{
  "text": "各位朋友早上好\n今天咱们就来看看本周AI界\n几件暗流涌动的军备竞赛\n从Openai秘密研发的超级新模型\n到阿里悄悄上架的Qwen新势力\n信息量直接拉满\n先看画面这份业内研报\n里面指明Gemini 3.5 Flash表现平平\n而Openai正研发规模超越\n而Openai正研发规模超越Astra的全新预训练模型\n与此同时\n阿里团队在LMArena\n秘密测试代号Kiana的模型\n其3D渲染能力让人惊艳\n此外图像生成领域有重磅进展\n带OpenAI since day水印的\n蒙娜丽莎已现身竞技场\n马斯克旗下的Grok image\n2.0也已经正式发布\n目前在文生图榜单高居第二\n综合表现同样不容小觑\n遗憾的是\n谷歌原本计划推出的\n120B超大体量模型Gemma V4 1,200亿\n近期爆出该项目已经被取消\n看完这轮大盘概览\n咱们先来深挖Openai的重磅核弹\n知名爆料人Chris在推特指出\nGPT5.5绝不是Openai今年\n最后一次大规模预训练\n爆料里特别强调\n虽然GPT6本身已经足够震撼\n但Openai在年底前还会掏出代号为\nbug的超级模型\n预训练规模将直接创下历史新高\n推文里甚至极其夸张的声称\nbug这款新模型的恐怖性能\n绝对会让之前的fable\n显得像个原始玩具\n不过对于它的发布\n时间线与代号对应关系\n行业内目前依然\n存在不少争论与疑虑\n比如知名博主\ntokengrammaring就公开反驳\n指出SEMI analysis在整理这份报告时\n把7月初的dog信息和\n后续的Astra给彻底搞混了\n在他看来\ndog早在Astra之前就已经\n在内部报告中出现过\n因此它极大概率只是\nAsha早期研发阶段的一个内部代号\n而不是什么全新的后续迭代产品\n毕竟Openai内部取代号向来都很随意\n此前就曾用过像\nMilthree Milfour这种混乱名称\n最后才统一叫Asha\n这也让外界很难做出精准判断\n所以说,咱们目前依然无法百分百确定,Dug到底代表着一个全新的超大模型预训练流,还是只属于同一批预训练中的不同阶段检查点,依旧充满神秘感。\n依旧充满神秘感\n而如果把视线转向谷歌,言报给出的结论就更残酷了。文章直接断言,Gimini 3 Pro已经是谷歌技术的高峰,近期Deepmind高层的变动更是加剧了内部焦虑。\n是加剧了内部焦虑\n报告甚至预测\n接下来的Gemini3.5Pro实力仅相当于Opus4.5\n甚至不如GLM5.2\n面对Anthropic与Openai的死斗\n很难招架\n不过其中最为引人注意的,仍然是这份言报尾部给出的重磅结论。OpenAI已经彻底克服了预训练的技术障碍,规模更为庞大的Dug模型正在全力推进,极可能在今年年底前彻底颠覆行业。\n极可能在今年年底前彻底颠覆行业\n聊完海外巨头的暗流涌动\n咱们把视角转回国内\n阿里团队最近在LLM Arena的code\nArena上悄悄上线了一款代号\n为piano的全新检查点模型\n不少开发者第一时间拿它\n跑起了硬核物理弹道测试\n将提亚娜与Claude OPPO S5以及\nGPT5.6放在同等提示词下\n进行3D前端代码生成的大比拼\n而从最终实测运行效果来看,Kiana用3GS生成的弩炮模拟动画可以说是极其逼真,不仅整体结构非常完整,渲染和发射的物理交互更是丝滑顺畅。\n交互更是丝滑顺畅\n在我看来\n这款神秘新模型\n在前端网页开发与复杂逻辑呈现上\n展现出的综合素质完全不\n逊色于目前市场上任何\n顶尖的第一梯队旗舰模型\n如果大家也想亲自体验它的实力\n可以去lmaArena的下拉菜单\n里选择Beta mode盲测模式\n随后进入code Arena\n输入你想要的prompt\n在提交具体需求之后\n系统会有两个完全匿名的\n模型同时在后台生成网页\n用户需要通过对比双边\n生成的结果进行盲测投票\n投票结束后就会揭晓\n代码到底属于哪两款模型\n不过由于抽取是随机的\n想碰上Kiana得运气好或多刷几次\n下面咱们就来看几个它生成\n的代表性高难度实战案例\n咱们先来看这个极其惊艳的案例\n提示词要求生成一个名为AIRO ONE的360度互动耳机展示页面\n提示词要求生成一个名为AIRO ONE的360度互动耳机展示页面\nKiana生成的UI显得极其精致\n黑金高级感直接拉满\n更绝的是\n这个页面不仅支持用鼠标\n全向拖拽旋转和自由缩放\n点击耳机上的touch control\nDISC按鈕\n还会及时弹出极其精准\n的技术参数标签与提示\n完成度非常高\n而最让人惊叹的则是页面\n底部的动态配色切换功能\n当点击不同颜色按钮\n3D耳机的表面材质与\n光照渲染会实时改变\n这种Three.js代码实现水平\n简直远远超越了大多数对手\n看完了单独展示\n咱们再来看一组\n直观的四方横向对比\n在同等提示词下\n生成蒸汽朋克飞艇\n四个画面分别由Claude Opus 5、Qwen\nQwen、Claude Fastr 5 High\n以及Qwen 3.8 Max渲染\n大家仔细看右上角\nTiana生成的这艘飞艇\n无论是气囊的网格\n复杂的金属管道\n还是悬挂吊舱的细节丰富度\n都明显碾压了左下角\nFastr 5 High生成的简化模型\n虽然左上角Opus5在材质与\n细节纹理上略微胜出一点点\n但Qwen展示出的设计合理性\n还有超强的复杂代码构建能力\n已经完全能跟顶尖的\nOpus模型正面交锋了\n结合刚才这一系列惊人表现\n业内普遍推测\n这个代号Qiana的模型\n极可能就是阿里即将发布的\nQwen 4.0或Qwen 3.9大更新的\n前哨检查点太值得期待了\n咱们再看这组黄昏\n庄园3D场景的对抗\n由Qiana與老牌\nQwen 3.8 Max展开直接PK\n这次给到的高难度渲染要求是\n生成一个极致拟真的黄昏乡村庄园\n大家可以清晰的看到\nQiana生成的庄园简直太惊艳了\n不仅精细的还原了小巧的\n彩灯串与温馨的屋顶光影\n周围还布局了层次\n丰富的花坛和森林\n光影渲染的整体质感\n远远超越了下方的老版本\n可以说是彻底坐实了它作为\n下一代大模型的绝对实力\n聊完了代码与3D\n咱们再把视线转向\n另一个卷到飞起的赛道\n图像生成\n最近在LMArena竞技场里\n也突然低调上线了一个代号叫\n蒙娜丽莎1的全新图像大模型\n表现同样非常惹眼\n而要说最锤的铁证\n就是用Openai官方的验证工具去\n检测蒙娜丽莎一跑出来的图片\n系统居然直接识别出了Openai\n自家专属的seed ID隐形水印\n这下可以说是当场破案了\n这就意味着Openai在图像\n生成领域也要掏出底牌了\n无论接下来叫ChatGPT image\n2.5还是直接升到3.0\n画质与控图能力都将\n迎来脱胎换骨般的质飞跃\n不过就在各大厂商猛砸算力的时候\n谷歌这边却传来了一条坏消息\n此前被开源社区寄予厚望的\n超大开源模型Gemma V4 120B\n最新爆料显示已被官方彻底砍掉\n根据泄露的后台界面显示\n这款原本拥有1,230亿参数以及\n262k超百万级上下文窗口的开源旗舰\n大模型已经被标记为了取消状态\n确定不会再面向公众发布了\n这也侧面印证了谷歌在开源与闭源\n战略之间正在经历剧烈的阵痛调整\n他们很可能是想把\n全部计算资源集中起来\n全力冲刺下一代闭源Gemini\n不得不说\n真挺令人唏嘘的\n而除了Openai与谷歌\n马斯克的Grok也不甘示弱\n咱们一起来看看\n这些精美的生成范例\n无论是波普风海报\n细腻纺织纹理\n还是素描步骤拆解\nGrok image 2.0在文字渲染和\n审美排版上都极其出色\n按照惯例\n看完软件层的神仙打架\n咱们在结尾再来看一个极其\n硬核的具身智能硬件项目\n东京大学最新展示了一款名为\nDragon的空中变形飞行机器人\n它最令人惊叹的地方就是彻底\n打破了传统无人机固定的形态限制\n能在空中完成极度\n复杂的自由变体运动\n大家仔细看\nDragon机器人在空中飞行时\n能精准改变关节构形\n甚至能用身体灵活抓取\n并包裹移动悬挂物体\n在结合了AI控制算法后\n它生动展示了具身智能在\n复杂空间下的无限潜力\n从底层大模型到具身智能\nAI的进化爆发正全面到来\n感谢你看到这里\n这里是大雷早上好\n关注频道\n咱们下期见",
  "language": "zh",
  "duration": 583.633563,
  "segments": [
    {
      "start": 0.1,
      "end": 1.5,
      "text": "各位朋友早上好",
      "source": "source_caption"
    },
    {
      "start": 1.6,
      "end": 3.933,
      "text": "今天咱们就来看看本周AI界",
      "source": "source_caption"
    },
    {
      "start": 3.933,
      "end": 6.2,
      "text": "几件暗流涌动的军备竞赛",
      "source": "source_caption"
    },
    {
      "start": 6.333,
      "end": 9.2,
      "text": "从Openai秘密研发的超级新模型",
      "source": "source_caption"
    },
    {
      "start": 9.3,
      "end": 12.133,
      "text": "到阿里悄悄上架的Qwen新势力",
      "source": "source_caption"
    },
    {
      "start": 12.333,
      "end": 13.833,
      "text": "信息量直接拉满",
      "source": "source_caption"
    },
    {
      "start": 13.833,
      "end": 15.9,
      "text": "先看画面这份业内研报",
      "source": "source_caption"
    },
    {
      "start": 16.133,
      "end": 19.3,
      "text": "里面指明Gemini 3.5 Flash表现平平",
      "source": "source_caption",
      "text_original": "里面指明Gemini3.5Flash表现平平",
      "correction_applied": true,
      "correction_source": "llm_correct_a_presegmentation"
    },
    {
      "start": 19.6,
      "end": 22.033,
      "text": "而Openai正研发规模超越",
      "source": "source_caption"
    },
    {
      "start": 22.033,
      "end": 24.0,
      "text": "而Openai正研发规模超越Astra的全新预训练模型",
      "source": "source_caption",
      "text_original": "Estrela的全新预训练模型",
      "correction_applied": true,
      "correction_source": "llm_correct_a_presegmentation"
    },
    {
      "start": 24.033,
      "end": 24.9,
      "text": "与此同时",
      "source": "source_caption"
    },
    {
      "start": 25.066,
      "end": 26.933,
      "text": "阿里团队在LMArena",
      "source": "source_caption",
      "text_original": "阿里团队在Almerina",
      "correction_applied": true,
      "correction_source": "llm_correct_a_presegmentation"
    },
    {
      "start": 26.933,
      "end": 29.333,
      "text": "秘密测试代号Kiana的模型",
      "source": "source_caption"
    },
    {
      "start": 29.6,
      "end": 32.033,
