{ "text": "更多时候,人民需要的是足够强,同时便宜、稳定、能部署,并且不会因为地区限制、产品分层或者政策变化突然断供的模型。\nHello大家好,我是Jacky。最近Anthropic,也就是Claude背后的公司,被曝在澳洲寻找大规模数据中心。\n它已经和澳大利亚政府签了AI的合作文件。据媒体报道,Anthropic还在接触大约1.4吉瓦级别的数据中心容量。\n这个新闻其实放在日新月异的AI圈很容易被掩埋\n但我觉得它真正有意义\n是把AI竞争从模型榜单拉回了物理世界\nAnthropic是一个最会写代码写文章做推理的AI公司\n但它下一个最关心的东西\n开始变成了哪里有地哪里有电\n哪里能接上数据中心\n过去我们总是在问谁的模型更加聪明\n谁的推理能力更强\n但随着AI的发展\n中美都发现\n如果AI最后变成一种\n可以被大规模部署\n持续调用\n按token计价的能力\n那竞争的范围就远不止是模型本身\n而是谁能把这种能力\n以低成本稳定大规模的供给出去\n那如果把模型\nAPI和智能体\n看成是AI公司卖出去的产品\n那背后的生产函数很清晰\n芯片是机器\n数据中心是厂房\n算法是工艺\n数据是配方\n而电力就是持续消耗的生产资料\n虚拟一个模型\n本质上是在把电\n芯片时间\n数据和算法工艺\n压缩成一组模型权重\n那么这个模型每推理一次\n本质上就是在把电和算力\n持续换成一个又一个token\n所以Anthropic去找数据中心\n找电力\n这件事它一点都不奇怪\n我们之前太习惯\n把AI想象成云上的魔法了\n好像模型只要足够聪明\n它就能无限复制\n无限扩张\n但现实根本不是这个样子\nIEA的报告里有个很直观的数字\n全球数据中心用电到2030年预计会翻倍\n到大约945泰瓦时\n美国能源部LBNL的报道也提到过\n美国数据中心在2023年\n已经消耗了美国大约4.4%的总用电量\n到了2028年可能会上升至6.7%到12%\n你再看Answerpick自己今年2月份的公开承诺\n他说愿意承担自家数据中心带来的消费者定价上涨\n还愿意支付电网升级的成本\n那以前的软件公司扩张用户多了\n无非就是服务器带宽云成本上去一点\n而现在一个前沿模型公司的扩张\n可能会直接改变一个市一个州\n甚至一个电网区域的复合结构\n那如果你在审视中美AI发展的时候\n带上这个视角\n那中美AI的差异就不只是谁的模型能力更强了\n而是谁更能把模型能力做成长期低成本稳定可用的供给\n先来说美国\n美国在AI领域确实是领跑者\nOpenAI\nAnswerpick\n谷歌这些公司仍然在定义最前沿必源模型的节奏\n更重要的是\n美国它拥有一整套前沿AI的产业配套\n英伟达\nAMD这样的芯片公司\nAWS\nGoogle Cloud这样的云平台\n包括成熟的开发者生态\n以及愿意持续砸大钱的资本市场\nStanford AI Index 2026 里也能看到\n美国在有影响力模型的数量和私人 AI 投资上仍然非常领先\n2025 年美国机构产生了59个有影响力的模型\n中国是35个\n私人 AI 投资这一块美国也明显是高得多\n所以如果你只是问最强的闭源模型是谁做出来的\n那我觉得在很长一段时间里\n这个答案大概率还是美国\n不过这条路线的代价也日渐凸显\n伯克利Lab今年的并网电力数据\n很能说明这个问题\n到2025年底\n美国有超过2060吉瓦的发电和储能项目\n在排队接入电网\n注意啊\n这不是已经能用的电\n而是还在排队等着接入电网的项目\n而在过去\n2000年到2020年申请进入电网的容量\n你到了2025年真正商业运营的只有13%\n美国的基础建设效率可见一斑\n真的是太慢了\n这个就是美国AI的现实处境\n它不是没有模型\n也不是完全见不到电力\n而是前沿AI的负载太大\n增长太快\n开始和美国分散的电网体系\n漫长的审批周期\n地方电价争议\n包括输电瓶颈发生冲突\n美国的AI公司很快可以出到钱\n可以买到芯片\n但基础设施的建设速度\n不会像模型版本一样\n几个月就迭代一次\n而是按照年甚至几年为单位的\n而中国的AI路线\n我认为很可能不会复制美国这条路\n中国的AI方面当然有巨大的差距\n尤其是高端GPU\nHBM\n先进制程\n云上高端算力的供给\n这些都是硬约束\n但中国模型这两年的进步\n有一个非常明显的特点\n如果美国的顶级大模型是100分\n那中国厂商在不断的\n把85分90分的开源模型能力\n做得明显更便宜\n更开放\n更容易部署\n还是斯坦福AI Index 2025\n有一个观点我觉得很值得看\n中美顶级模型的能力差距\n已经明显缩小了\n2025年以来\n中美模型多次交换领先位置\n他甚至提到了\n截至2026年3月\n美国顶级模型\n相对中国顶级模型的领先\n大约只有2.7%\n当然这个2.7%\n不是说中国模型\n已经全面追上美国了\n那不同榜单侧的东西也不一样\n但它至少说明了一件事\n就是中美的模型能力\n已经不是三年前那种一眼看不到边的差距了\n那如果你再看开放权重模型\nArtificial Analyst今年6月的文章里\nGLM 5.2在它的Intelligent Index上\n成为领先的开放权重模型之一\n并且在一些企业任务评测里\n它已经进入了高能力低成本的区间\n这个例子能说明一个方向\n就是中国模型不只是追赶\n它是在把接近前沿模型的能力\n用更开放更低成本的方式来释放出来\n而这条路线比很多人想象的重要\n因为市场并不总是只买最强的\n很多企业开发者出海场景\n其实不需要世界第一聪明的模型\n前段时间你但凡狠狠调过\n非GPT模型的人应该会有同感\n那个token的消耗真的可怕\n成本真的很高\n更多时候人们需要的是足够强\n同时便宜稳定能部署\n并且不会因为地区限制\n产品分层或者政策变化\n突然断供的模型\n这个就是中国开放模型的长期价值\n这不是低端替代\n而是中国模型更能够把AI能力去产业化和规模化\n而一旦你把AI看成工业化的供给\n那电力和基础设施就变成了中国这边非常重要的一张牌\n国家能源局2024年的数据很夸张\n2025年中国可再生能源新增装机4.52亿千瓦\n占全国电力新增装机的83%\n其中太阳能新增3.18亿千瓦\n风力新增1.2亿千瓦\n到2025年底\n中国可再生能源的装机会达到23.4亿千瓦\n大约占全国总装机的60%\n所以能感知到中国在能源\n所以能感知到,中國在能源和基礎設施上的動員能力非常的強。\n那隨著模型能力越來越強,訓練新模型肯定是仍然重要的。\n但推理過程的成本會變成AI發展的關鍵變量。\n因为只要AI真正进入企业办公教育等等越来越多的场景\n那AI公司未来最大的账单很可能不是训练一个模型要花多少钱\n而是每天几十亿几百亿次的调用单位成本能不能压下来\nToken听起来抽象\n但它就是AI能力被调用被计费被规模化交付的最小单位\n未来AI如果真的变成了基础设施\n那谁能把每一次Token的成本压下来\n谁就能让AI覆盖更多场景\n这也是为什么我会觉得\n未来中美AI很可能不是一个\nB元打败开元或者开元打败B元的故事\n更可能是两套市场长期并存\n最顶级的B元模型\n继续服务于科研复杂推理和最前沿的agent\n它们会很贵\n会分层\n会有访问门槛\n同时也会继续代表模型能力的天花板\n但更大的市场\n可能是企业内部的流程\n自由化的部署\n本地化的应用\n一些相对低成本的智能体\n大规模的推理和调用\n这里要的它不是世界第一聪明\n而是够强稳定便宜可控\nClaude最大的问题就在这里\n它不是不强\n但如果一个模型公司依赖的是最强模型\n可以一直卖出稀缺溢价\n它就会不断的被重新定价\n被OpenAI的下一代闭源模型重新定价\n被DeepSeek这样便宜量大的开源模型重新定价\n被中国一批又一批开源模型重新定价\n你最强当然是有很大价值的\n但如果别人用便宜很多的价格\n做到你八九成的能力\n还能私有部署\n还不受你这套访问规则的约束\n那很多真实的生产场景\n就会开始重新算账了\n那这个账一算\nAI竞争就已经从互联网逻辑\n进入了能源工业的逻辑\n而在能源工业的逻辑里\n最重要的是产能成本和供应稳定性\n所以Claude去欧洲找电这件事\n表面看是一个AI公司\n要扩数据中心\n但往深里看\n它是在提醒我们\n模型并不是浮在云上的魔法\nAI公司卖的是智能\n但它每天消耗的是电\n抢的是芯片\n租的是实实在在的厂房\n所以我认为中美的AI竞争\n不能只看谁的模型最聪明\n还要结合访问限制\n单位Token的价格\n还有要观察\n应用市场是不是真的愿意为最强模型\n付出高昂代价\n我认为从0到1的创新\n和从1到10的落地\n这两条路的竞争\n也许会一直持续下去\nOK\n那本期视频的主要内容就这么多了\n大家伙日常在用哪个模型最多\n也可以打在评论或者弹幕上\n我们分享一下\n如果喜欢本期视频的话\n请到点赞朋友订阅我的话\n我是HYK\n我们下期视频见\n祝你好运\n拜拜", "language": "zh", "duration": 627.733313, "segments": [ { "start": 0.0, "end": 9.0, "text": "更多时候,人民需要的是足够强,同时便宜、稳定、能部署,并且不会因为地区限制、产品分层或者政策变化突然断供的模型。", "chunk": 0, "language": "zh", "avg_logprob": -0.16780263426675962, "no_speech_prob": 8.249408968430405e-12, "compression_ratio": 1.1830985915492958, "source": "whisper", "text_original": "更多时候,人民需要的是足够强,同时便宜、稳定、能部署,并且不会因为地区限制、产品分层或者政策变化突然断工的模型。", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 9.0, "end": 16.5, "text": "Hello大家好,我是Jacky。最近Anthropic,也就是Claude背后的公司,被曝在澳洲寻找大规模数据中心。", "chunk": 0, "language": "zh", "avg_logprob": -0.16780263426675962, "no_speech_prob": 8.249408968430405e-12, "compression_ratio": 1.1830985915492958, "source": "whisper", "text_original": "Hello大家好,我是Jacky。最近Answerpik,也就是Cloud背后的公司,被曝在澳洲寻找大规模数据中心。", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 16.5, "end": 26.0, "text": "它已经和澳大利亚政府签了AI的合作文件。据媒体报道,Anthropic还在接触大约1.4吉瓦级别的数据中心容量。", "chunk": 0, "language": "zh", "avg_logprob": -0.16780263426675962, "no_speech_prob": 8.249408968430405e-12, "compression_ratio": 1.1830985915492958, "source": "whisper", "text_original": "它已经和澳大利亚政府签了AI的合作文件。据媒体报道,Answerpik还在接触大约1.4级瓦级别的数据中心容量。", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 26.0, "end": 29.76, "text": "这个新闻其实放在日新月异的AI圈很容易被掩埋", "chunk": 0, "language": "zh", "avg_logprob": -0.10323255202349495, "no_speech_prob": 1.5933372676801838e-11, "compression_ratio": 1.2879256965944272, "source": "whisper", "text_original": "这个新闻其实放在日新月异的AI全很容易被掩埋", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 29.76, "end": 31.46, "text": "但我觉得它真正有意义", "chunk": 0, "language": "zh", "avg_logprob": -0.10323255202349495, "no_speech_prob": 1.5933372676801838e-11, "compression_ratio": 1.2879256965944272, "source": "whisper", "text_original": "但我觉得它真正有一次地方", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 31.46, "end": 35.54, "text": "是把AI竞争从模型榜单拉回了物理世界", "chunk": 0, "language": "zh", "avg_logprob": -0.10323255202349495, "no_speech_prob": 1.5933372676801838e-11, "compression_ratio": 1.2879256965944272, "source": "whisper" }, { "start": 35.54, "end": 39.74, "text": "Anthropic是一个最会写代码写文章做推理的AI公司", "chunk": 0, "language": "zh", "avg_logprob": -0.10323255202349495, "no_speech_prob": 1.5933372676801838e-11, "compression_ratio": 1.2879256965944272, "source": "whisper", "text_original": "Anthorpik是一个最会写代码写文章做推理的AI公司", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 39.74, "end": 41.7, "text": "但它下一个最关心的东西", "chunk": 0, "language": "zh", "avg_logprob": -0.10323255202349495, "no_speech_prob": 1.5933372676801838e-11, "compression_ratio": 1.2879256965944272, "source": "whisper" }, { "start": 41.7, "end": 43.86, "text": "开始变成了哪里有地哪里有电", "chunk": 0, "language": "zh", "avg_logprob": -0.10323255202349495, "no_speech_prob": 1.5933372676801838e-11, "compression_ratio": 