
    {
      "start": 1332.26,
      "end": 1334.76,
      "text": "陷入无尽的价格战和算力内卷",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08202179128473455,
      "no_speech_prob": 1.4553388008597956e-11,
      "compression_ratio": 1.2899408284023668
    },
    {
      "start": 1334.76,
      "end": 1336.28,
      "text": "稍万融资就倒闭",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08202179128473455,
      "no_speech_prob": 1.4553388008597956e-11,
      "compression_ratio": 1.2899408284023668
    },
    {
      "start": 1336.28,
      "end": 1337.56,
      "text": "存活率极低",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08202179128473455,
      "no_speech_prob": 1.4553388008597956e-11,
      "compression_ratio": 1.2899408284023668
    },
    {
      "start": 1337.8,
      "end": 1340.32,
      "text": "而SY NTNG Mac King ES Lab",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.10015787661654278,
      "no_speech_prob": 1.532086263411614e-11,
      "compression_ratio": 1.2305295950155763
    },
    {
      "start": 1340.32,
      "end": 1341.84,
      "text": "彻底避开了这片红海",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.10015787661654278,
      "no_speech_prob": 1.532086263411614e-11,
      "compression_ratio": 1.2305295950155763
    },
    {
      "start": 1341.84,
      "end": 1343.94,
      "text": "它不跟任何人抢C端用户",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.10015787661654278,
      "no_speech_prob": 1.532086263411614e-11,
      "compression_ratio": 1.2305295950155763
    },
    {
      "start": 1343.94,
      "end": 1344.9,
      "text": "不卷跑分",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.10015787661654278,
      "no_speech_prob": 1.532086263411614e-11,
      "compression_ratio": 1.2305295950155763
    },
    {
      "start": 1344.9,
      "end": 1346.14,
      "text": "不卷花式功能",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.10015787661654278,
      "no_speech_prob": 1.532086263411614e-11,
      "compression_ratio": 1.2305295950155763
    },
    {
      "start": 1346.14,
      "end": 1347.86,
      "text": "直接切入最赚钱",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.10015787661654278,
      "no_speech_prob": 1.532086263411614e-11,
      "compression_ratio": 1.2305295950155763
    },
    {
      "start": 1347.86,
      "end": 1348.5,
      "text": "最稳定",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.10015787661654278,
      "no_speech_prob": 1.532086263411614e-11,
      "compression_ratio": 1.2305295950155763
    },
    {
      "start": 1348.5,
      "end": 1351.0,
      "text": "刚需最强的弊端企业服务市场",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.10015787661654278,
      "no_speech_prob": 1.532086263411614e-11,
      "compression_ratio": 1.2305295950155763
    },
    {
      "start": 1351.0,
      "end": 1352.72,
      "text": "对于大型企业来说",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.10015787661654278,
      "no_speech_prob": 1.532086263411614e-11,
      "compression_ratio": 1.2305295950155763
    },
    {
      "start": 1352.72,
      "end": 1353.6,
      "text": "AI成本",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.10015787661654278,
      "no_speech_prob": 1.532086263411614e-11,
      "compression_ratio": 1.2305295950155763
    },
    {
      "start": 1353.6,
      "end": 1354.54,
      "text": "数据安全",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.10015787661654278,
      "no_speech_prob": 1.532086263411614e-11,
      "compression_ratio": 1.2305295950155763
    },
    {
      "start": 1354.54,
      "end": 1355.62,
      "text": "私有化部署",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.10015787661654278,
      "no_speech_prob": 1.532086263411614e-11,
      "compression_ratio": 1.2305295950155763
    },
    {
      "start": 1355.62,
      "end": 1356.64,
      "text": "定制化适配",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.10015787661654278,
      "no_speech_prob": 1.532086263411614e-11,
      "compression_ratio": 1.2305295950155763
    },
    {
      "start": 1356.64,
      "end": 1358.34,
      "text": "是比模型聪不聪明",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.10015787661654278,
      "no_speech_prob": 1.532086263411614e-11,
      "compression_ratio": 1.2305295950155763
    },
    {
      "start": 1358.34,
      "end": 1359.56,
      "text": "更重要的刚需",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.10015787661654278,
      "no_speech_prob": 1.532086263411614e-11,
      "compression_ratio": 1.2305295950155763
    },
    {
      "start": 1359.56,
      "end": 1361.26,
      "text": "哪怕通用能力平庸",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.10015787661654278,
      "no_speech_prob": 1.532086263411614e-11,
      "compression_ratio": 1.2305295950155763
    },
    {
      "start": 1361.26,
      "end": 1362.24,
      "text": "只要足够稳",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.10015787661654278,
      "no_speech_prob": 1.532086263411614e-11,
      "compression_ratio": 1.2305295950155763
    },
    {
      "start": 1362.24,
      "end": 1363.18,