      "text": "其3D渲染能力让人惊艳",
      "source": "source_caption"
    },
    {
      "start": 32.033,
      "end": 34.733,
      "text": "此外图像生成领域有重磅进展",
      "source": "source_caption"
    },
    {
      "start": 35.033,
      "end": 37.3,
      "text": "带OpenAI since day水印的",
      "source": "source_caption",
      "text_original": "带Openai since day水印的",
      "correction_applied": true,
      "correction_source": "llm_correct_a_presegmentation"
    },
    {
      "start": 37.3,
      "end": 39.1,
      "text": "蒙娜丽莎已现身竞技场",
      "source": "source_caption"
    },
    {
      "start": 39.1,
      "end": 41.1,
      "text": "马斯克旗下的Grok image",
      "source": "source_caption"
    },
    {
      "start": 41.1,
      "end": 43.3,
      "text": "2.0也已经正式发布",
      "source": "source_caption"
    },
    {
      "start": 43.5,
      "end": 46.433,
      "text": "目前在文生图榜单高居第二",
      "source": "source_caption"
    },
    {
      "start": 46.7,
      "end": 48.9,
      "text": "综合表现同样不容小觑",
      "source": "source_caption"
    },
    {
      "start": 49.066,
      "end": 50.033,
      "text": "遗憾的是",
      "source": "source_caption"
    },
    {
      "start": 50.333,
      "end": 52.3,
      "text": "谷歌原本计划推出的",
      "source": "source_caption"
    },
    {
      "start": 52.3,
      "end": 56.633,
      "text": "120B超大体量模型Gemma V4 1,200亿",
      "source": "source_caption"
    },
    {
      "start": 56.833,
      "end": 59.6,
      "text": "近期爆出该项目已经被取消",
      "source": "source_caption"
    },
    {
      "start": 59.633,
      "end": 61.066,
      "text": "看完这轮大盘概览",
      "source": "source_caption"
    },
    {
      "start": 61.066,
      "end": 63.633,
      "text": "咱们先来深挖Openai的重磅核弹",
      "source": "source_caption"
    },
    {
      "start": 64.066,
      "end": 66.466,
      "text": "知名爆料人Chris在推特指出",
      "source": "source_caption"
    },
    {
      "start": 66.7,
      "end": 69.3,
      "text": "GPT5.5绝不是Openai今年",
      "source": "source_caption"
    },
    {
      "start": 69.3,
      "end": 71.1,
      "text": "最后一次大规模预训练",
      "source": "source_caption"
    },
    {
      "start": 71.1,
      "end": 72.433,
      "text": "爆料里特别强调",
      "source": "source_caption"
    },
    {
      "start": 72.533,
      "end": 75.333,
      "text": "虽然GPT6本身已经足够震撼",
      "source": "source_caption"
    },
    {
      "start": 75.4,
      "end": 78.666,
      "text": "但Openai在年底前还会掏出代号为",
      "source": "source_caption"
    },
    {
      "start": 78.933,
      "end": 80.066,
      "text": "bug的超级模型",
      "source": "source_caption"
    },
    {
      "start": 80.2,
      "end": 83.0,
      "text": "预训练规模将直接创下历史新高",
      "source": "source_caption"
    },
    {
      "start": 83.0,
      "end": 85.133,
      "text": "推文里甚至极其夸张的声称",
      "source": "source_caption"
    },
    {
      "start": 85.333,
      "end": 87.666,
      "text": "bug这款新模型的恐怖性能",
      "source": "source_caption"
    },
    {
      "start": 87.8,
      "end": 89.733,
      "text": "绝对会让之前的fable",
      "source": "source_caption"
    },
    {
      "start": 89.933,
      "end": 91.666,
      "text": "显得像个原始玩具",
      "source": "source_caption"
    },
    {
      "start": 92.133,
      "end": 93.233,
      "text": "不过对于它的发布",
      "source": "source_caption"
    },
    {
      "start": 93.233,
      "end": 95.633,
      "text": "时间线与代号对应关系",
      "source": "source_caption"
    },
    {
      "start": 95.7,
      "end": 97.033,
      "text": "行业内目前依然",
      "source": "source_caption"
    },
    {
      "start": 97.033,
      "end": 98.7,
      "text": "存在不少争论与疑虑",
      "source": "source_caption"
    },
    {
      "start": 98.7,
      "end": 99.866,
      "text": "比如知名博主",
      "source": "source_caption"
    },
    {
      "start": 99.866,
      "end": 101.833,
      "text": "tokengrammaring就公开反驳",
      "source": "source_caption"
    },
    {
      "start": 102.266,
      "end": 105.233,
      "text": "指出SEMI analysis在整理这份报告时",
      "source": "source_caption"
    },
    {
      "start": 105.3,
      "end": 107.433,
      "text": "把7月初的dog信息和",
      "source": "source_caption"
    },
    {
      "start": 107.433,
      "end": 109.666,
      "text": "后续的Astra给彻底搞混了",
      "source": "source_caption"
    },
    {
      "start": 109.666,
      "end": 110.466,
      "text": "在他看来",
      "source": "source_caption"
    },
    {
      "start": 110.466,
      "end": 113.066,
      "text": "dog早在Astra之前就已经",
      "source": "source_caption"
    },
    {
      "start": 113.066,
      "end": 114.866,
      "text": "在内部报告中出现过",
      "source": "source_caption"
    },
    {
      "start": 115.2,
      "end": 117.233,
      "text": "因此它极大概率只是",
      "source": "source_caption"
    },
    {
      "start": 117.3,
      "end": 120.3,
      "text": "Asha早期研发阶段的一个内部代号",
      "source": "source_caption"
    },
    {
      "start": 120.4,
      "end": 123.0,
      "text": "而不是什么全新的后续迭代产品",
      "source": "source_caption"
    },
    {
      "start": 123.0,
      "end": 126.433,
      "text": "毕竟Openai内部取代号向来都很随意",
      "source": "source_caption"
    },
    {
      "start": 126.6,
      "end": 128.433,
      "text": "此前就曾用过像",
      "source": "source_caption"
    },
    {
      "start": 128.433,
      "end": 130.9,
      "text": "Milthree Milfour这种混乱名称",
      "source": "source_caption"
    },
    {
      "start": 131.0,
      "end": 132.933,
      "text": "最后才统一叫Asha",
      "source": "source_caption"
    },
    {
      "start": 133.233,
      "end": 136.0,
      "text": "这也让外界很难做出精准判断",
      "source": "source_caption"
    },
    {
      "id": 15,
      "seek": 12284,
      "start": 136.0,