1.2879256965944272, "source": "whisper" }, { "start": 43.86, "end": 45.72, "text": "哪里能接上数据中心", "chunk": 0, "language": "zh", "avg_logprob": -0.10323255202349495, "no_speech_prob": 1.5933372676801838e-11, "compression_ratio": 1.2879256965944272, "source": "whisper" }, { "start": 45.72, "end": 49.02, "text": "过去我们总是在问谁的模型更加聪明", "chunk": 0, "language": "zh", "avg_logprob": -0.10323255202349495, "no_speech_prob": 1.5933372676801838e-11, "compression_ratio": 1.2879256965944272, "source": "whisper" }, { "start": 49.02, "end": 50.38, "text": "谁的推理能力更强", "chunk": 0, "language": "zh", "avg_logprob": -0.10323255202349495, "no_speech_prob": 1.5933372676801838e-11, "compression_ratio": 1.2879256965944272, "source": "whisper" }, { "start": 50.38, "end": 51.76, "text": "但随着AI的发展", "chunk": 0, "language": "zh", "avg_logprob": -0.10323255202349495, "no_speech_prob": 1.5933372676801838e-11, "compression_ratio": 1.2879256965944272, "source": "whisper" }, { "start": 51.76, "end": 53.0, "text": "中美都发现", "chunk": 0, "language": "zh", "avg_logprob": -0.10323255202349495, "no_speech_prob": 1.5933372676801838e-11, "compression_ratio": 1.2879256965944272, "source": "whisper" }, { "start": 53.0, "end": 55.16, "text": "如果AI最后变成一种", "chunk": 0, "language": "zh", "avg_logprob": -0.07292562150685801, "no_speech_prob": 1.098922862430296e-11, "compression_ratio": 1.2577319587628866, "source": "whisper" }, { "start": 55.16, "end": 56.6, "text": "可以被大规模部署", "chunk": 0, "language": "zh", "avg_logprob": -0.07292562150685801, "no_speech_prob": 1.098922862430296e-11, "compression_ratio": 1.2577319587628866, "source": "whisper" }, { "start": 56.6, "end": 57.74, "text": "持续调用", "chunk": 0, "language": "zh", "avg_logprob": -0.07292562150685801, "no_speech_prob": 1.098922862430296e-11, "compression_ratio": 1.2577319587628866, "source": "whisper" }, { "start": 57.74, "end": 59.32, "text": "按token计价的能力", "chunk": 0, "language": "zh", "avg_logprob": -0.07292562150685801, "no_speech_prob": 1.098922862430296e-11, "compression_ratio": 1.2577319587628866, "source": "whisper" }, { "start": 59.32, "end": 62.88, "text": "那竞争的范围就远不止是模型本身", "chunk": 0, "language": "zh", "avg_logprob": -0.07292562150685801, "no_speech_prob": 1.098922862430296e-11, "compression_ratio": 1.2577319587628866, "source": "whisper" }, { "start": 62.88, "end": 64.92, "text": "而是谁能把这种能力", "chunk": 0, "language": "zh", "avg_logprob": -0.07292562150685801, "no_speech_prob": 1.098922862430296e-11, "compression_ratio": 1.2577319587628866, "source": "whisper" }, { "start": 64.92, "end": 68.06, "text": "以低成本稳定大规模的供给出去", "chunk": 0, "language": "zh", "avg_logprob": -0.07292562150685801, "no_speech_prob": 1.098922862430296e-11, "compression_ratio": 1.2577319587628866, "source": "whisper" }, { "start": 68.06, "end": 69.48, "text": "那如果把模型", "chunk": 0, "language": "zh", "avg_logprob": -0.07292562150685801, "no_speech_prob": 1.098922862430296e-11, "compression_ratio": 1.2577319587628866, "source": "whisper" }, { "start": 69.48, "end": 70.86, "text": "API和智能体", "chunk": 0, "language": "zh", "avg_logprob": -0.07292562150685801, "no_speech_prob": 1.098922862430296e-11, "compression_ratio": 1.2577319587628866, "source": "whisper" }, { "start": 70.86, "end": 73.28, "text": "看成是AI公司卖出去的产品", "chunk": 0, "language": "zh", "avg_logprob": -0.07292562150685801, "no_speech_prob": 1.098922862430296e-11, "compression_ratio": 1.2577319587628866, "source": "whisper" }, { "start": 73.28, "end": 75.62, "text": "那背后的生产函数很清晰", "chunk": 0, "language": "zh", "avg_logprob": -0.07292562150685801, "no_speech_prob": 1.098922862430296e-11, "compression_ratio": 1.2577319587628866, "source": "whisper" }, { "start": 75.62, "end": 76.64, "text": "芯片是机器", "chunk": 0, "language": "zh", "avg_logprob": -0.07292562150685801, "no_speech_prob": 1.098922862430296e-11, "compression_ratio": 1.2577319587628866, "source": "whisper" }, { "start": 76.64, "end": 78.32, "text": "数据中心是厂房", "chunk": 0, "language": "zh", "avg_logprob": -0.07292562150685801, "no_speech_prob": 1.098922862430296e-11, "compression_ratio": 1.2577319587628866, "source": "whisper" }, { "start": 78.32, "end": 79.48, "text": "算法是工艺", "chunk": 0, "language": "zh", "avg_logprob": -0.07292562150685801, "no_speech_prob": 1.098922862430296e-11, "compression_ratio": 1.2577319587628866, "source": "whisper" }, { "start": 79.48, "end": 80.54, "text": "数据是配方", "chunk": 0, "language": "zh", "avg_logprob": -0.07292562150685801, "no_speech_prob": 1.098922862430296e-11, "compression_ratio": 1.2577319587628866, "source": "whisper" }, { "start": 80.54, "end": 84.08, "text": "而电力就是持续消耗的生产资料", "chunk": 0, "language": "zh", "avg_logprob": -0.1082883005991032, "no_speech_prob": 1.468449493946533e-11, "compression_ratio": 1.3219814241486068, "source": "whisper" }, { "start": 84.08, "end": 85.14, "text": "虚拟一个模型", "chunk": 0, "language": "zh", "avg_logprob": -0.1082883005991032, "no_speech_prob": 1.468449493946533e-11, "compression_ratio": 1.3219814241486068, "source": "whisper" }, { "start": 85.14, "end": 86.7, "text": "本质上是在把电", "chunk": 0, "language": "zh", "avg_logprob": -0.1082883005991032, "no_speech_prob": 1.468449493946533e-11, "compression_ratio": 1.3219814241486068, "source": "whisper" }, { "start": 86.7, "end": 87.74, "text": "芯片时间", "chunk": 0, "language": "zh", "avg_logprob": -0.1082883005991032, "no_speech_prob": 1.468449493946533e-11, "compression_ratio": 1.3219814241486068, "source": "whisper" }, { "start": 87.74, "end": 89.28, "text": "数据和算法工艺", "chunk": 0, "language": "zh", "avg_logprob": -0.1082883005991032, "no_speech_prob": 1.468449493946533e-11, "compression_ratio": 1.3219814241486068, "source": "whisper" }, { "start": 89.28, "end": 91.5, "text": "压缩成一组模型权重", "chunk": 0, "language": "zh", "avg_logprob": -0.1082883005991032, "no_speech_prob": 1.468449493946533e-11, "compression_ratio": 1.3219814241486068, "source": "whisper" }, { "start": 91.5, "end": 93.46, "text": "那么这个模型每推理一次", "chunk": 0, "language": "zh", "avg_logprob": -0.1082883005991032, "no_speech_prob": 1.468449493946533e-11, "compression_ratio": 1.3219814241486068, "source": "whisper" }, { "start": 93.46, "end": 95.64, "text": "本质上就是在把电和算力", "chunk": 0, "language": "zh", "avg_logprob": -0.1082883005991032, "no_speech_prob": 1.468449493946533e-11, "compression_ratio": 1.3219814241486068, "source": "whisper" }, { "start": 95.64, "end": 97.82, "text": "持续换成一个又一个token", "chunk": 0, "language": "zh", "avg_logprob": -0.1082883005991032, "no_speech_prob": 1.468449493946533e-11, "compression_ratio": 1.3219814241486068, "source": "whisper" }, { "start": 97.82, "end": 100.14, "text": "所以Anthropic去找数据中心", "chunk": 0, "language": "zh", "avg_logprob": -0.1082883005991032, "no_speech_prob": 