      "text": "足够便宜",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.10015787661654278,
      "no_speech_prob": 1.532086263411614e-11,
      "compression_ratio": 1.2305295950155763
    },
    {
      "start": 1363.18,
      "end": 1364.12,
      "text": "足够安全",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.10015787661654278,
      "no_speech_prob": 1.532086263411614e-11,
      "compression_ratio": 1.2305295950155763
    },
    {
      "start": 1364.12,
      "end": 1365.22,
      "text": "足够好微调",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.10015787661654278,
      "no_speech_prob": 1.532086263411614e-11,
      "compression_ratio": 1.2305295950155763
    },
    {
      "start": 1365.22,
      "end": 1366.54,
      "text": "就是企业愿意",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.10015787661654278,
      "no_speech_prob": 1.532086263411614e-11,
      "compression_ratio": 1.2305295950155763
    },
    {
      "start": 1366.54,
      "end": 1368.24,
      "text": "长期付费的核心资产",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07460912927850946,
      "no_speech_prob": 1.9076126095418644e-11,
      "compression_ratio": 1.2607449856733524
    },
    {
      "start": 1368.24,
      "end": 1369.94,
      "text": "这也是为什么A16Z",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07460912927850946,
      "no_speech_prob": 1.9076126095418644e-11,
      "compression_ratio": 1.2607449856733524
    },
    {
      "start": 1369.94,
      "end": 1371.82,
      "text": "敢于在没有成熟产品",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07460912927850946,
      "no_speech_prob": 1.9076126095418644e-11,
      "compression_ratio": 1.2607449856733524
    },
    {
      "start": 1371.82,
      "end": 1373.7,
      "text": "没有实际营收的情况下",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07460912927850946,
      "no_speech_prob": 1.9076126095418644e-11,
      "compression_ratio": 1.2607449856733524
    },
    {
      "start": 1373.7,
      "end": 1374.74,
      "text": "豪至20亿美金",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07460912927850946,
      "no_speech_prob": 1.9076126095418644e-11,
      "compression_ratio": 1.2607449856733524
    },
    {
      "start": 1374.74,
      "end": 1377.04,
      "text": "资本赌的不是一款模型的输赢",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07460912927850946,
      "no_speech_prob": 1.9076126095418644e-11,
      "compression_ratio": 1.2607449856733524
    },
    {
      "start": 1377.04,
      "end": 1379.36,
      "text": "而是赌未来企业AI私有化",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07460912927850946,
      "no_speech_prob": 1.9076126095418644e-11,
      "compression_ratio": 1.2607449856733524
    },
    {
      "start": 1379.36,
      "end": 1380.16,
      "text": "定制化",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07460912927850946,
      "no_speech_prob": 1.9076126095418644e-11,
      "compression_ratio": 1.2607449856733524
    },
    {
      "start": 1380.16,
      "end": 1382.02,
      "text": "合规化的万亿级风口",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07460912927850946,
      "no_speech_prob": 1.9076126095418644e-11,
      "compression_ratio": 1.2607449856733524
    },
    {
      "start": 1382.02,
      "end": 1384.32,
      "text": "反观行业内很多盲目堆算力",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07460912927850946,
      "no_speech_prob": 1.9076126095418644e-11,
      "compression_ratio": 1.2607449856733524
    },
    {
      "start": 1384.32,
      "end": 1385.1,
      "text": "卷参数",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07460912927850946,
      "no_speech_prob": 1.9076126095418644e-11,
      "compression_ratio": 1.2607449856733524
    },
    {
      "start": 1385.1,
      "end": 1386.22,
      "text": "卷跑分的团队",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07460912927850946,
      "no_speech_prob": 1.9076126095418644e-11,
      "compression_ratio": 1.2607449856733524
    },
    {
      "start": 1386.22,
      "end": 1387.48,
      "text": "看似热度爆棚",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07460912927850946,
      "no_speech_prob": 1.9076126095418644e-11,
      "compression_ratio": 1.2607449856733524
    },
    {
      "start": 1387.48,
      "end": 1388.22,
      "text": "数据亮眼",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07460912927850946,
      "no_speech_prob": 1.9076126095418644e-11,
      "compression_ratio": 1.2607449856733524
    },
    {
      "start": 1388.22,
      "end": 1390.68,
      "text": "实则完全偏离了真实商业需求",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07460912927850946,
      "no_speech_prob": 1.9076126095418644e-11,
      "compression_ratio": 1.2607449856733524
    },
    {
      "start": 1390.68,
      "end": 1391.86,
      "text": "最终只能沦为",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07460912927850946,
      "no_speech_prob": 1.9076126095418644e-11,
      "compression_ratio": 1.2607449856733524
    },
    {
      "start": 1391.86,
      "end": 1393.38,
      "text": "昙花一现的行业过客",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07460912927850946,