      "end": 148.84,
      "text": "所以说,咱们目前依然无法百分百确定,Dug到底代表着一个全新的超大模型预训练流,还是只属于同一批预训练中的不同阶段检查点,依旧充满神秘感。",
      "tokens": [
        51015,
        7239,
        8090,
        11,
        8975,
        109,
        9497,
        39004,
        3254,
        251,
        5823,
        31364,
        11148,
        31906,
        6627,
        31906,
        38114,
        106,
        12088,
        11,
        35,
        697,
        33883,
        19105,
        17571,
        20708,
        20182,
        11319,
        12560,
        1546,
        19869,
        3582,
        41908,
        39823,
        12501,
        226,
        7422,
        255,
        10115,
        225,
        27854,
        11,
        45726,
        14003,
        9636,
        252,
        37732,
        13089,
        2257,
        3416,
        117,
        12501,
        226,
        7422,
        255,
        10115,
        225,
        5975,
        1546,
        47123,
        10034,
        114,
        28427,
        16407,
        222,
        42623,
        12579,
        11,
        3254,
        251,
        4479,
        100,
        2347,
        227,
        15868,
        94,
        20432,
        8204,
        246,
        9709,
        1543,
        51665
      ],
      "temperature": 0,
      "avg_logprob": -0.04188870460756364,
      "compression_ratio": 1.1257861635220126,
      "no_speech_prob": 1.6567058488958786e-11,
      "source": "whisper_gap_repair"
    },
    {
      "start": 148.84,
      "end": 149.033,
      "text": "依旧充满神秘感",
      "source": "source_caption"
    },
    {
      "id": 16,
      "seek": 14884,
      "start": 149.033,
      "end": 161.84,
      "text": "而如果把视线转向谷歌,言报给出的结论就更残酷了。文章直接断言,Gimini 3 Pro已经是谷歌技术的高峰,近期Deepmind高层的变动更是加剧了内部焦虑。",
      "tokens": [
        50365,
        11070,
        13119,
        16075,
        40656,
        16853,
        123,
        17819,
        105,
        24282,
        8897,
        115,
        29582,
        11,
        12009,
        49817,
        23197,
        7781,
        1546,
        45641,
        7422,
        118,
        3111,
        19002,
        15976,
        233,
        16911,
        115,
        2289,
        1543,
        17174,
        11957,
        254,
        43297,
        4307,
        255,
        12009,
        11,
        38,
        332,
        3812,
        805,
        1705,
        49161,
        1541,
        8897,
        115,
        29582,
        32502,
        1474,
        107,
        1546,
        12979,
        26555,
        108,
        11,
        17463,
        16786,
        11089,
        595,
        13733,
        12979,
        9636,
        224,
        1546,
        2129,
        246,
        34961,
        19002,
        1541,
        9990,
        5935,
        100,
        2289,
        34742,
        13470,
        4971,
        99,
        12026,
        239,
        1543,
        51015
      ],
      "temperature": 0,
      "avg_logprob": -0.1170989331744966,
      "compression_ratio": 1.0268817204301075,
      "no_speech_prob": 1.5840022002389098e-11,
      "source": "whisper_gap_repair"
    },
    {
      "start": 161.84,
      "end": 162.433,
      "text": "是加剧了内部焦虑",
      "source": "source_caption"
    },
    {
      "start": 162.433,
      "end": 163.8,
      "text": "报告甚至预测",
      "source": "source_caption"
    },
    {
      "start": 163.933,
      "end": 168.2,
      "text": "接下来的Gemini3.5Pro实力仅相当于Opus4.5",
      "source": "source_caption"
    },
    {
      "start": 168.333,
      "end": 170.5,
      "text": "甚至不如GLM5.2",
      "source": "source_caption"
    },
    {
      "start": 170.6,
      "end": 173.133,
      "text": "面对Anthropic与Openai的死斗",
      "source": "source_caption"
    },
    {
      "start": 173.133,
      "end": 173.933,
      "text": "很难招架",
      "source": "source_caption"
    },
    {
      "id": 18,
      "seek": 17384,
      "start": 173.933,
      "end": 189.84,
      "text": "不过其中最为引人注意的,仍然是这份言报尾部给出的重磅结论。OpenAI已经彻底克服了预训练的技术障碍,规模更为庞大的Dug模型正在全力推进,极可能在今年年底前彻底颠覆行业。",
      "tokens": [
        50365,
        1960,
        16866,
        9572,
        5975,
        8661,
        13992,
        41301,
        4035,
        40794,
        1546,
        11,
        1550,
        235,
        5823,
        1541,
        5562,
        36266,
        12009,
        49817,
        1530,
        122,
        13470,
        23197,
        7781,
        1546,
        12624,
        163,
        96,
        227,
        45641,
        7422,
        118,
        1543,
        45569,
        48698,
        49161,
        7391,
        119,
        22816,
        24881,
        27408,
        2289,
        12501,
        226,
        7422,
        255,
        10115,
        225,
        1546,
        32502,
        1474,
        107,
        10673,
        250,
        16337,
        235,
        11,
        6758,
        226,
        41908,
        19002,
        13992,
        6346,
        252,
        39156,
        35,
        697,
        41908,
        39823,
        15789,
        3581,
        11319,
        13486,
        33597,
        36700,
        11,
        7360,
        223,
        16657,
        3581,
        38937,
        5157,
        22816,
        8945,
        7391,
        119,
        22816,
        12501,
        254,
        2888,
        228,
        8082,
        940,
        248,
        1543,
        51165
      ],
      "temperature": 0,
      "avg_logprob": -0.03848728025802458,
      "compression_ratio": 1.0684931506849316,
      "no_speech_prob": 1.978980521122331e-11,