1.468449493946533e-11, "compression_ratio": 1.3219814241486068, "source": "whisper", "text_original": "所以Answerpik去找数据中心", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 100.14, "end": 100.78, "text": "找电力", "chunk": 0, "language": "zh", "avg_logprob": -0.1082883005991032, "no_speech_prob": 1.468449493946533e-11, "compression_ratio": 1.3219814241486068, "source": "whisper" }, { "start": 100.78, "end": 102.2, "text": "这件事它一点都不奇怪", "chunk": 0, "language": "zh", "avg_logprob": -0.1082883005991032, "no_speech_prob": 1.468449493946533e-11, "compression_ratio": 1.3219814241486068, "source": "whisper" }, { "start": 102.2, "end": 103.7, "text": "我们之前太习惯", "chunk": 0, "language": "zh", "avg_logprob": -0.1082883005991032, "no_speech_prob": 1.468449493946533e-11, "compression_ratio": 1.3219814241486068, "source": "whisper" }, { "start": 103.7, "end": 106.2, "text": "把AI想象成云上的魔法了", "chunk": 0, "language": "zh", "avg_logprob": -0.1082883005991032, "no_speech_prob": 1.468449493946533e-11, "compression_ratio": 1.3219814241486068, "source": "whisper" }, { "start": 106.2, "end": 108.0, "text": "好像模型只要足够聪明", "chunk": 0, "language": "zh", "avg_logprob": -0.1082883005991032, "no_speech_prob": 1.468449493946533e-11, "compression_ratio": 1.3219814241486068, "source": "whisper" }, { "start": 108.0, "end": 109.36, "text": "它就能无限复制", "chunk": 0, "language": "zh", "avg_logprob": -0.1082883005991032, "no_speech_prob": 1.468449493946533e-11, "compression_ratio": 1.3219814241486068, "source": "whisper" }, { "start": 109.36, "end": 110.12, "text": "无限扩张", "chunk": 0, "language": "zh", "avg_logprob": -0.1082883005991032, "no_speech_prob": 1.468449493946533e-11, "compression_ratio": 1.3219814241486068, "source": "whisper" }, { "start": 110.12, "end": 112.4, "text": "但现实根本不是这个样子", "chunk": 0, "language": "zh", "avg_logprob": -0.09773959257663825, "no_speech_prob": 1.520640557917119e-11, "compression_ratio": 1.1627906976744187, "source": "whisper" }, { "start": 112.4, "end": 114.56, "text": "IEA的报告里有个很直观的数字", "chunk": 0, "language": "zh", "avg_logprob": -0.09773959257663825, "no_speech_prob": 1.520640557917119e-11, "compression_ratio": 1.1627906976744187, "source": "whisper" }, { "start": 114.56, "end": 118.56, "text": "全球数据中心用电到2030年预计会翻倍", "chunk": 0, "language": "zh", "avg_logprob": -0.09773959257663825, "no_speech_prob": 1.520640557917119e-11, "compression_ratio": 1.1627906976744187, "source": "whisper" }, { "start": 118.56, "end": 121.4, "text": "到大约945泰瓦时", "chunk": 0, "language": "zh", "avg_logprob": -0.09773959257663825, "no_speech_prob": 1.520640557917119e-11, "compression_ratio": 1.1627906976744187, "source": "whisper" }, { "start": 121.4, "end": 124.54, "text": "美国能源部LBNL的报道也提到过", "chunk": 0, "language": "zh", "avg_logprob": -0.09773959257663825, "no_speech_prob": 1.520640557917119e-11, "compression_ratio": 1.1627906976744187, "source": "whisper" }, { "start": 124.54, "end": 127.14, "text": "美国数据中心在2023年", "chunk": 0, "language": "zh", "avg_logprob": -0.09773959257663825, "no_speech_prob": 1.520640557917119e-11, "compression_ratio": 1.1627906976744187, "source": "whisper" }, { "start": 127.14, "end": 130.94, "text": "已经消耗了美国大约4.4%的总用电量", "chunk": 0, "language": "zh", "avg_logprob": -0.09773959257663825, "no_speech_prob": 1.520640557917119e-11, "compression_ratio": 1.1627906976744187, "source": "whisper" }, { "start": 130.94, "end": 135.58, "text": "到了2028年可能会上升至6.7%到12%", "chunk": 0, "language": "zh", "avg_logprob": -0.09773959257663825, "no_speech_prob": 1.520640557917119e-11, "compression_ratio": 1.1627906976744187, "source": "whisper" }, { "start": 135.58, "end": 139.16, "text": "你再看Answerpick自己今年2月份的公开承诺", "chunk": 0, "language": "zh", "avg_logprob": -0.09773959257663825, "no_speech_prob": 1.520640557917119e-11, "compression_ratio": 1.1627906976744187, "source": "whisper" }, { "start": 139.16, "end": 143.48, "text": "他说愿意承担自家数据中心带来的消费者定价上涨", "chunk": 0, "language": "zh", "avg_logprob": -0.05060821870413933, "no_speech_prob": 1.3622530187218373e-11, "compression_ratio": 1.370253164556962, "source": "whisper" }, { "start": 143.48, "end": 145.96, "text": "还愿意支付电网升级的成本", "chunk": 0, "language": "zh", "avg_logprob": -0.05060821870413933, "no_speech_prob": 1.3622530187218373e-11, "compression_ratio": 1.370253164556962, "source": "whisper" }, { "start": 145.96, "end": 148.82, "text": "那以前的软件公司扩张用户多了", "chunk": 0, "language": "zh", "avg_logprob": -0.05060821870413933, "no_speech_prob": 1.3622530187218373e-11, "compression_ratio": 1.370253164556962, "source": "whisper" }, { "start": 148.82, "end": 152.12, "text": "无非就是服务器带宽云成本上去一点", "chunk": 0, "language": "zh", "avg_logprob": -0.05060821870413933, "no_speech_prob": 1.3622530187218373e-11, "compression_ratio": 1.370253164556962, "source": "whisper" }, { "start": 152.12, "end": 154.86, "text": "而现在一个前沿模型公司的扩张", "chunk": 0, "language": "zh", "avg_logprob": -0.05060821870413933, "no_speech_prob": 1.3622530187218373e-11, "compression_ratio": 1.370253164556962, "source": "whisper" }, { "start": 154.86, "end": 157.38, "text": "可能会直接改变一个市一个州", "chunk": 0, "language": "zh", "avg_logprob": -0.05060821870413933, "no_speech_prob": 1.3622530187218373e-11, "compression_ratio": 1.370253164556962, "source": "whisper" }, { "start": 157.38, "end": 159.84, "text": "甚至一个电网区域的复合结构", "chunk": 0, "language": "zh", "avg_logprob": -0.05060821870413933, "no_speech_prob": 1.3622530187218373e-11, "compression_ratio": 1.370253164556962, "source": "whisper" }, { "start": 159.84, "end": 162.6, "text": "那如果你在审视中美AI发展的时候", "chunk": 0, "language": "zh", "avg_logprob": -0.05060821870413933, "no_speech_prob": 1.3622530187218373e-11, "compression_ratio": 1.370253164556962, "source": "whisper" }, { "start": 162.6, "end": 163.72, "text": "带上这个视角", "chunk": 0, "language": "zh", "avg_logprob": -0.05060821870413933, "no_speech_prob": 1.3622530187218373e-11, "compression_ratio": 1.370253164556962, "source": "whisper" }, { "start": 163.72, "end": 168.0, "text": "那中美AI的差异就不只是谁的模型能力更强了", "chunk": 0, "language": "zh", "avg_logprob": -0.05060821870413933, "no_speech_prob": 1.3622530187218373e-11, "compression_ratio": 1.370253164556962, "source": "whisper" }, { "start": 168.0, "end": 173.28, "text": "而是谁更能把模型能力做成长期低成本稳定可用的供给", "chunk": 0, "language": "zh", "avg_logprob": -0.11657566947733017, "no_speech_prob": 1.473117461348039e-11, "compression_ratio": 1.1899109792284865, "source": "whisper" }, { "start": 173.28, "end": 174.98, "text": "先来说美国", "chunk": 0, "language": "zh", "avg_logprob": -0.11657566947733017, "no_speech_prob": 1.473117461348039e-11, "compression_ratio": 1.1899109792284865, "source": "whisper" }, { "start": 174.98, "end": 