      "no_speech_prob": 1.9076126095418644e-11,
      "compression_ratio": 1.2607449856733524
    },
    {
      "start": 1393.38,
      "end": 1394.34,
      "text": "除此之外",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07460912927850946,
      "no_speech_prob": 1.9076126095418644e-11,
      "compression_ratio": 1.2607449856733524
    },
    {
      "start": 1394.34,
      "end": 1395.56,
      "text": "米拉团队的布局",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07460912927850946,
      "no_speech_prob": 1.9076126095418644e-11,
      "compression_ratio": 1.2607449856733524
    },
    {
      "start": 1395.56,
      "end": 1397.44,
      "text": "还有一个极具远见的细节",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.09317830029655905,
      "no_speech_prob": 1.4091793704151812e-11,
      "compression_ratio": 1.172106824925816
    },
    {
      "start": 1397.44,
      "end": 1399.04,
      "text": "那就是彻底放弃",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.09317830029655905,
      "no_speech_prob": 1.4091793704151812e-11,
      "compression_ratio": 1.172106824925816
    },
    {
      "start": 1399.04,
      "end": 1400.54,
      "text": "避原锁库的传统玩法",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.09317830029655905,
      "no_speech_prob": 1.4091793704151812e-11,
      "compression_ratio": 1.172106824925816
    },
    {
      "start": 1400.54,
      "end": 1402.54,
      "text": "坚持全开源 全透明",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.09317830029655905,
      "no_speech_prob": 1.4091793704151812e-11,
      "compression_ratio": 1.172106824925816
    },
    {
      "start": 1402.54,
      "end": 1404.6,
      "text": "在当下AI舆论环境中",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.09317830029655905,
      "no_speech_prob": 1.4091793704151812e-11,
      "compression_ratio": 1.172106824925816
    },
    {
      "start": 1404.6,
      "end": 1407.02,
      "text": "避原模型随时面临授权变动",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.09317830029655905,
      "no_speech_prob": 1.4091793704151812e-11,
      "compression_ratio": 1.172106824925816
    },
    {
      "start": 1407.02,
      "end": 1408.9,
      "text": "服务关停 接口涨价",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.09317830029655905,
      "no_speech_prob": 1.4091793704151812e-11,
      "compression_ratio": 1.172106824925816
    },
    {
      "start": 1408.9,
      "end": 1410.18,
      "text": "功能阉割的风险",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.09317830029655905,
      "no_speech_prob": 1.4091793704151812e-11,
      "compression_ratio": 1.172106824925816
    },
    {
      "start": 1410.18,
      "end": 1412.34,
      "text": "企业用户极其没有安全感",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.09317830029655905,
      "no_speech_prob": 1.4091793704151812e-11,
      "compression_ratio": 1.172106824925816
    },
    {
      "start": 1412.34,
      "end": 1414.82,
      "text": "而Inkling的Apache 2.0协议",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.09317830029655905,
      "no_speech_prob": 1.4091793704151812e-11,
      "compression_ratio": 1.172106824925816
    },
    {
      "start": 1414.82,
      "end": 1416.9,
      "text": "给予了企业最大的自由度",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.09317830029655905,
      "no_speech_prob": 1.4091793704151812e-11,
      "compression_ratio": 1.172106824925816
    },
    {
      "start": 1416.9,
      "end": 1418.84,
      "text": "可商用 可二次开发",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.09317830029655905,
      "no_speech_prob": 1.4091793704151812e-11,
      "compression_ratio": 1.172106824925816
    },
    {
      "start": 1418.84,
      "end": 1421.34,
      "text": "可私有化部署 无板权捆绑",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.09317830029655905,
      "no_speech_prob": 1.4091793704151812e-11,
      "compression_ratio": 1.172106824925816
    },
    {
      "start": 1421.34,
      "end": 1423.14,
      "text": "这种极致开放的姿态",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.09317830029655905,
      "no_speech_prob": 1.4091793704151812e-11,
      "compression_ratio": 1.172106824925816
    },
    {
      "start": 1423.14,
      "end": 1425.86,
      "text": "刚好击中了当下企业最核心的诉求",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08780728181203207,
      "no_speech_prob": 1.1562059469560904e-11,
      "compression_ratio": 1.1492537313432836
    },
    {
      "start": 1425.86,
      "end": 1427.08,
      "text": "免费拿模型",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08780728181203207,
      "no_speech_prob": 1.1562059469560904e-11,
      "compression_ratio": 1.1492537313432836
    },
    {
      "start": 1427.08,
      "end": 1428.18,
      "text": "免费改模型",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08780728181203207,
      "no_speech_prob": 1.1562059469560904e-11,
      "compression_ratio": 1.1492537313432836
    },
    {
      "start": 1428.18,
      "end": 1429.02,
      "text": "免费部署",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08780728181203207,