      "source": "whisper_gap_repair"
    },
    {
      "start": 189.84,
      "end": 190.066,
      "text": "极可能在今年年底前彻底颠覆行业",
      "source": "source_caption"
    },
    {
      "start": 190.066,
      "end": 192.3,
      "text": "聊完海外巨头的暗流涌动",
      "source": "source_caption"
    },
    {
      "start": 192.4,
      "end": 194.433,
      "text": "咱们把视角转回国内",
      "source": "source_caption"
    },
    {
      "start": 194.833,
      "end": 198.433,
      "text": "阿里团队最近在LLM Arena的code",
      "source": "source_caption"
    },
    {
      "start": 198.433,
      "end": 201.0,
      "text": "Arena上悄悄上线了一款代号",
      "source": "source_caption"
    },
    {
      "start": 201.0,
      "end": 203.833,
      "text": "为piano的全新检查点模型",
      "source": "source_caption"
    },
    {
      "start": 203.833,
      "end": 205.933,
      "text": "不少开发者第一时间拿它",
      "source": "source_caption"
    },
    {
      "start": 205.933,
      "end": 208.333,
      "text": "跑起了硬核物理弹道测试",
      "source": "source_caption"
    },
    {
      "start": 208.733,
      "end": 211.5,
      "text": "将提亚娜与Claude OPPO S5以及",
      "source": "source_caption"
    },
    {
      "start": 211.5,
      "end": 214.7,
      "text": "GPT5.6放在同等提示词下",
      "source": "source_caption"
    },
    {
      "start": 214.733,
      "end": 218.0,
      "text": "进行3D前端代码生成的大比拼",
      "source": "source_caption"
    },
    {
      "id": 21,
      "seek": 21784,
      "start": 218.0,
      "end": 232.44,
      "text": "而从最终实测运行效果来看,Kiana用3GS生成的弩炮模拟动画可以说是极其逼真,不仅整体结构非常完整,渲染和发射的物理交互更是丝滑顺畅。",
      "tokens": [
        50365,
        11070,
        35630,
        8661,
        10115,
        230,
        24726,
        11038,
        233,
        3316,
        238,
        8082,
        43076,
        9319,
        6912,
        4200,
        11,
        42,
        8497,
        9254,
        18,
        24446,
        8244,
        11336,
        1546,
        5702,
        102,
        4804,
        106,
        41908,
        6852,
        253,
        34961,
        27126,
        6723,
        8090,
        1541,
        7360,
        223,
        9572,
        2215,
        120,
        6303,
        11,
        1960,
        1550,
        227,
        27662,
        29485,
        45641,
        7360,
        226,
        14392,
        14128,
        27662,
        11,
        29717,
        1431,
        122,
        12565,
        28926,
        1530,
        226,
        1546,
        23516,
        13876,
        28455,
        1369,
        240,
        19002,
        1541,
        940,
        251,
        15868,
        239,
        10178,
        118,
        6904,
        227,
        1543,
        51095
      ],
      "temperature": 0,
      "avg_logprob": -0.043097673736003615,
      "compression_ratio": 1.1842105263157894,
      "no_speech_prob": 1.6516971818036907e-11,
      "source": "whisper_gap_repair",
      "text_original": "而从最终实测运行效果来看,Kiana用3GS生成的弩炮模拟动画可以说是极其逼真,不仅整体结构非常完整,拉显和发射的物理交互更是丝滑顺畅。",
      "correction_applied": true,
      "correction_source": "llm_correct_a_presegmentation"
    },
    {
      "start": 232.44,
      "end": 232.633,
      "text": "交互更是丝滑顺畅",
      "source": "source_caption"
    },
    {
      "start": 232.7,
      "end": 233.466,
      "text": "在我看来",
      "source": "source_caption"
    },
    {
      "start": 233.466,
      "end": 235.0,
      "text": "这款神秘新模型",
      "source": "source_caption"
    },
    {
      "start": 235.3,
      "end": 238.4,
      "text": "在前端网页开发与复杂逻辑呈现上",
      "source": "source_caption"
    },
    {
      "start": 238.466,
      "end": 241.0,
      "text": "展现出的综合素质完全不",
      "source": "source_caption"
    },
    {
      "start": 241.0,
      "end": 243.266,
      "text": "逊色于目前市场上任何",
      "source": "source_caption"
    },
    {
      "start": 243.266,
      "end": 245.6,
      "text": "顶尖的第一梯队旗舰模型",
      "source": "source_caption"
    },
    {
      "start": 245.6,
      "end": 248.066,
      "text": "如果大家也想亲自体验它的实力",
      "source": "source_caption"
    },
    {
      "start": 248.4,
      "end": 250.7,
      "text": "可以去lmaArena的下拉菜单",
      "source": "source_caption",
      "text_original": "可以去l美Arena的下拉菜单",
      "correction_applied": true,
      "correction_source": "llm_correct_a_presegmentation"
    },
    {
      "start": 250.7,
      "end": 253.1,
      "text": "里选择Beta mode盲测模式",
      "source": "source_caption"
    },
    {
      "start": 253.466,
      "end": 255.266,
      "text": "随后进入code Arena",
      "source": "source_caption"
    },
    {
      "start": 255.333,
      "end": 256.866,
      "text": "输入你想要的prompt",
      "source": "source_caption"
    },
    {
      "start": 256.866,
      "end": 258.9,
      "text": "在提交具体需求之后",
      "source": "source_caption"
    },
    {
      "start": 259.0,
      "end": 261.2,
      "text": "系统会有两个完全匿名的",
      "source": "source_caption"
    },
    {
      "start": 261.2,
      "end": 263.6,
      "text": "模型同时在后台生成网页",
      "source": "source_caption"
    },
    {
      "start": 263.9,
      "end": 265.866,
      "text": "用户需要通过对比双边",
      "source": "source_caption"
    },
    {
      "start": 265.866,
      "end": 268.233,
      "text": "生成的结果进行盲测投票",
      "source": "source_caption"
    },
    {
      "start": 268.4,
      "end": 270.3,
      "text": "投票结束后就会揭晓",
      "source": "source_caption"
    },
    {
      "start": 270.3,
      "end": 272.4,
      "text": "代码到底属于哪两款模型",
      "source": "source_caption"
    },
    {
      "start": 272.4,
      "end": 274.266,
      "text": "不过由于抽取是随机的",
      "source": "source_caption"
    },
