{
  "text": "你有没有想过,真正定义AI下一阶段走向的,未必是那些轰轰烈烈对外举办的大模型发布会。而是少数行业元老在创业社群内部的深度对话。很多足以改变后续行业走向的核心观点,最先都是诞生在,这类小规模交流现场,没有华丽的宣传包装。全部是几十年工程实践沉淀下来的硬核思考。前不久,谷歌的杰弗迪恩就在YC创业学校完成了一场信息密度极高的圆桌对话。\n整场交流把当下AI行业的现实瓶颈\n硬件底层逻辑\nagent发展边界\n还有普通创业者能够抓住的现实机遇\n全部完整剖析出来\n如果你正在接触AI相关项目\n或是想要看懂未来三五年\n行业真实的眼睛路线\n接下来的内容值得仔细看完\n觉得内容对你有启发\n可以点赞支持\n不错过更多硬核行业解读\n主持人开场抛出的第一个问题就相当尖锐\n直接回溯到一年之前的公开观点\n去年5月份杰弗迪恩曾经在AISNOT的活动上提出一个判断\nAI已经达到初级工程师的水准\n时隔一整年\n行业内外都在观望\n当初的这番论断到底有没有应验\n面对这个问题\n杰弗迪恩没有选择夸大成果\n也没有刻意淡化进展\n他给出了一份非常务实的复盘\n在Agent的体系配合长周期运行时的编程任务当中\n大模型能力的迭代速度远远超出他此前的预估\n放到初级工程师的评判标准之下\n当初的判断大体上已经得到验证\n但同时他也坦诚\n自己严重低估了复杂任务处理能力的上涨幅度\n实际提升速度比事前的预判要快得多\n能力的提升并不单单局限在代码编写\n这个单一赛道Agent这套运行范式\n正在其他各式各样的业务场景当中\n展现出实实在在的价值\n顺着这个话题\n主持人继续向前追问\n把目光投向2027年\n想要听听属于杰弗迪恩的新一轮大胆预判\n他思索之后给出了答案\n核心落在机器学习系统的自我自动化上面\n简单来讲\n就是让AI系统自主开展大批量的实验工作\n把复杂的大问题拆解成若干个更小的子任务\n搭建一套闭环的自动实验循环\n完成子任务的处理之后\n再把分散的结果重新整合汇总\n以此完成自身能力的迭代升级\n这套全自动问题拆解搭配自动实验的思路\n并不单单局限于机器学习本身\n只要某一个科学或者工程领域\n能够建立起一套可以量化衡量的目标\n就有机会依靠这套模式\n拿到巨大的突破\n讲到这里\n主持人抛出一个很有历史氛围感的类比\n2001年的时候\n谷歌搜索还完全依靠硬盘完成数据读取\n杰弗迪恩和长期搭档桑杰\n通过餐金值演算完成一轮成本测算\n得出一个结论\n未来某一个时间节点\n整套搜索索引\n完全可以全部放进机器的内存之中\n基于这个判断\n两个人短短几天就迭代出全新版本的谷歌搜索\n不再依赖硬盘读取数据\n全部运行在内存当中\n这也成为谷歌搜索体验实现跨越时提升的关键转折点\n历史总是会出现相似的韵律\n那么放到2026年\n属于当下这个时代\n整套系统全部搬进内存的标志性时刻又会是什么\n现场6000名创业者\n又应当盯住哪些信号去设计自己的产品\n杰弗迪恩的回答把落脚点放在硬件层面\n它坦言当下的情况和当年谷歌搜索的故事\n并不能完全等同\n但有一个趋势是确定的\n高性能同时兼顾低能耗的推理硬件系统\n会迎来大规模普及\n整个行业已经形成共识\n推理环节才是让agent大规模面\n向普通用户落地的核心基石\n而延迟又是决定用户体验好坏的关键指标\n硬件走向专业化\n是整套系统相比通用GPU TPU设备\n实现更低功耗\n更低延迟最重要的实现路径\n我们所有人都已经习惯等待模型返回输出结果\n但等待本身并不是良好的用户体验\n速度才是一切的王道\n主持人继续顺着这个方向推进\n如果延迟可以直接降低50倍\n整个行业会发生什么样的变化\n紧接着又抛出另外一个值得深思的问题\n时至今日\n行业当中还有哪些被大家全盘接受\n但实际上已经不再成立的固有AI假设\n杰弗迪恩给出的回答\n会刷新不少人的固有认知\n绝大多数从业者\n还没有真正接纳一件事\n依托能力足够强的底层基座模型\nAgent并不只能运行短短一两个小时\n连续运转几天甚至几周\n去完成复杂度极高的任务\n这件事已经具备现实可行性\n现在小部分从业者\n已经看到苗头\n可绝大多数人\n并没有真正把这个可能性\n内化到自己的产品设计逻辑当中\n而这件事未来会释放出巨大的影响力\n很多人会好奇到底什么样的业务场景\n需要Agent连续运航数周的时间\n杰弗迪恩举了一个很具象的案例\n我们可以交给Agent一项完整的软件版本重构工作\n把一整套软件完整迁移到另外一门编程语言\n重构之后的版本可以实现安全性的提升\n或是运行性能的优化\nAI完全可以持续推进整套复杂工程\n把这项工作落地完成\n话题就此切换到TPU诞生的过往\n杰弗迪恩在行业里面广为流传的习惯\n就是餐巾纸数学\n也就是在餐巾纸上快速完成数量级层面的粗略估算\n时间回到2013年前后\n谷歌语音识别的效果迎来实质性的飞跃\n他做过一笔测算\n如果每一位谷歌手机用户\n每天仅仅使用三分钟的语音识别功能\n谷歌整体的服务器集群规模就需要直接翻倍\n对应的成本会膨胀到完全无法接受的地步\n为了解决这个现实难题\n团队决定自研专属芯片\n这也就是TPU的起源\n他回忆\n那时候深度学习驱动的语音系统\n对比旧有方案\n算力开销高出一大截\n但是错误率直接减半\n短短几个月时间就拿到过去\n20年语音识别领域积累下来的全部进步\n初代TPU没有办法运行Chrome浏览器\n也跑不了Word这类通用软件\n但是在机器学习推理场景当中\n拥有极强的针对性\n最终落地的初代芯片\n对比CPU和GPU\n能效可以提升30到80倍\n延迟下降20到30倍\n打下了非常扎实的技术底子\n后续TPU才一步步迭代成如今的模样\n有意思的地方在于团队研发TPU的那个阶段\nTransformer架构还没有问世\n没有人能够预料到后续TPU会成为整个AI产业的基石硬件\n也正是因为这样\n当时团队的设计思路是打造一套通用线性怠数运算系统\n而不是死死绑定某一套特定算法做极致专用化\n机器学习算法永远处在持续演化的进程当中\n硬件不能过度定制\n但是又要保留足够的专用属性\n以此拿到巨大的性能收益\n比如做大规格乘法运算单元\n配置高速内存\n搭建高速互联通路\n后续几代TPU更是优化多芯片协同工作的能力\n让大量芯片可以联合处理同一个复杂任务\n主持人接着提问\n放到今天\n创业者又该做怎样的餐巾纸数学\n能不能依靠简单的数量级估算\n挖掘出下一代如同TPO一般重量级的创新\n杰弗蒂恩表示\n这件事没有统一标准答案\n但是底层思考逻辑是相通的\n首先找准自己所在赛道真正的瓶颈\n再去反问自己\n是否存在一套和当下主流方案完全不一样的解法\n能不能把核心指标抬升一个甚至两个数量级\n不要被市面上已经成熟的实现方案束缚住思路\n回归第一性原理去思考\n如果从零开始解决这个问题\n自己会选择什么样的路径\n往往就能够挖掘出别人看不见的机会\n随后主持人问到一个\n只有内行才会深究的话题\n多年以前杰弗迪恩写过一份流传极广的清单\n名字叫做每个工程师都应该知道的延迟数字\n里面记录缓存未命中的耗石\n磁盘寻道耗石\n杰弗迪恩列举出一系列至关重要的硬件数字\n储存到加速器片上内存之间的带宽\n通往乘法运算单元的带宽\n完成一次乘法运算需要消耗的能量\n芯片与芯片之间的互联带宽\n在该带宽条件之下可以联动多少颗芯片\n还有从500颗芯片扩展到1万颗芯片的规模过程当中\n网络带宽会出现多大程度的衰减\n这些底层数字会实实在在约束上层业务问题的解法\n紧接着他抛出一个足以改变很多人AI系统设计思路的数据\n完成一次计算消耗的能量大概在EP焦2级别\n但是完成一次数据搬运\n一次IO操作\n消耗的能量是前者的1000倍\n就算只是把数据从加速器的高带宽内存搬运到处理器内部\n参与运算\n能量开销同样会急剧拉高\n搬数据的代价远远高于做计算这条物理层面的硬约数\n塑造了机器学习领域大量的工程选择\n如果不存在如此悬殊的能耗差距\n我们就不需要对数据做批量处理\n可现实条件摆在眼前\n工程师只能把大量样本或者token攒成批次统一处理\n以此贪薄数据搬运带来的巨大能耗惩罚\n把1000倍的开销\n稀释为1000除以批次规模\n但如果我们追求极低延迟的推理表现\n批量处理这套逻辑就会变成阻碍\n底层硬件的能量约束\n会自上而下影响整套AI系统的每一层设计\n主持人继续追问\n那我们有没有机会实现单条数据的处理模式\n杰弗迪恩表示自己现阶段思考更多的是推理测的问题\n推理业务天然对低延迟有着强烈诉求\n训练环节对此的要求反而没有那么严苛\n在他看来\n推理硬件的专用化程度\n如今远远还没有抵达上线\n核心优化方向\n一方面是尽可能压缩数据搬运的频次\n另一方面去探索极低精度运算\n不需要硬件去兼容五花八门的精度规格\n把业务真正需要的精度\n直接固化到硬件设计当中\n聊完硬件层面的种种约束\n对话把视角向上拉高\n过去大家谈及AI进步\n惯性思维就是模型规模更大\n训练数据更多\n参数体量上涨\n训练时间拉长\n但是最近几年\n真正拉开差距的\n不再只有模型本体\n模型外围的整套配套体系\n价值正在持续走高\n不单单要看参数规模和训练数据集\n检索工具\n记忆模块\nAgent\n工具链\n全部都是不可或缺的组成部分\n杰弗迪恩认同这样的看法\n模型只是整套解决方案当中的\n其中一块拼图\n我们真正的目标\n是搭建一整套\n可以解决复杂现实问题的完整系统\n这就意味着模型要学会\n调用各式各样的外部工具\n检索外部相关信息\n留存过往处理其他任务\n沉淀下来的历史信息\n并且把这些信息\n带入当下的上下文当中\n这里有一个很关键的区别\n训练数据就像是一锅大杂烩\n数万亿token混杂在一起\n压缩近几千亿的模型参数内部\n与之对比\n模型在处理具体任务时拿到的上下文信息\n逻辑会清晰很多\n同时模型还要识别可用的工具\n判断哪些工具适配当前步骤\n如何把大任务拆解成一连串工具调用\n甚至尝试多条不同的解题路径\n再评估各个方案的实际效果\n这就是复杂agent以及dooragent编排的核心能力\n上下文工程这个方向很有意思的一点\n使它的准入门槛相对更低\n训练大模型需要海量算力\n海量数据资源\n普通团队很难触碰\n但是做上下文工程\n只需要调用Gemini这类模型的API接口\n再搭配自己搭建的检索逻辑\n工具调用流程\n工作流编排\n就可以动手实践\n想要把上下文工程做到足够强\n杰弗迪恩给出一套非常务实的实操思路\n就是实实在在拿模型和工具链\n去处理真实业务\n在实操当中观察模型的出错点位\n很多时候想要提升模型表现\n并不需要去改动底层模型权重\n而是提供质量更高的指引\n编写更完善的技能描述\n告诉模型面对某一类问题\n应当如何调用工具\n反复迭代之后\n你就会慢慢摸清楚\n模型还缺少哪些信息\n由此进入自我迭代优化的正向循环\n他分享了一段亲身经历\n几周之前\n他和桑杰一起做底层代码库的性能调优\n团队内部有自研的微型基准测试库\n可以统计各类操作的耗时\n这套数据结构被谷歌内部几百万个进程调用\n性能表现必须得到保障\n在没有Agent的时代\n完整流程是先测试现有性能\n修改代码\n重新跑基准测试\n观察性能变化\n还要执行更大范围的测试\n检查缓存占用等一系列指标\n后来他们专门编写对应的技能\n教会Agent完整复刻整套工作流\n执行基准测试\n修改代码\n重新叫验性能指标\n《循环迭代优化》\n这套方案在部分问题上拿到相当不错的效果\n本质上就是把人类工程师的工作流程\n转化成模型能够读懂执行的指令\n几个月之前\n他和桑杰共同发布过一份30页左右的公开文档\n名字叫做Performance Heightened TS\n记录大量性能优化的实操技巧\n已经有不少开发者把这份文档做总结提炼之后\n投入给不同的大模型\n最终测试结果显示\n模型分析代码性能问题的能力得到明显提升\n而且这份资料对所有人完全公开\n谈及agent\n有一个所有人都会碰到的现实痛点\n不少agent在前十几步执行都十分顺畅\n等到推进到第三十步\n四十步\n甚至五十步之后\n就会开始逻辑跑偏\n杰弗迪恩坦言\n这就是当下技术最大的约束条件\n一部分原因是模型处理训练机之外的任务\n一旦脱离熟悉的任务分布\n效果就会下滑\n偏离熟悉的边界越远\n失效概率就越高\n针对该问题\n有两条可行的解决路径\n第一种\n编写对应的技能和提示之意\n尽量让模型维持在自己熟悉\n成功率更高的执行路径\n第二种\n采用多agent的架构\n让多个独立agent\n并行尝试不同解题思路\n再交由另外一个模型或者agent\n去评判各个路径的可行性\n本质就是对全部可行解法\n做搜索筛选\n保留有希望的分支\n丢弃无效跑偏的分支\n依靠推理计算\n去便利解决方案空间\n以此提升常链路Agent的定性\n在谷歌内部就搭建了一整套调度框架\n以及海量技能库\nAgent可以调用内部编码工具\n执行代码审查\n完成性能测试\n读取日志文件\n只要技能定义足够合理\nAgent的实用价值就会被大幅释放\n紧接着\n对话落到在场每一位创业者都无比关心的命题\n谷歌能够做到从处理器到上层产品\n全链路协同设计\n那对于两三个人组成的小型创业团队\n还剩下哪些可以抓住的机会\n杰弗迪恩讲到\n谷歌以及通用大模型团队\n都在竭尽全力打造通用性极强\n几乎可以处理各类任务的基座模型\n但恰恰是追求通用这个前提\n导致它很难针对某一个垂直细分领域\n做深度打磨\n如果创业团队\n可以针对特定行业打磨优质的产品交互界面\n配套专属的技能体系\n甚至是垂直微调的专用模型\n完全有机会建立起明显的竞争优势\n打造出体验流畅\n准确率出色的垂直产品\n不过他也做出提醒\n通用模型的能力还在持续变强\n覆盖的场景不断扩张\n创业者必须做好判断\n自己选择的赛道\n是未来半年到12个月\n就会被通用模型覆盖\n还是两三年之内\n通用模型都难以企及\n以此判断赛道的抗风险能力\n那创业者应当如何筛选\n值得投入的方向\n他给出的第一条建议朴素\n但是分类\n往往代表存在机会\n如果已经可以完成一部分\n只是效果不够完美\n反而要提高警惕\n这代表能力的种子已经存在\n只需要补充训练数据\n扩大模型规模\n能力就会快速迭代\n所以要优先寻找当下通用模型\n成功率只有0%或者1%的问题\n而不是20%左右的赛道\n这类机会\n一部分属于训练分布之外的全新任务\n另一类场景\n是产品可以拿到通用模型\n接触不到的私有数据\n比如处理用户个人信息的产品\n通用模型本身无法获取用户私人资料\n这就创造天然壁垒\n还有一类机会\n针对高难度问题\n以可控成本训练垂直专用模型\n换取极高精度\nAlphaFold就是典型案例\n面向蛋白质折叠的专用模型取得巨大成功\n却并非通用大模型\n材料科学\n芯片设计领域也存在同类的落地可能性\n怎样成长为AI原生的创始人\n他提到\n和虚拟agent的协作的时候\n把需求描述的越清晰\n最终产出效果就越好\n需求定义足够明确\nAgent就拥有清晰的边界\n明白自身的目标\n从很早的计算机发展阶段\n工程师就被要求开发软件之前\n梳理清楚产品规格\n如今只是把执行工作交给Agent\n明确规格这件事反而变得更加关键\n人类工程师可以读懂语境\n主动反问确认细节\nAgent虽然部分场景可以追问\n但整体还是高度依赖清晰完备的需求描述\n现在代码迁移类的Coding Agent\n效果出众\n背后的原因也正在于此\n把一套软件从一门语言迁移到另一门语言\n原始代码本身就是完整的规格\n原有测试用力可以直接复用\n同时完成迁移\n持续比对新旧版本行为差异\n直到二者表现对齐\n正是规格足够明确\n才造就这类任务亮眼的表现\n未来大规模调度数百个Agent\n成为常态\n大量代码由Agent生成\n什么能力会变得最为稀缺\n杰弗·迪恩给出的答案是品位\n也就是判断哪些问题值得交给agent去解决\n放到研究领域大部分的成败\n从选择研究课题的那一刻就已经注定\n手握全套工具\n不如选对值得投入的问题\n把普通问题打磨得再精致\n也比不上攻克一个高价值的难题\n但是品位本身没有标准化的考核指标\n应当如何训练\n一部分来源于大量项目经验\n经手足够多不同类型的问题\n慢慢积累对机会的判断力\n还有一个实操办法\n写下未来12个月之内\n自己预判会迎来爆发的方向\n后续只挑选其中一部分落地\n12个月之后回头复盘\n看看哪些预判成真\n哪些被其他人实现\n哪些完全没有发生\n相当于给自己制造大量样本\n反向训练自己的判断力\n第三种方式就是做疯狂的思想实验\n不要把行业默认的前提全部当成真理\n他分享了一段和同事开展的思想实验\n过去60年\n芯片设计制造行业\n都在拼命缩小晶体管尺寸\n压低出错概率\n默认目标就是同一款芯片\n每一颗成品都应当完全一致\n不能出现比特翻转\n但是宏观层面\n大规模分布式系统的逻辑恰恰相反\n依靠不可靠的硬件组件\n搭建高可靠系统\n磁盘会损坏\n但是数据不能丢失\n于是做多副本\n使用纠缠码\n可是这套容错思路\n并没有下放到晶体管这一层\n如果设计一套芯片\n晶体管每天会出现20次错误\n而不是几百万年才出现一次故障\n整套系统会走向何方\n制造工艺或许会迎来全新可能性\n信号传输采用多条冗余通路\n保障至少一路信号可以送达\n这套逻辑和人脑神经信号传递十分相似\n大脑信号传输本身就不是绝对可靠\n重要信息就依靠多条通路同时传递\n它并不是说当下就要落地这套方案\n但是这类思想实验意义重大\n要定期重新审视所有人习以为常的预设\n回顾他的职业生涯\nTPU的诞生就是推翻固有假设之后接触的果实\n在问题还没有成为主流的时候\n思考小众场景的专用硬件\nMapReduce也是同理\n团队厌倦了手工编写大量并行\n容错相关的重复代码\n于是提出全新抽象框架\n上层只需要编写业务逻辑\n底层统一处理并行调度和容错\n最后成为谷歌大规模计算的核心底座\n而这一切的源头\n同样来自思想实验\n现在有Alpha Chip辅助芯片布局\n有Alpha UVO完成方案生成评估留存\n整套AI制造AI的体系\n正在逐步成型\n广义来讲\n这本质就是科学方法的自动化\n提出假设\n搭建环境\n执行实验\n评估结果\n循环往复推进迭代\n如今越来越多领域\n可以把这套闭环自动化\n一次性执行海量实验\n压缩循环延迟\n就变得格外关键\n前提是评估器本身要足够快速\n他举了一个十年之前的案例\n同事研究量子化学\n解析分子属性\n传统密度泛含理论模拟器\n跑一次就要耗费一整夜\n团队收集大量输入输出样本\n训练神经网络近似模拟器\n速度提升30万倍\n精度却几乎没有损失\n彻底改写科研模式\n从前筛选1000万候选分子\n要耗费6个月筹备算力\n现在一顿午饭的时间\n就可以完成筛选\n大量行业还有机会打造高速仿真模型\n一旦实现\n整个实验迭代周期就会被彻底重构\n它最看好的方向\n就是模型依靠大量实验\n实现地规式自我改进\n持续提升每一份算力\n可以产出的新发现\n对话推进到后半段\n谈及职业路上的挫折\n2014年杰弗蒂恩\n杰弗里辛顿\n奥里奥尔维尼亚尔斯\n联合发布知识蒸馏论文\n用大模型充当老师\n小模型充当学生\n得到体积更小\n成本更低\n同时效果优异的模型\n如今整个行业都在广泛使用\n但是这篇论文当年被NIPS聚稿\n评审的观点认为\n这项工作很难带来重大突破\n杰弗迪恩表示\n并不责怪评审委员会\n对方站在学术视角\n评判是否属于理论层面的重大突破\n而团队清楚这件事巨大的工程价值\n目标是降低模型服务成本\n扩大落地场景\n他们把论文上传到ARXV预印本平台\n行业从业者阅读之后\n大量落地使用\nGemini当中的Flash系列模型\n就有一部分来自大尺寸Pro模型的蒸馏\n在同规格模型当中\n基准测试表现十分靠前\n这件事带来的启示就是\n就算成果遭到拒绝\n也不要停下脚步\n如果把1999年\n刚刚加入只有20人规模谷歌的年轻杰弗迪恩\n直接传送到2026年\n他会选择大厂前沿实验室\n还是选择自主创业\n他表示没有标准答案\n最核心的判断表现\n创业公司风险更高\n必须拥有极强的热情\n但也可以收获极高的回报\n无论选择哪一条道路\n都要反问自己\n如果项目走到最好的结局\n世界会不会切实变好\n如果仅仅只是很酷\n就不值得投入大量时间\n谈及如何寻找优秀的合作者\n不仅要寻找专业能力过硬\n技能互补的伙伴\n也要选择相处起来足够愉快的人\n创业路上要共同面对大量棘手难题\n彼此的相处体验格外重要\n小团队的魅力就是每个人掌握其他人不具备的能力\n组合在一起可以完成单人无法实现的成果\n要把自己的工程或者科研生涯当成一条工具腰带\n持续补充新的工具\n你永远预料不到未来会碰到什么样的难题\n掌握的工具越多\n可以解决的问题边界就越宽广\n主持人抛出最后一个问题\n他希望新一代创作者把精力投入到哪些方向\n杰弗迪恩讲到\n值得探索的问题数不胜数\n他个人十分看好全新硬件路线\n高效推理硬件拥有巨大价值\n机器学习算法本身也还有巨大革新空间\n比如数据效率远远高于当下的系统\n现在大模型18岁阶段接触的数据总量\n比人类18岁接触的信息多出1000倍\n但是人类在大量任务当中\n依旧可以持平甚至优于大模型\n我们能不能打造数据效率更高\n可以从自身行动当中持续学习的系统\n持续学习本身就是极具价值的课题\n多agent交互同样充满想象空间\n把视角在向外延伸\n如何促成人类之间更良性的对话\n搭建工具帮助人们理性交流\n帮助志趣相投的人建立连接\n同样是值得深耕的方向\n整场对话听完\n不难感受到杰弗迪恩看待行业的独特视角\n他不会空谈AGI什么时候到来\n而是立足于能耗\n数据搬运\n延迟这些底层物理约束\n向上推导系统设计逻辑\n推导创业机会\n全行业疯狂内卷模型参数的时候\n他提醒大家数据搬运的成本是计算的1000倍\n推理硬件专用化是下一轮效率跃升的关键\n所有人焦虑通用模型吞噬创业机会\n他告诉大家去寻找成功率0%到1%的赛道\n挖掘私有数据\n垂直专用模型的机会\n所有人埋头打磨提示词的时候\n他点名未来最稀缺的资源是品位\n是筛选问题的判断力\n这些认知不是凭空想象\n来自几十年系统工程沉淀\n来自一次次参金值演算\n来自一次次推翻固有认知的思想实验\n以上就是整场访谈的核心精华\n如果你觉得这些来自一线工程大佬的思考\n对你有启发\n欢迎点赞\n也可以把这期内容分享给身边做AI开发创业的朋友\n一起跳出网络上的流量炒作\n看见行业更真实的底层逻辑\n我们下期再见",
  "language": "zh",
  "duration": 1564.2355,
  "segments": [
    {
      "start": 0.0,
      "end": 28.3,
      "text": "你有没有想过,真正定义AI下一阶段走向的,未必是那些轰轰烈烈对外举办的大模型发布会。而是少数行业元老在创业社群内部的深度对话。很多足以改变后续行业走向的核心观点,最先都是诞生在,这类小规模交流现场,没有华丽的宣传包装。全部是几十年工程实践沉淀下来的硬核思考。前不久,谷歌的杰弗迪恩就在YC创业学校完成了一场信息密度极高的圆桌对话。",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.05304001316879735,
      "no_speech_prob": 7.231804998591995e-12,
      "compression_ratio": 1.280323450134771,
      "source": "whisper"
    },
    {
      "start": 28.3,
      "end": 31.42,
      "text": "整场交流把当下AI行业的现实瓶颈",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06246225357055664,
      "no_speech_prob": 1.1748414741052926e-11,
      "compression_ratio": 1.2737430167597765,
      "source": "whisper"
    },
    {
      "start": 31.42,
      "end": 32.84,
      "text": "硬件底层逻辑",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06246225357055664,
      "no_speech_prob": 1.1748414741052926e-11,
      "compression_ratio": 1.2737430167597765,
      "source": "whisper"
    },
    {
      "start": 32.84,
      "end": 34.28,
      "text": "agent发展边界",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06246225357055664,
      "no_speech_prob": 1.1748414741052926e-11,
      "compression_ratio": 1.2737430167597765,
      "source": "whisper"
    },
    {
      "start": 34.28,
      "end": 37.02,
      "text": "还有普通创业者能够抓住的现实机遇",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06246225357055664,
      "no_speech_prob": 1.1748414741052926e-11,
      "compression_ratio": 1.2737430167597765,
      "source": "whisper"
    },
    {
      "start": 37.02,
      "end": 38.84,
      "text": "全部完整剖析出来",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06246225357055664,
      "no_speech_prob": 1.1748414741052926e-11,
      "compression_ratio": 1.2737430167597765,
      "source": "whisper"
    },
    {
      "start": 38.84,
      "end": 41.2,
      "text": "如果你正在接触AI相关项目",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06246225357055664,
      "no_speech_prob": 1.1748414741052926e-11,
      "compression_ratio": 1.2737430167597765,
      "source": "whisper"
    },
    {
      "start": 41.2,
      "end": 43.34,
      "text": "或是想要看懂未来三五年",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06246225357055664,
      "no_speech_prob": 1.1748414741052926e-11,
      "compression_ratio": 1.2737430167597765,
      "source": "whisper"
    },
    {
      "start": 43.34,
      "end": 45.02,
      "text": "行业真实的眼睛路线",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06246225357055664,
      "no_speech_prob": 1.1748414741052926e-11,
      "compression_ratio": 1.2737430167597765,
      "source": "whisper"
    },
    {
      "start": 45.02,
      "end": 47.28,
      "text": "接下来的内容值得仔细看完",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06246225357055664,
      "no_speech_prob": 1.1748414741052926e-11,
      "compression_ratio": 1.2737430167597765,
      "source": "whisper"
    },
    {
      "start": 47.28,
      "end": 49.18,
      "text": "觉得内容对你有启发",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06246225357055664,
      "no_speech_prob": 1.1748414741052926e-11,
      "compression_ratio": 1.2737430167597765,
      "source": "whisper"
    },
    {
      "start": 49.18,
      "end": 50.32,
      "text": "可以点赞支持",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06246225357055664,
      "no_speech_prob": 1.1748414741052926e-11,
      "compression_ratio": 1.2737430167597765,
      "source": "whisper"
    },
    {
      "start": 50.32,
      "end": 52.24,
      "text": "不错过更多硬核行业解读",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06246225357055664,
      "no_speech_prob": 1.1748414741052926e-11,
      "compression_ratio": 1.2737430167597765,
      "source": "whisper"
    },
    {
      "start": 52.24,
      "end": 55.34,
      "text": "主持人开场抛出的第一个问题就相当尖锐",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06246225357055664,
      "no_speech_prob": 1.1748414741052926e-11,
      "compression_ratio": 1.2737430167597765,
      "source": "whisper"
    },
    {
      "start": 55.34,
      "end": 58.1,
      "text": "直接回溯到一年之前的公开观点",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06246225357055664,
      "no_speech_prob": 1.1748414741052926e-11,
      "compression_ratio": 1.2737430167597765,
      "source": "whisper"
    },
    {
      "start": 58.1,
      "end": 62.46,
      "text": "去年5月份杰弗迪恩曾经在AISNOT的活动上提出一个判断",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07510175437570732,
      "no_speech_prob": 1.4845508503058547e-11,
      "compression_ratio": 1.2994505494505495,
      "source": "whisper"
    },
    {
      "start": 62.46,
      "end": 64.94,
      "text": "AI已经达到初级工程师的水准",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07510175437570732,
      "no_speech_prob": 1.4845508503058547e-11,
      "compression_ratio": 1.2994505494505495,
      "source": "whisper"
    },
    {
      "start": 64.94,
      "end": 66.1,
      "text": "时隔一整年",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07510175437570732,
      "no_speech_prob": 1.4845508503058547e-11,
      "compression_ratio": 1.2994505494505495,
      "source": "whisper"
    },
    {
      "start": 66.1,
      "end": 67.58,
      "text": "行业内外都在观望",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07510175437570732,
      "no_speech_prob": 1.4845508503058547e-11,
      "compression_ratio": 1.2994505494505495,
      "source": "whisper"
    },
    {
      "start": 67.58,
      "end": 69.98,
      "text": "当初的这番论断到底有没有应验",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07510175437570732,
      "no_speech_prob": 1.4845508503058547e-11,
      "compression_ratio": 1.2994505494505495,
      "source": "whisper"
    },
    {
      "start": 69.98,
      "end": 71.16,
      "text": "面对这个问题",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07510175437570732,
      "no_speech_prob": 1.4845508503058547e-11,
      "compression_ratio": 1.2994505494505495,
      "source": "whisper"
    },
    {
      "start": 71.16,
      "end": 73.4,
      "text": "杰弗迪恩没有选择夸大成果",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07510175437570732,
      "no_speech_prob": 1.4845508503058547e-11,
      "compression_ratio": 1.2994505494505495,
      "source": "whisper"
    },
    {
      "start": 73.4,
      "end": 75.1,
      "text": "也没有刻意淡化进展",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07510175437570732,
      "no_speech_prob": 1.4845508503058547e-11,
      "compression_ratio": 1.2994505494505495,
      "source": "whisper"
    },
    {
      "start": 75.1,
      "end": 77.3,
      "text": "他给出了一份非常务实的复盘",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07510175437570732,
      "no_speech_prob": 1.4845508503058547e-11,
      "compression_ratio": 1.2994505494505495,
      "source": "whisper"
    },
    {
      "start": 77.3,
      "end": 81.4,
      "text": "在Agent的体系配合长周期运行时的编程任务当中",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07510175437570732,
      "no_speech_prob": 1.4845508503058547e-11,
      "compression_ratio": 1.2994505494505495,
      "source": "whisper"
    },
    {
      "start": 81.4,
      "end": 85.24,
      "text": "大模型能力的迭代速度远远超出他此前的预估",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07510175437570732,
      "no_speech_prob": 1.4845508503058547e-11,
      "compression_ratio": 1.2994505494505495,
      "source": "whisper"
    },
    {
      "start": 85.24,
      "end": 87.74,
      "text": "放到初级工程师的评判标准之下",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07510175437570732,
      "no_speech_prob": 1.4845508503058547e-11,
      "compression_ratio": 1.2994505494505495,
      "source": "whisper"
    },
    {
      "start": 87.74,
      "end": 90.36,
      "text": "当初的判断大体上已经得到验证",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06715658223517587,
      "no_speech_prob": 1.444581954057611e-11,
      "compression_ratio": 1.2773109243697478,
      "source": "whisper"
    },
    {
      "start": 90.36,
      "end": 91.78,
      "text": "但同时他也坦诚",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06715658223517587,
      "no_speech_prob": 1.444581954057611e-11,
      "compression_ratio": 1.2773109243697478,
      "source": "whisper"
    },
    {
      "start": 91.78,
      "end": 95.36,
      "text": "自己严重低估了复杂任务处理能力的上涨幅度",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06715658223517587,
      "no_speech_prob": 1.444581954057611e-11,
      "compression_ratio": 1.2773109243697478,
      "source": "whisper"
    },
    {
      "start": 95.36,
      "end": 98.52,
      "text": "实际提升速度比事前的预判要快得多",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06715658223517587,
      "no_speech_prob": 1.444581954057611e-11,
      "compression_ratio": 1.2773109243697478,
      "source": "whisper"
    },
    {
      "start": 98.52,
      "end": 101.58,
      "text": "能力的提升并不单单局限在代码编写",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06715658223517587,
      "no_speech_prob": 1.444581954057611e-11,
      "compression_ratio": 1.2773109243697478,
      "source": "whisper"
    },
    {
      "start": 101.58,
      "end": 104.2,
      "text": "这个单一赛道Agent这套运行范式",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06715658223517587,
      "no_speech_prob": 1.444581954057611e-11,
      "compression_ratio": 1.2773109243697478,
      "source": "whisper"
    },
    {
      "start": 104.2,
      "end": 106.98,
      "text": "正在其他各式各样的业务场景当中",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06715658223517587,
      "no_speech_prob": 1.444581954057611e-11,
      "compression_ratio": 1.2773109243697478,
      "source": "whisper"
    },
    {
      "start": 106.98,
      "end": 109.06,
      "text": "展现出实实在在的价值",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06715658223517587,
      "no_speech_prob": 1.444581954057611e-11,
      "compression_ratio": 1.2773109243697478,
      "source": "whisper"
    },
    {
      "start": 109.06,
      "end": 110.3,
      "text": "顺着这个话题",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06715658223517587,
      "no_speech_prob": 1.444581954057611e-11,
      "compression_ratio": 1.2773109243697478,
      "source": "whisper"
    },
    {
      "start": 110.3,
      "end": 111.86,
      "text": "主持人继续向前追问",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06715658223517587,
      "no_speech_prob": 1.444581954057611e-11,
      "compression_ratio": 1.2773109243697478,
      "source": "whisper"
    },
    {
      "start": 111.86,
      "end": 113.78,
      "text": "把目光投向2027年",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06715658223517587,
      "no_speech_prob": 1.444581954057611e-11,
      "compression_ratio": 1.2773109243697478,
      "source": "whisper"
    },
    {
      "start": 113.78,
      "end": 116.78,
      "text": "想要听听属于杰弗迪恩的新一轮大胆预判",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06715658223517587,
      "no_speech_prob": 1.444581954057611e-11,
      "compression_ratio": 1.2773109243697478,
      "source": "whisper"
    },
    {
      "start": 116.78,
      "end": 118.58,
      "text": "他思索之后给出了答案",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06604987443095506,
      "no_speech_prob": 1.1734540422692064e-11,
      "compression_ratio": 1.3850931677018634,
      "source": "whisper"
    },
    {
      "start": 118.58,
      "end": 121.74,
      "text": "核心落在机器学习系统的自我自动化上面",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06604987443095506,
      "no_speech_prob": 1.1734540422692064e-11,
      "compression_ratio": 1.3850931677018634,
      "source": "whisper"
    },
    {
      "start": 121.74,
      "end": 122.68,
      "text": "简单来讲",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06604987443095506,
      "no_speech_prob": 1.1734540422692064e-11,
      "compression_ratio": 1.3850931677018634,
      "source": "whisper"
    },
    {
      "start": 122.68,
      "end": 126.12,
      "text": "就是让AI系统自主开展大批量的实验工作",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06604987443095506,
      "no_speech_prob": 1.1734540422692064e-11,
      "compression_ratio": 1.3850931677018634,
      "source": "whisper"
    },
    {
      "start": 126.12,
      "end": 129.92,
      "text": "把复杂的大问题拆解成若干个更小的子任务",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06604987443095506,
      "no_speech_prob": 1.1734540422692064e-11,
      "compression_ratio": 1.3850931677018634,
      "source": "whisper"
    },
    {
      "start": 129.92,
      "end": 132.14,
      "text": "搭建一套闭环的自动实验循环",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06604987443095506,
      "no_speech_prob": 1.1734540422692064e-11,
      "compression_ratio": 1.3850931677018634,
      "source": "whisper"
    },
    {
      "start": 132.14,
      "end": 134.08,
      "text": "完成子任务的处理之后",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06604987443095506,
      "no_speech_prob": 1.1734540422692064e-11,
      "compression_ratio": 1.3850931677018634,
      "source": "whisper"
    },
    {
      "start": 134.08,
      "end": 136.52,
      "text": "再把分散的结果重新整合汇总",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06604987443095506,
      "no_speech_prob": 1.1734540422692064e-11,
      "compression_ratio": 1.3850931677018634,
      "source": "whisper"
    },
    {
      "start": 136.52,
      "end": 138.96,
      "text": "以此完成自身能力的迭代升级",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06604987443095506,
      "no_speech_prob": 1.1734540422692064e-11,
      "compression_ratio": 1.3850931677018634,
      "source": "whisper"
    },
    {
      "start": 138.96,
      "end": 142.2,
      "text": "这套全自动问题拆解搭配自动实验的思路",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06604987443095506,
      "no_speech_prob": 1.1734540422692064e-11,
      "compression_ratio": 1.3850931677018634,
      "source": "whisper"
    },
    {
      "start": 142.2,
      "end": 144.68,
      "text": "并不单单局限于机器学习本身",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06604987443095506,
      "no_speech_prob": 1.1734540422692064e-11,
      "compression_ratio": 1.3850931677018634,
      "source": "whisper"
    },
    {
      "start": 144.68,
      "end": 147.22,
      "text": "只要某一个科学或者工程领域",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06950577061350753,
      "no_speech_prob": 1.1893250273509182e-11,
      "compression_ratio": 1.3236151603498543,
      "source": "whisper"
    },
    {
      "start": 147.22,
      "end": 150.12,
      "text": "能够建立起一套可以量化衡量的目标",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06950577061350753,
      "no_speech_prob": 1.1893250273509182e-11,
      "compression_ratio": 1.3236151603498543,
      "source": "whisper"
    },
    {
      "start": 150.12,
      "end": 152.04,
      "text": "就有机会依靠这套模式",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06950577061350753,
      "no_speech_prob": 1.1893250273509182e-11,
      "compression_ratio": 1.3236151603498543,
      "source": "whisper"
    },
    {
      "start": 152.04,
      "end": 153.38,
      "text": "拿到巨大的突破",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06950577061350753,
      "no_speech_prob": 1.1893250273509182e-11,
      "compression_ratio": 1.3236151603498543,
      "source": "whisper"
    },
    {
      "start": 153.38,
      "end": 154.28,
      "text": "讲到这里",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06950577061350753,
      "no_speech_prob": 1.1893250273509182e-11,
      "compression_ratio": 1.3236151603498543,
      "source": "whisper"
    },
    {
      "start": 154.28,
      "end": 157.1,
      "text": "主持人抛出一个很有历史氛围感的类比",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06950577061350753,
      "no_speech_prob": 1.1893250273509182e-11,
      "compression_ratio": 1.3236151603498543,
      "source": "whisper"
    },
    {
      "start": 157.1,
      "end": 158.64,
      "text": "2001年的时候",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06950577061350753,
      "no_speech_prob": 1.1893250273509182e-11,
      "compression_ratio": 1.3236151603498543,