      "no_speech_prob": 1.1562059469560904e-11,
      "compression_ratio": 1.1492537313432836
    },
    {
      "start": 1429.02,
      "end": 1431.36,
      "text": "最后只为精细化微调服务",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08780728181203207,
      "no_speech_prob": 1.1562059469560904e-11,
      "compression_ratio": 1.1492537313432836
    },
    {
      "start": 1431.36,
      "end": 1433.06,
      "text": "自动化运为工具付费",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08780728181203207,
      "no_speech_prob": 1.1562059469560904e-11,
      "compression_ratio": 1.1492537313432836
    },
    {
      "start": 1433.06,
      "end": 1434.3,
      "text": "这种先让利",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08780728181203207,
      "no_speech_prob": 1.1562059469560904e-11,
      "compression_ratio": 1.1492537313432836
    },
    {
      "start": 1434.3,
      "end": 1435.98,
      "text": "后所刻的商业模式",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08780728181203207,
      "no_speech_prob": 1.1562059469560904e-11,
      "compression_ratio": 1.1492537313432836
    },
    {
      "start": 1435.98,
      "end": 1438.3,
      "text": "远比一次性售卖API要长久",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08780728181203207,
      "no_speech_prob": 1.1562059469560904e-11,
      "compression_ratio": 1.1492537313432836
    },
    {
      "start": 1438.3,
      "end": 1439.16,
      "text": "稳固得多",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08780728181203207,
      "no_speech_prob": 1.1562059469560904e-11,
      "compression_ratio": 1.1492537313432836
    },
    {
      "start": 1440.0,
      "end": 1452.44,
      "text": "很多人始終想不通一個核心問題,為什麼無數大廠耗費巨資打造的頂級通用大模型,落地企業服務時頻頻翻車,反而是Inclin這種跑分中庸的模型,精準踩中了行業痛點。",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.1524671945512665,
      "no_speech_prob": 7.841343026282477e-12,
      "compression_ratio": 1.2604790419161678
    },
    {
      "start": 1452.44,
      "end": 1455.4,
      "text": "答案其实藏在企业真实的AI落地困境里",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08178435133096758,
      "no_speech_prob": 1.3058302703039537e-11,
      "compression_ratio": 1.290625
    },
    {
      "start": 1455.4,
      "end": 1457.16,
      "text": "绝大多数传统企业",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08178435133096758,
      "no_speech_prob": 1.3058302703039537e-11,
      "compression_ratio": 1.290625
    },
    {
      "start": 1457.16,
      "end": 1458.1,
      "text": "金融机构",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08178435133096758,
      "no_speech_prob": 1.3058302703039537e-11,
      "compression_ratio": 1.290625
    },
    {
      "start": 1458.1,
      "end": 1459.08,
      "text": "医疗集团",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08178435133096758,
      "no_speech_prob": 1.3058302703039537e-11,
      "compression_ratio": 1.290625
    },
    {
      "start": 1459.08,
      "end": 1460.22,
      "text": "制造业工厂",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08178435133096758,
      "no_speech_prob": 1.3058302703039537e-11,
      "compression_ratio": 1.290625
    },
    {
      "start": 1460.22,
      "end": 1462.2,
      "text": "根本不需要模型创作诗歌",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08178435133096758,
      "no_speech_prob": 1.3058302703039537e-11,
      "compression_ratio": 1.290625
    },
    {
      "start": 1462.2,
      "end": 1463.02,
      "text": "撰写文案",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08178435133096758,
      "no_speech_prob": 1.3058302703039537e-11,
      "compression_ratio": 1.290625
    },
    {
      "start": 1463.02,
      "end": 1464.14,
      "text": "完成趣味问答",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08178435133096758,
      "no_speech_prob": 1.3058302703039537e-11,
      "compression_ratio": 1.290625
    },
    {
      "start": 1464.14,
      "end": 1466.54,
      "text": "也不需要极致的顶尖推力跑分",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08178435133096758,
      "no_speech_prob": 1.3058302703039537e-11,
      "compression_ratio": 1.290625
    },
    {
      "start": 1466.54,
      "end": 1468.6,
      "text": "他们的核心需求极其朴素",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08178435133096758,
      "no_speech_prob": 1.3058302703039537e-11,
      "compression_ratio": 1.290625
    },
    {
      "start": 1468.6,
      "end": 1470.4,
      "text": "就是稳定低成本",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08178435133096758,
      "no_speech_prob": 1.3058302703039537e-11,
      "compression_ratio": 1.290625
    },
    {
      "start": 1470.4,
      "end": 1470.98,
      "text": "可管控",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08178435133096758,
      "no_speech_prob": 1.3058302703039537e-11,
      "compression_ratio": 1.290625
    },
    {
      "start": 1470.98,
      "end": 1471.68,
      "text": "不幻觉",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08178435133096758,
      "no_speech_prob": 1.3058302703039537e-11,
      "compression_ratio": 1.290625
    },
    {
      "start": 1471.68,