    {
      "start": 274.266,
      "end": 277.1,
      "text": "想碰上Kiana得运气好或多刷几次",
      "source": "source_caption"
    },
    {
      "start": 277.5,
      "end": 279.833,
      "text": "下面咱们就来看几个它生成",
      "source": "source_caption"
    },
    {
      "start": 279.833,
      "end": 282.233,
      "text": "的代表性高难度实战案例",
      "source": "source_caption"
    },
    {
      "start": 282.233,
      "end": 284.866,
      "text": "咱们先来看这个极其惊艳的案例",
      "source": "source_caption"
    },
    {
      "start": 285.233,
      "end": 288.2,
      "text": "提示词要求生成一个名为AIRO ONE的360度互动耳机展示页面",
      "source": "source_caption",
      "text_original": "提示词要求生成一个名为AIRO",
      "correction_applied": true,
      "correction_source": "llm_correct_a_presegmentation"
    },
    {
      "start": 288.2,
      "end": 290.933,
      "text": "提示词要求生成一个名为AIRO ONE的360度互动耳机展示页面",
      "source": "source_caption",
      "text_original": "one的360度互动耳机展示页面",
      "correction_applied": true,
      "correction_source": "llm_correct_a_presegmentation"
    },
    {
      "start": 291.066,
      "end": 293.433,
      "text": "Kiana生成的UI显得极其精致",
      "source": "source_caption",
      "text_original": "Tianna生成的UI显得极其精致",
      "correction_applied": true,
      "correction_source": "llm_correct_a_presegmentation"
    },
    {
      "start": 293.666,
      "end": 295.533,
      "text": "黑金高级感直接拉满",
      "source": "source_caption"
    },
    {
      "start": 295.533,
      "end": 296.533,
      "text": "更绝的是",
      "source": "source_caption"
    },
    {
      "start": 296.6,
      "end": 298.7,
      "text": "这个页面不仅支持用鼠标",
      "source": "source_caption"
    },
    {
      "start": 298.7,
      "end": 301.333,
      "text": "全向拖拽旋转和自由缩放",
      "source": "source_caption"
    },
    {
      "start": 301.7,
      "end": 303.7,
      "text": "点击耳机上的touch control",
      "source": "source_caption"
    },
    {
      "start": 303.833,
      "end": 304.633,
      "text": "DISC按鈕",
      "source": "source_caption",
      "text_original": "DISC按钮",
      "correction_applied": true,
      "correction_source": "llm_correct_a_presegmentation"
    },
    {
      "start": 304.933,
      "end": 307.033,
      "text": "还会及时弹出极其精准",
      "source": "source_caption"
    },
    {
      "start": 307.033,
      "end": 309.066,
      "text": "的技术参数标签与提示",
      "source": "source_caption"
    },
    {
      "start": 309.4,
      "end": 310.633,
      "text": "完成度非常高",
      "source": "source_caption"
    },
    {
      "start": 310.633,
      "end": 313.133,
      "text": "而最让人惊叹的则是页面",
      "source": "source_caption"
    },
    {
      "start": 313.133,
      "end": 315.633,
      "text": "底部的动态配色切换功能",
      "source": "source_caption"
    },
    {
      "start": 316.0,
      "end": 317.733,
      "text": "当点击不同颜色按钮",
      "source": "source_caption"
    },
    {
      "start": 317.9,
      "end": 319.866,
      "text": "3D耳机的表面材质与",
      "source": "source_caption"
    },
    {
      "start": 319.866,
      "end": 321.7,
      "text": "光照渲染会实时改变",
      "source": "source_caption"
    },
    {
      "start": 321.9,
      "end": 324.433,
      "text": "这种Three.js代码实现水平",
      "source": "source_caption",
      "text_original": "这种threejs代码实现水平",
      "correction_applied": true,
      "correction_source": "llm_correct_a_presegmentation"
    },
    {
      "start": 324.6,
      "end": 326.9,
      "text": "简直远远超越了大多数对手",
      "source": "source_caption"
    },
    {
      "start": 326.9,
      "end": 328.366,
      "text": "看完了单独展示",
      "source": "source_caption"
    },
    {
      "start": 328.5,
      "end": 329.666,
      "text": "咱们再来看一组",
      "source": "source_caption"
    },
    {
      "start": 329.666,
      "end": 331.533,
      "text": "直观的四方横向对比",
      "source": "source_caption"
    },
    {
      "start": 332.066,
      "end": 333.533,
      "text": "在同等提示词下",
      "source": "source_caption"
    },
    {
      "start": 333.533,
      "end": 335.166,
      "text": "生成蒸汽朋克飞艇",
      "source": "source_caption"
    },
    {
      "start": 335.433,
      "end": 338.3,
      "text": "四个画面分别由Claude Opus 5、Qwen",
      "source": "source_caption",
      "text_original": "四个画面分别由Claude OPS 5米x",
      "correction_applied": true,
      "correction_source": "llm_correct_a_presegmentation"
    },
    {
      "start": 338.7,
      "end": 340.633,
      "text": "Qwen、Claude Fastr 5 High",
      "source": "source_caption",
      "text_original": "Tiana Claude Fabo5嗨",
      "correction_applied": true,
      "correction_source": "llm_correct_a_presegmentation"
    },
    {
      "start": 340.733,
      "end": 342.9,
      "text": "以及Qwen 3.8 Max渲染",
      "source": "source_caption",
      "text_original": "以及昆仑3.8Max渲染",
      "correction_applied": true,
      "correction_source": "llm_correct_a_presegmentation"
    },
    {
      "start": 342.9,
      "end": 344.3,
      "text": "大家仔细看右上角",
      "source": "source_caption"
    },
    {
      "start": 344.3,
      "end": 346.433,
      "text": "Tiana生成的这艘飞艇",
      "source": "source_caption"
    },
    {
      "start": 346.833,
      "end": 348.366,
      "text": "无论是气囊的网格",
      "source": "source_caption"
    },
    {
      "start": 348.566,
      "end": 350.133,
      "text": "复杂的金属管道",
      "source": "source_caption"
    },
    {
      "start": 350.133,
      "end": 352.666,
      "text": "还是悬挂吊舱的细节丰富度",
      "source": "source_caption"
    },
    {
      "start": 352.7,
      "end": 354.533,
      "text": "都明显碾压了左下角",
      "source": "source_caption"
    },
    {
      "start": 354.533,
      "end": 356.666,
      "text": "Fastr 5 High生成的简化模型",