      "source": "whisper"
    },
    {
      "start": 158.64,
      "end": 162.04,
      "text": "谷歌搜索还完全依靠硬盘完成数据读取",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06950577061350753,
      "no_speech_prob": 1.1893250273509182e-11,
      "compression_ratio": 1.3236151603498543,
      "source": "whisper"
    },
    {
      "start": 162.04,
      "end": 164.1,
      "text": "杰弗迪恩和长期搭档桑杰",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06950577061350753,
      "no_speech_prob": 1.1893250273509182e-11,
      "compression_ratio": 1.3236151603498543,
      "source": "whisper"
    },
    {
      "start": 164.1,
      "end": 166.98,
      "text": "通过餐金值演算完成一轮成本测算",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06950577061350753,
      "no_speech_prob": 1.1893250273509182e-11,
      "compression_ratio": 1.3236151603498543,
      "source": "whisper"
    },
    {
      "start": 166.98,
      "end": 168.08,
      "text": "得出一个结论",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06950577061350753,
      "no_speech_prob": 1.1893250273509182e-11,
      "compression_ratio": 1.3236151603498543,
      "source": "whisper"
    },
    {
      "start": 168.08,
      "end": 169.8,
      "text": "未来某一个时间节点",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06950577061350753,
      "no_speech_prob": 1.1893250273509182e-11,
      "compression_ratio": 1.3236151603498543,
      "source": "whisper"
    },
    {
      "start": 169.8,
      "end": 171.16,
      "text": "整套搜索索引",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06950577061350753,
      "no_speech_prob": 1.1893250273509182e-11,
      "compression_ratio": 1.3236151603498543,
      "source": "whisper"
    },
    {
      "start": 171.16,
      "end": 173.78,
      "text": "完全可以全部放进机器的内存之中",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06950577061350753,
      "no_speech_prob": 1.1893250273509182e-11,
      "compression_ratio": 1.3236151603498543,
      "source": "whisper"
    },
    {
      "start": 173.78,
      "end": 174.82,
      "text": "基于这个判断",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0752577223219313,
      "no_speech_prob": 1.2318657378884712e-11,
      "compression_ratio": 1.298342541436464,
      "source": "whisper"
    },
    {
      "start": 174.82,
      "end": 178.72,
      "text": "两个人短短几天就迭代出全新版本的谷歌搜索",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0752577223219313,
      "no_speech_prob": 1.2318657378884712e-11,
      "compression_ratio": 1.298342541436464,
      "source": "whisper"
    },
    {
      "start": 178.72,
      "end": 180.78,
      "text": "不再依赖硬盘读取数据",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0752577223219313,
      "no_speech_prob": 1.2318657378884712e-11,
      "compression_ratio": 1.298342541436464,
      "source": "whisper"
    },
    {
      "start": 180.78,
      "end": 182.56,
      "text": "全部运行在内存当中",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0752577223219313,
      "no_speech_prob": 1.2318657378884712e-11,
      "compression_ratio": 1.298342541436464,
      "source": "whisper"
    },
    {
      "start": 182.56,
      "end": 186.78,
      "text": "这也成为谷歌搜索体验实现跨越时提升的关键转折点",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0752577223219313,
      "no_speech_prob": 1.2318657378884712e-11,
      "compression_ratio": 1.298342541436464,
      "source": "whisper"
    },
    {
      "start": 186.78,
      "end": 189.06,
      "text": "历史总是会出现相似的韵律",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0752577223219313,
      "no_speech_prob": 1.2318657378884712e-11,
      "compression_ratio": 1.298342541436464,
      "source": "whisper"
    },
    {
      "start": 189.06,
      "end": 190.76,
      "text": "那么放到2026年",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0752577223219313,
      "no_speech_prob": 1.2318657378884712e-11,
      "compression_ratio": 1.298342541436464,
      "source": "whisper"
    },
    {
      "start": 190.76,
      "end": 192.26,
      "text": "属于当下这个时代",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0752577223219313,
      "no_speech_prob": 1.2318657378884712e-11,
      "compression_ratio": 1.298342541436464,
      "source": "whisper"
    },
    {
      "start": 192.26,
      "end": 195.5,
      "text": "整套系统全部搬进内存的标志性时刻又会是什么",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0752577223219313,
      "no_speech_prob": 1.2318657378884712e-11,
      "compression_ratio": 1.298342541436464,
      "source": "whisper"
    },
    {
      "start": 195.5,
      "end": 197.34,
      "text": "现场6000名创业者",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0752577223219313,
      "no_speech_prob": 1.2318657378884712e-11,
      "compression_ratio": 1.298342541436464,
      "source": "whisper"
    },
    {
      "start": 197.34,
      "end": 200.38,
      "text": "又应当盯住哪些信号去设计自己的产品",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0752577223219313,
      "no_speech_prob": 1.2318657378884712e-11,
      "compression_ratio": 1.298342541436464,
      "source": "whisper"
    },
    {
      "start": 200.38,
      "end": 203.46,
      "text": "杰弗迪恩的回答把落脚点放在硬件层面",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0752577223219313,
      "no_speech_prob": 1.2318657378884712e-11,
      "compression_ratio": 1.298342541436464,
      "source": "whisper"
    },
    {
      "start": 203.46,
      "end": 206.98,
      "text": "它坦言当下的情况和当年谷歌搜索的故事",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06131814706205117,
      "no_speech_prob": 1.3244232099918207e-11,
      "compression_ratio": 1.2447761194029852,
      "source": "whisper"
    },
    {
      "start": 206.98,
      "end": 208.32,
      "text": "并不能完全等同",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06131814706205117,
      "no_speech_prob": 1.3244232099918207e-11,
      "compression_ratio": 1.2447761194029852,
      "source": "whisper"
    },
    {
      "start": 208.32,
      "end": 210.24,
      "text": "但有一个趋势是确定的",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06131814706205117,
      "no_speech_prob": 1.3244232099918207e-11,
      "compression_ratio": 1.2447761194029852,
      "source": "whisper"
    },
    {
      "start": 210.24,
      "end": 213.22,
      "text": "高性能同时兼顾低能耗的推理硬件系统",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06131814706205117,
      "no_speech_prob": 1.3244232099918207e-11,
      "compression_ratio": 1.2447761194029852,
      "source": "whisper"
    },
    {
      "start": 213.22,
      "end": 214.86,
      "text": "会迎来大规模普及",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06131814706205117,
      "no_speech_prob": 1.3244232099918207e-11,
      "compression_ratio": 1.2447761194029852,
      "source": "whisper"
    },
    {
      "start": 214.86,
      "end": 216.82,
      "text": "整个行业已经形成共识",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06131814706205117,
      "no_speech_prob": 1.3244232099918207e-11,
      "compression_ratio": 1.2447761194029852,
      "source": "whisper"
    },
    {
      "start": 216.82,
      "end": 219.34,
      "text": "推理环节才是让agent大规模面",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06131814706205117,
      "no_speech_prob": 1.3244232099918207e-11,
      "compression_ratio": 1.2447761194029852,
      "source": "whisper"
    },
    {
      "start": 219.34,
      "end": 221.72,
      "text": "向普通用户落地的核心基石",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06131814706205117,
      "no_speech_prob": 1.3244232099918207e-11,
      "compression_ratio": 1.2447761194029852,
      "source": "whisper"
    },
    {
      "start": 221.72,
      "end": 225.1,
      "text": "而延迟又是决定用户体验好坏的关键指标",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06131814706205117,
      "no_speech_prob": 1.3244232099918207e-11,
      "compression_ratio": 1.2447761194029852,
      "source": "whisper"
    },
    {
      "start": 225.1,
      "end": 226.7,
      "text": "硬件走向专业化",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06131814706205117,
      "no_speech_prob": 1.3244232099918207e-11,
      "compression_ratio": 1.2447761194029852,
      "source": "whisper"
    },
    {
      "start": 226.7,
      "end": 229.34,
      "text": "是整套系统相比通用GPU TPU设备",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06131814706205117,
      "no_speech_prob": 1.3244232099918207e-11,
      "compression_ratio": 1.2447761194029852,
      "source": "whisper"
    },
    {
      "start": 229.34,
      "end": 231.24,
      "text": "实现更低功耗",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06131814706205117,
      "no_speech_prob": 1.3244232099918207e-11,
      "compression_ratio": 1.2447761194029852,
      "source": "whisper"
    },
    {
      "start": 231.24,
      "end": 233.46,
      "text": "更低延迟最重要的实现路径",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07009432441309879,
      "no_speech_prob": 1.3215095684415701e-11,
      "compression_ratio": 1.2754820936639117,
      "source": "whisper"
    },
    {
      "start": 233.46,
      "end": 237.0,
      "text": "我们所有人都已经习惯等待模型返回输出结果",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07009432441309879,
      "no_speech_prob": 1.3215095684415701e-11,
      "compression_ratio": 1.2754820936639117,
      "source": "whisper"
    },
    {
      "start": 237.0,
      "end": 239.8,
      "text": "但等待本身并不是良好的用户体验",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07009432441309879,
      "no_speech_prob": 1.3215095684415701e-11,
      "compression_ratio": 1.2754820936639117,
      "source": "whisper"
    },
    {
      "start": 239.8,
      "end": 241.36,
      "text": "速度才是一切的王道",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07009432441309879,
      "no_speech_prob": 1.3215095684415701e-11,
      "compression_ratio": 1.2754820936639117,
      "source": "whisper"
    },
    {
      "start": 241.36,
      "end": 243.7,
      "text": "主持人继续顺着这个方向推进",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07009432441309879,
      "no_speech_prob": 1.3215095684415701e-11,
      "compression_ratio": 1.2754820936639117,
      "source": "whisper"
    },
    {
      "start": 243.7,
      "end": 246.12,
      "text": "如果延迟可以直接降低50倍",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07009432441309879,
      "no_speech_prob": 1.3215095684415701e-11,
      "compression_ratio": 1.2754820936639117,
      "source": "whisper"
    },
    {
      "start": 246.12,
      "end": 248.7,
      "text": "整个行业会发生什么样的变化",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07009432441309879,
      "no_speech_prob": 1.3215095684415701e-11,
      "compression_ratio": 1.2754820936639117,
      "source": "whisper"
    },
    {
      "start": 248.7,
      "end": 251.24,
      "text": "紧接着又抛出另外一个值得深思的问题",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07009432441309879,
      "no_speech_prob": 1.3215095684415701e-11,
      "compression_ratio": 1.2754820936639117,
      "source": "whisper"
    },
    {
      "start": 251.24,
      "end": 252.22,
      "text": "时至今日",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07009432441309879,
      "no_speech_prob": 1.3215095684415701e-11,
      "compression_ratio": 1.2754820936639117,
      "source": "whisper"
    },
    {
      "start": 252.22,
      "end": 255.1,
      "text": "行业当中还有哪些被大家全盘接受",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07009432441309879,
      "no_speech_prob": 1.3215095684415701e-11,
      "compression_ratio": 1.2754820936639117,
      "source": "whisper"
    },
    {
      "start": 255.1,
      "end": 258.32,
      "text": "但实际上已经不再成立的固有AI假设",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07009432441309879,
      "no_speech_prob": 1.3215095684415701e-11,
      "compression_ratio": 1.2754820936639117,
      "source": "whisper"
    },
    {
      "start": 258.32,
      "end": 259.92,
      "text": "杰弗迪恩给出的回答",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07009432441309879,
      "no_speech_prob": 1.3215095684415701e-11,
      "compression_ratio": 1.2754820936639117,
      "source": "whisper"
    },
    {
      "start": 259.92,
      "end": 261.66,
      "text": "会刷新不少人的固有认知",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0827026733985314,
      "no_speech_prob": 1.45583961552731e-11,
      "compression_ratio": 1.3372781065088757,
      "source": "whisper"
    },
    {
      "start": 261.66,
      "end": 263.2,
      "text": "绝大多数从业者",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0827026733985314,
      "no_speech_prob": 1.45583961552731e-11,
      "compression_ratio": 1.3372781065088757,
      "source": "whisper"
    },
    {
      "start": 263.2,
      "end": 264.8,
      "text": "还没有真正接纳一件事",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0827026733985314,
      "no_speech_prob": 1.45583961552731e-11,
      "compression_ratio": 1.3372781065088757,
      "source": "whisper"
    },
    {
      "start": 264.8,
      "end": 267.46,
      "text": "依托能力足够强的底层基座模型",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0827026733985314,
      "no_speech_prob": 1.45583961552731e-11,
      "compression_ratio": 1.3372781065088757,
      "source": "whisper"
    },
    {
      "start": 267.46,
      "end": 270.34,
      "text": "Agent并不只能运行短短一两个小时",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0827026733985314,
      "no_speech_prob": 1.45583961552731e-11,
      "compression_ratio": 1.3372781065088757,
      "source": "whisper"
    },
    {
      "start": 270.34,
      "end": 272.5,
      "text": "连续运转几天甚至几周",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0827026733985314,
      "no_speech_prob": 1.45583961552731e-11,
      "compression_ratio": 1.3372781065088757,
      "source": "whisper"
    },
    {
      "start": 272.5,
      "end": 274.76,
      "text": "去完成复杂度极高的任务",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0827026733985314,
      "no_speech_prob": 1.45583961552731e-11,
      "compression_ratio": 1.3372781065088757,
      "source": "whisper"
    },
    {
      "start": 274.76,
      "end": 277.16,
      "text": "这件事已经具备现实可行性",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0827026733985314,
      "no_speech_prob": 1.45583961552731e-11,
      "compression_ratio": 1.3372781065088757,
      "source": "whisper"
    },
    {
      "start": 277.16,
      "end": 278.8,
      "text": "现在小部分从业者",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0827026733985314,
      "no_speech_prob": 1.45583961552731e-11,
      "compression_ratio": 1.3372781065088757,
      "source": "whisper"
    },
    {
      "start": 278.8,
      "end": 280.04,
      "text": "已经看到苗头",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0827026733985314,
      "no_speech_prob": 1.45583961552731e-11,
      "compression_ratio": 1.3372781065088757,
      "source": "whisper"
    },
    {
      "start": 280.04,
      "end": 281.3,
      "text": "可绝大多数人",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0827026733985314,
      "no_speech_prob": 1.45583961552731e-11,
      "compression_ratio": 1.3372781065088757,
      "source": "whisper"
    },
    {
      "start": 281.3,
      "end": 283.18,
      "text": "并没有真正把这个可能性",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0827026733985314,
      "no_speech_prob": 1.45583961552731e-11,
      "compression_ratio": 1.3372781065088757,
      "source": "whisper"
    },
    {
      "start": 283.18,
      "end": 285.98,
      "text": "内化到自己的产品设计逻辑当中",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0827026733985314,
      "no_speech_prob": 1.45583961552731e-11,
      "compression_ratio": 1.3372781065088757,
      "source": "whisper"
    },
    {
      "start": 285.98,
      "end": 288.72,
      "text": "而这件事未来会释放出巨大的影响力",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0827026733985314,
      "no_speech_prob": 1.45583961552731e-11,
      "compression_ratio": 1.3372781065088757,
      "source": "whisper"
    },
    {
      "start": 288.72,
      "end": 291.56,
      "text": "很多人会好奇到底什么样的业务场景",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08960349703095649,
      "no_speech_prob": 1.260387974544308e-11,
      "compression_ratio": 1.3063063063063063,
      "source": "whisper"
    },
    {
      "start": 291.56,
      "end": 294.18,
      "text": "需要Agent连续运航数周的时间",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08960349703095649,
      "no_speech_prob": 1.260387974544308e-11,
      "compression_ratio": 1.3063063063063063,
      "source": "whisper"
    },
    {
      "start": 294.18,
      "end": 296.54,
      "text": "杰弗迪恩举了一个很具象的案例",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08960349703095649,
      "no_speech_prob": 1.260387974544308e-11,
      "compression_ratio": 1.3063063063063063,
      "source": "whisper"
    },
    {
      "start": 296.54,
      "end": 300.68,
      "text": "我们可以交给Agent一项完整的软件版本重构工作",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08960349703095649,
      "no_speech_prob": 1.260387974544308e-11,
      "compression_ratio": 1.3063063063063063,
      "source": "whisper"
    },
    {
      "start": 300.68,
      "end": 304.68,
      "text": "把一整套软件完整迁移到另外一门编程语言",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08960349703095649,
      "no_speech_prob": 1.260387974544308e-11,
      "compression_ratio": 1.3063063063063063,
      "source": "whisper"
    },
    {
      "start": 304.68,
      "end": 307.68,
      "text": "重构之后的版本可以实现安全性的提升",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08960349703095649,
      "no_speech_prob": 1.260387974544308e-11,
      "compression_ratio": 1.3063063063063063,
      "source": "whisper"
    },
    {
      "start": 307.68,
      "end": 309.4,
      "text": "或是运行性能的优化",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08960349703095649,
      "no_speech_prob": 1.260387974544308e-11,
      "compression_ratio": 1.3063063063063063,
      "source": "whisper"
    },
    {
      "start": 309.4,
      "end": 312.56,
      "text": "AI完全可以持续推进整套复杂工程",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08960349703095649,
      "no_speech_prob": 1.260387974544308e-11,
      "compression_ratio": 1.3063063063063063,
      "source": "whisper"
    },
    {
      "start": 312.56,
      "end": 314.4,
      "text": "把这项工作落地完成",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08960349703095649,
      "no_speech_prob": 1.260387974544308e-11,
      "compression_ratio": 1.3063063063063063,
      "source": "whisper"
    },
    {
      "start": 314.4,
      "end": 317.36,
      "text": "话题就此切换到TPU诞生的过往",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08960349703095649,
      "no_speech_prob": 1.260387974544308e-11,
      "compression_ratio": 1.3063063063063063,
      "source": "whisper"
    },
    {
      "start": 317.36,
      "end": 320.22,
      "text": "杰弗迪恩在行业里面广为流传的习惯",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.059893404642740886,
      "no_speech_prob": 1.3012550238722387e-11,
      "compression_ratio": 1.3010752688172043,
      "source": "whisper"
    },
    {
      "start": 320.22,
      "end": 321.84,
      "text": "就是餐巾纸数学",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.059893404642740886,
      "no_speech_prob": 1.3012550238722387e-11,
      "compression_ratio": 1.3010752688172043,
      "source": "whisper"
    },
    {
      "start": 321.84,
      "end": 325.78,
      "text": "也就是在餐巾纸上快速完成数量级层面的粗略估算",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.059893404642740886,
      "no_speech_prob": 1.3012550238722387e-11,
      "compression_ratio": 1.3010752688172043,
      "source": "whisper"
    },
    {
      "start": 325.78,
      "end": 327.92,
      "text": "时间回到2013年前后",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.059893404642740886,
      "no_speech_prob": 1.3012550238722387e-11,
      "compression_ratio": 1.3010752688172043,
      "source": "whisper"
    },
    {
      "start": 327.92,
      "end": 330.54,
      "text": "谷歌语音识别的效果迎来实质性的飞跃",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.059893404642740886,
      "no_speech_prob": 1.3012550238722387e-11,
      "compression_ratio": 1.3010752688172043,
      "source": "whisper"
    },
    {
      "start": 330.54,
      "end": 331.94,
      "text": "他做过一笔测算",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.059893404642740886,
      "no_speech_prob": 1.3012550238722387e-11,
      "compression_ratio": 1.3010752688172043,
      "source": "whisper"
    },
    {
      "start": 331.94,
      "end": 334.0,
      "text": "如果每一位谷歌手机用户",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.059893404642740886,
      "no_speech_prob": 1.3012550238722387e-11,
      "compression_ratio": 1.3010752688172043,
      "source": "whisper"
    },
    {
      "start": 334.0,
      "end": 336.9,
      "text": "每天仅仅使用三分钟的语音识别功能",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.059893404642740886,
      "no_speech_prob": 1.3012550238722387e-11,
      "compression_ratio": 1.3010752688172043,
      "source": "whisper"
    },
    {
      "start": 336.9,
      "end": 340.16,
      "text": "谷歌整体的服务器集群规模就需要直接翻倍",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.059893404642740886,
      "no_speech_prob": 1.3012550238722387e-11,
      "compression_ratio": 1.3010752688172043,
      "source": "whisper"
    },
    {
      "start": 340.16,
      "end": 343.36,
      "text": "对应的成本会膨胀到完全无法接受的地步",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.059893404642740886,
      "no_speech_prob": 1.3012550238722387e-11,
      "compression_ratio": 1.3010752688172043,
      "source": "whisper"
    },
    {
      "start": 343.36,
      "end": 345.18,
      "text": "为了解决这个现实难题",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.059893404642740886,
      "no_speech_prob": 1.3012550238722387e-11,
      "compression_ratio": 1.3010752688172043,
      "source": "whisper"
    },
    {
      "start": 345.18,
      "end": 347.18,
      "text": "团队决定自研专属芯片",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.059893404642740886,
      "no_speech_prob": 1.3012550238722387e-11,
      "compression_ratio": 1.3010752688172043,
      "source": "whisper"
    },
    {
      "start": 347.18,
      "end": 349.02,
      "text": "这也就是TPU的起源",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0747088845243159,
      "no_speech_prob": 1.2755768663552658e-11,
      "compression_ratio": 1.2,
      "source": "whisper"
    },
    {
      "start": 349.02,
      "end": 349.88,
      "text": "他回忆",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0747088845243159,
      "no_speech_prob": 1.2755768663552658e-11,
      "compression_ratio": 1.2,
      "source": "whisper"
    },
    {
      "start": 349.88,
      "end": 352.44,
      "text": "那时候深度学习驱动的语音系统",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0747088845243159,
      "no_speech_prob": 1.2755768663552658e-11,
      "compression_ratio": 1.2,
      "source": "whisper"
    },
    {
      "start": 352.44,
      "end": 353.8,
      "text": "对比旧有方案",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0747088845243159,
      "no_speech_prob": 1.2755768663552658e-11,
      "compression_ratio": 1.2,
      "source": "whisper"
    },
    {
      "start": 353.8,
      "end": 355.84,
      "text": "算力开销高出一大截",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0747088845243159,
      "no_speech_prob": 1.2755768663552658e-11,
      "compression_ratio": 1.2,
      "source": "whisper"
    },
    {
      "start": 355.84,
      "end": 357.96,
      "text": "但是错误率直接减半",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0747088845243159,
      "no_speech_prob": 1.2755768663552658e-11,
      "compression_ratio": 1.2,
      "source": "whisper"
    },
    {
      "start": 357.96,
      "end": 360.12,
      "text": "短短几个月时间就拿到过去",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0747088845243159,
      "no_speech_prob": 1.2755768663552658e-11,
      "compression_ratio": 1.2,
      "source": "whisper"
    },
    {
      "start": 360.12,
      "end": 362.96,
      "text": "20年语音识别领域积累下来的全部进步",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0747088845243159,
      "no_speech_prob": 1.2755768663552658e-11,
      "compression_ratio": 1.2,
      "source": "whisper"
    },
    {
      "start": 362.96,
      "end": 365.78,
      "text": "初代TPU没有办法运行Chrome浏览器",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0747088845243159,
      "no_speech_prob": 1.2755768663552658e-11,
      "compression_ratio": 1.2,
      "source": "whisper"
    },
    {
      "start": 365.78,
      "end": 367.84,
      "text": "也跑不了Word这类通用软件",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0747088845243159,
      "no_speech_prob": 1.2755768663552658e-11,
      "compression_ratio": 1.2,
      "source": "whisper"
    },
    {
      "start": 367.84,
      "end": 370.02,
      "text": "但是在机器学习推理场景当中",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0747088845243159,
      "no_speech_prob": 1.2755768663552658e-11,
      "compression_ratio": 1.2,
      "source": "whisper"
    },
    {
      "start": 370.02,
      "end": 371.3,
      "text": "拥有极强的针对性",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0747088845243159,
      "no_speech_prob": 1.2755768663552658e-11,
      "compression_ratio": 1.2,
      "source": "whisper"
    },
    {
      "start": 371.3,
      "end": 373.26,
      "text": "最终落地的初代芯片",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0747088845243159,
      "no_speech_prob": 1.2755768663552658e-11,
      "compression_ratio": 1.2,
      "source": "whisper"
    },
    {
      "start": 373.26,
      "end": 375.08,