      "end": 1473.48,
      "text": "能对接自身业务数据",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08178435133096758,
      "no_speech_prob": 1.3058302703039537e-11,
      "compression_ratio": 1.290625
    },
    {
      "start": 1473.48,
      "end": 1475.28,
      "text": "同时全程保障数据",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08178435133096758,
      "no_speech_prob": 1.3058302703039537e-11,
      "compression_ratio": 1.290625
    },
    {
      "start": 1475.28,
      "end": 1475.92,
      "text": "不外泄",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08178435133096758,
      "no_speech_prob": 1.3058302703039537e-11,
      "compression_ratio": 1.290625
    },
    {
      "start": 1475.92,
      "end": 1476.92,
      "text": "过去几年",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08178435133096758,
      "no_speech_prob": 1.3058302703039537e-11,
      "compression_ratio": 1.290625
    },
    {
      "start": 1476.92,
      "end": 1480.22,
      "text": "大量企业跟风接入各类顶尖臂源大模型API",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08178435133096758,
      "no_speech_prob": 1.3058302703039537e-11,
      "compression_ratio": 1.290625
    },
    {
      "start": 1480.22,
      "end": 1482.78,
      "text": "看似快速完成了AI数字化升级",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08832093412652979,
      "no_speech_prob": 1.1501766553068116e-11,
      "compression_ratio": 1.275229357798165
    },
    {
      "start": 1482.78,
      "end": 1484.54,
      "text": "实则埋下了无数隐患",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08832093412652979,
      "no_speech_prob": 1.1501766553068116e-11,
      "compression_ratio": 1.275229357798165
    },
    {
      "start": 1484.54,
      "end": 1487.0,
      "text": "首先是持续走高的调用成本",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08832093412652979,
      "no_speech_prob": 1.1501766553068116e-11,
      "compression_ratio": 1.275229357798165
    },
    {
      "start": 1487.0,
      "end": 1491.1,
      "text": "企业日均数万数十万次的重复性基础调用",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08832093412652979,
      "no_speech_prob": 1.1501766553068116e-11,
      "compression_ratio": 1.275229357798165
    },
    {
      "start": 1491.1,
      "end": 1492.7,
      "text": "累积下来的算力开支",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08832093412652979,
      "no_speech_prob": 1.1501766553068116e-11,
      "compression_ratio": 1.275229357798165
    },
    {
      "start": 1492.7,
      "end": 1494.8,
      "text": "每年动辄数百万甚至上千万",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08832093412652979,
      "no_speech_prob": 1.1501766553068116e-11,
      "compression_ratio": 1.275229357798165
    },
    {
      "start": 1494.8,
      "end": 1496.6,
      "text": "长期投入完全没有上线",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08832093412652979,
      "no_speech_prob": 1.1501766553068116e-11,
      "compression_ratio": 1.275229357798165
    },
    {
      "start": 1496.6,
      "end": 1498.56,
      "text": "其次是数据泄露风险",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08832093412652979,
      "no_speech_prob": 1.1501766553068116e-11,
      "compression_ratio": 1.275229357798165
    },
    {
      "start": 1498.56,
      "end": 1500.46,
      "text": "企业内部的客户资料",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08832093412652979,
      "no_speech_prob": 1.1501766553068116e-11,
      "compression_ratio": 1.275229357798165
    },
    {
      "start": 1500.46,
      "end": 1501.46,
      "text": "财务报表",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08832093412652979,
      "no_speech_prob": 1.1501766553068116e-11,
      "compression_ratio": 1.275229357798165
    },
    {
      "start": 1501.46,
      "end": 1502.54,
      "text": "生产数据",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08832093412652979,
      "no_speech_prob": 1.1501766553068116e-11,
      "compression_ratio": 1.275229357798165
    },
    {
      "start": 1502.54,
      "end": 1503.56,
      "text": "并立信息",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08832093412652979,
      "no_speech_prob": 1.1501766553068116e-11,
      "compression_ratio": 1.275229357798165
    },
    {
      "start": 1503.56,
      "end": 1506.7,
      "text": "一旦上传至第三方B元模型云端服务器",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08832093412652979,
      "no_speech_prob": 1.1501766553068116e-11,
      "compression_ratio": 1.275229357798165
    },
    {
      "start": 1506.7,
      "end": 1508.6,
      "text": "就彻底脱离了自身管控",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.08832093412652979,
      "no_speech_prob": 1.1501766553068116e-11,
      "compression_ratio": 1.275229357798165
    },
    {
      "start": 1508.6,
      "end": 1510.76,
      "text": "随时可能面临数据抓取",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06795436143875122,
      "no_speech_prob": 1.2032417597007683e-11,
      "compression_ratio": 1.3714285714285714
    },
    {
      "start": 1510.76,
      "end": 1511.96,
      "text": "缓存留存",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06795436143875122,