      "source": "source_caption",
      "text_original": "FIB五h生成的简化模型",
      "correction_applied": true,
      "correction_source": "llm_correct_a_presegmentation"
    },
    {
      "start": 356.666,
      "end": 359.333,
      "text": "虽然左上角Opus5在材质与",
      "source": "source_caption"
    },
    {
      "start": 359.333,
      "end": 361.5,
      "text": "细节纹理上略微胜出一点点",
      "source": "source_caption"
    },
    {
      "start": 361.733,
      "end": 364.266,
      "text": "但Qwen展示出的设计合理性",
      "source": "source_caption",
      "text_original": "但Kiara展示出的设计合理性",
      "correction_applied": true,
      "correction_source": "llm_correct_a_presegmentation"
    },
    {
      "start": 364.3,
      "end": 366.9,
      "text": "还有超强的复杂代码构建能力",
      "source": "source_caption"
    },
    {
      "start": 367.033,
      "end": 368.7,
      "text": "已经完全能跟顶尖的",
      "source": "source_caption"
    },
    {
      "start": 368.7,
      "end": 370.466,
      "text": "Opus模型正面交锋了",
      "source": "source_caption"
    },
    {
      "start": 370.666,
      "end": 372.766,
      "text": "结合刚才这一系列惊人表现",
      "source": "source_caption"
    },
    {
      "start": 372.9,
      "end": 374.3,
      "text": "业内普遍推测",
      "source": "source_caption"
    },
    {
      "start": 374.366,
      "end": 376.266,
      "text": "这个代号Qiana的模型",
      "source": "source_caption",
      "text_original": "这个代号QIANA的模型",
      "correction_applied": true,
      "correction_source": "llm_correct_a_presegmentation"
    },
    {
      "start": 376.366,
      "end": 378.733,
      "text": "极可能就是阿里即将发布的",
      "source": "source_caption"
    },
    {
      "start": 378.866,
      "end": 382.033,
      "text": "Qwen 4.0或Qwen 3.9大更新的",
      "source": "source_caption",
      "text_original": "QWEN4.0或Qwen三点九大更新的",
      "correction_applied": true,
      "correction_source": "llm_correct_a_presegmentation"
    },
    {
      "start": 382.233,
      "end": 384.533,
      "text": "前哨检查点太值得期待了",
      "source": "source_caption"
    },
    {
      "start": 384.533,
      "end": 386.133,
      "text": "咱们再看这组黄昏",
      "source": "source_caption"
    },
    {
      "start": 386.133,
      "end": 388.466,
      "text": "庄园3D场景的对抗",
      "source": "source_caption"
    },
    {
      "start": 388.566,
      "end": 390.466,
      "text": "由Qiana與老牌",
      "source": "source_caption",
      "text_original": "由克亚娜与老牌",
      "correction_applied": true,
      "correction_source": "llm_correct_a_presegmentation"
    },
    {
      "start": 390.466,
      "end": 393.533,
      "text": "Qwen 3.8 Max展开直接PK",
      "source": "source_caption",
      "text_original": "葵文3.8Max shy展开直接PK",
      "correction_applied": true,
      "correction_source": "llm_correct_a_presegmentation"
    },
    {
      "start": 394.066,
      "end": 396.633,
      "text": "这次给到的高难度渲染要求是",
      "source": "source_caption"
    },
    {
      "start": 396.633,
      "end": 399.433,
      "text": "生成一个极致拟真的黄昏乡村庄园",
      "source": "source_caption"
    },
    {
      "start": 399.433,
      "end": 400.933,
      "text": "大家可以清晰的看到",
      "source": "source_caption"
    },
    {
      "start": 400.966,
      "end": 403.633,
      "text": "Qiana生成的庄园简直太惊艳了",
      "source": "source_caption",
      "text_original": "Tianna生成的庄园简直太惊艳了",
      "correction_applied": true,
      "correction_source": "llm_correct_a_presegmentation"
    },
    {
      "start": 403.7,
      "end": 406.133,
      "text": "不仅精细的还原了小巧的",
      "source": "source_caption"
    },
    {
      "start": 406.133,
      "end": 408.633,
      "text": "彩灯串与温馨的屋顶光影",
      "source": "source_caption"
    },
    {
      "start": 408.766,
      "end": 410.333,
      "text": "周围还布局了层次",
      "source": "source_caption"
    },
    {
      "start": 410.333,
      "end": 411.933,
      "text": "丰富的花坛和森林",
      "source": "source_caption"
    },
    {
      "start": 412.466,
      "end": 414.3,
      "text": "光影渲染的整体质感",
      "source": "source_caption"
    },
    {
      "start": 414.333,
      "end": 416.233,
      "text": "远远超越了下方的老版本",
      "source": "source_caption"
    },
    {
      "start": 416.233,
      "end": 418.366,
      "text": "可以说是彻底坐实了它作为",
      "source": "source_caption"
    },
    {
      "start": 418.366,
      "end": 420.366,
      "text": "下一代大模型的绝对实力",
      "source": "source_caption"
    },
    {
      "start": 420.366,
      "end": 421.833,
      "text": "聊完了代码与3D",
      "source": "source_caption"
    },
    {
      "start": 421.9,
      "end": 423.366,
      "text": "咱们再把视线转向",
      "source": "source_caption"
    },
    {
      "start": 423.366,
      "end": 425.133,
      "text": "另一个卷到飞起的赛道",
      "source": "source_caption"
    },
    {
      "start": 425.266,
      "end": 426.3,
      "text": "图像生成",
      "source": "source_caption"
    },
    {
      "start": 426.733,
      "end": 429.166,
      "text": "最近在LMArena竞技场里",
      "source": "source_caption",
      "text_original": "最近在Lrmina竞技场里",
      "correction_applied": true,
      "correction_source": "llm_correct_a_presegmentation"
    },
    {
      "start": 429.166,
      "end": 431.466,
      "text": "也突然低调上线了一个代号叫",
      "source": "source_caption"
    },
    {
      "start": 431.466,
      "end": 433.9,
      "text": "蒙娜丽莎1的全新图像大模型",
      "source": "source_caption",
      "text_original": "蒙娜丽莎一的全新图像大模型",
      "correction_applied": true,
      "correction_source": "llm_correct_a_presegmentation"
    },
    {
      "start": 434.133,
      "end": 435.633,
      "text": "表现同样非常惹眼",
      "source": "source_caption"
    },
    {
      "start": 435.633,
      "end": 437.566,