      "text": "对比CPU和GPU",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.0747088845243159,
      "no_speech_prob": 1.2755768663552658e-11,
      "compression_ratio": 1.2,
      "source": "whisper"
    },
    {
      "start": 375.08,
      "end": 377.52,
      "text": "能效可以提升30到80倍",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07475976554714904,
      "no_speech_prob": 1.2018317764594943e-11,
      "compression_ratio": 1.2284122562674096,
      "source": "whisper"
    },
    {
      "start": 377.52,
      "end": 379.6,
      "text": "延迟下降20到30倍",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07475976554714904,
      "no_speech_prob": 1.2018317764594943e-11,
      "compression_ratio": 1.2284122562674096,
      "source": "whisper"
    },
    {
      "start": 379.6,
      "end": 381.88,
      "text": "打下了非常扎实的技术底子",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07475976554714904,
      "no_speech_prob": 1.2018317764594943e-11,
      "compression_ratio": 1.2284122562674096,
      "source": "whisper"
    },
    {
      "start": 381.88,
      "end": 384.82,
      "text": "后续TPU才一步步迭代成如今的模样",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07475976554714904,
      "no_speech_prob": 1.2018317764594943e-11,
      "compression_ratio": 1.2284122562674096,
      "source": "whisper"
    },
    {
      "start": 384.82,
      "end": 388.2,
      "text": "有意思的地方在于团队研发TPU的那个阶段",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07475976554714904,
      "no_speech_prob": 1.2018317764594943e-11,
      "compression_ratio": 1.2284122562674096,
      "source": "whisper"
    },
    {
      "start": 388.2,
      "end": 390.16,
      "text": "Transformer架构还没有问世",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07475976554714904,
      "no_speech_prob": 1.2018317764594943e-11,
      "compression_ratio": 1.2284122562674096,
      "source": "whisper"
    },
    {
      "start": 390.16,
      "end": 394.72,
      "text": "没有人能够预料到后续TPU会成为整个AI产业的基石硬件",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07475976554714904,
      "no_speech_prob": 1.2018317764594943e-11,
      "compression_ratio": 1.2284122562674096,
      "source": "whisper"
    },
    {
      "start": 394.72,
      "end": 396.04,
      "text": "也正是因为这样",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07475976554714904,
      "no_speech_prob": 1.2018317764594943e-11,
      "compression_ratio": 1.2284122562674096,
      "source": "whisper"
    },
    {
      "start": 396.04,
      "end": 400.56,
      "text": "当时团队的设计思路是打造一套通用线性怠数运算系统",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07475976554714904,
      "no_speech_prob": 1.2018317764594943e-11,
      "compression_ratio": 1.2284122562674096,
      "source": "whisper"
    },
    {
      "start": 400.56,
      "end": 404.32,
      "text": "而不是死死绑定某一套特定算法做极致专用化",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.07475976554714904,
      "no_speech_prob": 1.2018317764594943e-11,
      "compression_ratio": 1.2284122562674096,
      "source": "whisper"
    },
    {
      "start": 404.32,
      "end": 407.62,
      "text": "机器学习算法永远处在持续演化的进程当中",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06384702162309126,
      "no_speech_prob": 1.2610753587216639e-11,
      "compression_ratio": 1.2737430167597765,
      "source": "whisper"
    },
    {
      "start": 407.62,
      "end": 409.22,
      "text": "硬件不能过度定制",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06384702162309126,
      "no_speech_prob": 1.2610753587216639e-11,
      "compression_ratio": 1.2737430167597765,
      "source": "whisper"
    },
    {
      "start": 409.22,
      "end": 411.48,
      "text": "但是又要保留足够的专用属性",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06384702162309126,
      "no_speech_prob": 1.2610753587216639e-11,
      "compression_ratio": 1.2737430167597765,
      "source": "whisper"
    },
    {
      "start": 411.48,
      "end": 413.3,
      "text": "以此拿到巨大的性能收益",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06384702162309126,
      "no_speech_prob": 1.2610753587216639e-11,
      "compression_ratio": 1.2737430167597765,
      "source": "whisper"
    },
    {
      "start": 413.3,
      "end": 415.4,
      "text": "比如做大规格乘法运算单元",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06384702162309126,
      "no_speech_prob": 1.2610753587216639e-11,
      "compression_ratio": 1.2737430167597765,
      "source": "whisper"
    },
    {
      "start": 415.4,
      "end": 416.7,
      "text": "配置高速内存",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06384702162309126,
      "no_speech_prob": 1.2610753587216639e-11,
      "compression_ratio": 1.2737430167597765,
      "source": "whisper"
    },
    {
      "start": 416.7,
      "end": 418.18,
      "text": "搭建高速互联通路",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06384702162309126,
      "no_speech_prob": 1.2610753587216639e-11,
      "compression_ratio": 1.2737430167597765,
      "source": "whisper"
    },
    {
      "start": 418.18,
      "end": 421.98,
      "text": "后续几代TPU更是优化多芯片协同工作的能力",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06384702162309126,
      "no_speech_prob": 1.2610753587216639e-11,
      "compression_ratio": 1.2737430167597765,
      "source": "whisper"
    },
    {
      "start": 421.98,
      "end": 425.38,
      "text": "让大量芯片可以联合处理同一个复杂任务",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06384702162309126,
      "no_speech_prob": 1.2610753587216639e-11,
      "compression_ratio": 1.2737430167597765,
      "source": "whisper"
    },
    {
      "start": 425.38,
      "end": 426.48,
      "text": "主持人接着提问",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06384702162309126,
      "no_speech_prob": 1.2610753587216639e-11,
      "compression_ratio": 1.2737430167597765,
      "source": "whisper"
    },
    {
      "start": 426.48,
      "end": 427.48,
      "text": "放到今天",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06384702162309126,
      "no_speech_prob": 1.2610753587216639e-11,
      "compression_ratio": 1.2737430167597765,
      "source": "whisper"
    },
    {
      "start": 427.48,
      "end": 430.16,
      "text": "创业者又该做怎样的餐巾纸数学",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06384702162309126,
      "no_speech_prob": 1.2610753587216639e-11,
      "compression_ratio": 1.2737430167597765,
      "source": "whisper"
    },
    {
      "start": 430.16,
      "end": 432.72,
      "text": "能不能依靠简单的数量级估算",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.06384702162309126,
      "no_speech_prob": 1.2610753587216639e-11,
      "compression_ratio": 1.2737430167597765,
      "source": "whisper"
    },
    {
      "start": 432.72,
      "end": 436.24,
      "text": "挖掘出下一代如同TPO一般重量级的创新",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08594183535189243,
      "no_speech_prob": 1.2858742716448379e-11,
      "compression_ratio": 1.2919075144508672,
      "source": "whisper"
    },
    {
      "start": 436.24,
      "end": 437.42,
      "text": "杰弗蒂恩表示",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08594183535189243,
      "no_speech_prob": 1.2858742716448379e-11,
      "compression_ratio": 1.2919075144508672,
      "source": "whisper"
    },
    {
      "start": 437.42,
      "end": 439.48,
      "text": "这件事没有统一标准答案",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08594183535189243,
      "no_speech_prob": 1.2858742716448379e-11,
      "compression_ratio": 1.2919075144508672,
      "source": "whisper"
    },
    {
      "start": 439.48,
      "end": 441.88,
      "text": "但是底层思考逻辑是相通的",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08594183535189243,
      "no_speech_prob": 1.2858742716448379e-11,
      "compression_ratio": 1.2919075144508672,
      "source": "whisper"
    },
    {
      "start": 441.88,
      "end": 444.8,
      "text": "首先找准自己所在赛道真正的瓶颈",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08594183535189243,
      "no_speech_prob": 1.2858742716448379e-11,
      "compression_ratio": 1.2919075144508672,
      "source": "whisper"
    },
    {
      "start": 444.8,
      "end": 446.12,
      "text": "再去反问自己",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08594183535189243,
      "no_speech_prob": 1.2858742716448379e-11,
      "compression_ratio": 1.2919075144508672,
      "source": "whisper"
    },
    {
      "start": 446.12,
      "end": 449.92,
      "text": "是否存在一套和当下主流方案完全不一样的解法",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08594183535189243,
      "no_speech_prob": 1.2858742716448379e-11,
      "compression_ratio": 1.2919075144508672,
      "source": "whisper"
    },
    {
      "start": 449.92,
      "end": 453.44,
      "text": "能不能把核心指标抬升一个甚至两个数量级",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08594183535189243,
      "no_speech_prob": 1.2858742716448379e-11,
      "compression_ratio": 1.2919075144508672,
      "source": "whisper"
    },
    {
      "start": 453.44,
      "end": 456.98,
      "text": "不要被市面上已经成熟的实现方案束缚住思路",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08594183535189243,
      "no_speech_prob": 1.2858742716448379e-11,
      "compression_ratio": 1.2919075144508672,
      "source": "whisper"
    },
    {
      "start": 456.98,
      "end": 458.94,
      "text": "回归第一性原理去思考",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08594183535189243,
      "no_speech_prob": 1.2858742716448379e-11,
      "compression_ratio": 1.2919075144508672,
      "source": "whisper"
    },
    {
      "start": 458.94,
      "end": 461.04,
      "text": "如果从零开始解决这个问题",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.08594183535189243,
      "no_speech_prob": 1.2858742716448379e-11,
      "compression_ratio": 1.2919075144508672,
      "source": "whisper"
    },
    {
      "start": 461.04,
      "end": 462.86,
      "text": "自己会选择什么样的路径",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09456344242513615,
      "no_speech_prob": 1.485497141962e-11,
      "compression_ratio": 1.2213438735177866,
      "source": "whisper"
    },
    {
      "start": 462.86,
      "end": 465.46,
      "text": "往往就能够挖掘出别人看不见的机会",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09456344242513615,
      "no_speech_prob": 1.485497141962e-11,
      "compression_ratio": 1.2213438735177866,
      "source": "whisper"
    },
    {
      "start": 465.46,
      "end": 467.06,
      "text": "随后主持人问到一个",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09456344242513615,
      "no_speech_prob": 1.485497141962e-11,
      "compression_ratio": 1.2213438735177866,
      "source": "whisper"
    },
    {
      "start": 467.06,
      "end": 469.14,
      "text": "只有内行才会深究的话题",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09456344242513615,
      "no_speech_prob": 1.485497141962e-11,
      "compression_ratio": 1.2213438735177866,
      "source": "whisper"
    },
    {
      "start": 469.14,
      "end": 472.6,
      "text": "多年以前杰弗迪恩写过一份流传极广的清单",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09456344242513615,
      "no_speech_prob": 1.485497141962e-11,
      "compression_ratio": 1.2213438735177866,
      "source": "whisper"
    },
    {
      "start": 472.6,
      "end": 476.16,
      "text": "名字叫做每个工程师都应该知道的延迟数字",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09456344242513615,
      "no_speech_prob": 1.485497141962e-11,
      "compression_ratio": 1.2213438735177866,
      "source": "whisper"
    },
    {
      "start": 476.16,
      "end": 478.56,
      "text": "里面记录缓存未命中的耗石",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09456344242513615,
      "no_speech_prob": 1.485497141962e-11,
      "compression_ratio": 1.2213438735177866,
      "source": "whisper"
    },
    {
      "start": 478.56,
      "end": 480.0,
      "text": "磁盘寻道耗石",
      "chunk": 0,
      "language": "zh",
      "avg_logprob": -0.09456344242513615,
      "no_speech_prob": 1.485497141962e-11,
      "compression_ratio": 1.2213438735177866,
      "source": "whisper"
    },
    {
      "start": 494.28,
      "end": 497.66,
      "text": "杰弗迪恩列举出一系列至关重要的硬件数字",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.054761942971516125,
      "no_speech_prob": 1.41700921829635e-11,
      "compression_ratio": 1.4115755627009647,
      "source": "whisper"
    },
    {
      "start": 497.66,
      "end": 500.6,
      "text": "储存到加速器片上内存之间的带宽",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.054761942971516125,
      "no_speech_prob": 1.41700921829635e-11,
      "compression_ratio": 1.4115755627009647,
      "source": "whisper"
    },
    {
      "start": 500.6,
      "end": 502.88,
      "text": "通往乘法运算单元的带宽",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.054761942971516125,
      "no_speech_prob": 1.41700921829635e-11,
      "compression_ratio": 1.4115755627009647,
      "source": "whisper"
    },
    {
      "start": 502.88,
      "end": 505.7,
      "text": "完成一次乘法运算需要消耗的能量",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.054761942971516125,
      "no_speech_prob": 1.41700921829635e-11,
      "compression_ratio": 1.4115755627009647,
      "source": "whisper"
    },
    {
      "start": 505.7,
      "end": 507.98,
      "text": "芯片与芯片之间的互联带宽",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.054761942971516125,
      "no_speech_prob": 1.41700921829635e-11,
      "compression_ratio": 1.4115755627009647,
      "source": "whisper"
    },
    {
      "start": 507.98,
      "end": 511.22,
      "text": "在该带宽条件之下可以联动多少颗芯片",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.054761942971516125,
      "no_speech_prob": 1.41700921829635e-11,
      "compression_ratio": 1.4115755627009647,
      "source": "whisper"
    },
    {
      "start": 511.22,
      "end": 515.24,
      "text": "还有从500颗芯片扩展到1万颗芯片的规模过程当中",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.054761942971516125,
      "no_speech_prob": 1.41700921829635e-11,
      "compression_ratio": 1.4115755627009647,
      "source": "whisper"
    },
    {
      "start": 515.24,
      "end": 517.98,
      "text": "网络带宽会出现多大程度的衰减",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.054761942971516125,
      "no_speech_prob": 1.41700921829635e-11,
      "compression_ratio": 1.4115755627009647,
      "source": "whisper"
    },
    {
      "start": 517.98,
      "end": 522.36,
      "text": "这些底层数字会实实在在约束上层业务问题的解法",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.054761942971516125,
      "no_speech_prob": 1.41700921829635e-11,
      "compression_ratio": 1.4115755627009647,
      "source": "whisper"
    },
    {
      "start": 522.36,
      "end": 526.44,
      "text": "紧接着他抛出一个足以改变很多人AI系统设计思路的数据",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07002168101042949,
      "no_speech_prob": 1.0985907496208203e-11,
      "compression_ratio": 1.338368580060423,
      "source": "whisper"
    },
    {
      "start": 526.44,
      "end": 529.94,
      "text": "完成一次计算消耗的能量大概在EP焦2级别",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07002168101042949,
      "no_speech_prob": 1.0985907496208203e-11,
      "compression_ratio": 1.338368580060423,
      "source": "whisper"
    },
    {
      "start": 529.94,
      "end": 531.8,
      "text": "但是完成一次数据搬运",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07002168101042949,
      "no_speech_prob": 1.0985907496208203e-11,
      "compression_ratio": 1.338368580060423,
      "source": "whisper"
    },
    {
      "start": 531.8,
      "end": 532.98,
      "text": "一次IO操作",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07002168101042949,
      "no_speech_prob": 1.0985907496208203e-11,
      "compression_ratio": 1.338368580060423,
      "source": "whisper"
    },
    {
      "start": 532.98,
      "end": 535.3,
      "text": "消耗的能量是前者的1000倍",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07002168101042949,
      "no_speech_prob": 1.0985907496208203e-11,
      "compression_ratio": 1.338368580060423,
      "source": "whisper"
    },
    {
      "start": 535.3,
      "end": 539.78,
      "text": "就算只是把数据从加速器的高带宽内存搬运到处理器内部",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07002168101042949,
      "no_speech_prob": 1.0985907496208203e-11,
      "compression_ratio": 1.338368580060423,
      "source": "whisper"
    },
    {
      "start": 539.78,
      "end": 540.86,
      "text": "参与运算",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07002168101042949,
      "no_speech_prob": 1.0985907496208203e-11,
      "compression_ratio": 1.338368580060423,
      "source": "whisper"
    },
    {
      "start": 540.86,
      "end": 543.16,
      "text": "能量开销同样会急剧拉高",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07002168101042949,
      "no_speech_prob": 1.0985907496208203e-11,
      "compression_ratio": 1.338368580060423,
      "source": "whisper"
    },
    {
      "start": 543.16,
      "end": 547.6,
      "text": "搬数据的代价远远高于做计算这条物理层面的硬约数",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07002168101042949,
      "no_speech_prob": 1.0985907496208203e-11,
      "compression_ratio": 1.338368580060423,
      "source": "whisper"
    },
    {
      "start": 547.6,
      "end": 550.54,
      "text": "塑造了机器学习领域大量的工程选择",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07002168101042949,
      "no_speech_prob": 1.0985907496208203e-11,
      "compression_ratio": 1.338368580060423,
      "source": "whisper"
    },
    {
      "start": 550.54,
      "end": 553.16,
      "text": "如果不存在如此悬殊的能耗差距",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08323018891470772,
      "no_speech_prob": 1.1751593621822654e-11,
      "compression_ratio": 1.3072625698324023,
      "source": "whisper"
    },
    {
      "start": 553.16,
      "end": 555.6,
      "text": "我们就不需要对数据做批量处理",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08323018891470772,
      "no_speech_prob": 1.1751593621822654e-11,
      "compression_ratio": 1.3072625698324023,
      "source": "whisper"
    },
    {
      "start": 555.6,
      "end": 557.48,
      "text": "可现实条件摆在眼前",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08323018891470772,
      "no_speech_prob": 1.1751593621822654e-11,
      "compression_ratio": 1.3072625698324023,
      "source": "whisper"
    },
    {
      "start": 557.48,
      "end": 561.62,
      "text": "工程师只能把大量样本或者token攒成批次统一处理",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08323018891470772,
      "no_speech_prob": 1.1751593621822654e-11,
      "compression_ratio": 1.3072625698324023,
      "source": "whisper"
    },
    {
      "start": 561.62,
      "end": 564.68,
      "text": "以此贪薄数据搬运带来的巨大能耗惩罚",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08323018891470772,
      "no_speech_prob": 1.1751593621822654e-11,
      "compression_ratio": 1.3072625698324023,
      "source": "whisper"
    },
    {
      "start": 564.68,
      "end": 566.16,
      "text": "把1000倍的开销",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08323018891470772,
      "no_speech_prob": 1.1751593621822654e-11,
      "compression_ratio": 1.3072625698324023,
      "source": "whisper"
    },
    {
      "start": 566.16,
      "end": 568.38,
      "text": "稀释为1000除以批次规模",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08323018891470772,
      "no_speech_prob": 1.1751593621822654e-11,
      "compression_ratio": 1.3072625698324023,
      "source": "whisper"
    },
    {
      "start": 568.38,
      "end": 571.38,
      "text": "但如果我们追求极低延迟的推理表现",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08323018891470772,
      "no_speech_prob": 1.1751593621822654e-11,
      "compression_ratio": 1.3072625698324023,
      "source": "whisper"
    },
    {
      "start": 571.38,
      "end": 574.1,
      "text": "批量处理这套逻辑就会变成阻碍",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08323018891470772,
      "no_speech_prob": 1.1751593621822654e-11,
      "compression_ratio": 1.3072625698324023,
      "source": "whisper"
    },
    {
      "start": 574.1,
      "end": 575.88,
      "text": "底层硬件的能量约束",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08323018891470772,
      "no_speech_prob": 1.1751593621822654e-11,
      "compression_ratio": 1.3072625698324023,
      "source": "whisper"
    },
    {
      "start": 575.88,
      "end": 579.18,
      "text": "会自上而下影响整套AI系统的每一层设计",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08323018891470772,
      "no_speech_prob": 1.1751593621822654e-11,
      "compression_ratio": 1.3072625698324023,
      "source": "whisper"
    },
    {
      "start": 579.18,
      "end": 580.46,
      "text": "主持人继续追问",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08323018891470772,
      "no_speech_prob": 1.1751593621822654e-11,
      "compression_ratio": 1.3072625698324023,
      "source": "whisper"
    },
    {
      "start": 580.46,
      "end": 583.96,
      "text": "那我们有没有机会实现单条数据的处理模式",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.05626670116653091,
      "no_speech_prob": 1.3050733237152112e-11,
      "compression_ratio": 1.3806818181818181,
      "source": "whisper"
    },
    {
      "start": 583.96,
      "end": 587.54,
      "text": "杰弗迪恩表示自己现阶段思考更多的是推理测的问题",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.05626670116653091,
      "no_speech_prob": 1.3050733237152112e-11,
      "compression_ratio": 1.3806818181818181,
      "source": "whisper"
    },
    {
      "start": 587.54,
      "end": 590.76,
      "text": "推理业务天然对低延迟有着强烈诉求",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.05626670116653091,
      "no_speech_prob": 1.3050733237152112e-11,
      "compression_ratio": 1.3806818181818181,
      "source": "whisper"
    },
    {
      "start": 590.76,
      "end": 593.7,
      "text": "训练环节对此的要求反而没有那么严苛",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.05626670116653091,
      "no_speech_prob": 1.3050733237152112e-11,
      "compression_ratio": 1.3806818181818181,
      "source": "whisper"
    },
    {
      "start": 593.7,
      "end": 594.8,
      "text": "在他看来",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.05626670116653091,
      "no_speech_prob": 1.3050733237152112e-11,
      "compression_ratio": 1.3806818181818181,
      "source": "whisper"
    },
    {
      "start": 594.8,
      "end": 596.6,
      "text": "推理硬件的专用化程度",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.05626670116653091,
      "no_speech_prob": 1.3050733237152112e-11,
      "compression_ratio": 1.3806818181818181,
      "source": "whisper"
    },
    {
      "start": 596.6,
      "end": 598.58,
      "text": "如今远远还没有抵达上线",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.05626670116653091,
      "no_speech_prob": 1.3050733237152112e-11,
      "compression_ratio": 1.3806818181818181,
      "source": "whisper"
    },
    {
      "start": 598.58,
      "end": 599.92,
      "text": "核心优化方向",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.05626670116653091,
      "no_speech_prob": 1.3050733237152112e-11,
      "compression_ratio": 1.3806818181818181,
      "source": "whisper"
    },
    {
      "start": 599.92,
      "end": 602.76,
      "text": "一方面是尽可能压缩数据搬运的频次",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.05626670116653091,
      "no_speech_prob": 1.3050733237152112e-11,
      "compression_ratio": 1.3806818181818181,
      "source": "whisper"
    },
    {
      "start": 602.76,
      "end": 605.22,
      "text": "另一方面去探索极低精度运算",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.05626670116653091,
      "no_speech_prob": 1.3050733237152112e-11,
      "compression_ratio": 1.3806818181818181,
      "source": "whisper"
    },
    {
      "start": 605.22,
      "end": 608.4,
      "text": "不需要硬件去兼容五花八门的精度规格",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.05626670116653091,
      "no_speech_prob": 1.3050733237152112e-11,
      "compression_ratio": 1.3806818181818181,
      "source": "whisper"
    },
    {
      "start": 608.4,
      "end": 610.36,
      "text": "把业务真正需要的精度",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.05626670116653091,
      "no_speech_prob": 1.3050733237152112e-11,
      "compression_ratio": 1.3806818181818181,
      "source": "whisper"
    },
    {
      "start": 610.36,
      "end": 612.44,
      "text": "直接固化到硬件设计当中",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08143887800328872,
      "no_speech_prob": 1.736130510188172e-11,
      "compression_ratio": 1.307936507936508,
      "source": "whisper"
    },
    {
      "start": 612.44,
      "end": 614.52,
      "text": "聊完硬件层面的种种约束",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08143887800328872,
      "no_speech_prob": 1.736130510188172e-11,
      "compression_ratio": 1.307936507936508,
      "source": "whisper"
    },
    {
      "start": 614.52,
      "end": 616.42,
      "text": "对话把视角向上拉高",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08143887800328872,
      "no_speech_prob": 1.736130510188172e-11,
      "compression_ratio": 1.307936507936508,
      "source": "whisper"
    },
    {
      "start": 616.42,
      "end": 618.46,
      "text": "过去大家谈及AI进步",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08143887800328872,
      "no_speech_prob": 1.736130510188172e-11,
      "compression_ratio": 1.307936507936508,
      "source": "whisper"
    },
    {
      "start": 618.46,
      "end": 620.76,
      "text": "惯性思维就是模型规模更大",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08143887800328872,
      "no_speech_prob": 1.736130510188172e-11,
      "compression_ratio": 1.307936507936508,
      "source": "whisper"
    },
    {
      "start": 620.76,
      "end": 622.24,
      "text": "训练数据更多",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08143887800328872,
      "no_speech_prob": 1.736130510188172e-11,