177.54, "text": "美国在AI领域确实是领跑者", "chunk": 0, "language": "zh", "avg_logprob": -0.11657566947733017, "no_speech_prob": 1.473117461348039e-11, "compression_ratio": 1.1899109792284865, "source": "whisper" }, { "start": 177.54, "end": 178.26, "text": "OpenAI", "chunk": 0, "language": "zh", "avg_logprob": -0.11657566947733017, "no_speech_prob": 1.473117461348039e-11, "compression_ratio": 1.1899109792284865, "source": "whisper" }, { "start": 178.26, "end": 179.1, "text": "Answerpick", "chunk": 0, "language": "zh", "avg_logprob": -0.11657566947733017, "no_speech_prob": 1.473117461348039e-11, "compression_ratio": 1.1899109792284865, "source": "whisper" }, { "start": 179.1, "end": 183.62, "text": "谷歌这些公司仍然在定义最前沿必源模型的节奏", "chunk": 0, "language": "zh", "avg_logprob": -0.11657566947733017, "no_speech_prob": 1.473117461348039e-11, "compression_ratio": 1.1899109792284865, "source": "whisper" }, { "start": 183.62, "end": 184.7, "text": "更重要的是", "chunk": 0, "language": "zh", "avg_logprob": -0.11657566947733017, "no_speech_prob": 1.473117461348039e-11, "compression_ratio": 1.1899109792284865, "source": "whisper" }, { "start": 184.7, "end": 188.1, "text": "美国它拥有一整套前沿AI的产业配套", "chunk": 0, "language": "zh", "avg_logprob": -0.11657566947733017, "no_speech_prob": 1.473117461348039e-11, "compression_ratio": 1.1899109792284865, "source": "whisper" }, { "start": 188.1, "end": 188.54, "text": "英伟达", "chunk": 0, "language": "zh", "avg_logprob": -0.11657566947733017, "no_speech_prob": 1.473117461348039e-11, "compression_ratio": 1.1899109792284865, "source": "whisper" }, { "start": 188.54, "end": 190.16, "text": "AMD这样的芯片公司", "chunk": 0, "language": "zh", "avg_logprob": -0.11657566947733017, "no_speech_prob": 1.473117461348039e-11, "compression_ratio": 1.1899109792284865, "source": "whisper" }, { "start": 190.16, "end": 190.9, "text": "AWS", "chunk": 0, "language": "zh", "avg_logprob": -0.11657566947733017, "no_speech_prob": 1.473117461348039e-11, "compression_ratio": 1.1899109792284865, "source": "whisper" }, { "start": 190.9, "end": 192.54, "text": "Google Cloud这样的云平台", "chunk": 0, "language": "zh", "avg_logprob": -0.11657566947733017, "no_speech_prob": 1.473117461348039e-11, "compression_ratio": 1.1899109792284865, "source": "whisper" }, { "start": 192.54, "end": 194.2, "text": "包括成熟的开发者生态", "chunk": 0, "language": "zh", "avg_logprob": -0.11657566947733017, "no_speech_prob": 1.473117461348039e-11, "compression_ratio": 1.1899109792284865, "source": "whisper" }, { "start": 194.2, "end": 197.16, "text": "以及愿意持续砸大钱的资本市场", "chunk": 0, "language": "zh", "avg_logprob": -0.11657566947733017, "no_speech_prob": 1.473117461348039e-11, "compression_ratio": 1.1899109792284865, "source": "whisper" }, { "start": 197.16, "end": 199.84, "text": "Stanford AI Index 2026 里也能看到", "chunk": 0, "language": "zh", "avg_logprob": -0.12020343282948369, "no_speech_prob": 1.4004943905965295e-11, "compression_ratio": 1.2546012269938651, "source": "whisper" }, { "start": 199.84, "end": 204.8, "text": "美国在有影响力模型的数量和私人 AI 投资上仍然非常领先", "chunk": 0, "language": "zh", "avg_logprob": -0.12020343282948369, "no_speech_prob": 1.4004943905965295e-11, "compression_ratio": 1.2546012269938651, "source": "whisper" }, { "start": 204.8, "end": 208.84, "text": "2025 年美国机构产生了59个有影响力的模型", "chunk": 0, "language": "zh", "avg_logprob": -0.12020343282948369, "no_speech_prob": 1.4004943905965295e-11, "compression_ratio": 1.2546012269938651, "source": "whisper" }, { "start": 208.84, "end": 210.34, "text": "中国是35个", "chunk": 0, "language": "zh", "avg_logprob": -0.12020343282948369, "no_speech_prob": 1.4004943905965295e-11, "compression_ratio": 1.2546012269938651, "source": "whisper" }, { "start": 210.34, "end": 213.54, "text": "私人 AI 投资这一块美国也明显是高得多", "chunk": 0, "language": "zh", "avg_logprob": -0.12020343282948369, "no_speech_prob": 1.4004943905965295e-11, "compression_ratio": 1.2546012269938651, "source": "whisper" }, { "start": 213.54, "end": 217.1, "text": "所以如果你只是问最强的闭源模型是谁做出来的", "chunk": 0, "language": "zh", "avg_logprob": -0.12020343282948369, "no_speech_prob": 1.4004943905965295e-11, "compression_ratio": 1.2546012269938651, "source": "whisper", "text_original": "所以如果你只是问最强的必源模型是谁做出来的", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 217.1, "end": 219.1, "text": "那我觉得在很长一段时间里", "chunk": 0, "language": "zh", "avg_logprob": -0.12020343282948369, "no_speech_prob": 1.4004943905965295e-11, "compression_ratio": 1.2546012269938651, "source": "whisper" }, { "start": 219.1, "end": 221.12, "text": "这个答案大概率还是美国", "chunk": 0, "language": "zh", "avg_logprob": -0.12020343282948369, "no_speech_prob": 1.4004943905965295e-11, "compression_ratio": 1.2546012269938651, "source": "whisper" }, { "start": 221.12, "end": 223.58, "text": "不过这条路线的代价也日渐凸显", "chunk": 0, "language": "zh", "avg_logprob": -0.12020343282948369, "no_speech_prob": 1.4004943905965295e-11, "compression_ratio": 1.2546012269938651, "source": "whisper" }, { "start": 223.58, "end": 227.02, "text": "伯克利Lab今年的并网电力数据", "chunk": 0, "language": "zh", "avg_logprob": -0.10174260488370569, "no_speech_prob": 1.2955449213425396e-11, "compression_ratio": 1.2822299651567943, "source": "whisper", "text_original": "伯克利Lab今年的并广对立数据", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 227.02, "end": 228.06, "text": "很能说明这个问题", "chunk": 0, "language": "zh", "avg_logprob": -0.10174260488370569, "no_speech_prob": 1.2955449213425396e-11, "compression_ratio": 1.2822299651567943, "source": "whisper" }, { "start": 228.06, "end": 229.52, "text": "到2025年底", "chunk": 0, "language": "zh", "avg_logprob": -0.10174260488370569, "no_speech_prob": 1.2955449213425396e-11, "compression_ratio": 1.2822299651567943, "source": "whisper" }, { "start": 229.52, "end": 234.72, "text": "美国有超过2060吉瓦的发电和储能项目", "chunk": 0, "language": "zh", "avg_logprob": -0.10174260488370569, "no_speech_prob": 1.2955449213425396e-11, "compression_ratio": 1.2822299651567943, "source": "whisper", "text_original": "美国有超过2060几瓦的发电和储能项目", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 234.72, "end": 235.84, "text": "在排队接入电网", "chunk": 0, "language": "zh", "avg_logprob": -0.10174260488370569, "no_speech_prob": 1.2955449213425396e-11, "compression_ratio": 1.2822299651567943, "source": "whisper" }, { "start": 235.84, "end": 236.74, "text": "注意啊", "chunk": 0, "language": "zh", "avg_logprob": -0.10174260488370569, "no_speech_prob": 1.2955449213425396e-11, "compression_ratio": 1.2822299651567943, "source": "whisper" }, { "start": 236.74, "end": 238.18, "text": "这不是已经能用的电", "chunk": 0, "language": "zh", "avg_logprob": -0.10174260488370569, "no_speech_prob": 1.2955449213425396e-11, "compression_ratio": 1.2822299651567943, "source": "whisper" }, { "start": 