      "no_speech_prob": 1.2032417597007683e-11,
      "compression_ratio": 1.3714285714285714
    },
    {
      "start": 1511.96,
      "end": 1513.44,
      "text": "合规溯源的风险",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06795436143875122,
      "no_speech_prob": 1.2032417597007683e-11,
      "compression_ratio": 1.3714285714285714
    },
    {
      "start": 1513.44,
      "end": 1514.62,
      "text": "更关键的是",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06795436143875122,
      "no_speech_prob": 1.2032417597007683e-11,
      "compression_ratio": 1.3714285714285714
    },
    {
      "start": 1514.62,
      "end": 1516.78,
      "text": "通用大模型的过度聪明",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06795436143875122,
      "no_speech_prob": 1.2032417597007683e-11,
      "compression_ratio": 1.3714285714285714
    },
    {
      "start": 1516.78,
      "end": 1518.8,
      "text": "反而成为企业落地的累赘",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06795436143875122,
      "no_speech_prob": 1.2032417597007683e-11,
      "compression_ratio": 1.3714285714285714
    },
    {
      "start": 1518.8,
      "end": 1521.04,
      "text": "很多企业业务只需要标准化",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06795436143875122,
      "no_speech_prob": 1.2032417597007683e-11,
      "compression_ratio": 1.3714285714285714
    },
    {
      "start": 1521.04,
      "end": 1522.4,
      "text": "流程化的固定输出",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06795436143875122,
      "no_speech_prob": 1.2032417597007683e-11,
      "compression_ratio": 1.3714285714285714
    },
    {
      "start": 1522.4,
      "end": 1524.22,
      "text": "不需要模型自由发挥",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06795436143875122,
      "no_speech_prob": 1.2032417597007683e-11,
      "compression_ratio": 1.3714285714285714
    },
    {
      "start": 1524.22,
      "end": 1525.34,
      "text": "发散创作",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06795436143875122,
      "no_speech_prob": 1.2032417597007683e-11,
      "compression_ratio": 1.3714285714285714
    },
    {
      "start": 1525.34,
      "end": 1528.04,
      "text": "通用模型自带的高自由度推理能力",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06795436143875122,
      "no_speech_prob": 1.2032417597007683e-11,
      "compression_ratio": 1.3714285714285714
    },
    {
      "start": 1528.04,
      "end": 1529.7,
      "text": "不仅会增加算力损耗",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06795436143875122,
      "no_speech_prob": 1.2032417597007683e-11,
      "compression_ratio": 1.3714285714285714
    },
    {
      "start": 1529.7,
      "end": 1532.3,
      "text": "还容易输出不符合企业规章制度",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06795436143875122,
      "no_speech_prob": 1.2032417597007683e-11,
      "compression_ratio": 1.3714285714285714
    },
    {
      "start": 1532.3,
      "end": 1533.48,
      "text": "行业规范的内容",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06795436143875122,
      "no_speech_prob": 1.2032417597007683e-11,
      "compression_ratio": 1.3714285714285714
    },
    {
      "start": 1533.48,
      "end": 1535.48,
      "text": "需要人工二次审核修正",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06795436143875122,
      "no_speech_prob": 1.2032417597007683e-11,
      "compression_ratio": 1.3714285714285714
    },
    {
      "start": 1535.48,
      "end": 1537.18,
      "text": "反而降低了工作效率",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06795436143875122,
      "no_speech_prob": 1.2032417597007683e-11,
      "compression_ratio": 1.3714285714285714
    },
    {
      "start": 1537.18,
      "end": 1538.92,
      "text": "而Inclin的全套设计",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.07185758924270425,
      "no_speech_prob": 1.3367668950936551e-11,
      "compression_ratio": 1.2477611940298508
    },
    {
      "start": 1538.92,
      "end": 1541.48,
      "text": "恰好针对性解决了这些行业顽疾",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.07185758924270425,
      "no_speech_prob": 1.3367668950936551e-11,
      "compression_ratio": 1.2477611940298508
    },
    {
      "start": 1541.48,
      "end": 1544.16,
      "text": "可调节的思考力度杜绝算力浪费",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.07185758924270425,
      "no_speech_prob": 1.3367668950936551e-11,
      "compression_ratio": 1.2477611940298508
    },
    {
      "start": 1544.16,
      "end": 1546.1,
      "text": "极致诚实的输出特性",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.07185758924270425,
      "no_speech_prob": 1.3367668950936551e-11,
      "compression_ratio": 1.2477611940298508
    },
    {
      "start": 1546.1,
      "end": 1547.96,
      "text": "规避业务幻觉风险",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.07185758924270425,
      "no_speech_prob": 1.3367668950936551e-11,
      "compression_ratio": 1.2477611940298508
    },
    {
      "start": 1547.96,
      "end": 1550.56,
      "text": "轻量化部署适配各类企业硬件条件",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.07185758924270425,