      "compression_ratio": 1.307936507936508,
      "source": "whisper"
    },
    {
      "start": 622.24,
      "end": 623.52,
      "text": "参数体量上涨",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08143887800328872,
      "no_speech_prob": 1.736130510188172e-11,
      "compression_ratio": 1.307936507936508,
      "source": "whisper"
    },
    {
      "start": 623.52,
      "end": 624.8,
      "text": "训练时间拉长",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08143887800328872,
      "no_speech_prob": 1.736130510188172e-11,
      "compression_ratio": 1.307936507936508,
      "source": "whisper"
    },
    {
      "start": 624.8,
      "end": 625.96,
      "text": "但是最近几年",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08143887800328872,
      "no_speech_prob": 1.736130510188172e-11,
      "compression_ratio": 1.307936507936508,
      "source": "whisper"
    },
    {
      "start": 625.96,
      "end": 627.4,
      "text": "真正拉开差距的",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08143887800328872,
      "no_speech_prob": 1.736130510188172e-11,
      "compression_ratio": 1.307936507936508,
      "source": "whisper"
    },
    {
      "start": 627.4,
      "end": 628.82,
      "text": "不再只有模型本体",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08143887800328872,
      "no_speech_prob": 1.736130510188172e-11,
      "compression_ratio": 1.307936507936508,
      "source": "whisper"
    },
    {
      "start": 628.82,
      "end": 631.1,
      "text": "模型外围的整套配套体系",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08143887800328872,
      "no_speech_prob": 1.736130510188172e-11,
      "compression_ratio": 1.307936507936508,
      "source": "whisper"
    },
    {
      "start": 631.1,
      "end": 632.8,
      "text": "价值正在持续走高",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08143887800328872,
      "no_speech_prob": 1.736130510188172e-11,
      "compression_ratio": 1.307936507936508,
      "source": "whisper"
    },
    {
      "start": 632.8,
      "end": 635.56,
      "text": "不单单要看参数规模和训练数据集",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08143887800328872,
      "no_speech_prob": 1.736130510188172e-11,
      "compression_ratio": 1.307936507936508,
      "source": "whisper"
    },
    {
      "start": 635.56,
      "end": 636.48,
      "text": "检索工具",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08143887800328872,
      "no_speech_prob": 1.736130510188172e-11,
      "compression_ratio": 1.307936507936508,
      "source": "whisper"
    },
    {
      "start": 636.48,
      "end": 637.4,
      "text": "记忆模块",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08143887800328872,
      "no_speech_prob": 1.736130510188172e-11,
      "compression_ratio": 1.307936507936508,
      "source": "whisper"
    },
    {
      "start": 637.4,
      "end": 638.04,
      "text": "Agent",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08143887800328872,
      "no_speech_prob": 1.736130510188172e-11,
      "compression_ratio": 1.307936507936508,
      "source": "whisper"
    },
    {
      "start": 638.04,
      "end": 638.76,
      "text": "工具链",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08143887800328872,
      "no_speech_prob": 1.736130510188172e-11,
      "compression_ratio": 1.307936507936508,
      "source": "whisper"
    },
    {
      "start": 638.76,
      "end": 641.04,
      "text": "全部都是不可或缺的组成部分",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06920879696487287,
      "no_speech_prob": 1.3574736820731736e-11,
      "compression_ratio": 1.3667621776504297,
      "source": "whisper"
    },
    {
      "start": 641.04,
      "end": 642.88,
      "text": "杰弗迪恩认同这样的看法",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06920879696487287,
      "no_speech_prob": 1.3574736820731736e-11,
      "compression_ratio": 1.3667621776504297,
      "source": "whisper"
    },
    {
      "start": 642.88,
      "end": 645.32,
      "text": "模型只是整套解决方案当中的",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06920879696487287,
      "no_speech_prob": 1.3574736820731736e-11,
      "compression_ratio": 1.3667621776504297,
      "source": "whisper"
    },
    {
      "start": 645.32,
      "end": 646.32,
      "text": "其中一块拼图",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06920879696487287,
      "no_speech_prob": 1.3574736820731736e-11,
      "compression_ratio": 1.3667621776504297,
      "source": "whisper"
    },
    {
      "start": 646.32,
      "end": 647.54,
      "text": "我们真正的目标",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06920879696487287,
      "no_speech_prob": 1.3574736820731736e-11,
      "compression_ratio": 1.3667621776504297,
      "source": "whisper"
    },
    {
      "start": 647.54,
      "end": 649.14,
      "text": "是搭建一整套",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06920879696487287,
      "no_speech_prob": 1.3574736820731736e-11,
      "compression_ratio": 1.3667621776504297,
      "source": "whisper"
    },
    {
      "start": 649.14,
      "end": 652.02,
      "text": "可以解决复杂现实问题的完整系统",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06920879696487287,
      "no_speech_prob": 1.3574736820731736e-11,
      "compression_ratio": 1.3667621776504297,
      "source": "whisper"
    },
    {
      "start": 652.02,
      "end": 653.9,
      "text": "这就意味着模型要学会",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06920879696487287,
      "no_speech_prob": 1.3574736820731736e-11,
      "compression_ratio": 1.3667621776504297,
      "source": "whisper"
    },
    {
      "start": 653.9,
      "end": 655.56,
      "text": "调用各式各样的外部工具",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06920879696487287,
      "no_speech_prob": 1.3574736820731736e-11,
      "compression_ratio": 1.3667621776504297,
      "source": "whisper"
    },
    {
      "start": 655.56,
      "end": 657.54,
      "text": "检索外部相关信息",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06920879696487287,
      "no_speech_prob": 1.3574736820731736e-11,
      "compression_ratio": 1.3667621776504297,
      "source": "whisper"
    },
    {
      "start": 657.54,
      "end": 659.1,
      "text": "留存过往处理其他任务",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06920879696487287,
      "no_speech_prob": 1.3574736820731736e-11,
      "compression_ratio": 1.3667621776504297,
      "source": "whisper"
    },
    {
      "start": 659.1,
      "end": 660.48,
      "text": "沉淀下来的历史信息",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06920879696487287,
      "no_speech_prob": 1.3574736820731736e-11,
      "compression_ratio": 1.3667621776504297,
      "source": "whisper"
    },
    {
      "start": 660.48,
      "end": 661.9,
      "text": "并且把这些信息",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06920879696487287,
      "no_speech_prob": 1.3574736820731736e-11,
      "compression_ratio": 1.3667621776504297,
      "source": "whisper"
    },
    {
      "start": 661.9,
      "end": 663.78,
      "text": "带入当下的上下文当中",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06920879696487287,
      "no_speech_prob": 1.3574736820731736e-11,
      "compression_ratio": 1.3667621776504297,
      "source": "whisper"
    },
    {
      "start": 663.78,
      "end": 665.66,
      "text": "这里有一个很关键的区别",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06920879696487287,
      "no_speech_prob": 1.3574736820731736e-11,
      "compression_ratio": 1.3667621776504297,
      "source": "whisper"
    },
    {
      "start": 665.66,
      "end": 668.08,
      "text": "训练数据就像是一锅大杂烩",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06920879696487287,
      "no_speech_prob": 1.3574736820731736e-11,
      "compression_ratio": 1.3667621776504297,
      "source": "whisper"
    },
    {
      "start": 668.08,
      "end": 670.36,
      "text": "数万亿token混杂在一起",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07582448732734907,
      "no_speech_prob": 1.2594436778201601e-11,
      "compression_ratio": 1.287323943661972,
      "source": "whisper"
    },
    {
      "start": 670.36,
      "end": 673.28,
      "text": "压缩近几千亿的模型参数内部",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07582448732734907,
      "no_speech_prob": 1.2594436778201601e-11,
      "compression_ratio": 1.287323943661972,
      "source": "whisper"
    },
    {
      "start": 673.28,
      "end": 673.92,
      "text": "与之对比",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07582448732734907,
      "no_speech_prob": 1.2594436778201601e-11,
      "compression_ratio": 1.287323943661972,
      "source": "whisper"
    },
    {
      "start": 673.92,
      "end": 677.06,
      "text": "模型在处理具体任务时拿到的上下文信息",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07582448732734907,
      "no_speech_prob": 1.2594436778201601e-11,
      "compression_ratio": 1.287323943661972,
      "source": "whisper"
    },
    {
      "start": 677.06,
      "end": 678.44,
      "text": "逻辑会清晰很多",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07582448732734907,
      "no_speech_prob": 1.2594436778201601e-11,
      "compression_ratio": 1.287323943661972,
      "source": "whisper"
    },
    {
      "start": 678.44,
      "end": 680.9,
      "text": "同时模型还要识别可用的工具",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07582448732734907,
      "no_speech_prob": 1.2594436778201601e-11,
      "compression_ratio": 1.287323943661972,
      "source": "whisper"
    },
    {
      "start": 680.9,
      "end": 683.32,
      "text": "判断哪些工具适配当前步骤",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07582448732734907,
      "no_speech_prob": 1.2594436778201601e-11,
      "compression_ratio": 1.287323943661972,
      "source": "whisper"
    },
    {
      "start": 683.32,
      "end": 686.28,
      "text": "如何把大任务拆解成一连串工具调用",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07582448732734907,
      "no_speech_prob": 1.2594436778201601e-11,
      "compression_ratio": 1.287323943661972,
      "source": "whisper"
    },
    {
      "start": 686.28,
      "end": 688.92,
      "text": "甚至尝试多条不同的解题路径",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07582448732734907,
      "no_speech_prob": 1.2594436778201601e-11,
      "compression_ratio": 1.287323943661972,
      "source": "whisper"
    },
    {
      "start": 688.92,
      "end": 690.98,
      "text": "再评估各个方案的实际效果",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07582448732734907,
      "no_speech_prob": 1.2594436778201601e-11,
      "compression_ratio": 1.287323943661972,
      "source": "whisper"
    },
    {
      "start": 690.98,
      "end": 694.92,
      "text": "这就是复杂agent以及dooragent编排的核心能力",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07582448732734907,
      "no_speech_prob": 1.2594436778201601e-11,
      "compression_ratio": 1.287323943661972,
      "source": "whisper"
    },
    {
      "start": 694.92,
      "end": 697.44,
      "text": "上下文工程这个方向很有意思的一点",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07582448732734907,
      "no_speech_prob": 1.2594436778201601e-11,
      "compression_ratio": 1.287323943661972,
      "source": "whisper"
    },
    {
      "start": 697.44,
      "end": 699.72,
      "text": "使它的准入门槛相对更低",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06962668727821028,
      "no_speech_prob": 1.1844190558885082e-11,
      "compression_ratio": 1.3076923076923077,
      "source": "whisper"
    },
    {
      "start": 699.72,
      "end": 701.84,
      "text": "训练大模型需要海量算力",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06962668727821028,
      "no_speech_prob": 1.1844190558885082e-11,
      "compression_ratio": 1.3076923076923077,
      "source": "whisper"
    },
    {
      "start": 701.84,
      "end": 703.34,
      "text": "海量数据资源",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06962668727821028,
      "no_speech_prob": 1.1844190558885082e-11,
      "compression_ratio": 1.3076923076923077,
      "source": "whisper"
    },
    {
      "start": 703.34,
      "end": 704.9,
      "text": "普通团队很难触碰",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06962668727821028,
      "no_speech_prob": 1.1844190558885082e-11,
      "compression_ratio": 1.3076923076923077,
      "source": "whisper"
    },
    {
      "start": 704.9,
      "end": 706.46,
      "text": "但是做上下文工程",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06962668727821028,
      "no_speech_prob": 1.1844190558885082e-11,
      "compression_ratio": 1.3076923076923077,
      "source": "whisper"
    },
    {
      "start": 706.46,
      "end": 709.34,
      "text": "只需要调用Gemini这类模型的API接口",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06962668727821028,
      "no_speech_prob": 1.1844190558885082e-11,
      "compression_ratio": 1.3076923076923077,
      "source": "whisper"
    },
    {
      "start": 709.34,
      "end": 711.68,
      "text": "再搭配自己搭建的检索逻辑",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06962668727821028,
      "no_speech_prob": 1.1844190558885082e-11,
      "compression_ratio": 1.3076923076923077,
      "source": "whisper"
    },
    {
      "start": 711.68,
      "end": 713.04,
      "text": "工具调用流程",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06962668727821028,
      "no_speech_prob": 1.1844190558885082e-11,
      "compression_ratio": 1.3076923076923077,
      "source": "whisper"
    },
    {
      "start": 713.04,
      "end": 714.16,
      "text": "工作流编排",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06962668727821028,
      "no_speech_prob": 1.1844190558885082e-11,
      "compression_ratio": 1.3076923076923077,
      "source": "whisper"
    },
    {
      "start": 714.16,
      "end": 715.54,
      "text": "就可以动手实践",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06962668727821028,
      "no_speech_prob": 1.1844190558885082e-11,
      "compression_ratio": 1.3076923076923077,
      "source": "whisper"
    },
    {
      "start": 715.54,
      "end": 717.96,
      "text": "想要把上下文工程做到足够强",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06962668727821028,
      "no_speech_prob": 1.1844190558885082e-11,
      "compression_ratio": 1.3076923076923077,
      "source": "whisper"
    },
    {
      "start": 717.96,
      "end": 721.06,
      "text": "杰弗迪恩给出一套非常务实的实操思路",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06962668727821028,
      "no_speech_prob": 1.1844190558885082e-11,
      "compression_ratio": 1.3076923076923077,
      "source": "whisper"
    },
    {
      "start": 721.06,
      "end": 723.36,
      "text": "就是实实在在拿模型和工具链",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06962668727821028,
      "no_speech_prob": 1.1844190558885082e-11,
      "compression_ratio": 1.3076923076923077,
      "source": "whisper"
    },
    {
      "start": 723.36,
      "end": 724.58,
      "text": "去处理真实业务",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06962668727821028,
      "no_speech_prob": 1.1844190558885082e-11,
      "compression_ratio": 1.3076923076923077,
      "source": "whisper"
    },
    {
      "start": 724.58,
      "end": 727.32,
      "text": "在实操当中观察模型的出错点位",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06962668727821028,
      "no_speech_prob": 1.1844190558885082e-11,
      "compression_ratio": 1.3076923076923077,
      "source": "whisper"
    },
    {
      "start": 727.32,
      "end": 729.66,
      "text": "很多时候想要提升模型表现",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0675343625685748,
      "no_speech_prob": 1.4453885137377664e-11,
      "compression_ratio": 1.3032069970845481,
      "source": "whisper"
    },
    {
      "start": 729.66,
      "end": 732.18,
      "text": "并不需要去改动底层模型权重",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0675343625685748,
      "no_speech_prob": 1.4453885137377664e-11,
      "compression_ratio": 1.3032069970845481,
      "source": "whisper"
    },
    {
      "start": 732.18,
      "end": 734.36,
      "text": "而是提供质量更高的指引",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0675343625685748,
      "no_speech_prob": 1.4453885137377664e-11,
      "compression_ratio": 1.3032069970845481,
      "source": "whisper"
    },
    {
      "start": 734.36,
      "end": 736.4,
      "text": "编写更完善的技能描述",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0675343625685748,
      "no_speech_prob": 1.4453885137377664e-11,
      "compression_ratio": 1.3032069970845481,
      "source": "whisper"
    },
    {
      "start": 736.4,
      "end": 738.44,
      "text": "告诉模型面对某一类问题",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0675343625685748,
      "no_speech_prob": 1.4453885137377664e-11,
      "compression_ratio": 1.3032069970845481,
      "source": "whisper"
    },
    {
      "start": 738.44,
      "end": 739.94,
      "text": "应当如何调用工具",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0675343625685748,
      "no_speech_prob": 1.4453885137377664e-11,
      "compression_ratio": 1.3032069970845481,
      "source": "whisper"
    },
    {
      "start": 739.94,
      "end": 741.28,
      "text": "反复迭代之后",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0675343625685748,
      "no_speech_prob": 1.4453885137377664e-11,
      "compression_ratio": 1.3032069970845481,
      "source": "whisper"
    },
    {
      "start": 741.28,
      "end": 742.78,
      "text": "你就会慢慢摸清楚",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0675343625685748,
      "no_speech_prob": 1.4453885137377664e-11,
      "compression_ratio": 1.3032069970845481,
      "source": "whisper"
    },
    {
      "start": 742.78,
      "end": 744.48,
      "text": "模型还缺少哪些信息",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0675343625685748,
      "no_speech_prob": 1.4453885137377664e-11,
      "compression_ratio": 1.3032069970845481,
      "source": "whisper"
    },
    {
      "start": 744.48,
      "end": 747.4,
      "text": "由此进入自我迭代优化的正向循环",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0675343625685748,
      "no_speech_prob": 1.4453885137377664e-11,
      "compression_ratio": 1.3032069970845481,
      "source": "whisper"
    },
    {
      "start": 747.4,
      "end": 749.4,
      "text": "他分享了一段亲身经历",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0675343625685748,
      "no_speech_prob": 1.4453885137377664e-11,
      "compression_ratio": 1.3032069970845481,
      "source": "whisper"
    },
    {
      "start": 749.4,
      "end": 750.3,
      "text": "几周之前",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0675343625685748,
      "no_speech_prob": 1.4453885137377664e-11,
      "compression_ratio": 1.3032069970845481,
      "source": "whisper"
    },
    {
      "start": 750.3,
      "end": 753.24,
      "text": "他和桑杰一起做底层代码库的性能调优",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0675343625685748,
      "no_speech_prob": 1.4453885137377664e-11,
      "compression_ratio": 1.3032069970845481,
      "source": "whisper"
    },
    {
      "start": 753.24,
      "end": 756.56,
      "text": "团队内部有自研的微型基准测试库",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0675343625685748,
      "no_speech_prob": 1.4453885137377664e-11,
      "compression_ratio": 1.3032069970845481,
      "source": "whisper"
    },
    {
      "start": 756.56,
      "end": 758.64,
      "text": "可以统计各类操作的耗时",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08361771371629503,
      "no_speech_prob": 1.3203707224795913e-11,
      "compression_ratio": 1.3354037267080745,
      "source": "whisper"
    },
    {
      "start": 758.64,
      "end": 762.28,
      "text": "这套数据结构被谷歌内部几百万个进程调用",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08361771371629503,
      "no_speech_prob": 1.3203707224795913e-11,
      "compression_ratio": 1.3354037267080745,
      "source": "whisper"
    },
    {
      "start": 762.28,
      "end": 764.38,
      "text": "性能表现必须得到保障",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08361771371629503,
      "no_speech_prob": 1.3203707224795913e-11,
      "compression_ratio": 1.3354037267080745,
      "source": "whisper"
    },
    {
      "start": 764.38,
      "end": 766.02,
      "text": "在没有Agent的时代",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08361771371629503,
      "no_speech_prob": 1.3203707224795913e-11,
      "compression_ratio": 1.3354037267080745,
      "source": "whisper"
    },
    {
      "start": 766.02,
      "end": 768.5,
      "text": "完整流程是先测试现有性能",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08361771371629503,
      "no_speech_prob": 1.3203707224795913e-11,
      "compression_ratio": 1.3354037267080745,
      "source": "whisper"
    },
    {
      "start": 768.5,
      "end": 769.48,
      "text": "修改代码",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08361771371629503,
      "no_speech_prob": 1.3203707224795913e-11,
      "compression_ratio": 1.3354037267080745,
      "source": "whisper"
    },
    {
      "start": 769.48,
      "end": 771.12,
      "text": "重新跑基准测试",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08361771371629503,
      "no_speech_prob": 1.3203707224795913e-11,
      "compression_ratio": 1.3354037267080745,
      "source": "whisper"
    },
    {
      "start": 771.12,
      "end": 772.6,
      "text": "观察性能变化",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08361771371629503,
      "no_speech_prob": 1.3203707224795913e-11,
      "compression_ratio": 1.3354037267080745,
      "source": "whisper"
    },
    {
      "start": 772.6,
      "end": 774.56,
      "text": "还要执行更大范围的测试",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08361771371629503,
      "no_speech_prob": 1.3203707224795913e-11,
      "compression_ratio": 1.3354037267080745,
      "source": "whisper"
    },
    {
      "start": 774.56,
      "end": 776.8,
      "text": "检查缓存占用等一系列指标",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08361771371629503,
      "no_speech_prob": 1.3203707224795913e-11,
      "compression_ratio": 1.3354037267080745,
      "source": "whisper"
    },
    {
      "start": 776.8,
      "end": 779.06,
      "text": "后来他们专门编写对应的技能",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08361771371629503,
      "no_speech_prob": 1.3203707224795913e-11,
      "compression_ratio": 1.3354037267080745,
      "source": "whisper"
    },
    {
      "start": 779.06,
      "end": 781.86,
      "text": "教会Agent完整复刻整套工作流",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08361771371629503,
      "no_speech_prob": 1.3203707224795913e-11,
      "compression_ratio": 1.3354037267080745,
      "source": "whisper"
    },
    {
      "start": 781.86,
      "end": 783.26,
      "text": "执行基准测试",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08361771371629503,
      "no_speech_prob": 1.3203707224795913e-11,
      "compression_ratio": 1.3354037267080745,
      "source": "whisper"
    },
    {
      "start": 783.26,
      "end": 784.46,
      "text": "修改代码",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08361771371629503,
      "no_speech_prob": 1.3203707224795913e-11,
      "compression_ratio": 1.3354037267080745,
      "source": "whisper"
    },
    {
      "start": 784.46,
      "end": 785.9,
      "text": "重新叫验性能指标",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.08361771371629503,
      "no_speech_prob": 1.3203707224795913e-11,
      "compression_ratio": 1.3354037267080745,
      "source": "whisper"
    },
    {
      "start": 785.9,
      "end": 787.48,
      "text": "《循环迭代优化》",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07956562871518343,
      "no_speech_prob": 1.6005247474581985e-11,
      "compression_ratio": 1.2255639097744362,
      "source": "whisper"
    },
    {
      "start": 787.48,
      "end": 790.5,
      "text": "这套方案在部分问题上拿到相当不错的效果",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07956562871518343,
      "no_speech_prob": 1.6005247474581985e-11,
      "compression_ratio": 1.2255639097744362,
      "source": "whisper"
    },
    {
      "start": 790.5,
      "end": 793.32,
      "text": "本质上就是把人类工程师的工作流程",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07956562871518343,
      "no_speech_prob": 1.6005247474581985e-11,
      "compression_ratio": 1.2255639097744362,
      "source": "whisper"
    },
    {
      "start": 793.32,
      "end": 796.08,
      "text": "转化成模型能够读懂执行的指令",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07956562871518343,
      "no_speech_prob": 1.6005247474581985e-11,
      "compression_ratio": 1.2255639097744362,
      "source": "whisper"
    },
    {
      "start": 796.08,
      "end": 797.1,
      "text": "几个月之前",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07956562871518343,
      "no_speech_prob": 1.6005247474581985e-11,
      "compression_ratio": 1.2255639097744362,
      "source": "whisper"
    },
    {
      "start": 797.1,
      "end": 800.44,
      "text": "他和桑杰共同发布过一份30页左右的公开文档",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07956562871518343,
      "no_speech_prob": 1.6005247474581985e-11,
      "compression_ratio": 1.2255639097744362,
      "source": "whisper"
    },
    {
      "start": 800.44,
      "end": 802.98,
      "text": "名字叫做Performance Heightened TS",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07956562871518343,
      "no_speech_prob": 1.6005247474581985e-11,
      "compression_ratio": 1.2255639097744362,
      "source": "whisper"
    },
    {
      "start": 802.98,
      "end": 805.48,
      "text": "记录大量性能优化的实操技巧",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07956562871518343,
      "no_speech_prob": 1.6005247474581985e-11,
      "compression_ratio": 1.2255639097744362,
      "source": "whisper"
    },
    {