238.18, "end": 240.96, "text": "而是还在排队等着接入电网的项目", "chunk": 0, "language": "zh", "avg_logprob": -0.10174260488370569, "no_speech_prob": 1.2955449213425396e-11, "compression_ratio": 1.2822299651567943, "source": "whisper" }, { "start": 240.96, "end": 241.68, "text": "而在过去", "chunk": 0, "language": "zh", "avg_logprob": -0.10174260488370569, "no_speech_prob": 1.2955449213425396e-11, "compression_ratio": 1.2822299651567943, "source": "whisper" }, { "start": 241.68, "end": 245.62, "text": "2000年到2020年申请进入电网的容量", "chunk": 0, "language": "zh", "avg_logprob": -0.10174260488370569, "no_speech_prob": 1.2955449213425396e-11, "compression_ratio": 1.2822299651567943, "source": "whisper", "text_original": "2000年到2020年申请进入对立的容量", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 245.62, "end": 249.5, "text": "你到了2025年真正商业运营的只有13%", "chunk": 0, "language": "zh", "avg_logprob": -0.10174260488370569, "no_speech_prob": 1.2955449213425396e-11, "compression_ratio": 1.2822299651567943, "source": "whisper" }, { "start": 249.5, "end": 252.82, "text": "美国的基础建设效率可见一斑", "chunk": 0, "language": "zh", "avg_logprob": -0.10174260488370569, "no_speech_prob": 1.2955449213425396e-11, "compression_ratio": 1.2822299651567943, "source": "whisper" }, { "start": 252.82, "end": 254.1, "text": "真的是太慢了", "chunk": 0, "language": "zh", "avg_logprob": -0.10773728756194419, "no_speech_prob": 1.7572405336396812e-11, "compression_ratio": 1.3396226415094339, "source": "whisper" }, { "start": 254.1, "end": 256.64, "text": "这个就是美国AI的现实处境", "chunk": 0, "language": "zh", "avg_logprob": -0.10773728756194419, "no_speech_prob": 1.7572405336396812e-11, "compression_ratio": 1.3396226415094339, "source": "whisper" }, { "start": 256.64, "end": 258.1, "text": "它不是没有模型", "chunk": 0, "language": "zh", "avg_logprob": -0.10773728756194419, "no_speech_prob": 1.7572405336396812e-11, "compression_ratio": 1.3396226415094339, "source": "whisper" }, { "start": 258.1, "end": 260.2, "text": "也不是完全见不到电力", "chunk": 0, "language": "zh", "avg_logprob": -0.10773728756194419, "no_speech_prob": 1.7572405336396812e-11, "compression_ratio": 1.3396226415094339, "source": "whisper", "text_original": "也不是完全见不动电力", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 260.2, "end": 262.82, "text": "而是前沿AI的负载太大", "chunk": 0, "language": "zh", "avg_logprob": -0.10773728756194419, "no_speech_prob": 1.7572405336396812e-11, "compression_ratio": 1.3396226415094339, "source": "whisper" }, { "start": 262.82, "end": 263.8, "text": "增长太快", "chunk": 0, "language": "zh", "avg_logprob": -0.10773728756194419, "no_speech_prob": 1.7572405336396812e-11, "compression_ratio": 1.3396226415094339, "source": "whisper" }, { "start": 263.8, "end": 266.44, "text": "开始和美国分散的电网体系", "chunk": 0, "language": "zh", "avg_logprob": -0.10773728756194419, "no_speech_prob": 1.7572405336396812e-11, "compression_ratio": 1.3396226415094339, "source": "whisper" }, { "start": 266.44, "end": 267.9, "text": "漫长的审批周期", "chunk": 0, "language": "zh", "avg_logprob": -0.10773728756194419, "no_speech_prob": 1.7572405336396812e-11, "compression_ratio": 1.3396226415094339, "source": "whisper" }, { "start": 267.9, "end": 269.28, "text": "地方电价争议", "chunk": 0, "language": "zh", "avg_logprob": -0.10773728756194419, "no_speech_prob": 1.7572405336396812e-11, "compression_ratio": 1.3396226415094339, "source": "whisper" }, { "start": 269.28, "end": 271.14, "text": "包括输电瓶颈发生冲突", "chunk": 0, "language": "zh", "avg_logprob": -0.10773728756194419, "no_speech_prob": 1.7572405336396812e-11, "compression_ratio": 1.3396226415094339, "source": "whisper" }, { "start": 271.14, "end": 273.6, "text": "美国的AI公司很快可以出到钱", "chunk": 0, "language": "zh", "avg_logprob": -0.10773728756194419, "no_speech_prob": 1.7572405336396812e-11, "compression_ratio": 1.3396226415094339, "source": "whisper" }, { "start": 273.6, "end": 274.68, "text": "可以买到芯片", "chunk": 0, "language": "zh", "avg_logprob": -0.10773728756194419, "no_speech_prob": 1.7572405336396812e-11, "compression_ratio": 1.3396226415094339, "source": "whisper" }, { "start": 274.68, "end": 276.42, "text": "但基础设施的建设速度", "chunk": 0, "language": "zh", "avg_logprob": -0.10773728756194419, "no_speech_prob": 1.7572405336396812e-11, "compression_ratio": 1.3396226415094339, "source": "whisper" }, { "start": 276.42, "end": 278.44, "text": "不会像模型版本一样", "chunk": 0, "language": "zh", "avg_logprob": -0.10773728756194419, "no_speech_prob": 1.7572405336396812e-11, "compression_ratio": 1.3396226415094339, "source": "whisper" }, { "start": 278.44, "end": 279.8, "text": "几个月就迭代一次", "chunk": 0, "language": "zh", "avg_logprob": -0.10773728756194419, "no_speech_prob": 1.7572405336396812e-11, "compression_ratio": 1.3396226415094339, "source": "whisper", "text_original": "几个月就结在一次", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 279.8, "end": 282.34, "text": "而是按照年甚至几年为单位的", "chunk": 0, "language": "zh", "avg_logprob": -0.10773728756194419, "no_speech_prob": 1.7572405336396812e-11, "compression_ratio": 1.3396226415094339, "source": "whisper" }, { "start": 282.34, "end": 284.62, "text": "而中国的AI路线", "chunk": 0, "language": "zh", "avg_logprob": -0.11285503932407924, "no_speech_prob": 1.4568502654244142e-11, "compression_ratio": 1.2875816993464053, "source": "whisper" }, { "start": 284.62, "end": 286.82, "text": "我认为很可能不会复制美国这条路", "chunk": 0, "language": "zh", "avg_logprob": -0.11285503932407924, "no_speech_prob": 1.4568502654244142e-11, "compression_ratio": 1.2875816993464053, "source": "whisper" }, { "start": 286.82, "end": 289.2, "text": "中国的AI方面当然有巨大的差距", "chunk": 0, "language": "zh", "avg_logprob": -0.11285503932407924, "no_speech_prob": 1.4568502654244142e-11, "compression_ratio": 1.2875816993464053, "source": "whisper" }, { "start": 289.2, "end": 290.9, "text": "尤其是高端GPU", "chunk": 0, "language": "zh", "avg_logprob": -0.11285503932407924, "no_speech_prob": 1.4568502654244142e-11, "compression_ratio": 1.2875816993464053, "source": "whisper" }, { "start": 290.9, "end": 291.56, "text": "HBM", "chunk": 0, "language": "zh", "avg_logprob": -0.11285503932407924, "no_speech_prob": 1.4568502654244142e-11, "compression_ratio": 1.2875816993464053, "source": "whisper" }, { "start": 291.56, "end": 292.4, "text": "先进制程", "chunk": 0, "language": "zh", "avg_logprob": -0.11285503932407924, "no_speech_prob": 1.4568502654244142e-11, "compression_ratio": 1.2875816993464053, "source": "whisper" }, { "start": 292.4, "end": 294.0, "text": "云上高端算力的供给", "chunk": 0, "language": "zh", "avg_logprob": -0.11285503932407924, "no_speech_prob": 1.4568502654244142e-11, "compression_ratio": 1.2875816993464053, "source": "whisper", "text_original": "云上高端酸力的供给", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 