      "text": "而要说最锤的铁证",
      "source": "source_caption"
    },
    {
      "start": 437.7,
      "end": 440.633,
      "text": "就是用Openai官方的验证工具去",
      "source": "source_caption"
    },
    {
      "start": 440.633,
      "end": 443.366,
      "text": "检测蒙娜丽莎一跑出来的图片",
      "source": "source_caption"
    },
    {
      "start": 443.7,
      "end": 446.266,
      "text": "系统居然直接识别出了Openai",
      "source": "source_caption"
    },
    {
      "start": 446.266,
      "end": 449.133,
      "text": "自家专属的seed ID隐形水印",
      "source": "source_caption"
    },
    {
      "start": 449.366,
      "end": 451.3,
      "text": "这下可以说是当场破案了",
      "source": "source_caption"
    },
    {
      "start": 451.3,
      "end": 453.466,
      "text": "这就意味着Openai在图像",
      "source": "source_caption"
    },
    {
      "start": 453.466,
      "end": 455.566,
      "text": "生成领域也要掏出底牌了",
      "source": "source_caption"
    },
    {
      "start": 455.766,
      "end": 458.233,
      "text": "无论接下来叫ChatGPT image",
      "source": "source_caption"
    },
    {
      "start": 458.233,
      "end": 460.533,
      "text": "2.5还是直接升到3.0",
      "source": "source_caption"
    },
    {
      "start": 460.7,
      "end": 462.566,
      "text": "画质与控图能力都将",
      "source": "source_caption"
    },
    {
      "start": 462.566,
      "end": 464.733,
      "text": "迎来脱胎换骨般的质飞跃",
      "source": "source_caption"
    },
    {
      "start": 464.733,
      "end": 467.633,
      "text": "不过就在各大厂商猛砸算力的时候",
      "source": "source_caption"
    },
    {
      "start": 467.833,
      "end": 470.166,
      "text": "谷歌这边却传来了一条坏消息",
      "source": "source_caption"
    },
    {
      "start": 470.533,
      "end": 472.833,
      "text": "此前被开源社区寄予厚望的",
      "source": "source_caption"
    },
    {
      "start": 472.833,
      "end": 476.066,
      "text": "超大开源模型Gemma V4 120B",
      "source": "source_caption",
      "text_original": "超大开源模型Gemma V41,200亿",
      "correction_applied": true,
      "correction_source": "llm_correct_a_presegmentation"
    },
    {
      "start": 476.133,
      "end": 479.133,
      "text": "最新爆料显示已被官方彻底砍掉",
      "source": "source_caption"
    },
    {
      "start": 479.133,
      "end": 481.333,
      "text": "根据泄露的后台界面显示",
      "source": "source_caption"
    },
    {
      "start": 481.5,
      "end": 485.233,
      "text": "这款原本拥有1,230亿参数以及",
      "source": "source_caption"
    },
    {
      "start": 485.533,
      "end": 489.1,
      "text": "262k超百万级上下文窗口的开源旗舰",
      "source": "source_caption"
    },
    {
      "start": 489.1,
      "end": 492.3,
      "text": "大模型已经被标记为了取消状态",
      "source": "source_caption"
    },
    {
      "start": 492.466,
      "end": 494.766,
      "text": "确定不会再面向公众发布了",
      "source": "source_caption"
    },
    {
      "start": 494.766,
      "end": 497.7,
      "text": "这也侧面印证了谷歌在开源与闭源",
      "source": "source_caption"
    },
    {
      "start": 497.7,
      "end": 500.366,
      "text": "战略之间正在经历剧烈的阵痛调整",
      "source": "source_caption"
    },
    {
      "start": 500.7,
      "end": 502.266,
      "text": "他们很可能是想把",
      "source": "source_caption"
    },
    {
      "start": 502.266,
      "end": 503.966,
      "text": "全部计算资源集中起来",
      "source": "source_caption"
    },
    {
      "start": 503.966,
      "end": 506.433,
      "text": "全力冲刺下一代闭源Gemini",
      "source": "source_caption"
    },
    {
      "start": 506.7,
      "end": 507.366,
      "text": "不得不说",
      "source": "source_caption"
    },
    {
      "start": 507.366,
      "end": 508.666,
      "text": "真挺令人唏嘘的",
      "source": "source_caption"
    },
    {
      "start": 508.666,
      "end": 510.5,
      "text": "而除了Openai与谷歌",
      "source": "source_caption"
    },
    {
      "start": 510.566,
      "end": 513.1,
      "text": "马斯克的Grok也不甘示弱",
      "source": "source_caption"
    },
    {
      "start": 513.233,
      "end": 514.233,
      "text": "咱们一起来看看",
      "source": "source_caption"
    },
    {
      "start": 514.233,
      "end": 516.1,
      "text": "这些精美的生成范例",
      "source": "source_caption"
    },
    {
      "start": 516.633,
      "end": 518.033,
      "text": "无论是波普风海报",
      "source": "source_caption"
    },
    {
      "start": 518.166,
      "end": 519.5,
      "text": "细腻纺织纹理",
      "source": "source_caption"
    },
    {
      "start": 519.666,
      "end": 521.466,
      "text": "还是素描步骤拆解",
      "source": "source_caption"
    },
    {
      "start": 521.733,
      "end": 524.166,
      "text": "Grok image 2.0在文字渲染和",
      "source": "source_caption"
    },
    {
      "start": 524.166,
      "end": 526.433,
      "text": "审美排版上都极其出色",
      "source": "source_caption"
    },
    {
      "start": 526.433,
      "end": 527.266,
      "text": "按照惯例",
      "source": "source_caption"
    },
    {
      "start": 527.333,
      "end": 529.633,
      "text": "看完软件层的神仙打架",
      "source": "source_caption"
    },
    {
      "start": 529.666,
      "end": 532.333,
      "text": "咱们在结尾再来看一个极其",
      "source": "source_caption"
    },
    {
      "start": 532.333,
      "end": 535.1,
      "text": "硬核的具身智能硬件项目",
      "source": "source_caption"
    },
    {
      "start": 535.366,
      "end": 538.166,
      "text": "东京大学最新展示了一款名为",
      "source": "source_caption"
    },
    {
      "start": 538.566,
      "end": 541.5,
      "text": "Dragon的空中变形飞行机器人",
      "source": "source_caption"
    },
    {
      "start": 541.566,
      "end": 543.966,
      "text": "它最令人惊叹的地方就是彻底",
      "source": "source_caption"
    },
    {
      "start": 543.966,
      "end": 546.966,
      "text": "打破了传统无人机固定的形态限制",