      "start": 805.48,
      "end": 809.06,
      "text": "已经有不少开发者把这份文档做总结提炼之后",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07956562871518343,
      "no_speech_prob": 1.6005247474581985e-11,
      "compression_ratio": 1.2255639097744362,
      "source": "whisper"
    },
    {
      "start": 809.06,
      "end": 810.52,
      "text": "投入给不同的大模型",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07956562871518343,
      "no_speech_prob": 1.6005247474581985e-11,
      "compression_ratio": 1.2255639097744362,
      "source": "whisper"
    },
    {
      "start": 810.52,
      "end": 812.48,
      "text": "最终测试结果显示",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07956562871518343,
      "no_speech_prob": 1.6005247474581985e-11,
      "compression_ratio": 1.2255639097744362,
      "source": "whisper"
    },
    {
      "start": 812.48,
      "end": 815.76,
      "text": "模型分析代码性能问题的能力得到明显提升",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07956562871518343,
      "no_speech_prob": 1.6005247474581985e-11,
      "compression_ratio": 1.2255639097744362,
      "source": "whisper"
    },
    {
      "start": 815.76,
      "end": 818.64,
      "text": "而且这份资料对所有人完全公开",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09500239372253418,
      "no_speech_prob": 1.3157344137815219e-11,
      "compression_ratio": 1.3343465045592706,
      "source": "whisper"
    },
    {
      "start": 818.64,
      "end": 819.78,
      "text": "谈及agent",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09500239372253418,
      "no_speech_prob": 1.3157344137815219e-11,
      "compression_ratio": 1.3343465045592706,
      "source": "whisper"
    },
    {
      "start": 819.78,
      "end": 822.24,
      "text": "有一个所有人都会碰到的现实痛点",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09500239372253418,
      "no_speech_prob": 1.3157344137815219e-11,
      "compression_ratio": 1.3343465045592706,
      "source": "whisper"
    },
    {
      "start": 822.24,
      "end": 825.94,
      "text": "不少agent在前十几步执行都十分顺畅",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09500239372253418,
      "no_speech_prob": 1.3157344137815219e-11,
      "compression_ratio": 1.3343465045592706,
      "source": "whisper"
    },
    {
      "start": 825.94,
      "end": 827.4,
      "text": "等到推进到第三十步",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09500239372253418,
      "no_speech_prob": 1.3157344137815219e-11,
      "compression_ratio": 1.3343465045592706,
      "source": "whisper"
    },
    {
      "start": 827.4,
      "end": 828.12,
      "text": "四十步",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09500239372253418,
      "no_speech_prob": 1.3157344137815219e-11,
      "compression_ratio": 1.3343465045592706,
      "source": "whisper"
    },
    {
      "start": 828.12,
      "end": 829.36,
      "text": "甚至五十步之后",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09500239372253418,
      "no_speech_prob": 1.3157344137815219e-11,
      "compression_ratio": 1.3343465045592706,
      "source": "whisper"
    },
    {
      "start": 829.36,
      "end": 831.04,
      "text": "就会开始逻辑跑偏",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09500239372253418,
      "no_speech_prob": 1.3157344137815219e-11,
      "compression_ratio": 1.3343465045592706,
      "source": "whisper"
    },
    {
      "start": 831.04,
      "end": 832.42,
      "text": "杰弗迪恩坦言",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09500239372253418,
      "no_speech_prob": 1.3157344137815219e-11,
      "compression_ratio": 1.3343465045592706,
      "source": "whisper"
    },
    {
      "start": 832.42,
      "end": 834.98,
      "text": "这就是当下技术最大的约束条件",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09500239372253418,
      "no_speech_prob": 1.3157344137815219e-11,
      "compression_ratio": 1.3343465045592706,
      "source": "whisper"
    },
    {
      "start": 834.98,
      "end": 838.36,
      "text": "一部分原因是模型处理训练机之外的任务",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09500239372253418,
      "no_speech_prob": 1.3157344137815219e-11,
      "compression_ratio": 1.3343465045592706,
      "source": "whisper"
    },
    {
      "start": 838.36,
      "end": 840.12,
      "text": "一旦脱离熟悉的任务分布",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09500239372253418,
      "no_speech_prob": 1.3157344137815219e-11,
      "compression_ratio": 1.3343465045592706,
      "source": "whisper"
    },
    {
      "start": 840.12,
      "end": 841.42,
      "text": "效果就会下滑",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09500239372253418,
      "no_speech_prob": 1.3157344137815219e-11,
      "compression_ratio": 1.3343465045592706,
      "source": "whisper"
    },
    {
      "start": 841.42,
      "end": 843.62,
      "text": "偏离熟悉的边界越远",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09500239372253418,
      "no_speech_prob": 1.3157344137815219e-11,
      "compression_ratio": 1.3343465045592706,
      "source": "whisper"
    },
    {
      "start": 843.62,
      "end": 844.76,
      "text": "失效概率就越高",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09500239372253418,
      "no_speech_prob": 1.3157344137815219e-11,
      "compression_ratio": 1.3343465045592706,
      "source": "whisper"
    },
    {
      "start": 844.76,
      "end": 845.78,
      "text": "针对该问题",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06384218863721164,
      "no_speech_prob": 1.2259980357309797e-11,
      "compression_ratio": 1.3284457478005864,
      "source": "whisper"
    },
    {
      "start": 845.78,
      "end": 847.88,
      "text": "有两条可行的解决路径",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06384218863721164,
      "no_speech_prob": 1.2259980357309797e-11,
      "compression_ratio": 1.3284457478005864,
      "source": "whisper"
    },
    {
      "start": 847.88,
      "end": 848.56,
      "text": "第一种",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06384218863721164,
      "no_speech_prob": 1.2259980357309797e-11,
      "compression_ratio": 1.3284457478005864,
      "source": "whisper"
    },
    {
      "start": 848.56,
      "end": 850.92,
      "text": "编写对应的技能和提示之意",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06384218863721164,
      "no_speech_prob": 1.2259980357309797e-11,
      "compression_ratio": 1.3284457478005864,
      "source": "whisper"
    },
    {
      "start": 850.92,
      "end": 853.24,
      "text": "尽量让模型维持在自己熟悉",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06384218863721164,
      "no_speech_prob": 1.2259980357309797e-11,
      "compression_ratio": 1.3284457478005864,
      "source": "whisper"
    },
    {
      "start": 853.24,
      "end": 855.06,
      "text": "成功率更高的执行路径",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06384218863721164,
      "no_speech_prob": 1.2259980357309797e-11,
      "compression_ratio": 1.3284457478005864,
      "source": "whisper"
    },
    {
      "start": 855.06,
      "end": 855.84,
      "text": "第二种",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06384218863721164,
      "no_speech_prob": 1.2259980357309797e-11,
      "compression_ratio": 1.3284457478005864,
      "source": "whisper"
    },
    {
      "start": 855.84,
      "end": 857.4,
      "text": "采用多agent的架构",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06384218863721164,
      "no_speech_prob": 1.2259980357309797e-11,
      "compression_ratio": 1.3284457478005864,
      "source": "whisper"
    },
    {
      "start": 857.4,
      "end": 858.88,
      "text": "让多个独立agent",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06384218863721164,
      "no_speech_prob": 1.2259980357309797e-11,
      "compression_ratio": 1.3284457478005864,
      "source": "whisper"
    },
    {
      "start": 858.88,
      "end": 860.78,
      "text": "并行尝试不同解题思路",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06384218863721164,
      "no_speech_prob": 1.2259980357309797e-11,
      "compression_ratio": 1.3284457478005864,
      "source": "whisper"
    },
    {
      "start": 860.78,
      "end": 863.1,
      "text": "再交由另外一个模型或者agent",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06384218863721164,
      "no_speech_prob": 1.2259980357309797e-11,
      "compression_ratio": 1.3284457478005864,
      "source": "whisper"
    },
    {
      "start": 863.1,
      "end": 864.92,
      "text": "去评判各个路径的可行性",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06384218863721164,
      "no_speech_prob": 1.2259980357309797e-11,
      "compression_ratio": 1.3284457478005864,
      "source": "whisper"
    },
    {
      "start": 864.92,
      "end": 867.12,
      "text": "本质就是对全部可行解法",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06384218863721164,
      "no_speech_prob": 1.2259980357309797e-11,
      "compression_ratio": 1.3284457478005864,
      "source": "whisper"
    },
    {
      "start": 867.12,
      "end": 868.08,
      "text": "做搜索筛选",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06384218863721164,
      "no_speech_prob": 1.2259980357309797e-11,
      "compression_ratio": 1.3284457478005864,
      "source": "whisper"
    },
    {
      "start": 868.08,
      "end": 869.64,
      "text": "保留有希望的分支",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06384218863721164,
      "no_speech_prob": 1.2259980357309797e-11,
      "compression_ratio": 1.3284457478005864,
      "source": "whisper"
    },
    {
      "start": 869.64,
      "end": 871.54,
      "text": "丢弃无效跑偏的分支",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06384218863721164,
      "no_speech_prob": 1.2259980357309797e-11,
      "compression_ratio": 1.3284457478005864,
      "source": "whisper"
    },
    {
      "start": 871.54,
      "end": 872.78,
      "text": "依靠推理计算",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06384218863721164,
      "no_speech_prob": 1.2259980357309797e-11,
      "compression_ratio": 1.3284457478005864,
      "source": "whisper"
    },
    {
      "start": 872.78,
      "end": 874.46,
      "text": "去便利解决方案空间",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.06384218863721164,
      "no_speech_prob": 1.2259980357309797e-11,
      "compression_ratio": 1.3284457478005864,
      "source": "whisper"
    },
    {
      "start": 874.46,
      "end": 876.88,
      "text": "以此提升常链路Agent的定性",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07274852980167494,
      "no_speech_prob": 1.2059676174347445e-11,
      "compression_ratio": 1.2919075144508672,
      "source": "whisper"
    },
    {
      "start": 876.88,
      "end": 880.0,
      "text": "在谷歌内部就搭建了一整套调度框架",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07274852980167494,
      "no_speech_prob": 1.2059676174347445e-11,
      "compression_ratio": 1.2919075144508672,
      "source": "whisper"
    },
    {
      "start": 880.0,
      "end": 881.22,
      "text": "以及海量技能库",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07274852980167494,
      "no_speech_prob": 1.2059676174347445e-11,
      "compression_ratio": 1.2919075144508672,
      "source": "whisper"
    },
    {
      "start": 881.22,
      "end": 883.34,
      "text": "Agent可以调用内部编码工具",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07274852980167494,
      "no_speech_prob": 1.2059676174347445e-11,
      "compression_ratio": 1.2919075144508672,
      "source": "whisper"
    },
    {
      "start": 883.34,
      "end": 884.66,
      "text": "执行代码审查",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07274852980167494,
      "no_speech_prob": 1.2059676174347445e-11,
      "compression_ratio": 1.2919075144508672,
      "source": "whisper"
    },
    {
      "start": 884.66,
      "end": 885.98,
      "text": "完成性能测试",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07274852980167494,
      "no_speech_prob": 1.2059676174347445e-11,
      "compression_ratio": 1.2919075144508672,
      "source": "whisper"
    },
    {
      "start": 885.98,
      "end": 887.3,
      "text": "读取日志文件",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07274852980167494,
      "no_speech_prob": 1.2059676174347445e-11,
      "compression_ratio": 1.2919075144508672,
      "source": "whisper"
    },
    {
      "start": 887.3,
      "end": 889.04,
      "text": "只要技能定义足够合理",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07274852980167494,
      "no_speech_prob": 1.2059676174347445e-11,
      "compression_ratio": 1.2919075144508672,
      "source": "whisper"
    },
    {
      "start": 889.04,
      "end": 891.68,
      "text": "Agent的实用价值就会被大幅释放",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07274852980167494,
      "no_speech_prob": 1.2059676174347445e-11,
      "compression_ratio": 1.2919075144508672,
      "source": "whisper"
    },
    {
      "start": 891.68,
      "end": 892.48,
      "text": "紧接着",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07274852980167494,
      "no_speech_prob": 1.2059676174347445e-11,
      "compression_ratio": 1.2919075144508672,
      "source": "whisper"
    },
    {
      "start": 892.48,
      "end": 895.9,
      "text": "对话落到在场每一位创业者都无比关心的命题",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07274852980167494,
      "no_speech_prob": 1.2059676174347445e-11,
      "compression_ratio": 1.2919075144508672,
      "source": "whisper"
    },
    {
      "start": 895.9,
      "end": 898.96,
      "text": "谷歌能够做到从处理器到上层产品",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07274852980167494,
      "no_speech_prob": 1.2059676174347445e-11,
      "compression_ratio": 1.2919075144508672,
      "source": "whisper"
    },
    {
      "start": 898.96,
      "end": 900.48,
      "text": "全链路协同设计",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07274852980167494,
      "no_speech_prob": 1.2059676174347445e-11,
      "compression_ratio": 1.2919075144508672,
      "source": "whisper"
    },
    {
      "start": 900.48,
      "end": 903.34,
      "text": "那对于两三个人组成的小型创业团队",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.07274852980167494,
      "no_speech_prob": 1.2059676174347445e-11,
      "compression_ratio": 1.2919075144508672,
      "source": "whisper"
    },
    {
      "start": 903.34,
      "end": 905.6,
      "text": "还剩下哪些可以抓住的机会",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0628199081075677,
      "no_speech_prob": 1.3212583804822486e-11,
      "compression_ratio": 1.344213649851632,
      "source": "whisper"
    },
    {
      "start": 905.6,
      "end": 906.98,
      "text": "杰弗迪恩讲到",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0628199081075677,
      "no_speech_prob": 1.3212583804822486e-11,
      "compression_ratio": 1.344213649851632,
      "source": "whisper"
    },
    {
      "start": 906.98,
      "end": 909.32,
      "text": "谷歌以及通用大模型团队",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0628199081075677,
      "no_speech_prob": 1.3212583804822486e-11,
      "compression_ratio": 1.344213649851632,
      "source": "whisper"
    },
    {
      "start": 909.32,
      "end": 911.88,
      "text": "都在竭尽全力打造通用性极强",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0628199081075677,
      "no_speech_prob": 1.3212583804822486e-11,
      "compression_ratio": 1.344213649851632,
      "source": "whisper"
    },
    {
      "start": 911.88,
      "end": 914.48,
      "text": "几乎可以处理各类任务的基座模型",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0628199081075677,
      "no_speech_prob": 1.3212583804822486e-11,
      "compression_ratio": 1.344213649851632,
      "source": "whisper"
    },
    {
      "start": 914.48,
      "end": 917.02,
      "text": "但恰恰是追求通用这个前提",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0628199081075677,
      "no_speech_prob": 1.3212583804822486e-11,
      "compression_ratio": 1.344213649851632,
      "source": "whisper"
    },
    {
      "start": 917.02,
      "end": 919.98,
      "text": "导致它很难针对某一个垂直细分领域",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0628199081075677,
      "no_speech_prob": 1.3212583804822486e-11,
      "compression_ratio": 1.344213649851632,
      "source": "whisper"
    },
    {
      "start": 919.98,
      "end": 921.06,
      "text": "做深度打磨",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0628199081075677,
      "no_speech_prob": 1.3212583804822486e-11,
      "compression_ratio": 1.344213649851632,
      "source": "whisper"
    },
    {
      "start": 921.06,
      "end": 922.28,
      "text": "如果创业团队",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0628199081075677,
      "no_speech_prob": 1.3212583804822486e-11,
      "compression_ratio": 1.344213649851632,
      "source": "whisper"
    },
    {
      "start": 922.28,
      "end": 925.64,
      "text": "可以针对特定行业打磨优质的产品交互界面",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0628199081075677,
      "no_speech_prob": 1.3212583804822486e-11,
      "compression_ratio": 1.344213649851632,
      "source": "whisper"
    },
    {
      "start": 925.64,
      "end": 927.46,
      "text": "配套专属的技能体系",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0628199081075677,
      "no_speech_prob": 1.3212583804822486e-11,
      "compression_ratio": 1.344213649851632,
      "source": "whisper"
    },
    {
      "start": 927.46,
      "end": 929.82,
      "text": "甚至是垂直微调的专用模型",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0628199081075677,
      "no_speech_prob": 1.3212583804822486e-11,
      "compression_ratio": 1.344213649851632,
      "source": "whisper"
    },
    {
      "start": 929.82,
      "end": 932.6,
      "text": "完全有机会建立起明显的竞争优势",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.0628199081075677,
      "no_speech_prob": 1.3212583804822486e-11,
      "compression_ratio": 1.344213649851632,
      "source": "whisper"
    },
    {
      "start": 932.6,
      "end": 934.12,
      "text": "打造出体验流畅",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09054155257141706,
      "no_speech_prob": 1.3364943700355791e-11,
      "compression_ratio": 1.3396825396825396,
      "source": "whisper"
    },
    {
      "start": 934.12,
      "end": 936.06,
      "text": "准确率出色的垂直产品",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09054155257141706,
      "no_speech_prob": 1.3364943700355791e-11,
      "compression_ratio": 1.3396825396825396,
      "source": "whisper"
    },
    {
      "start": 936.06,
      "end": 937.8,
      "text": "不过他也做出提醒",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09054155257141706,
      "no_speech_prob": 1.3364943700355791e-11,
      "compression_ratio": 1.3396825396825396,
      "source": "whisper"
    },
    {
      "start": 937.8,
      "end": 940.4,
      "text": "通用模型的能力还在持续变强",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09054155257141706,
      "no_speech_prob": 1.3364943700355791e-11,
      "compression_ratio": 1.3396825396825396,
      "source": "whisper"
    },
    {
      "start": 940.4,
      "end": 942.34,
      "text": "覆盖的场景不断扩张",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09054155257141706,
      "no_speech_prob": 1.3364943700355791e-11,
      "compression_ratio": 1.3396825396825396,
      "source": "whisper"
    },
    {
      "start": 942.34,
      "end": 944.2,
      "text": "创业者必须做好判断",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09054155257141706,
      "no_speech_prob": 1.3364943700355791e-11,
      "compression_ratio": 1.3396825396825396,
      "source": "whisper"
    },
    {
      "start": 944.2,
      "end": 945.52,
      "text": "自己选择的赛道",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09054155257141706,
      "no_speech_prob": 1.3364943700355791e-11,
      "compression_ratio": 1.3396825396825396,
      "source": "whisper"
    },
    {
      "start": 945.52,
      "end": 947.36,
      "text": "是未来半年到12个月",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09054155257141706,
      "no_speech_prob": 1.3364943700355791e-11,
      "compression_ratio": 1.3396825396825396,
      "source": "whisper"
    },
    {
      "start": 947.36,
      "end": 948.88,
      "text": "就会被通用模型覆盖",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09054155257141706,
      "no_speech_prob": 1.3364943700355791e-11,
      "compression_ratio": 1.3396825396825396,
      "source": "whisper"
    },
    {
      "start": 948.88,
      "end": 950.22,
      "text": "还是两三年之内",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09054155257141706,
      "no_speech_prob": 1.3364943700355791e-11,
      "compression_ratio": 1.3396825396825396,
      "source": "whisper"
    },
    {
      "start": 950.22,
      "end": 951.82,
      "text": "通用模型都难以企及",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09054155257141706,
      "no_speech_prob": 1.3364943700355791e-11,
      "compression_ratio": 1.3396825396825396,
      "source": "whisper"
    },
    {
      "start": 951.82,
      "end": 954.22,
      "text": "以此判断赛道的抗风险能力",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09054155257141706,
      "no_speech_prob": 1.3364943700355791e-11,
      "compression_ratio": 1.3396825396825396,
      "source": "whisper"
    },
    {
      "start": 954.22,
      "end": 956.24,
      "text": "那创业者应当如何筛选",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09054155257141706,
      "no_speech_prob": 1.3364943700355791e-11,
      "compression_ratio": 1.3396825396825396,
      "source": "whisper"
    },
    {
      "start": 956.24,
      "end": 957.62,
      "text": "值得投入的方向",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09054155257141706,
      "no_speech_prob": 1.3364943700355791e-11,
      "compression_ratio": 1.3396825396825396,
      "source": "whisper"
    },
    {
      "start": 957.62,
      "end": 959.4,
      "text": "他给出的第一条建议朴素",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09054155257141706,
      "no_speech_prob": 1.3364943700355791e-11,
      "compression_ratio": 1.3396825396825396,
      "source": "whisper"
    },
    {
      "start": 959.4,
      "end": 960.0,
      "text": "但是分类",
      "chunk": 1,
      "language": "zh",
      "avg_logprob": -0.09054155257141706,
      "no_speech_prob": 1.3364943700355791e-11,
      "compression_ratio": 1.3396825396825396,
      "source": "whisper"
    },
    {
      "start": 975.0,
      "end": 976.64,
      "text": "往往代表存在机会",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06611669631231398,
      "no_speech_prob": 2.977109184176818e-11,
      "compression_ratio": 1.368421052631579,
      "source": "whisper"
    },
    {
      "start": 976.64,
      "end": 978.46,
      "text": "如果已经可以完成一部分",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06611669631231398,
      "no_speech_prob": 2.977109184176818e-11,
      "compression_ratio": 1.368421052631579,
      "source": "whisper"
    },
    {
      "start": 978.46,
      "end": 980.04,
      "text": "只是效果不够完美",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06611669631231398,
      "no_speech_prob": 2.977109184176818e-11,
      "compression_ratio": 1.368421052631579,
      "source": "whisper"
    },
    {
      "start": 980.04,
      "end": 981.38,
      "text": "反而要提高警惕",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06611669631231398,
      "no_speech_prob": 2.977109184176818e-11,
      "compression_ratio": 1.368421052631579,
      "source": "whisper"
    },
    {
      "start": 981.38,
      "end": 983.76,
      "text": "这代表能力的种子已经存在",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06611669631231398,
      "no_speech_prob": 2.977109184176818e-11,
      "compression_ratio": 1.368421052631579,
      "source": "whisper"
    },
    {
      "start": 983.76,
      "end": 985.56,
      "text": "只需要补充训练数据",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06611669631231398,
      "no_speech_prob": 2.977109184176818e-11,
      "compression_ratio": 1.368421052631579,
      "source": "whisper"
    },
    {
      "start": 985.56,
      "end": 986.98,
      "text": "扩大模型规模",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06611669631231398,
      "no_speech_prob": 2.977109184176818e-11,
      "compression_ratio": 1.368421052631579,
      "source": "whisper"
    },
    {
      "start": 986.98,
      "end": 988.64,
      "text": "能力就会快速迭代",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06611669631231398,
      "no_speech_prob": 2.977109184176818e-11,
      "compression_ratio": 1.368421052631579,
      "source": "whisper"
    },
    {
      "start": 988.64,
      "end": 991.16,
      "text": "所以要优先寻找当下通用模型",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06611669631231398,
      "no_speech_prob": 2.977109184176818e-11,
      "compression_ratio": 1.368421052631579,
      "source": "whisper"
    },
    {
      "start": 991.16,
      "end": 993.98,
      "text": "成功率只有0%或者1%的问题",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06611669631231398,
      "no_speech_prob": 2.977109184176818e-11,
      "compression_ratio": 1.368421052631579,
      "source": "whisper"
    },
    {
      "start": 993.98,
      "end": 996.22,
      "text": "而不是20%左右的赛道",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06611669631231398,
      "no_speech_prob": 2.977109184176818e-11,
      "compression_ratio": 1.368421052631579,
      "source": "whisper"
    },
    {
      "start": 996.22,
      "end": 997.28,
      "text": "这类机会",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06611669631231398,
      "no_speech_prob": 2.977109184176818e-11,
      "compression_ratio": 1.368421052631579,
      "source": "whisper"
    },
    {
      "start": 997.28,
      "end": 1000.3,
      "text": "一部分属于训练分布之外的全新任务",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06611669631231398,