294.0, "end": 296.08, "text": "这些都是硬约束", "chunk": 0, "language": "zh", "avg_logprob": -0.11285503932407924, "no_speech_prob": 1.4568502654244142e-11, "compression_ratio": 1.2875816993464053, "source": "whisper", "text_original": "这些都是应约数", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 296.08, "end": 297.78, "text": "但中国模型这两年的进步", "chunk": 0, "language": "zh", "avg_logprob": -0.11285503932407924, "no_speech_prob": 1.4568502654244142e-11, "compression_ratio": 1.2875816993464053, "source": "whisper" }, { "start": 297.78, "end": 298.94, "text": "有一个非常明显的特点", "chunk": 0, "language": "zh", "avg_logprob": -0.11285503932407924, "no_speech_prob": 1.4568502654244142e-11, "compression_ratio": 1.2875816993464053, "source": "whisper" }, { "start": 298.94, "end": 302.42, "text": "如果美国的顶级大模型是100分", "chunk": 0, "language": "zh", "avg_logprob": -0.11285503932407924, "no_speech_prob": 1.4568502654244142e-11, "compression_ratio": 1.2875816993464053, "source": "whisper", "text_original": "如果美国的顶级BN模型是100分", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 302.42, "end": 304.82, "text": "那中国厂商在不断的", "chunk": 0, "language": "zh", "avg_logprob": -0.11285503932407924, "no_speech_prob": 1.4568502654244142e-11, "compression_ratio": 1.2875816993464053, "source": "whisper" }, { "start": 304.82, "end": 307.78, "text": "把85分90分的开源模型能力", "chunk": 0, "language": "zh", "avg_logprob": -0.11285503932407924, "no_speech_prob": 1.4568502654244142e-11, "compression_ratio": 1.2875816993464053, "source": "whisper" }, { "start": 307.78, "end": 309.88, "text": "做得明显更便宜", "chunk": 0, "language": "zh", "avg_logprob": -0.11285503932407924, "no_speech_prob": 1.4568502654244142e-11, "compression_ratio": 1.2875816993464053, "source": "whisper", "text_original": "做的明显更便宜", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 309.88, "end": 310.62, "text": "更开放", "chunk": 0, "language": "zh", "avg_logprob": -0.11285503932407924, "no_speech_prob": 1.4568502654244142e-11, "compression_ratio": 1.2875816993464053, "source": "whisper" }, { "start": 310.62, "end": 311.64, "text": "更容易部署", "chunk": 0, "language": "zh", "avg_logprob": -0.11285503932407924, "no_speech_prob": 1.4568502654244142e-11, "compression_ratio": 1.2875816993464053, "source": "whisper" }, { "start": 311.64, "end": 313.78, "text": "还是斯坦福AI Index 2025", "chunk": 0, "language": "zh", "avg_logprob": -0.11128545611092214, "no_speech_prob": 1.6440349082103012e-11, "compression_ratio": 1.2884615384615385, "source": "whisper", "text_original": "还是斯坦福AI Index 2026", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 313.78, "end": 315.44, "text": "有一个观点我觉得很值得看", "chunk": 0, "language": "zh", "avg_logprob": -0.11128545611092214, "no_speech_prob": 1.6440349082103012e-11, "compression_ratio": 1.2884615384615385, "source": "whisper" }, { "start": 315.44, "end": 317.72, "text": "中美顶级模型的能力差距", "chunk": 0, "language": "zh", "avg_logprob": -0.11128545611092214, "no_speech_prob": 1.6440349082103012e-11, "compression_ratio": 1.2884615384615385, "source": "whisper" }, { "start": 317.72, "end": 319.06, "text": "已经明显缩小了", "chunk": 0, "language": "zh", "avg_logprob": -0.11128545611092214, "no_speech_prob": 1.6440349082103012e-11, "compression_ratio": 1.2884615384615385, "source": "whisper" }, { "start": 319.06, "end": 320.22, "text": "2025年以来", "chunk": 0, "language": "zh", "avg_logprob": -0.11128545611092214, "no_speech_prob": 1.6440349082103012e-11, "compression_ratio": 1.2884615384615385, "source": "whisper" }, { "start": 320.22, "end": 322.56, "text": "中美模型多次交换领先位置", "chunk": 0, "language": "zh", "avg_logprob": -0.11128545611092214, "no_speech_prob": 1.6440349082103012e-11, "compression_ratio": 1.2884615384615385, "source": "whisper" }, { "start": 322.56, "end": 323.88, "text": "他甚至提到了", "chunk": 0, "language": "zh", "avg_logprob": -0.11128545611092214, "no_speech_prob": 1.6440349082103012e-11, "compression_ratio": 1.2884615384615385, "source": "whisper" }, { "start": 323.88, "end": 325.64, "text": "截至2026年3月", "chunk": 0, "language": "zh", "avg_logprob": -0.11128545611092214, "no_speech_prob": 1.6440349082103012e-11, "compression_ratio": 1.2884615384615385, "source": "whisper" }, { "start": 325.64, "end": 327.44, "text": "美国顶级模型", "chunk": 0, "language": "zh", "avg_logprob": -0.11128545611092214, "no_speech_prob": 1.6440349082103012e-11, "compression_ratio": 1.2884615384615385, "source": "whisper" }, { "start": 327.44, "end": 329.5, "text": "相对中国顶级模型的领先", "chunk": 0, "language": "zh", "avg_logprob": -0.11128545611092214, "no_speech_prob": 1.6440349082103012e-11, "compression_ratio": 1.2884615384615385, "source": "whisper" }, { "start": 329.5, "end": 331.7, "text": "大约只有2.7%", "chunk": 0, "language": "zh", "avg_logprob": -0.11128545611092214, "no_speech_prob": 1.6440349082103012e-11, "compression_ratio": 1.2884615384615385, "source": "whisper" }, { "start": 331.7, "end": 333.28, "text": "当然这个2.7%", "chunk": 0, "language": "zh", "avg_logprob": -0.11128545611092214, "no_speech_prob": 1.6440349082103012e-11, "compression_ratio": 1.2884615384615385, "source": "whisper" }, { "start": 333.28, "end": 334.38, "text": "不是说中国模型", "chunk": 0, "language": "zh", "avg_logprob": -0.11128545611092214, "no_speech_prob": 1.6440349082103012e-11, "compression_ratio": 1.2884615384615385, "source": "whisper" }, { "start": 334.38, "end": 336.0, "text": "已经全面追上美国了", "chunk": 0, "language": "zh", "avg_logprob": -0.11128545611092214, "no_speech_prob": 1.6440349082103012e-11, "compression_ratio": 1.2884615384615385, "source": "whisper" }, { "start": 336.0, "end": 337.98, "text": "那不同榜单侧的东西也不一样", "chunk": 0, "language": "zh", "avg_logprob": -0.11128545611092214, "no_speech_prob": 1.6440349082103012e-11, "compression_ratio": 1.2884615384615385, "source": "whisper" }, { "start": 337.98, "end": 339.48, "text": "但它至少说明了一件事", "chunk": 0, "language": "zh", "avg_logprob": -0.11128545611092214, "no_speech_prob": 1.6440349082103012e-11, "compression_ratio": 1.2884615384615385, "source": "whisper" }, { "start": 339.48, "end": 341.72, "text": "就是中美的模型能力", "chunk": 0, "language": "zh", "avg_logprob": -0.09592158117412049, "no_speech_prob": 1.2511488373112556e-11, "compression_ratio": 1.2567164179104477, "source": "whisper" }, { "start": 341.72, "end": 345.1, "text": "已经不是三年前那种一眼看不到边的差距了", "chunk": 0, "language": "zh", "avg_logprob": -0.09592158117412049, "no_speech_prob": 1.2511488373112556e-11, "compression_ratio": 1.2567164179104477, "source": "whisper" }, { "start": 345.1, "end": 347.54, "text": "那如果你再看开放权重模型", "chunk": 0, "language": "zh", "avg_logprob": -0.09592158117412049, "no_speech_prob": 1.2511488373112556e-11, "compression_ratio": 