      "source": "source_caption"
    },
    {
      "start": 547.266,
      "end": 549.166,
      "text": "能在空中完成极度",
      "source": "source_caption"
    },
    {
      "start": 549.166,
      "end": 551.166,
      "text": "复杂的自由变体运动",
      "source": "source_caption"
    },
    {
      "start": 551.266,
      "end": 552.266,
      "text": "大家仔细看",
      "source": "source_caption"
    },
    {
      "start": 552.466,
      "end": 554.9,
      "text": "Dragon机器人在空中飞行时",
      "source": "source_caption",
      "text_original": "Dirogan机器人在空中飞行时",
      "correction_applied": true,
      "correction_source": "llm_correct_a_presegmentation"
    },
    {
      "start": 554.9,
      "end": 556.933,
      "text": "能精准改变关节构形",
      "source": "source_caption"
    },
    {
      "start": 557.133,
      "end": 559.3,
      "text": "甚至能用身体灵活抓取",
      "source": "source_caption"
    },
    {
      "start": 559.333,
      "end": 561.333,
      "text": "并包裹移动悬挂物体",
      "source": "source_caption"
    },
    {
      "start": 561.666,
      "end": 563.833,
      "text": "在结合了AI控制算法后",
      "source": "source_caption"
    },
    {
      "start": 563.866,
      "end": 566.266,
      "text": "它生动展示了具身智能在",
      "source": "source_caption"
    },
    {
      "start": 566.266,
      "end": 568.233,
      "text": "复杂空间下的无限潜力",
      "source": "source_caption"
    },
    {
      "start": 568.3,
      "end": 570.933,
      "text": "从底层大模型到具身智能",
      "source": "source_caption"
    },
    {
      "start": 571.066,
      "end": 573.766,
      "text": "AI的进化爆发正全面到来",
      "source": "source_caption"
    },
    {
      "start": 574.033,
      "end": 575.166,
      "text": "感谢你看到这里",
      "source": "source_caption"
    },
    {
      "start": 575.166,
      "end": 576.666,
      "text": "这里是大雷早上好",
      "source": "source_caption"
    },
    {
      "start": 576.666,
      "end": 577.533,
      "text": "关注频道",
      "source": "source_caption"
    },
    {
      "start": 577.533,
      "end": 578.633,
      "text": "咱们下期见",
      "source": "source_caption"
    }
  ],
  "transcript_selection": {
    "schema": "videonote.transcript_selection.v1",
    "status": "SELECTED",
    "requested_language": "zh",
    "duration": 583.633563,
    "reference_path": "/Users/sagawa/AI_WORK/video_notes/url/20260812_092935__比GPT-6更可怕_OpenAI秘密模型Doug曝光_AI圈炸裂_Scarier_Than_GPT-6_OpenAI_s_Secret_Model_Doug_L/reference_subtitle.vtt",
    "reference_kind": "manual",
    "reference_language": "zh",
    "selected_source": "source_with_whisper_gap_repair",
    "decision_reason": "trusted_partial_source_with_clean_whisper_gap_repair",
    "whisper_rejected": false,
    "source_profile": {
      "segment_count": 257,
      "multilingual": false,
      "first_start": 0.1,
      "final_end": 578.633,
      "duration_coverage": 0.991432,
      "maximum_gap_seconds": 0.533,
      "inverted_count": 0,
      "language": {
        "requested": "zh",
        "matched": true,
        "latin": 537,
        "han": 2332,
        "kana": 0,
        "hangul": 0
      },
      "hallucination_region_count": 0,
      "hallucination_regions": [],
      "usable_full": true,
      "usable_partial": true
    },
    "whisper_profile": {
      "segment_count": 74,
      "multilingual": false,
      "first_start": 0.0,
      "final_end": 578.26,
      "duration_coverage": 0.990793,
      "maximum_gap_seconds": 0.7,
      "inverted_count": 0,
      "language": {
        "requested": "zh",
        "matched": true,
        "latin": 570,
        "han": 2303,
        "kana": 0,
        "hangul": 0
      },
      "hallucination_region_count": 0,
      "hallucination_regions": [],
      "usable_full": true,
      "usable_partial": true
    },
    "comparison": {
      "window_seconds": 30.0,
      "compared_window_count": 20,
      "median_score": 0.8735,
      "divergent_window_count": 0,
      "divergent_ratio": 0.0,
      "divergent_windows": []
    },
    "timestamp_alignment": {
      "applied": false,
      "estimated_offset_seconds": 0.0,
      "matched_segment_count": 21,
      "match_scores": [
        0.6141,
        0.7359,
        0.6781,
        0.6795,
        0.6452,
        0.6875,
        0.6207,
        0.7097,
        0.7551,
        0.6857,
        0.6471,
        1.0,
        0.6072,
        0.6733,
        0.9091,
        1.0,
        1.0,
        0.674,
        0.674,
        0.6941,
        0.6364,
        0.6957,
        0.6048,
        0.651,
        0.6366,
        0.6622,
        0.6397,
        0.6566,
        0.6976,
        0.674,
        0.6977,
        0.6341,
        0.672,
        0.6147,
        0.6667,
        0.7011,
        0.6483,
        0.6822,
        0.6011,
        0.6863
      ]
    },
    "filled_source_profile": {
      "segment_count": 237,
      "multilingual": false,
      "first_start": 0.1,
      "final_end": 578.633,
      "duration_coverage": 0.991432,
      "maximum_gap_seconds": 0.533,
      "inverted_count": 0,
      "language": {
        "requested": "zh",
        "matched": true,
        "latin": 539,
        "han": 2366,
        "kana": 0,
        "hangul": 0
      },
      "hallucination_region_count": 0,
      "hallucination_regions": [],
      "usable_full": true,
      "usable_partial": true
    },
    "whisper_duration_guard": {
      "media_duration": 583.634,
      "rejected_segment_count": 1,
      "rejected_segments": [
        {
          "index": 74,
          "start": 581.04,
          "end": 611.02,
          "reason": "material_end_timestamp_overrun"
        }
      ]
    }
  },
  "presegmentation_correction": {
    "applied": true,
    "correction_count": 25
  }
}