      "no_speech_prob": 2.977109184176818e-11,
      "compression_ratio": 1.368421052631579,
      "source": "whisper"
    },
    {
      "start": 1000.3,
      "end": 1001.3,
      "text": "另一类场景",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06611669631231398,
      "no_speech_prob": 2.977109184176818e-11,
      "compression_ratio": 1.368421052631579,
      "source": "whisper"
    },
    {
      "start": 1001.3,
      "end": 1003.3,
      "text": "是产品可以拿到通用模型",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06611669631231398,
      "no_speech_prob": 2.977109184176818e-11,
      "compression_ratio": 1.368421052631579,
      "source": "whisper"
    },
    {
      "start": 1003.3,
      "end": 1004.9,
      "text": "接触不到的私有数据",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06611669631231398,
      "no_speech_prob": 2.977109184176818e-11,
      "compression_ratio": 1.368421052631579,
      "source": "whisper"
    },
    {
      "start": 1005.0,
      "end": 1007.1,
      "text": "比如处理用户个人信息的产品",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07852886540229331,
      "no_speech_prob": 1.3458918007858145e-11,
      "compression_ratio": 1.2493074792243768,
      "source": "whisper"
    },
    {
      "start": 1007.1,
      "end": 1010.16,
      "text": "通用模型本身无法获取用户私人资料",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07852886540229331,
      "no_speech_prob": 1.3458918007858145e-11,
      "compression_ratio": 1.2493074792243768,
      "source": "whisper"
    },
    {
      "start": 1010.16,
      "end": 1011.72,
      "text": "这就创造天然壁垒",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07852886540229331,
      "no_speech_prob": 1.3458918007858145e-11,
      "compression_ratio": 1.2493074792243768,
      "source": "whisper"
    },
    {
      "start": 1011.72,
      "end": 1012.96,
      "text": "还有一类机会",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07852886540229331,
      "no_speech_prob": 1.3458918007858145e-11,
      "compression_ratio": 1.2493074792243768,
      "source": "whisper"
    },
    {
      "start": 1012.96,
      "end": 1014.32,
      "text": "针对高难度问题",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07852886540229331,
      "no_speech_prob": 1.3458918007858145e-11,
      "compression_ratio": 1.2493074792243768,
      "source": "whisper"
    },
    {
      "start": 1014.32,
      "end": 1017.02,
      "text": "以可控成本训练垂直专用模型",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07852886540229331,
      "no_speech_prob": 1.3458918007858145e-11,
      "compression_ratio": 1.2493074792243768,
      "source": "whisper"
    },
    {
      "start": 1017.02,
      "end": 1018.52,
      "text": "换取极高精度",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07852886540229331,
      "no_speech_prob": 1.3458918007858145e-11,
      "compression_ratio": 1.2493074792243768,
      "source": "whisper"
    },
    {
      "start": 1018.52,
      "end": 1020.68,
      "text": "AlphaFold就是典型案例",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07852886540229331,
      "no_speech_prob": 1.3458918007858145e-11,
      "compression_ratio": 1.2493074792243768,
      "source": "whisper"
    },
    {
      "start": 1020.68,
      "end": 1023.86,
      "text": "面向蛋白质折叠的专用模型取得巨大成功",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07852886540229331,
      "no_speech_prob": 1.3458918007858145e-11,
      "compression_ratio": 1.2493074792243768,
      "source": "whisper"
    },
    {
      "start": 1023.86,
      "end": 1025.48,
      "text": "却并非通用大模型",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07852886540229331,
      "no_speech_prob": 1.3458918007858145e-11,
      "compression_ratio": 1.2493074792243768,
      "source": "whisper"
    },
    {
      "start": 1025.48,
      "end": 1026.44,
      "text": "材料科学",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07852886540229331,
      "no_speech_prob": 1.3458918007858145e-11,
      "compression_ratio": 1.2493074792243768,
      "source": "whisper"
    },
    {
      "start": 1026.44,
      "end": 1029.92,
      "text": "芯片设计领域也存在同类的落地可能性",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07852886540229331,
      "no_speech_prob": 1.3458918007858145e-11,
      "compression_ratio": 1.2493074792243768,
      "source": "whisper"
    },
    {
      "start": 1029.92,
      "end": 1032.34,
      "text": "怎样成长为AI原生的创始人",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07852886540229331,
      "no_speech_prob": 1.3458918007858145e-11,
      "compression_ratio": 1.2493074792243768,
      "source": "whisper"
    },
    {
      "start": 1032.34,
      "end": 1033.14,
      "text": "他提到",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07852886540229331,
      "no_speech_prob": 1.3458918007858145e-11,
      "compression_ratio": 1.2493074792243768,
      "source": "whisper"
    },
    {
      "start": 1033.14,
      "end": 1034.92,
      "text": "和虚拟agent的协作的时候",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07852886540229331,
      "no_speech_prob": 1.3458918007858145e-11,
      "compression_ratio": 1.2493074792243768,
      "source": "whisper"
    },
    {
      "start": 1034.92,
      "end": 1036.84,
      "text": "把需求描述的越清晰",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06117100899036114,
      "no_speech_prob": 1.3886273207697197e-11,
      "compression_ratio": 1.3402366863905326,
      "source": "whisper"
    },
    {
      "start": 1036.84,
      "end": 1038.72,
      "text": "最终产出效果就越好",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06117100899036114,
      "no_speech_prob": 1.3886273207697197e-11,
      "compression_ratio": 1.3402366863905326,
      "source": "whisper"
    },
    {
      "start": 1038.72,
      "end": 1040.46,
      "text": "需求定义足够明确",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06117100899036114,
      "no_speech_prob": 1.3886273207697197e-11,
      "compression_ratio": 1.3402366863905326,
      "source": "whisper"
    },
    {
      "start": 1040.46,
      "end": 1042.54,
      "text": "Agent就拥有清晰的边界",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06117100899036114,
      "no_speech_prob": 1.3886273207697197e-11,
      "compression_ratio": 1.3402366863905326,
      "source": "whisper"
    },
    {
      "start": 1042.54,
      "end": 1043.88,
      "text": "明白自身的目标",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06117100899036114,
      "no_speech_prob": 1.3886273207697197e-11,
      "compression_ratio": 1.3402366863905326,
      "source": "whisper"
    },
    {
      "start": 1043.88,
      "end": 1045.98,
      "text": "从很早的计算机发展阶段",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06117100899036114,
      "no_speech_prob": 1.3886273207697197e-11,
      "compression_ratio": 1.3402366863905326,
      "source": "whisper"
    },
    {
      "start": 1045.98,
      "end": 1048.42,
      "text": "工程师就被要求开发软件之前",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06117100899036114,
      "no_speech_prob": 1.3886273207697197e-11,
      "compression_ratio": 1.3402366863905326,
      "source": "whisper"
    },
    {
      "start": 1048.42,
      "end": 1049.88,
      "text": "梳理清楚产品规格",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06117100899036114,
      "no_speech_prob": 1.3886273207697197e-11,
      "compression_ratio": 1.3402366863905326,
      "source": "whisper"
    },
    {
      "start": 1049.88,
      "end": 1052.54,
      "text": "如今只是把执行工作交给Agent",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06117100899036114,
      "no_speech_prob": 1.3886273207697197e-11,
      "compression_ratio": 1.3402366863905326,
      "source": "whisper"
    },
    {
      "start": 1052.54,
      "end": 1055.14,
      "text": "明确规格这件事反而变得更加关键",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06117100899036114,
      "no_speech_prob": 1.3886273207697197e-11,
      "compression_ratio": 1.3402366863905326,
      "source": "whisper"
    },
    {
      "start": 1055.14,
      "end": 1057.16,
      "text": "人类工程师可以读懂语境",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06117100899036114,
      "no_speech_prob": 1.3886273207697197e-11,
      "compression_ratio": 1.3402366863905326,
      "source": "whisper"
    },
    {
      "start": 1057.16,
      "end": 1058.78,
      "text": "主动反问确认细节",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06117100899036114,
      "no_speech_prob": 1.3886273207697197e-11,
      "compression_ratio": 1.3402366863905326,
      "source": "whisper"
    },
    {
      "start": 1058.78,
      "end": 1061.08,
      "text": "Agent虽然部分场景可以追问",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06117100899036114,
      "no_speech_prob": 1.3886273207697197e-11,
      "compression_ratio": 1.3402366863905326,
      "source": "whisper"
    },
    {
      "start": 1061.08,
      "end": 1064.3,
      "text": "但整体还是高度依赖清晰完备的需求描述",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06117100899036114,
      "no_speech_prob": 1.3886273207697197e-11,
      "compression_ratio": 1.3402366863905326,
      "source": "whisper"
    },
    {
      "start": 1064.3,
      "end": 1066.68,
      "text": "现在代码迁移类的Coding Agent",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07115980121196729,
      "no_speech_prob": 1.2643877264628678e-11,
      "compression_ratio": 1.3072463768115943,
      "source": "whisper"
    },
    {
      "start": 1066.68,
      "end": 1067.7,
      "text": "效果出众",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07115980121196729,
      "no_speech_prob": 1.2643877264628678e-11,
      "compression_ratio": 1.3072463768115943,
      "source": "whisper"
    },
    {
      "start": 1067.7,
      "end": 1069.74,
      "text": "背后的原因也正在于此",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07115980121196729,
      "no_speech_prob": 1.2643877264628678e-11,
      "compression_ratio": 1.3072463768115943,
      "source": "whisper"
    },
    {
      "start": 1069.74,
      "end": 1073.02,
      "text": "把一套软件从一门语言迁移到另一门语言",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07115980121196729,
      "no_speech_prob": 1.2643877264628678e-11,
      "compression_ratio": 1.3072463768115943,
      "source": "whisper"
    },
    {
      "start": 1073.02,
      "end": 1075.42,
      "text": "原始代码本身就是完整的规格",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07115980121196729,
      "no_speech_prob": 1.2643877264628678e-11,
      "compression_ratio": 1.3072463768115943,
      "source": "whisper"
    },
    {
      "start": 1075.42,
      "end": 1077.64,
      "text": "原有测试用力可以直接复用",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07115980121196729,
      "no_speech_prob": 1.2643877264628678e-11,
      "compression_ratio": 1.3072463768115943,
      "source": "whisper"
    },
    {
      "start": 1077.64,
      "end": 1079.0,
      "text": "同时完成迁移",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07115980121196729,
      "no_speech_prob": 1.2643877264628678e-11,
      "compression_ratio": 1.3072463768115943,
      "source": "whisper"
    },
    {
      "start": 1079.0,
      "end": 1081.38,
      "text": "持续比对新旧版本行为差异",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07115980121196729,
      "no_speech_prob": 1.2643877264628678e-11,
      "compression_ratio": 1.3072463768115943,
      "source": "whisper"
    },
    {
      "start": 1081.38,
      "end": 1083.1,
      "text": "直到二者表现对齐",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07115980121196729,
      "no_speech_prob": 1.2643877264628678e-11,
      "compression_ratio": 1.3072463768115943,
      "source": "whisper"
    },
    {
      "start": 1083.1,
      "end": 1084.88,
      "text": "正是规格足够明确",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07115980121196729,
      "no_speech_prob": 1.2643877264628678e-11,
      "compression_ratio": 1.3072463768115943,
      "source": "whisper"
    },
    {
      "start": 1084.88,
      "end": 1087.1,
      "text": "才造就这类任务亮眼的表现",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07115980121196729,
      "no_speech_prob": 1.2643877264628678e-11,
      "compression_ratio": 1.3072463768115943,
      "source": "whisper"
    },
    {
      "start": 1087.1,
      "end": 1089.22,
      "text": "未来大规模调度数百个Agent",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07115980121196729,
      "no_speech_prob": 1.2643877264628678e-11,
      "compression_ratio": 1.3072463768115943,
      "source": "whisper"
    },
    {
      "start": 1089.22,
      "end": 1090.08,
      "text": "成为常态",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07115980121196729,
      "no_speech_prob": 1.2643877264628678e-11,
      "compression_ratio": 1.3072463768115943,
      "source": "whisper"
    },
    {
      "start": 1090.08,
      "end": 1091.88,
      "text": "大量代码由Agent生成",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07115980121196729,
      "no_speech_prob": 1.2643877264628678e-11,
      "compression_ratio": 1.3072463768115943,
      "source": "whisper"
    },
    {
      "start": 1091.88,
      "end": 1093.74,
      "text": "什么能力会变得最为稀缺",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07115980121196729,
      "no_speech_prob": 1.2643877264628678e-11,
      "compression_ratio": 1.3072463768115943,
      "source": "whisper"
    },
    {
      "start": 1093.74,
      "end": 1095.8,
      "text": "杰弗·迪恩给出的答案是品位",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08197415049483137,
      "no_speech_prob": 1.598936434643594e-11,
      "compression_ratio": 1.3147632311977715,
      "source": "whisper"
    },
    {
      "start": 1095.8,
      "end": 1099.44,
      "text": "也就是判断哪些问题值得交给agent去解决",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08197415049483137,
      "no_speech_prob": 1.598936434643594e-11,
      "compression_ratio": 1.3147632311977715,
      "source": "whisper"
    },
    {
      "start": 1099.44,
      "end": 1101.9,
      "text": "放到研究领域大部分的成败",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08197415049483137,
      "no_speech_prob": 1.598936434643594e-11,
      "compression_ratio": 1.3147632311977715,
      "source": "whisper"
    },
    {
      "start": 1101.9,
      "end": 1104.76,
      "text": "从选择研究课题的那一刻就已经注定",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08197415049483137,
      "no_speech_prob": 1.598936434643594e-11,
      "compression_ratio": 1.3147632311977715,
      "source": "whisper"
    },
    {
      "start": 1104.76,
      "end": 1106.1,
      "text": "手握全套工具",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08197415049483137,
      "no_speech_prob": 1.598936434643594e-11,
      "compression_ratio": 1.3147632311977715,
      "source": "whisper"
    },
    {
      "start": 1106.1,
      "end": 1108.12,
      "text": "不如选对值得投入的问题",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08197415049483137,
      "no_speech_prob": 1.598936434643594e-11,
      "compression_ratio": 1.3147632311977715,
      "source": "whisper"
    },
    {
      "start": 1108.12,
      "end": 1110.18,
      "text": "把普通问题打磨得再精致",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08197415049483137,
      "no_speech_prob": 1.598936434643594e-11,
      "compression_ratio": 1.3147632311977715,
      "source": "whisper"
    },
    {
      "start": 1110.18,
      "end": 1112.72,
      "text": "也比不上攻克一个高价值的难题",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08197415049483137,
      "no_speech_prob": 1.598936434643594e-11,
      "compression_ratio": 1.3147632311977715,
      "source": "whisper"
    },
    {
      "start": 1112.72,
      "end": 1115.54,
      "text": "但是品位本身没有标准化的考核指标",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08197415049483137,
      "no_speech_prob": 1.598936434643594e-11,
      "compression_ratio": 1.3147632311977715,
      "source": "whisper"
    },
    {
      "start": 1115.54,
      "end": 1116.88,
      "text": "应当如何训练",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08197415049483137,
      "no_speech_prob": 1.598936434643594e-11,
      "compression_ratio": 1.3147632311977715,
      "source": "whisper"
    },
    {
      "start": 1116.88,
      "end": 1119.12,
      "text": "一部分来源于大量项目经验",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08197415049483137,
      "no_speech_prob": 1.598936434643594e-11,
      "compression_ratio": 1.3147632311977715,
      "source": "whisper"
    },
    {
      "start": 1119.12,
      "end": 1121.48,
      "text": "经手足够多不同类型的问题",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08197415049483137,
      "no_speech_prob": 1.598936434643594e-11,
      "compression_ratio": 1.3147632311977715,
      "source": "whisper"
    },
    {
      "start": 1121.48,
      "end": 1123.62,
      "text": "慢慢积累对机会的判断力",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08197415049483137,
      "no_speech_prob": 1.598936434643594e-11,
      "compression_ratio": 1.3147632311977715,
      "source": "whisper"
    },
    {
      "start": 1123.62,
      "end": 1125.06,
      "text": "还有一个实操办法",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.09794780482416568,
      "no_speech_prob": 1.2644569419295593e-11,
      "compression_ratio": 1.3476923076923077,
      "source": "whisper"
    },
    {
      "start": 1125.06,
      "end": 1126.88,
      "text": "写下未来12个月之内",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.09794780482416568,
      "no_speech_prob": 1.2644569419295593e-11,
      "compression_ratio": 1.3476923076923077,
      "source": "whisper"
    },
    {
      "start": 1126.88,
      "end": 1129.26,
      "text": "自己预判会迎来爆发的方向",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.09794780482416568,
      "no_speech_prob": 1.2644569419295593e-11,
      "compression_ratio": 1.3476923076923077,
      "source": "whisper"
    },
    {
      "start": 1129.26,
      "end": 1132.0,
      "text": "后续只挑选其中一部分落地",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.09794780482416568,
      "no_speech_prob": 1.2644569419295593e-11,
      "compression_ratio": 1.3476923076923077,
      "source": "whisper"
    },
    {
      "start": 1132.0,
      "end": 1133.82,
      "text": "12个月之后回头复盘",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.09794780482416568,
      "no_speech_prob": 1.2644569419295593e-11,
      "compression_ratio": 1.3476923076923077,
      "source": "whisper"
    },
    {
      "start": 1133.82,
      "end": 1135.46,
      "text": "看看哪些预判成真",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.09794780482416568,
      "no_speech_prob": 1.2644569419295593e-11,
      "compression_ratio": 1.3476923076923077,
      "source": "whisper"
    },
    {
      "start": 1135.46,
      "end": 1137.14,
      "text": "哪些被其他人实现",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.09794780482416568,
      "no_speech_prob": 1.2644569419295593e-11,
      "compression_ratio": 1.3476923076923077,
      "source": "whisper"
    },
    {
      "start": 1137.14,
      "end": 1138.64,
      "text": "哪些完全没有发生",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.09794780482416568,
      "no_speech_prob": 1.2644569419295593e-11,
      "compression_ratio": 1.3476923076923077,
      "source": "whisper"
    },
    {
      "start": 1138.64,
      "end": 1140.96,
      "text": "相当于给自己制造大量样本",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.09794780482416568,
      "no_speech_prob": 1.2644569419295593e-11,
      "compression_ratio": 1.3476923076923077,
      "source": "whisper"
    },
    {
      "start": 1140.96,
      "end": 1142.66,
      "text": "反向训练自己的判断力",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.09794780482416568,
      "no_speech_prob": 1.2644569419295593e-11,
      "compression_ratio": 1.3476923076923077,
      "source": "whisper"
    },
    {
      "start": 1142.66,
      "end": 1145.72,
      "text": "第三种方式就是做疯狂的思想实验",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.09794780482416568,
      "no_speech_prob": 1.2644569419295593e-11,
      "compression_ratio": 1.3476923076923077,
      "source": "whisper"
    },
    {
      "start": 1145.72,
      "end": 1148.78,
      "text": "不要把行业默认的前提全部当成真理",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.09794780482416568,
      "no_speech_prob": 1.2644569419295593e-11,
      "compression_ratio": 1.3476923076923077,
      "source": "whisper"
    },
    {
      "start": 1148.78,
      "end": 1151.54,
      "text": "他分享了一段和同事开展的思想实验",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.09794780482416568,
      "no_speech_prob": 1.2644569419295593e-11,
      "compression_ratio": 1.3476923076923077,
      "source": "whisper"
    },
    {
      "start": 1151.54,
      "end": 1152.66,
      "text": "过去60年",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.09794780482416568,
      "no_speech_prob": 1.2644569419295593e-11,
      "compression_ratio": 1.3476923076923077,
      "source": "whisper"
    },
    {
      "start": 1152.66,
      "end": 1154.22,
      "text": "芯片设计制造行业",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0626948971218533,
      "no_speech_prob": 1.4337441824052721e-11,
      "compression_ratio": 1.331306990881459,
      "source": "whisper"
    },
    {
      "start": 1154.22,
      "end": 1156.42,
      "text": "都在拼命缩小晶体管尺寸",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0626948971218533,
      "no_speech_prob": 1.4337441824052721e-11,
      "compression_ratio": 1.331306990881459,
      "source": "whisper"
    },
    {
      "start": 1156.42,
      "end": 1157.88,
      "text": "压低出错概率",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0626948971218533,
      "no_speech_prob": 1.4337441824052721e-11,
      "compression_ratio": 1.331306990881459,
      "source": "whisper"
    },
    {
      "start": 1157.88,
      "end": 1159.9,
      "text": "默认目标就是同一款芯片",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0626948971218533,
      "no_speech_prob": 1.4337441824052721e-11,
      "compression_ratio": 1.331306990881459,
      "source": "whisper"
    },
    {
      "start": 1159.9,
      "end": 1162.04,
      "text": "每一颗成品都应当完全一致",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0626948971218533,
      "no_speech_prob": 1.4337441824052721e-11,
      "compression_ratio": 1.331306990881459,
      "source": "whisper"
    },
    {
      "start": 1162.04,
      "end": 1163.6,
      "text": "不能出现比特翻转",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0626948971218533,
      "no_speech_prob": 1.4337441824052721e-11,
      "compression_ratio": 1.331306990881459,
      "source": "whisper"
    },
    {
      "start": 1163.6,
      "end": 1164.94,
      "text": "但是宏观层面",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0626948971218533,
      "no_speech_prob": 1.4337441824052721e-11,
      "compression_ratio": 1.331306990881459,
      "source": "whisper"
    },
    {
      "start": 1164.94,
      "end": 1167.52,
      "text": "大规模分布式系统的逻辑恰恰相反",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0626948971218533,
      "no_speech_prob": 1.4337441824052721e-11,
      "compression_ratio": 1.331306990881459,
      "source": "whisper"
    },
    {
      "start": 1167.52,
      "end": 1169.44,
      "text": "依靠不可靠的硬件组件",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0626948971218533,
      "no_speech_prob": 1.4337441824052721e-11,
      "compression_ratio": 1.331306990881459,
      "source": "whisper"
    },
    {
      "start": 1169.44,
      "end": 1171.04,
      "text": "搭建高可靠系统",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0626948971218533,
      "no_speech_prob": 1.4337441824052721e-11,
      "compression_ratio": 1.331306990881459,
      "source": "whisper"
    },
    {
      "start": 1171.04,
      "end": 1172.18,
      "text": "磁盘会损坏",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0626948971218533,
      "no_speech_prob": 1.4337441824052721e-11,
      "compression_ratio": 1.331306990881459,
      "source": "whisper"
    },
    {
      "start": 1172.18,
      "end": 1173.88,
      "text": "但是数据不能丢失",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0626948971218533,
      "no_speech_prob": 1.4337441824052721e-11,
      "compression_ratio": 1.331306990881459,
      "source": "whisper"
    },
    {
      "start": 1173.88,
      "end": 1175.42,
      "text": "于是做多副本",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0626948971218533,
      "no_speech_prob": 1.4337441824052721e-11,
      "compression_ratio": 1.331306990881459,
      "source": "whisper"
    },
    {
      "start": 1175.42,
      "end": 1176.68,
      "text": "使用纠缠码",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0626948971218533,
      "no_speech_prob": 1.4337441824052721e-11,
      "compression_ratio": 1.331306990881459,
      "source": "whisper"
    },
    {
      "start": 1176.68,
      "end": 1178.24,
      "text": "可是这套容错思路",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0626948971218533,
      "no_speech_prob": 1.4337441824052721e-11,
      "compression_ratio": 1.331306990881459,
      "source": "whisper"
    },
    {
      "start": 1178.24,
      "end": 1180.34,
      "text": "并没有下放到晶体管这一层",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0626948971218533,
      "no_speech_prob": 1.4337441824052721e-11,
      "compression_ratio": 1.331306990881459,
      "source": "whisper"
    },
    {
      "start": 1180.34,
      "end": 1181.9,
      "text": "如果设计一套芯片",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.0626948971218533,
      "no_speech_prob": 1.4337441824052721e-11,
      "compression_ratio": 1.331306990881459,
      "source": "whisper"
    },
    {
      "start": 1181.9,
      "end": 1184.58,