1.2567164179104477, "source": "whisper" }, { "start": 347.54, "end": 350.44, "text": "Artificial Analyst今年6月的文章里", "chunk": 0, "language": "zh", "avg_logprob": -0.09592158117412049, "no_speech_prob": 1.2511488373112556e-11, "compression_ratio": 1.2567164179104477, "source": "whisper" }, { "start": 350.44, "end": 353.36, "text": "GLM 5.2在它的Intelligent Index上", "chunk": 0, "language": "zh", "avg_logprob": -0.09592158117412049, "no_speech_prob": 1.2511488373112556e-11, "compression_ratio": 1.2567164179104477, "source": "whisper" }, { "start": 353.36, "end": 356.48, "text": "成为领先的开放权重模型之一", "chunk": 0, "language": "zh", "avg_logprob": -0.09592158117412049, "no_speech_prob": 1.2511488373112556e-11, "compression_ratio": 1.2567164179104477, "source": "whisper" }, { "start": 356.48, "end": 358.32, "text": "并且在一些企业任务评测里", "chunk": 0, "language": "zh", "avg_logprob": -0.09592158117412049, "no_speech_prob": 1.2511488373112556e-11, "compression_ratio": 1.2567164179104477, "source": "whisper" }, { "start": 358.32, "end": 360.98, "text": "它已经进入了高能力低成本的区间", "chunk": 0, "language": "zh", "avg_logprob": -0.09592158117412049, "no_speech_prob": 1.2511488373112556e-11, "compression_ratio": 1.2567164179104477, "source": "whisper" }, { "start": 360.98, "end": 362.6, "text": "这个例子能说明一个方向", "chunk": 0, "language": "zh", "avg_logprob": -0.09592158117412049, "no_speech_prob": 1.2511488373112556e-11, "compression_ratio": 1.2567164179104477, "source": "whisper" }, { "start": 362.6, "end": 365.0, "text": "就是中国模型不只是追赶", "chunk": 0, "language": "zh", "avg_logprob": -0.09592158117412049, "no_speech_prob": 1.2511488373112556e-11, "compression_ratio": 1.2567164179104477, "source": "whisper", "text_original": "就是中国模型不只是追宝", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 365.0, "end": 367.76, "text": "它是在把接近前沿模型的能力", "chunk": 0, "language": "zh", "avg_logprob": -0.09592158117412049, "no_speech_prob": 1.2511488373112556e-11, "compression_ratio": 1.2567164179104477, "source": "whisper" }, { "start": 367.76, "end": 371.38, "text": "用更开放更低成本的方式来释放出来", "chunk": 0, "language": "zh", "avg_logprob": -0.1220403646493887, "no_speech_prob": 1.3927224824794582e-11, "compression_ratio": 1.2662337662337662, "source": "whisper" }, { "start": 371.38, "end": 374.28, "text": "而这条路线比很多人想象的重要", "chunk": 0, "language": "zh", "avg_logprob": -0.1220403646493887, "no_speech_prob": 1.3927224824794582e-11, "compression_ratio": 1.2662337662337662, "source": "whisper" }, { "start": 374.28, "end": 377.72, "text": "因为市场并不总是只买最强的", "chunk": 0, "language": "zh", "avg_logprob": -0.1220403646493887, "no_speech_prob": 1.3927224824794582e-11, "compression_ratio": 1.2662337662337662, "source": "whisper", "text_original": "因为市场并不总是直卖最强的", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 377.72, "end": 380.42, "text": "很多企业开发者出海场景", "chunk": 0, "language": "zh", "avg_logprob": -0.1220403646493887, "no_speech_prob": 1.3927224824794582e-11, "compression_ratio": 1.2662337662337662, "source": "whisper", "text_original": "很多企业开发者出这场景", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 380.42, "end": 384.0, "text": "其实不需要世界第一聪明的模型", "chunk": 0, "language": "zh", "avg_logprob": -0.1220403646493887, "no_speech_prob": 1.3927224824794582e-11, "compression_ratio": 1.2662337662337662, "source": "whisper" }, { "start": 384.0, "end": 386.48, "text": "前段时间你但凡狠狠调过", "chunk": 0, "language": "zh", "avg_logprob": -0.1220403646493887, "no_speech_prob": 1.3927224824794582e-11, "compression_ratio": 1.2662337662337662, "source": "whisper", "text_original": "前段时间你但凡狠狠登过", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 386.48, "end": 388.72, "text": "非GPT模型的人应该会有同感", "chunk": 0, "language": "zh", "avg_logprob": -0.1220403646493887, "no_speech_prob": 1.3927224824794582e-11, "compression_ratio": 1.2662337662337662, "source": "whisper", "text_original": "非波5模型的人应该会有同感", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 388.72, "end": 390.92, "text": "那个token的消耗真的可怕", "chunk": 0, "language": "zh", "avg_logprob": -0.1220403646493887, "no_speech_prob": 1.3927224824794582e-11, "compression_ratio": 1.2662337662337662, "source": "whisper" }, { "start": 390.92, "end": 392.18, "text": "成本真的很高", "chunk": 0, "language": "zh", "avg_logprob": -0.1220403646493887, "no_speech_prob": 1.3927224824794582e-11, "compression_ratio": 1.2662337662337662, "source": "whisper" }, { "start": 392.18, "end": 395.06, "text": "更多时候人们需要的是足够强", "chunk": 0, "language": "zh", "avg_logprob": -0.1220403646493887, "no_speech_prob": 1.3927224824794582e-11, "compression_ratio": 1.2662337662337662, "source": "whisper" }, { "start": 395.06, "end": 397.4, "text": "同时便宜稳定能部署", "chunk": 0, "language": "zh", "avg_logprob": -0.1220403646493887, "no_speech_prob": 1.3927224824794582e-11, "compression_ratio": 1.2662337662337662, "source": "whisper" }, { "start": 397.4, "end": 399.38, "text": "并且不会因为地区限制", "chunk": 0, "language": "zh", "avg_logprob": -0.059910068264255274, "no_speech_prob": 1.2081496393590019e-11, "compression_ratio": 1.2836879432624113, "source": "whisper" }, { "start": 399.38, "end": 401.26, "text": "产品分层或者政策变化", "chunk": 0, "language": "zh", "avg_logprob": -0.059910068264255274, "no_speech_prob": 1.2081496393590019e-11, "compression_ratio": 1.2836879432624113, "source": "whisper" }, { "start": 401.26, "end": 402.54, "text": "突然断供的模型", "chunk": 0, "language": "zh", "avg_logprob": -0.059910068264255274, "no_speech_prob": 1.2081496393590019e-11, "compression_ratio": 1.2836879432624113, "source": "whisper", "text_original": "突然断工的模型", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 402.54, "end": 405.58, "text": "这个就是中国开放模型的长期价值", "chunk": 0, "language": "zh", "avg_logprob": -0.059910068264255274, "no_speech_prob": 1.2081496393590019e-11, "compression_ratio": 1.2836879432624113, "source": "whisper" }, { "start": 405.58, "end": 407.06, "text": "这不是低端替代", "chunk": 0, "language": "zh", "avg_logprob": -0.059910068264255274, "no_speech_prob": 1.2081496393590019e-11, "compression_ratio": 1.2836879432624113, "source": "whisper" }, { "start": 407.06, "end": 411.2, "text": "而是中国模型更能够把AI能力去产业化和规模化", "chunk": 0, "language": "zh", "avg_logprob": -0.059910068264255274, "no_speech_prob": 1.2081496393590019e-11, "compression_ratio": 1.2836879432624113, "source": "whisper" }, { "start": 411.2, "end": 414.78, "text": "而一旦你把AI看成工业化的供给", "chunk": 0, "language": "zh", "avg_logprob": -0.059910068264255274, "no_speech_prob": 1.2081496393590019e-11, "compression_ratio": 1.2836879432624113, "source": "whisper" }, { "start": 414.78, "end": 419.72, "text": "那电力和基础设施就变成了中国这边非常重要的一张牌", "chunk": 0, "language": "zh", "avg_logprob": -0.059910068264255274, "no_speech_prob": 