      "no_speech_prob": 1.3367668950936551e-11,
      "compression_ratio": 1.2477611940298508
    },
    {
      "start": 1550.56,
      "end": 1552.1,
      "text": "完全开源的属性",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.07185758924270425,
      "no_speech_prob": 1.3367668950936551e-11,
      "compression_ratio": 1.2477611940298508
    },
    {
      "start": 1552.1,
      "end": 1553.8,
      "text": "满足私有化部署需求",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.07185758924270425,
      "no_speech_prob": 1.3367668950936551e-11,
      "compression_ratio": 1.2477611940298508
    },
    {
      "start": 1553.8,
      "end": 1556.26,
      "text": "再搭配Tinker全自动微调平台",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.07185758924270425,
      "no_speech_prob": 1.3367668950936551e-11,
      "compression_ratio": 1.2477611940298508
    },
    {
      "start": 1556.26,
      "end": 1559.16,
      "text": "彻底抹平了企业AI落地的技术鸿沟",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.07185758924270425,
      "no_speech_prob": 1.3367668950936551e-11,
      "compression_ratio": 1.2477611940298508
    },
    {
      "start": 1559.16,
      "end": 1562.08,
      "text": "哪怕是没有专业算法团队的传统企业",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.07185758924270425,
      "no_speech_prob": 1.3367668950936551e-11,
      "compression_ratio": 1.2477611940298508
    },
    {
      "start": 1562.08,
      "end": 1565.06,
      "text": "也能轻松搭建专属行业AI系统",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.07185758924270425,
      "no_speech_prob": 1.3367668950936551e-11,
      "compression_ratio": 1.2477611940298508
    },
    {
      "start": 1565.06,
      "end": 1567.52,
      "text": "这种精准贴合商业落地的产品逻辑",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.059561874352249446,
      "no_speech_prob": 1.675324982908233e-11,
      "compression_ratio": 1.3231707317073171
    },
    {
      "start": 1567.52,
      "end": 1571.32,
      "text": "也是它能够碾压一众跑分更强模型的核心底气",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.059561874352249446,
      "no_speech_prob": 1.675324982908233e-11,
      "compression_ratio": 1.3231707317073171
    },
    {
      "start": 1571.32,
      "end": 1572.28,
      "text": "除此之外",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.059561874352249446,
      "no_speech_prob": 1.675324982908233e-11,
      "compression_ratio": 1.3231707317073171
    },
    {
      "start": 1572.28,
      "end": 1576.02,
      "text": "我们还能从行业迭代周期看懂这套布局的前瞻性",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.059561874352249446,
      "no_speech_prob": 1.675324982908233e-11,
      "compression_ratio": 1.3231707317073171
    },
    {
      "start": 1576.02,
      "end": 1579.52,
      "text": "如今通用大模型的技术迭代已经进入瓶颈期",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.059561874352249446,
      "no_speech_prob": 1.675324982908233e-11,
      "compression_ratio": 1.3231707317073171
    },
    {
      "start": 1579.52,
      "end": 1582.0,
      "text": "性能提升的边际效益持续递减",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.059561874352249446,
      "no_speech_prob": 1.675324982908233e-11,
      "compression_ratio": 1.3231707317073171
    },
    {
      "start": 1582.0,
      "end": 1585.74,
      "text": "单纯比拼参数和算力已经没有实质商业意义",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.059561874352249446,
      "no_speech_prob": 1.675324982908233e-11,
      "compression_ratio": 1.3231707317073171
    },
    {
      "start": 1585.74,
      "end": 1590.46,
      "text": "未来AI行业的竞争早已从技术参数竞赛转向商业落地竞赛",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.059561874352249446,
      "no_speech_prob": 1.675324982908233e-11,
      "compression_ratio": 1.3231707317073171
    },
    {
      "start": 1590.46,
      "end": 1592.54,
      "text": "谁能更低成本更合规",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.059561874352249446,
      "no_speech_prob": 1.675324982908233e-11,
      "compression_ratio": 1.3231707317073171
    },
    {
      "start": 1592.54,
      "end": 1594.42,
      "text": "更便捷地帮助各行各业",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0643341676244196,
      "no_speech_prob": 1.148448263571522e-11,
      "compression_ratio": 1.3130699088145896
    },
    {
      "start": 1594.42,
      "end": 1595.94,
      "text": "完成AI私有化",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0643341676244196,
      "no_speech_prob": 1.148448263571522e-11,
      "compression_ratio": 1.3130699088145896
    },
    {
      "start": 1595.94,
      "end": 1596.94,
      "text": "定制化改造",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0643341676244196,
      "no_speech_prob": 1.148448263571522e-11,
      "compression_ratio": 1.3130699088145896
    },
    {
      "start": 1596.94,
      "end": 1598.74,
      "text": "谁就能掌握行业话语权",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0643341676244196,
      "no_speech_prob": 1.148448263571522e-11,
      "compression_ratio": 1.3130699088145896
    },
    {
      "start": 1598.74,
      "end": 1601.32,