      "text": "晶体管每天会出现20次错误",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07579784297463882,
      "no_speech_prob": 1.3055457756538935e-11,
      "compression_ratio": 1.3622291021671826,
      "source": "whisper"
    },
    {
      "start": 1184.58,
      "end": 1186.92,
      "text": "而不是几百万年才出现一次故障",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07579784297463882,
      "no_speech_prob": 1.3055457756538935e-11,
      "compression_ratio": 1.3622291021671826,
      "source": "whisper"
    },
    {
      "start": 1186.92,
      "end": 1188.94,
      "text": "整套系统会走向何方",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07579784297463882,
      "no_speech_prob": 1.3055457756538935e-11,
      "compression_ratio": 1.3622291021671826,
      "source": "whisper"
    },
    {
      "start": 1188.94,
      "end": 1191.68,
      "text": "制造工艺或许会迎来全新可能性",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07579784297463882,
      "no_speech_prob": 1.3055457756538935e-11,
      "compression_ratio": 1.3622291021671826,
      "source": "whisper"
    },
    {
      "start": 1191.68,
      "end": 1193.98,
      "text": "信号传输采用多条冗余通路",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07579784297463882,
      "no_speech_prob": 1.3055457756538935e-11,
      "compression_ratio": 1.3622291021671826,
      "source": "whisper"
    },
    {
      "start": 1193.98,
      "end": 1196.28,
      "text": "保障至少一路信号可以送达",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07579784297463882,
      "no_speech_prob": 1.3055457756538935e-11,
      "compression_ratio": 1.3622291021671826,
      "source": "whisper"
    },
    {
      "start": 1196.28,
      "end": 1199.5,
      "text": "这套逻辑和人脑神经信号传递十分相似",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07579784297463882,
      "no_speech_prob": 1.3055457756538935e-11,
      "compression_ratio": 1.3622291021671826,
      "source": "whisper"
    },
    {
      "start": 1199.5,
      "end": 1202.22,
      "text": "大脑信号传输本身就不是绝对可靠",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07579784297463882,
      "no_speech_prob": 1.3055457756538935e-11,
      "compression_ratio": 1.3622291021671826,
      "source": "whisper"
    },
    {
      "start": 1202.22,
      "end": 1205.22,
      "text": "重要信息就依靠多条通路同时传递",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07579784297463882,
      "no_speech_prob": 1.3055457756538935e-11,
      "compression_ratio": 1.3622291021671826,
      "source": "whisper"
    },
    {
      "start": 1205.22,
      "end": 1208.02,
      "text": "它并不是说当下就要落地这套方案",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07579784297463882,
      "no_speech_prob": 1.3055457756538935e-11,
      "compression_ratio": 1.3622291021671826,
      "source": "whisper"
    },
    {
      "start": 1208.02,
      "end": 1210.3,
      "text": "但是这类思想实验意义重大",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07579784297463882,
      "no_speech_prob": 1.3055457756538935e-11,
      "compression_ratio": 1.3622291021671826,
      "source": "whisper"
    },
    {
      "start": 1210.3,
      "end": 1213.5,
      "text": "要定期重新审视所有人习以为常的预设",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06558776472173809,
      "no_speech_prob": 1.2555237231814953e-11,
      "compression_ratio": 1.2387640449438202,
      "source": "whisper"
    },
    {
      "start": 1213.5,
      "end": 1215.12,
      "text": "回顾他的职业生涯",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06558776472173809,
      "no_speech_prob": 1.2555237231814953e-11,
      "compression_ratio": 1.2387640449438202,
      "source": "whisper"
    },
    {
      "start": 1215.12,
      "end": 1219.1,
      "text": "TPU的诞生就是推翻固有假设之后接触的果实",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06558776472173809,
      "no_speech_prob": 1.2555237231814953e-11,
      "compression_ratio": 1.2387640449438202,
      "source": "whisper"
    },
    {
      "start": 1219.1,
      "end": 1221.4,
      "text": "在问题还没有成为主流的时候",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06558776472173809,
      "no_speech_prob": 1.2555237231814953e-11,
      "compression_ratio": 1.2387640449438202,
      "source": "whisper"
    },
    {
      "start": 1221.4,
      "end": 1223.72,
      "text": "思考小众场景的专用硬件",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06558776472173809,
      "no_speech_prob": 1.2555237231814953e-11,
      "compression_ratio": 1.2387640449438202,
      "source": "whisper"
    },
    {
      "start": 1223.72,
      "end": 1225.42,
      "text": "MapReduce也是同理",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06558776472173809,
      "no_speech_prob": 1.2555237231814953e-11,
      "compression_ratio": 1.2387640449438202,
      "source": "whisper"
    },
    {
      "start": 1225.42,
      "end": 1228.0,
      "text": "团队厌倦了手工编写大量并行",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06558776472173809,
      "no_speech_prob": 1.2555237231814953e-11,
      "compression_ratio": 1.2387640449438202,
      "source": "whisper"
    },
    {
      "start": 1228.0,
      "end": 1229.84,
      "text": "容错相关的重复代码",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06558776472173809,
      "no_speech_prob": 1.2555237231814953e-11,
      "compression_ratio": 1.2387640449438202,
      "source": "whisper"
    },
    {
      "start": 1229.84,
      "end": 1231.92,
      "text": "于是提出全新抽象框架",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06558776472173809,
      "no_speech_prob": 1.2555237231814953e-11,
      "compression_ratio": 1.2387640449438202,
      "source": "whisper"
    },
    {
      "start": 1231.92,
      "end": 1234.34,
      "text": "上层只需要编写业务逻辑",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06558776472173809,
      "no_speech_prob": 1.2555237231814953e-11,
      "compression_ratio": 1.2387640449438202,
      "source": "whisper"
    },
    {
      "start": 1234.34,
      "end": 1236.92,
      "text": "底层统一处理并行调度和容错",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06558776472173809,
      "no_speech_prob": 1.2555237231814953e-11,
      "compression_ratio": 1.2387640449438202,
      "source": "whisper"
    },
    {
      "start": 1236.92,
      "end": 1240.04,
      "text": "最后成为谷歌大规模计算的核心底座",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.06558776472173809,
      "no_speech_prob": 1.2555237231814953e-11,
      "compression_ratio": 1.2387640449438202,
      "source": "whisper"
    },
    {
      "start": 1240.04,
      "end": 1241.64,
      "text": "而这一切的源头",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07731399904702596,
      "no_speech_prob": 1.179024586295263e-11,
      "compression_ratio": 1.2462006079027355,
      "source": "whisper"
    },
    {
      "start": 1241.64,
      "end": 1243.34,
      "text": "同样来自思想实验",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07731399904702596,
      "no_speech_prob": 1.179024586295263e-11,
      "compression_ratio": 1.2462006079027355,
      "source": "whisper"
    },
    {
      "start": 1243.34,
      "end": 1245.7,
      "text": "现在有Alpha Chip辅助芯片布局",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07731399904702596,
      "no_speech_prob": 1.179024586295263e-11,
      "compression_ratio": 1.2462006079027355,
      "source": "whisper"
    },
    {
      "start": 1245.7,
      "end": 1248.56,
      "text": "有Alpha UVO完成方案生成评估留存",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07731399904702596,
      "no_speech_prob": 1.179024586295263e-11,
      "compression_ratio": 1.2462006079027355,
      "source": "whisper"
    },
    {
      "start": 1248.56,
      "end": 1250.62,
      "text": "整套AI制造AI的体系",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07731399904702596,
      "no_speech_prob": 1.179024586295263e-11,
      "compression_ratio": 1.2462006079027355,
      "source": "whisper"
    },
    {
      "start": 1250.62,
      "end": 1251.9,
      "text": "正在逐步成型",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07731399904702596,
      "no_speech_prob": 1.179024586295263e-11,
      "compression_ratio": 1.2462006079027355,
      "source": "whisper"
    },
    {
      "start": 1251.9,
      "end": 1252.88,
      "text": "广义来讲",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07731399904702596,
      "no_speech_prob": 1.179024586295263e-11,
      "compression_ratio": 1.2462006079027355,
      "source": "whisper"
    },
    {
      "start": 1252.88,
      "end": 1255.32,
      "text": "这本质就是科学方法的自动化",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07731399904702596,
      "no_speech_prob": 1.179024586295263e-11,
      "compression_ratio": 1.2462006079027355,
      "source": "whisper"
    },
    {
      "start": 1255.32,
      "end": 1256.38,
      "text": "提出假设",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07731399904702596,
      "no_speech_prob": 1.179024586295263e-11,
      "compression_ratio": 1.2462006079027355,
      "source": "whisper"
    },
    {
      "start": 1256.38,
      "end": 1257.44,
      "text": "搭建环境",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07731399904702596,
      "no_speech_prob": 1.179024586295263e-11,
      "compression_ratio": 1.2462006079027355,
      "source": "whisper"
    },
    {
      "start": 1257.44,
      "end": 1258.36,
      "text": "执行实验",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07731399904702596,
      "no_speech_prob": 1.179024586295263e-11,
      "compression_ratio": 1.2462006079027355,
      "source": "whisper"
    },
    {
      "start": 1258.36,
      "end": 1259.38,
      "text": "评估结果",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07731399904702596,
      "no_speech_prob": 1.179024586295263e-11,
      "compression_ratio": 1.2462006079027355,
      "source": "whisper"
    },
    {
      "start": 1259.38,
      "end": 1261.04,
      "text": "循环往复推进迭代",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07731399904702596,
      "no_speech_prob": 1.179024586295263e-11,
      "compression_ratio": 1.2462006079027355,
      "source": "whisper"
    },
    {
      "start": 1261.04,
      "end": 1262.56,
      "text": "如今越来越多领域",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07731399904702596,
      "no_speech_prob": 1.179024586295263e-11,
      "compression_ratio": 1.2462006079027355,
      "source": "whisper"
    },
    {
      "start": 1262.56,
      "end": 1264.2,
      "text": "可以把这套闭环自动化",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07731399904702596,
      "no_speech_prob": 1.179024586295263e-11,
      "compression_ratio": 1.2462006079027355,
      "source": "whisper"
    },
    {
      "start": 1264.2,
      "end": 1266.1,
      "text": "一次性执行海量实验",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07731399904702596,
      "no_speech_prob": 1.179024586295263e-11,
      "compression_ratio": 1.2462006079027355,
      "source": "whisper"
    },
    {
      "start": 1266.1,
      "end": 1267.52,
      "text": "压缩循环延迟",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07731399904702596,
      "no_speech_prob": 1.179024586295263e-11,
      "compression_ratio": 1.2462006079027355,
      "source": "whisper"
    },
    {
      "start": 1267.52,
      "end": 1268.78,
      "text": "就变得格外关键",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07731399904702596,
      "no_speech_prob": 1.179024586295263e-11,
      "compression_ratio": 1.2462006079027355,
      "source": "whisper"
    },
    {
      "start": 1268.78,
      "end": 1271.62,
      "text": "前提是评估器本身要足够快速",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08244279925332686,
      "no_speech_prob": 1.2699332904708704e-11,
      "compression_ratio": 1.2900302114803626,
      "source": "whisper"
    },
    {
      "start": 1271.62,
      "end": 1273.62,
      "text": "他举了一个十年之前的案例",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08244279925332686,
      "no_speech_prob": 1.2699332904708704e-11,
      "compression_ratio": 1.2900302114803626,
      "source": "whisper"
    },
    {
      "start": 1273.62,
      "end": 1275.3,
      "text": "同事研究量子化学",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08244279925332686,
      "no_speech_prob": 1.2699332904708704e-11,
      "compression_ratio": 1.2900302114803626,
      "source": "whisper"
    },
    {
      "start": 1275.3,
      "end": 1276.62,
      "text": "解析分子属性",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08244279925332686,
      "no_speech_prob": 1.2699332904708704e-11,
      "compression_ratio": 1.2900302114803626,
      "source": "whisper"
    },
    {
      "start": 1276.62,
      "end": 1278.48,
      "text": "传统密度泛含理论模拟器",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08244279925332686,
      "no_speech_prob": 1.2699332904708704e-11,
      "compression_ratio": 1.2900302114803626,
      "source": "whisper"
    },
    {
      "start": 1278.48,
      "end": 1280.06,
      "text": "跑一次就要耗费一整夜",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08244279925332686,
      "no_speech_prob": 1.2699332904708704e-11,
      "compression_ratio": 1.2900302114803626,
      "source": "whisper"
    },
    {
      "start": 1280.06,
      "end": 1282.54,
      "text": "团队收集大量输入输出样本",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08244279925332686,
      "no_speech_prob": 1.2699332904708704e-11,
      "compression_ratio": 1.2900302114803626,
      "source": "whisper"
    },
    {
      "start": 1282.54,
      "end": 1284.76,
      "text": "训练神经网络近似模拟器",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08244279925332686,
      "no_speech_prob": 1.2699332904708704e-11,
      "compression_ratio": 1.2900302114803626,
      "source": "whisper"
    },
    {
      "start": 1284.76,
      "end": 1286.5,
      "text": "速度提升30万倍",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08244279925332686,
      "no_speech_prob": 1.2699332904708704e-11,
      "compression_ratio": 1.2900302114803626,
      "source": "whisper"
    },
    {
      "start": 1286.5,
      "end": 1288.48,
      "text": "精度却几乎没有损失",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08244279925332686,
      "no_speech_prob": 1.2699332904708704e-11,
      "compression_ratio": 1.2900302114803626,
      "source": "whisper"
    },
    {
      "start": 1288.48,
      "end": 1290.22,
      "text": "彻底改写科研模式",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08244279925332686,
      "no_speech_prob": 1.2699332904708704e-11,
      "compression_ratio": 1.2900302114803626,
      "source": "whisper"
    },
    {
      "start": 1290.22,
      "end": 1292.32,
      "text": "从前筛选1000万候选分子",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08244279925332686,
      "no_speech_prob": 1.2699332904708704e-11,
      "compression_ratio": 1.2900302114803626,
      "source": "whisper"
    },
    {
      "start": 1292.32,
      "end": 1294.04,
      "text": "要耗费6个月筹备算力",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08244279925332686,
      "no_speech_prob": 1.2699332904708704e-11,
      "compression_ratio": 1.2900302114803626,
      "source": "whisper"
    },
    {
      "start": 1294.04,
      "end": 1295.58,
      "text": "现在一顿午饭的时间",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08244279925332686,
      "no_speech_prob": 1.2699332904708704e-11,
      "compression_ratio": 1.2900302114803626,
      "source": "whisper"
    },
    {
      "start": 1295.58,
      "end": 1296.76,
      "text": "就可以完成筛选",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.08244279925332686,
      "no_speech_prob": 1.2699332904708704e-11,
      "compression_ratio": 1.2900302114803626,
      "source": "whisper"
    },
    {
      "start": 1296.76,
      "end": 1299.78,
      "text": "大量行业还有机会打造高速仿真模型",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07741910601974628,
      "no_speech_prob": 1.5084137064413916e-11,
      "compression_ratio": 1.2988165680473374,
      "source": "whisper"
    },
    {
      "start": 1299.78,
      "end": 1300.72,
      "text": "一旦实现",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07741910601974628,
      "no_speech_prob": 1.5084137064413916e-11,
      "compression_ratio": 1.2988165680473374,
      "source": "whisper"
    },
    {
      "start": 1300.72,
      "end": 1303.62,
      "text": "整个实验迭代周期就会被彻底重构",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07741910601974628,
      "no_speech_prob": 1.5084137064413916e-11,
      "compression_ratio": 1.2988165680473374,
      "source": "whisper"
    },
    {
      "start": 1303.62,
      "end": 1305.14,
      "text": "它最看好的方向",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07741910601974628,
      "no_speech_prob": 1.5084137064413916e-11,
      "compression_ratio": 1.2988165680473374,
      "source": "whisper"
    },
    {
      "start": 1305.14,
      "end": 1307.08,
      "text": "就是模型依靠大量实验",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07741910601974628,
      "no_speech_prob": 1.5084137064413916e-11,
      "compression_ratio": 1.2988165680473374,
      "source": "whisper"
    },
    {
      "start": 1307.08,
      "end": 1308.9,
      "text": "实现地规式自我改进",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07741910601974628,
      "no_speech_prob": 1.5084137064413916e-11,
      "compression_ratio": 1.2988165680473374,
      "source": "whisper"
    },
    {
      "start": 1308.9,
      "end": 1310.88,
      "text": "持续提升每一份算力",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07741910601974628,
      "no_speech_prob": 1.5084137064413916e-11,
      "compression_ratio": 1.2988165680473374,
      "source": "whisper"
    },
    {
      "start": 1310.88,
      "end": 1312.3,
      "text": "可以产出的新发现",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07741910601974628,
      "no_speech_prob": 1.5084137064413916e-11,
      "compression_ratio": 1.2988165680473374,
      "source": "whisper"
    },
    {
      "start": 1312.3,
      "end": 1313.9,
      "text": "对话推进到后半段",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07741910601974628,
      "no_speech_prob": 1.5084137064413916e-11,
      "compression_ratio": 1.2988165680473374,
      "source": "whisper"
    },
    {
      "start": 1313.9,
      "end": 1315.46,
      "text": "谈及职业路上的挫折",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07741910601974628,
      "no_speech_prob": 1.5084137064413916e-11,
      "compression_ratio": 1.2988165680473374,
      "source": "whisper"
    },
    {
      "start": 1315.46,
      "end": 1316.9,
      "text": "2014年杰弗蒂恩",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07741910601974628,
      "no_speech_prob": 1.5084137064413916e-11,
      "compression_ratio": 1.2988165680473374,
      "source": "whisper"
    },
    {
      "start": 1316.9,
      "end": 1317.96,
      "text": "杰弗里辛顿",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07741910601974628,
      "no_speech_prob": 1.5084137064413916e-11,
      "compression_ratio": 1.2988165680473374,
      "source": "whisper"
    },
    {
      "start": 1317.96,
      "end": 1319.4,
      "text": "奥里奥尔维尼亚尔斯",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07741910601974628,
      "no_speech_prob": 1.5084137064413916e-11,
      "compression_ratio": 1.2988165680473374,
      "source": "whisper"
    },
    {
      "start": 1319.4,
      "end": 1321.4,
      "text": "联合发布知识蒸馏论文",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07741910601974628,
      "no_speech_prob": 1.5084137064413916e-11,
      "compression_ratio": 1.2988165680473374,
      "source": "whisper"
    },
    {
      "start": 1321.4,
      "end": 1323.34,
      "text": "用大模型充当老师",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07741910601974628,
      "no_speech_prob": 1.5084137064413916e-11,
      "compression_ratio": 1.2988165680473374,
      "source": "whisper"
    },
    {
      "start": 1323.34,
      "end": 1324.86,
      "text": "小模型充当学生",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07741910601974628,
      "no_speech_prob": 1.5084137064413916e-11,
      "compression_ratio": 1.2988165680473374,
      "source": "whisper"
    },
    {
      "start": 1324.86,
      "end": 1326.12,
      "text": "得到体积更小",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07741910601974628,
      "no_speech_prob": 1.5084137064413916e-11,
      "compression_ratio": 1.2988165680473374,
      "source": "whisper"
    },
    {
      "start": 1326.12,
      "end": 1327.0,
      "text": "成本更低",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07964169762351296,
      "no_speech_prob": 1.5689442800503883e-11,
      "compression_ratio": 1.2612359550561798,
      "source": "whisper"
    },
    {
      "start": 1327.0,
      "end": 1328.6,
      "text": "同时效果优异的模型",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07964169762351296,
      "no_speech_prob": 1.5689442800503883e-11,
      "compression_ratio": 1.2612359550561798,
      "source": "whisper"
    },
    {
      "start": 1328.6,
      "end": 1331.0,
      "text": "如今整个行业都在广泛使用",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07964169762351296,
      "no_speech_prob": 1.5689442800503883e-11,
      "compression_ratio": 1.2612359550561798,
      "source": "whisper"
    },
    {
      "start": 1331.0,
      "end": 1333.7,
      "text": "但是这篇论文当年被NIPS聚稿",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07964169762351296,
      "no_speech_prob": 1.5689442800503883e-11,
      "compression_ratio": 1.2612359550561798,
      "source": "whisper"
    },
    {
      "start": 1333.7,
      "end": 1335.0,
      "text": "评审的观点认为",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07964169762351296,
      "no_speech_prob": 1.5689442800503883e-11,
      "compression_ratio": 1.2612359550561798,
      "source": "whisper"
    },
    {
      "start": 1335.0,
      "end": 1336.9,
      "text": "这项工作很难带来重大突破",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07964169762351296,
      "no_speech_prob": 1.5689442800503883e-11,
      "compression_ratio": 1.2612359550561798,
      "source": "whisper"
    },
    {
      "start": 1336.9,
      "end": 1338.06,
      "text": "杰弗迪恩表示",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07964169762351296,
      "no_speech_prob": 1.5689442800503883e-11,
      "compression_ratio": 1.2612359550561798,
      "source": "whisper"
    },
    {
      "start": 1338.06,
      "end": 1339.66,
      "text": "并不责怪评审委员会",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07964169762351296,
      "no_speech_prob": 1.5689442800503883e-11,
      "compression_ratio": 1.2612359550561798,
      "source": "whisper"
    },
    {
      "start": 1339.66,
      "end": 1341.32,
      "text": "对方站在学术视角",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07964169762351296,
      "no_speech_prob": 1.5689442800503883e-11,
      "compression_ratio": 1.2612359550561798,
      "source": "whisper"
    },
    {
      "start": 1341.32,
      "end": 1344.1,
      "text": "评判是否属于理论层面的重大突破",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07964169762351296,
      "no_speech_prob": 1.5689442800503883e-11,
      "compression_ratio": 1.2612359550561798,
      "source": "whisper"
    },
    {
      "start": 1344.1,
      "end": 1346.98,
      "text": "而团队清楚这件事巨大的工程价值",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07964169762351296,
      "no_speech_prob": 1.5689442800503883e-11,
      "compression_ratio": 1.2612359550561798,
      "source": "whisper"
    },
    {
      "start": 1346.98,
      "end": 1349.28,
      "text": "目标是降低模型服务成本",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07964169762351296,
      "no_speech_prob": 1.5689442800503883e-11,
      "compression_ratio": 1.2612359550561798,
      "source": "whisper"
    },
    {
      "start": 1349.28,
      "end": 1350.68,
      "text": "扩大落地场景",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07964169762351296,
      "no_speech_prob": 1.5689442800503883e-11,
      "compression_ratio": 1.2612359550561798,
      "source": "whisper"
    },
    {
      "start": 1350.68,
      "end": 1353.8,
      "text": "他们把论文上传到ARXV预印本平台",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07964169762351296,
      "no_speech_prob": 1.5689442800503883e-11,
      "compression_ratio": 1.2612359550561798,
      "source": "whisper"
    },
    {
      "start": 1353.8,
      "end": 1355.58,
      "text": "行业从业者阅读之后",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.07964169762351296,
      "no_speech_prob": 1.5689442800503883e-11,
      "compression_ratio": 1.2612359550561798,
      "source": "whisper"
    },
    {
      "start": 1355.58,
      "end": 1356.64,
      "text": "大量落地使用",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.1036291075457493,
      "no_speech_prob": 3.327954925524956e-11,
      "compression_ratio": 1.2118644067796611,
      "source": "whisper"
    },
    {
      "start": 1356.64,
      "end": 1358.8,
      "text": "Gemini当中的Flash系列模型",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.1036291075457493,
      "no_speech_prob": 3.327954925524956e-11,
      "compression_ratio": 1.2118644067796611,
      "source": "whisper"
    },
    {
      "start": 1358.8,
      "end": 1361.92,
      "text": "就有一部分来自大尺寸Pro模型的蒸馏",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.1036291075457493,
      "no_speech_prob": 3.327954925524956e-11,
      "compression_ratio": 1.2118644067796611,
      "source": "whisper"
    },
    {
      "start": 1361.92,
      "end": 1363.66,
      "text": "在同规格模型当中",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.1036291075457493,
      "no_speech_prob": 3.327954925524956e-11,
      "compression_ratio": 1.2118644067796611,
      "source": "whisper"
    },