      "text": "米拉团队放弃无意义的跑分内卷",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0643341676244196,
      "no_speech_prob": 1.148448263571522e-11,
      "compression_ratio": 1.3130699088145896
    },
    {
      "start": 1601.32,
      "end": 1602.88,
      "text": "深耕企业服务赛道",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0643341676244196,
      "no_speech_prob": 1.148448263571522e-11,
      "compression_ratio": 1.3130699088145896
    },
    {
      "start": 1602.88,
      "end": 1606.3,
      "text": "本质是提前预判了AI行业的终局形态",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0643341676244196,
      "no_speech_prob": 1.148448263571522e-11,
      "compression_ratio": 1.3130699088145896
    },
    {
      "start": 1606.3,
      "end": 1607.88,
      "text": "放弃短期的流量热度",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0643341676244196,
      "no_speech_prob": 1.148448263571522e-11,
      "compression_ratio": 1.3130699088145896
    },
    {
      "start": 1607.88,
      "end": 1610.1,
      "text": "锁定了长期的行业垄断地位",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0643341676244196,
      "no_speech_prob": 1.148448263571522e-11,
      "compression_ratio": 1.3130699088145896
    },
    {
      "start": 1610.1,
      "end": 1611.34,
      "text": "看完这期视频",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0643341676244196,
      "no_speech_prob": 1.148448263571522e-11,
      "compression_ratio": 1.3130699088145896
    },
    {
      "start": 1611.34,
      "end": 1614.78,
      "text": "你如何看待大模型一味比拼跑分的内卷现状",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0643341676244196,
      "no_speech_prob": 1.148448263571522e-11,
      "compression_ratio": 1.3130699088145896
    },
    {
      "start": 1614.78,
      "end": 1617.04,
      "text": "又觉得未来企业AI定制服务",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0643341676244196,
      "no_speech_prob": 1.148448263571522e-11,
      "compression_ratio": 1.3130699088145896
    },
    {
      "start": 1617.04,
      "end": 1618.54,
      "text": "会走向怎样的格局",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0643341676244196,
      "no_speech_prob": 1.148448263571522e-11,
      "compression_ratio": 1.3130699088145896
    },
    {
      "start": 1618.54,
      "end": 1620.58,
      "text": "欢迎在评论区理性交流",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0643341676244196,
      "no_speech_prob": 1.148448263571522e-11,
      "compression_ratio": 1.3130699088145896
    },
    {
      "start": 1620.58,
      "end": 1623.68,
      "text": "如果你觉得本期深度行业解读干货充足",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.044717411662257,
      "no_speech_prob": 8.65671336697238e-12,
      "compression_ratio": 1.058139534883721
    },
    {
      "start": 1623.68,
      "end": 1625.68,
      "text": "记得点赞并且订阅频道",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.044717411662257,
      "no_speech_prob": 8.65671336697238e-12,
      "compression_ratio": 1.058139534883721
    },
    {
      "start": 1625.68,
      "end": 1628.6,
      "text": "后续持续跟踪全球AI巨头的战略博弈",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.044717411662257,
      "no_speech_prob": 8.65671336697238e-12,
      "compression_ratio": 1.058139534883721
    },
    {
      "start": 1628.6,
      "end": 1630.7,
      "text": "不要错过更多前沿产业分析",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.044717411662257,
      "no_speech_prob": 8.65671336697238e-12,
      "compression_ratio": 1.058139534883721
    },
    {
      "start": 1630.7,
      "end": 1632.04,
      "text": "我们下期再见",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.044717411662257,
      "no_speech_prob": 8.65671336697238e-12,
      "compression_ratio": 1.058139534883721
    }
  ],
  "segmented_transcription": {
    "status": "SEGMENTED_TRANSCRIPT_OK",
    "created_at": "2026-08-25T17:49:55",
    "manifest": "/Users/sagawa/AI_WORK/video_notes/url/20260825_174846__砸_20_億推出爭議大模型_離開_OpenAI_的頂尖人才_跳出跑分競賽另辟賽道/segmented_work/chunk_manifest.json",
    "audio": "/Users/sagawa/AI_WORK/video_notes/url/20260825_174846__砸_20_億推出爭議大模型_離開_OpenAI_的頂尖人才_跳出跑分競賽另辟賽道/audio_16k_mono.wav",
    "audio_sha256": "da39f1cc946a4a8f25f7119398981530983a8a77c77e35f62beb74d6ba746ad3",
    "audio_duration": 1632.455688,
    "model": "/Users/sagawa/AI_WORK/cache/huggingface/hub/models--mlx-community--whisper-large-v3-turbo/snapshots/a4aaeec0636e6fef84abdcbe3544cb2bf7e9f6fb",
    "language_hint": "zh",
    "condition_on_previous_text": false,
    "chunk_count": 4,
    "resumed_chunks": 0,
    "merged_segment_count": 700,
    "cleaned_segment_count": 700,
    "removed_hallucination_segment_count": 0,
    "final_segment_end": 1632.04,
    "global_repeat_run": 1,
    "global_repeat_text": "如果你持续关注硅谷AI产业动态,一定听过这场堪称传奇的人才出走事件。手握20亿美元顶级风头,集结一批从OpenAI出走的核心研发人才。秘密打磨接近两年,所有人都默认他们会拿出一款能够横扫行业的顶尖大模型。可最终的结果却让整个科技圈大跌眼镜。在前不久的发布会结束之后,他们推出的全新模型在主流权威基准评测榜单当中甚至无法冲进前五。",
    "tail_repeat_run": 1,
    "tail_repeat_text": "哪怕是没有专业算法团队的传统企业",
    "hallucination_region_count": 0,
    "hallucination_regions": [],
    "qc_ok": true,
    "qc_errors": []
  }
}