    {
      "start": 1363.66,
      "end": 1365.52,
      "text": "基准测试表现十分靠前",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.1036291075457493,
      "no_speech_prob": 3.327954925524956e-11,
      "compression_ratio": 1.2118644067796611,
      "source": "whisper"
    },
    {
      "start": 1365.52,
      "end": 1367.36,
      "text": "这件事带来的启示就是",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.1036291075457493,
      "no_speech_prob": 3.327954925524956e-11,
      "compression_ratio": 1.2118644067796611,
      "source": "whisper"
    },
    {
      "start": 1367.36,
      "end": 1369.04,
      "text": "就算成果遭到拒绝",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.1036291075457493,
      "no_speech_prob": 3.327954925524956e-11,
      "compression_ratio": 1.2118644067796611,
      "source": "whisper"
    },
    {
      "start": 1369.04,
      "end": 1370.28,
      "text": "也不要停下脚步",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.1036291075457493,
      "no_speech_prob": 3.327954925524956e-11,
      "compression_ratio": 1.2118644067796611,
      "source": "whisper"
    },
    {
      "start": 1370.28,
      "end": 1371.82,
      "text": "如果把1999年",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.1036291075457493,
      "no_speech_prob": 3.327954925524956e-11,
      "compression_ratio": 1.2118644067796611,
      "source": "whisper"
    },
    {
      "start": 1371.82,
      "end": 1375.14,
      "text": "刚刚加入只有20人规模谷歌的年轻杰弗迪恩",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.1036291075457493,
      "no_speech_prob": 3.327954925524956e-11,
      "compression_ratio": 1.2118644067796611,
      "source": "whisper"
    },
    {
      "start": 1375.14,
      "end": 1377.08,
      "text": "直接传送到2026年",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.1036291075457493,
      "no_speech_prob": 3.327954925524956e-11,
      "compression_ratio": 1.2118644067796611,
      "source": "whisper"
    },
    {
      "start": 1377.08,
      "end": 1379.34,
      "text": "他会选择大厂前沿实验室",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.1036291075457493,
      "no_speech_prob": 3.327954925524956e-11,
      "compression_ratio": 1.2118644067796611,
      "source": "whisper"
    },
    {
      "start": 1379.34,
      "end": 1381.02,
      "text": "还是选择自主创业",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.1036291075457493,
      "no_speech_prob": 3.327954925524956e-11,
      "compression_ratio": 1.2118644067796611,
      "source": "whisper"
    },
    {
      "start": 1381.02,
      "end": 1382.84,
      "text": "他表示没有标准答案",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.1036291075457493,
      "no_speech_prob": 3.327954925524956e-11,
      "compression_ratio": 1.2118644067796611,
      "source": "whisper"
    },
    {
      "start": 1382.84,
      "end": 1384.236,
      "text": "最核心的判断表现",
      "chunk": 2,
      "language": "zh",
      "avg_logprob": -0.1036291075457493,
      "no_speech_prob": 3.327954925524956e-11,
      "compression_ratio": 1.2118644067796611,
      "source": "whisper"
    },
    {
      "start": 1397.036,
      "end": 1398.616,
      "text": "创业公司风险更高",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06233117715367731,
      "no_speech_prob": 1.5316733992243314e-11,
      "compression_ratio": 1.3054755043227666,
      "source": "whisper"
    },
    {
      "start": 1398.616,
      "end": 1400.196,
      "text": "必须拥有极强的热情",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06233117715367731,
      "no_speech_prob": 1.5316733992243314e-11,
      "compression_ratio": 1.3054755043227666,
      "source": "whisper"
    },
    {
      "start": 1400.196,
      "end": 1402.036,
      "text": "但也可以收获极高的回报",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06233117715367731,
      "no_speech_prob": 1.5316733992243314e-11,
      "compression_ratio": 1.3054755043227666,
      "source": "whisper"
    },
    {
      "start": 1402.036,
      "end": 1403.776,
      "text": "无论选择哪一条道路",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06233117715367731,
      "no_speech_prob": 1.5316733992243314e-11,
      "compression_ratio": 1.3054755043227666,
      "source": "whisper"
    },
    {
      "start": 1403.776,
      "end": 1404.976,
      "text": "都要反问自己",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06233117715367731,
      "no_speech_prob": 1.5316733992243314e-11,
      "compression_ratio": 1.3054755043227666,
      "source": "whisper"
    },
    {
      "start": 1404.976,
      "end": 1407.016,
      "text": "如果项目走到最好的结局",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06233117715367731,
      "no_speech_prob": 1.5316733992243314e-11,
      "compression_ratio": 1.3054755043227666,
      "source": "whisper"
    },
    {
      "start": 1407.016,
      "end": 1408.876,
      "text": "世界会不会切实变好",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06233117715367731,
      "no_speech_prob": 1.5316733992243314e-11,
      "compression_ratio": 1.3054755043227666,
      "source": "whisper"
    },
    {
      "start": 1408.876,
      "end": 1410.436,
      "text": "如果仅仅只是很酷",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06233117715367731,
      "no_speech_prob": 1.5316733992243314e-11,
      "compression_ratio": 1.3054755043227666,
      "source": "whisper"
    },
    {
      "start": 1410.436,
      "end": 1412.396,
      "text": "就不值得投入大量时间",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06233117715367731,
      "no_speech_prob": 1.5316733992243314e-11,
      "compression_ratio": 1.3054755043227666,
      "source": "whisper"
    },
    {
      "start": 1412.396,
      "end": 1414.836,
      "text": "谈及如何寻找优秀的合作者",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06233117715367731,
      "no_speech_prob": 1.5316733992243314e-11,
      "compression_ratio": 1.3054755043227666,
      "source": "whisper"
    },
    {
      "start": 1414.836,
      "end": 1416.876,
      "text": "不仅要寻找专业能力过硬",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06233117715367731,
      "no_speech_prob": 1.5316733992243314e-11,
      "compression_ratio": 1.3054755043227666,
      "source": "whisper"
    },
    {
      "start": 1416.876,
      "end": 1418.196,
      "text": "技能互补的伙伴",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06233117715367731,
      "no_speech_prob": 1.5316733992243314e-11,
      "compression_ratio": 1.3054755043227666,
      "source": "whisper"
    },
    {
      "start": 1418.196,
      "end": 1420.896,
      "text": "也要选择相处起来足够愉快的人",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06233117715367731,
      "no_speech_prob": 1.5316733992243314e-11,
      "compression_ratio": 1.3054755043227666,
      "source": "whisper"
    },
    {
      "start": 1420.896,
      "end": 1423.836,
      "text": "创业路上要共同面对大量棘手难题",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06233117715367731,
      "no_speech_prob": 1.5316733992243314e-11,
      "compression_ratio": 1.3054755043227666,
      "source": "whisper"
    },
    {
      "start": 1423.836,
      "end": 1425.796,
      "text": "彼此的相处体验格外重要",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06233117715367731,
      "no_speech_prob": 1.5316733992243314e-11,
      "compression_ratio": 1.3054755043227666,
      "source": "whisper"
    },
    {
      "start": 1425.796,
      "end": 1429.816,
      "text": "小团队的魅力就是每个人掌握其他人不具备的能力",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0788673568017704,
      "no_speech_prob": 1.2649914102325077e-11,
      "compression_ratio": 1.3832335329341316,
      "source": "whisper"
    },
    {
      "start": 1429.816,
      "end": 1433.016,
      "text": "组合在一起可以完成单人无法实现的成果",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0788673568017704,
      "no_speech_prob": 1.2649914102325077e-11,
      "compression_ratio": 1.3832335329341316,
      "source": "whisper"
    },
    {
      "start": 1433.016,
      "end": 1436.236,
      "text": "要把自己的工程或者科研生涯当成一条工具腰带",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0788673568017704,
      "no_speech_prob": 1.2649914102325077e-11,
      "compression_ratio": 1.3832335329341316,
      "source": "whisper"
    },
    {
      "start": 1436.236,
      "end": 1437.956,
      "text": "持续补充新的工具",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0788673568017704,
      "no_speech_prob": 1.2649914102325077e-11,
      "compression_ratio": 1.3832335329341316,
      "source": "whisper"
    },
    {
      "start": 1437.956,
      "end": 1441.016,
      "text": "你永远预料不到未来会碰到什么样的难题",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0788673568017704,
      "no_speech_prob": 1.2649914102325077e-11,
      "compression_ratio": 1.3832335329341316,
      "source": "whisper"
    },
    {
      "start": 1441.016,
      "end": 1442.436,
      "text": "掌握的工具越多",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0788673568017704,
      "no_speech_prob": 1.2649914102325077e-11,
      "compression_ratio": 1.3832335329341316,
      "source": "whisper"
    },
    {
      "start": 1442.436,
      "end": 1444.816,
      "text": "可以解决的问题边界就越宽广",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0788673568017704,
      "no_speech_prob": 1.2649914102325077e-11,
      "compression_ratio": 1.3832335329341316,
      "source": "whisper"
    },
    {
      "start": 1444.816,
      "end": 1446.736,
      "text": "主持人抛出最后一个问题",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0788673568017704,
      "no_speech_prob": 1.2649914102325077e-11,
      "compression_ratio": 1.3832335329341316,
      "source": "whisper"
    },
    {
      "start": 1446.736,
      "end": 1450.436,
      "text": "他希望新一代创作者把精力投入到哪些方向",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0788673568017704,
      "no_speech_prob": 1.2649914102325077e-11,
      "compression_ratio": 1.3832335329341316,
      "source": "whisper"
    },
    {
      "start": 1450.436,
      "end": 1451.716,
      "text": "杰弗迪恩讲到",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0788673568017704,
      "no_speech_prob": 1.2649914102325077e-11,
      "compression_ratio": 1.3832335329341316,
      "source": "whisper"
    },
    {
      "start": 1451.716,
      "end": 1453.696,
      "text": "值得探索的问题数不胜数",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0788673568017704,
      "no_speech_prob": 1.2649914102325077e-11,
      "compression_ratio": 1.3832335329341316,
      "source": "whisper"
    },
    {
      "start": 1453.696,
      "end": 1456.336,
      "text": "他个人十分看好全新硬件路线",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06461550675186456,
      "no_speech_prob": 1.2761063906963077e-11,
      "compression_ratio": 1.4454828660436136,
      "source": "whisper"
    },
    {
      "start": 1456.336,
      "end": 1458.936,
      "text": "高效推理硬件拥有巨大价值",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06461550675186456,
      "no_speech_prob": 1.2761063906963077e-11,
      "compression_ratio": 1.4454828660436136,
      "source": "whisper"
    },
    {
      "start": 1458.936,
      "end": 1461.976,
      "text": "机器学习算法本身也还有巨大革新空间",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06461550675186456,
      "no_speech_prob": 1.2761063906963077e-11,
      "compression_ratio": 1.4454828660436136,
      "source": "whisper"
    },
    {
      "start": 1461.976,
      "end": 1465.196,
      "text": "比如数据效率远远高于当下的系统",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06461550675186456,
      "no_speech_prob": 1.2761063906963077e-11,
      "compression_ratio": 1.4454828660436136,
      "source": "whisper"
    },
    {
      "start": 1465.196,
      "end": 1467.836,
      "text": "现在大模型18岁阶段接触的数据总量",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06461550675186456,
      "no_speech_prob": 1.2761063906963077e-11,
      "compression_ratio": 1.4454828660436136,
      "source": "whisper"
    },
    {
      "start": 1467.836,
      "end": 1470.596,
      "text": "比人类18岁接触的信息多出1000倍",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06461550675186456,
      "no_speech_prob": 1.2761063906963077e-11,
      "compression_ratio": 1.4454828660436136,
      "source": "whisper"
    },
    {
      "start": 1470.596,
      "end": 1472.676,
      "text": "但是人类在大量任务当中",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06461550675186456,
      "no_speech_prob": 1.2761063906963077e-11,
      "compression_ratio": 1.4454828660436136,
      "source": "whisper"
    },
    {
      "start": 1472.676,
      "end": 1474.816,
      "text": "依旧可以持平甚至优于大模型",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06461550675186456,
      "no_speech_prob": 1.2761063906963077e-11,
      "compression_ratio": 1.4454828660436136,
      "source": "whisper"
    },
    {
      "start": 1474.816,
      "end": 1477.176,
      "text": "我们能不能打造数据效率更高",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06461550675186456,
      "no_speech_prob": 1.2761063906963077e-11,
      "compression_ratio": 1.4454828660436136,
      "source": "whisper"
    },
    {
      "start": 1477.176,
      "end": 1480.516,
      "text": "可以从自身行动当中持续学习的系统",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06461550675186456,
      "no_speech_prob": 1.2761063906963077e-11,
      "compression_ratio": 1.4454828660436136,
      "source": "whisper"
    },
    {
      "start": 1480.516,
      "end": 1483.196,
      "text": "持续学习本身就是极具价值的课题",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.06461550675186456,
      "no_speech_prob": 1.2761063906963077e-11,
      "compression_ratio": 1.4454828660436136,
      "source": "whisper"
    },
    {
      "start": 1483.196,
      "end": 1486.176,
      "text": "多agent交互同样充满想象空间",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0717157683349008,
      "no_speech_prob": 1.3632329640134166e-11,
      "compression_ratio": 1.2648809523809523,
      "source": "whisper"
    },
    {
      "start": 1486.176,
      "end": 1487.896,
      "text": "把视角在向外延伸",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0717157683349008,
      "no_speech_prob": 1.3632329640134166e-11,
      "compression_ratio": 1.2648809523809523,
      "source": "whisper"
    },
    {
      "start": 1487.896,
      "end": 1490.396,
      "text": "如何促成人类之间更良性的对话",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0717157683349008,
      "no_speech_prob": 1.3632329640134166e-11,
      "compression_ratio": 1.2648809523809523,
      "source": "whisper"
    },
    {
      "start": 1490.396,
      "end": 1492.736,
      "text": "搭建工具帮助人们理性交流",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0717157683349008,
      "no_speech_prob": 1.3632329640134166e-11,
      "compression_ratio": 1.2648809523809523,
      "source": "whisper"
    },
    {
      "start": 1492.736,
      "end": 1495.276,
      "text": "帮助志趣相投的人建立连接",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0717157683349008,
      "no_speech_prob": 1.3632329640134166e-11,
      "compression_ratio": 1.2648809523809523,
      "source": "whisper"
    },
    {
      "start": 1495.276,
      "end": 1497.436,
      "text": "同样是值得深耕的方向",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0717157683349008,
      "no_speech_prob": 1.3632329640134166e-11,
      "compression_ratio": 1.2648809523809523,
      "source": "whisper"
    },
    {
      "start": 1497.436,
      "end": 1498.536,
      "text": "整场对话听完",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0717157683349008,
      "no_speech_prob": 1.3632329640134166e-11,
      "compression_ratio": 1.2648809523809523,
      "source": "whisper"
    },
    {
      "start": 1498.536,
      "end": 1501.376,
      "text": "不难感受到杰弗迪恩看待行业的独特视角",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0717157683349008,
      "no_speech_prob": 1.3632329640134166e-11,
      "compression_ratio": 1.2648809523809523,
      "source": "whisper"
    },
    {
      "start": 1501.376,
      "end": 1504.016,
      "text": "他不会空谈AGI什么时候到来",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0717157683349008,
      "no_speech_prob": 1.3632329640134166e-11,
      "compression_ratio": 1.2648809523809523,
      "source": "whisper"
    },
    {
      "start": 1504.016,
      "end": 1505.116,
      "text": "而是立足于能耗",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0717157683349008,
      "no_speech_prob": 1.3632329640134166e-11,
      "compression_ratio": 1.2648809523809523,
      "source": "whisper"
    },
    {
      "start": 1505.116,
      "end": 1505.936,
      "text": "数据搬运",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0717157683349008,
      "no_speech_prob": 1.3632329640134166e-11,
      "compression_ratio": 1.2648809523809523,
      "source": "whisper"
    },
    {
      "start": 1505.936,
      "end": 1507.796,
      "text": "延迟这些底层物理约束",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0717157683349008,
      "no_speech_prob": 1.3632329640134166e-11,
      "compression_ratio": 1.2648809523809523,
      "source": "whisper"
    },
    {
      "start": 1507.796,
      "end": 1509.676,
      "text": "向上推导系统设计逻辑",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0717157683349008,
      "no_speech_prob": 1.3632329640134166e-11,
      "compression_ratio": 1.2648809523809523,
      "source": "whisper"
    },
    {
      "start": 1509.676,
      "end": 1510.996,
      "text": "推导创业机会",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.0717157683349008,
      "no_speech_prob": 1.3632329640134166e-11,
      "compression_ratio": 1.2648809523809523,
      "source": "whisper"
    },
    {
      "start": 1510.996,
      "end": 1513.556,
      "text": "全行业疯狂内卷模型参数的时候",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.07731571489450884,
      "no_speech_prob": 1.1754457650281491e-11,
      "compression_ratio": 1.3209876543209877,
      "source": "whisper"
    },
    {
      "start": 1513.556,
      "end": 1517.156,
      "text": "他提醒大家数据搬运的成本是计算的1000倍",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.07731571489450884,
      "no_speech_prob": 1.1754457650281491e-11,
      "compression_ratio": 1.3209876543209877,
      "source": "whisper"
    },
    {
      "start": 1517.156,
      "end": 1520.636,
      "text": "推理硬件专用化是下一轮效率跃升的关键",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.07731571489450884,
      "no_speech_prob": 1.1754457650281491e-11,
      "compression_ratio": 1.3209876543209877,
      "source": "whisper"
    },
    {
      "start": 1520.636,
      "end": 1523.696,
      "text": "所有人焦虑通用模型吞噬创业机会",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.07731571489450884,
      "no_speech_prob": 1.1754457650281491e-11,
      "compression_ratio": 1.3209876543209877,
      "source": "whisper"
    },
    {
      "start": 1523.696,
      "end": 1527.676,
      "text": "他告诉大家去寻找成功率0%到1%的赛道",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.07731571489450884,
      "no_speech_prob": 1.1754457650281491e-11,
      "compression_ratio": 1.3209876543209877,
      "source": "whisper"
    },
    {
      "start": 1527.676,
      "end": 1528.996,
      "text": "挖掘私有数据",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.07731571489450884,
      "no_speech_prob": 1.1754457650281491e-11,
      "compression_ratio": 1.3209876543209877,
      "source": "whisper"
    },
    {
      "start": 1528.996,
      "end": 1530.756,
      "text": "垂直专用模型的机会",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.07731571489450884,
      "no_speech_prob": 1.1754457650281491e-11,
      "compression_ratio": 1.3209876543209877,
      "source": "whisper"
    },
    {
      "start": 1530.756,
      "end": 1533.116,
      "text": "所有人埋头打磨提示词的时候",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.07731571489450884,
      "no_speech_prob": 1.1754457650281491e-11,
      "compression_ratio": 1.3209876543209877,
      "source": "whisper"
    },
    {
      "start": 1533.116,
      "end": 1535.856,
      "text": "他点名未来最稀缺的资源是品位",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.07731571489450884,
      "no_speech_prob": 1.1754457650281491e-11,
      "compression_ratio": 1.3209876543209877,
      "source": "whisper"
    },
    {
      "start": 1535.856,
      "end": 1537.796,
      "text": "是筛选问题的判断力",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.07731571489450884,
      "no_speech_prob": 1.1754457650281491e-11,
      "compression_ratio": 1.3209876543209877,
      "source": "whisper"
    },
    {
      "start": 1537.796,
      "end": 1539.996,
      "text": "这些认知不是凭空想象",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.07731571489450884,
      "no_speech_prob": 1.1754457650281491e-11,
      "compression_ratio": 1.3209876543209877,
      "source": "whisper"
    },
    {
      "start": 1539.996,
      "end": 1542.096,
      "text": "来自几十年系统工程沉淀",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.05651626469176493,
      "no_speech_prob": 1.104431216619739e-11,
      "compression_ratio": 1.3084745762711865,
      "source": "whisper"
    },
    {
      "start": 1542.096,
      "end": 1544.176,
      "text": "来自一次次参金值演算",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.05651626469176493,
      "no_speech_prob": 1.104431216619739e-11,
      "compression_ratio": 1.3084745762711865,
      "source": "whisper"
    },
    {
      "start": 1544.176,
      "end": 1546.976,
      "text": "来自一次次推翻固有认知的思想实验",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.05651626469176493,
      "no_speech_prob": 1.104431216619739e-11,
      "compression_ratio": 1.3084745762711865,
      "source": "whisper"
    },
    {
      "start": 1546.976,
      "end": 1549.376,
      "text": "以上就是整场访谈的核心精华",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.05651626469176493,
      "no_speech_prob": 1.104431216619739e-11,
      "compression_ratio": 1.3084745762711865,
      "source": "whisper"
    },
    {
      "start": 1549.376,
      "end": 1552.436,
      "text": "如果你觉得这些来自一线工程大佬的思考",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.05651626469176493,
      "no_speech_prob": 1.104431216619739e-11,
      "compression_ratio": 1.3084745762711865,
      "source": "whisper"
    },
    {
      "start": 1552.436,
      "end": 1553.236,
      "text": "对你有启发",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.05651626469176493,
      "no_speech_prob": 1.104431216619739e-11,
      "compression_ratio": 1.3084745762711865,
      "source": "whisper"
    },
    {
      "start": 1553.236,
      "end": 1554.336,
      "text": "欢迎点赞",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.05651626469176493,
      "no_speech_prob": 1.104431216619739e-11,
      "compression_ratio": 1.3084745762711865,
      "source": "whisper"
    },
    {
      "start": 1554.336,
      "end": 1557.996,
      "text": "也可以把这期内容分享给身边做AI开发创业的朋友",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.05651626469176493,
      "no_speech_prob": 1.104431216619739e-11,
      "compression_ratio": 1.3084745762711865,
      "source": "whisper"
    },
    {
      "start": 1557.996,
      "end": 1560.236,
      "text": "一起跳出网络上的流量炒作",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.05651626469176493,
      "no_speech_prob": 1.104431216619739e-11,
      "compression_ratio": 1.3084745762711865,
      "source": "whisper"
    },
    {
      "start": 1560.236,
      "end": 1562.636,
      "text": "看见行业更真实的底层逻辑",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.05651626469176493,
      "no_speech_prob": 1.104431216619739e-11,
      "compression_ratio": 1.3084745762711865,
      "source": "whisper"
    },
    {
      "start": 1562.636,
      "end": 1563.836,
      "text": "我们下期再见",
      "chunk": 3,
      "language": "zh",
      "avg_logprob": -0.05651626469176493,
      "no_speech_prob": 1.104431216619739e-11,
      "compression_ratio": 1.3084745762711865,
      "source": "whisper"
    }
  ],
  "transcript_selection": {
    "schema": "videonote.transcript_selection.v1",
    "status": "SELECTED",
    "requested_language": "zh",
    "duration": 1564.2355,
    "reference_path": null,
    "reference_kind": null,
    "reference_language": null,
    "selected_source": "whisper_segmented",
    "decision_reason": "source_unavailable_or_untrusted_whisper_passed",
    "whisper_rejected": false,
    "source_profile": {
      "segment_count": 0,
      "duration_coverage": 0.0,
      "usable_full": false,
      "usable_partial": false,
      "hallucination_region_count": 0,
      "hallucination_regions": [],
      "language": {
        "requested": "zh",
        "matched": false
      }
    },
    "whisper_profile": {
      "segment_count": 693,
      "multilingual": false,
      "first_start": 0.0,
      "final_end": 1563.836,
      "duration_coverage": 0.999745,
      "maximum_gap_seconds": 15.0,
      "inverted_count": 0,
      "language": {
        "requested": "zh",
        "matched": true,
        "latin": 359,
        "han": 7689,
        "kana": 0,
        "hangul": 0
      },
      "hallucination_region_count": 0,
      "hallucination_regions": [],
      "usable_full": true,
      "usable_partial": true
    },
    "comparison": {
      "window_seconds": 30.0,
      "compared_window_count": 0,
      "median_score": 0.0,
      "divergent_window_count": 0,
      "divergent_ratio": 0.0,
      "divergent_windows": []
    },
    "timestamp_alignment": {
      "applied": false,
      "estimated_offset_seconds": 0.0,
      "matched_segment_count": 0,
      "match_scores": []
    },
    "caption_timing_repair": null,
    "filled_source_profile": null,
    "whisper_duration_guard": {
      "media_duration": 1564.236,
      "rejected_segment_count": 0,
      "rejected_segments": []
    }
  }
}