1.2081496393590019e-11, "compression_ratio": 1.2836879432624113, "source": "whisper" }, { "start": 419.72, "end": 421.98, "text": "国家能源局2024年的数据很夸张", "chunk": 0, "language": "zh", "avg_logprob": -0.059910068264255274, "no_speech_prob": 1.2081496393590019e-11, "compression_ratio": 1.2836879432624113, "source": "whisper", "text_original": "国家能源局2025年的数据很夸张", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 421.98, "end": 428.1, "text": "2025年中国可再生能源新增装机4.52亿千瓦", "chunk": 0, "language": "zh", "avg_logprob": -0.08627857921792449, "no_speech_prob": 1.3162147587120199e-11, "compression_ratio": 1.5280898876404494, "source": "whisper" }, { "start": 428.1, "end": 431.48, "text": "占全国电力新增装机的83%", "chunk": 0, "language": "zh", "avg_logprob": -0.08627857921792449, "no_speech_prob": 1.3162147587120199e-11, "compression_ratio": 1.5280898876404494, "source": "whisper" }, { "start": 431.48, "end": 434.48, "text": "其中太阳能新增3.18亿千瓦", "chunk": 0, "language": "zh", "avg_logprob": -0.08627857921792449, "no_speech_prob": 1.3162147587120199e-11, "compression_ratio": 1.5280898876404494, "source": "whisper" }, { "start": 434.48, "end": 436.88, "text": "风力新增1.2亿千瓦", "chunk": 0, "language": "zh", "avg_logprob": -0.08627857921792449, "no_speech_prob": 1.3162147587120199e-11, "compression_ratio": 1.5280898876404494, "source": "whisper" }, { "start": 436.88, "end": 438.06, "text": "到2025年底", "chunk": 0, "language": "zh", "avg_logprob": -0.08627857921792449, "no_speech_prob": 1.3162147587120199e-11, "compression_ratio": 1.5280898876404494, "source": "whisper" }, { "start": 438.06, "end": 442.02, "text": "中国可再生能源的装机会达到23.4亿千瓦", "chunk": 0, "language": "zh", "avg_logprob": -0.08627857921792449, "no_speech_prob": 1.3162147587120199e-11, "compression_ratio": 1.5280898876404494, "source": "whisper" }, { "start": 442.02, "end": 445.26, "text": "大约占全国总装机的60%", "chunk": 0, "language": "zh", "avg_logprob": -0.08627857921792449, "no_speech_prob": 1.3162147587120199e-11, "compression_ratio": 1.5280898876404494, "source": "whisper" }, { "start": 445.26, "end": 447.733, "text": "所以能感知到中国在能源", "chunk": 0, "language": "zh", "avg_logprob": -0.08627857921792449, "no_speech_prob": 1.3162147587120199e-11, "compression_ratio": 1.5280898876404494, "source": "whisper" }, { "start": 447.733, "end": 450.913, "text": "所以能感知到,中國在能源和基礎設施上的動員能力非常的強。", "chunk": 1, "language": "zh", "avg_logprob": -0.1306719010876071, "no_speech_prob": 8.829321822279024e-12, "compression_ratio": 1.2912280701754386, "source": "whisper" }, { "start": 450.913, "end": 455.253, "text": "那隨著模型能力越來越強,訓練新模型肯定是仍然重要的。", "chunk": 1, "language": "zh", "avg_logprob": -0.1306719010876071, "no_speech_prob": 8.829321822279024e-12, "compression_ratio": 1.2912280701754386, "source": "whisper" }, { "start": 455.253, "end": 458.853, "text": "但推理過程的成本會變成AI發展的關鍵變量。", "chunk": 1, "language": "zh", "avg_logprob": -0.1306719010876071, "no_speech_prob": 8.829321822279024e-12, "compression_ratio": 1.2912280701754386, "source": "whisper" }, { "start": 458.853, "end": 463.453, "text": "因为只要AI真正进入企业办公教育等等越来越多的场景", "chunk": 1, "language": "zh", "avg_logprob": -0.07653871974589661, "no_speech_prob": 1.456389002452152e-11, "compression_ratio": 1.2912621359223302, "source": "whisper" }, { "start": 463.453, "end": 469.073, "text": "那AI公司未来最大的账单很可能不是训练一个模型要花多少钱", "chunk": 1, "language": "zh", "avg_logprob": -0.07653871974589661, "no_speech_prob": 1.456389002452152e-11, "compression_ratio": 1.2912621359223302, "source": "whisper" }, { "start": 469.073, "end": 473.973, "text": "而是每天几十亿几百亿次的调用单位成本能不能压下来", "chunk": 1, "language": "zh", "avg_logprob": -0.07653871974589661, "no_speech_prob": 1.456389002452152e-11, "compression_ratio": 1.2912621359223302, "source": "whisper", "text_original": "而是每天几十亿几百亿丝的调用单位成本能不能压下来", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 473.973, "end": 475.253, "text": "Token听起来抽象", "chunk": 1, "language": "zh", "avg_logprob": -0.07653871974589661, "no_speech_prob": 1.456389002452152e-11, "compression_ratio": 1.2912621359223302, "source": "whisper" }, { "start": 475.253, "end": 480.333, "text": "但它就是AI能力被调用被计费被规模化交付的最小单位", "chunk": 1, "language": "zh", "avg_logprob": -0.07653871974589661, "no_speech_prob": 1.456389002452152e-11, "compression_ratio": 1.2912621359223302, "source": "whisper" }, { "start": 480.333, "end": 483.353, "text": "未来AI如果真的变成了基础设施", "chunk": 1, "language": "zh", "avg_logprob": -0.07653871974589661, "no_speech_prob": 1.456389002452152e-11, "compression_ratio": 1.2912621359223302, "source": "whisper" }, { "start": 483.353, "end": 486.933, "text": "那谁能把每一次Token的成本压下来", "chunk": 1, "language": "zh", "avg_logprob": -0.07653871974589661, "no_speech_prob": 1.456389002452152e-11, "compression_ratio": 1.2912621359223302, "source": "whisper" }, { "start": 486.933, "end": 488.993, "text": "谁就能让AI覆盖更多场景", "chunk": 1, "language": "zh", "avg_logprob": -0.10011584165865782, "no_speech_prob": 1.6179212483646843e-11, "compression_ratio": 1.3148148148148149, "source": "whisper" }, { "start": 488.993, "end": 490.593, "text": "这也是为什么我会觉得", "chunk": 1, "language": "zh", "avg_logprob": -0.10011584165865782, "no_speech_prob": 1.6179212483646843e-11, "compression_ratio": 1.3148148148148149, "source": "whisper" }, { "start": 490.593, "end": 493.533, "text": "未来中美AI很可能不是一个", "chunk": 1, "language": "zh", "avg_logprob": -0.10011584165865782, "no_speech_prob": 1.6179212483646843e-11, "compression_ratio": 1.3148148148148149, "source": "whisper" }, { "start": 493.533, "end": 497.233, "text": "B元打败开元或者开元打败B元的故事", "chunk": 1, "language": "zh", "avg_logprob": -0.10011584165865782, "no_speech_prob": 1.6179212483646843e-11, "compression_ratio": 1.3148148148148149, "source": "whisper" }, { "start": 497.233, "end": 500.033, "text": "更可能是两套市场长期并存", "chunk": 1, "language": "zh", "avg_logprob": -0.10011584165865782, "no_speech_prob": 1.6179212483646843e-11, "compression_ratio": 1.3148148148148149, "source": "whisper" }, { "start": 500.033, "end": 501.593, "text": "最顶级的B元模型", "chunk": 1, "language": "zh", "avg_logprob": -0.10011584165865782, "no_speech_prob": 1.6179212483646843e-11, "compression_ratio": 1.3148148148148149, "source": "whisper" }, { "start": 501.593, "end": 505.033, "text": "继续服务于科研复杂推理和最前沿的agent", "chunk": 1, "language": "zh", "avg_logprob": -0.10011584165865782, "no_speech_prob": 1.6179212483646843e-11, "compression_ratio": 1.3148148148148149, "source": "whisper", "text_original": "继续服于科研复杂推理和最前沿的agent", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 505.033, "end": 506.293, "text": "它们会很贵", "chunk": 1, "language": "zh", "avg_logprob": -0.10011584165865782, "no_speech_prob": 1.6179212483646843e-11, "compression_ratio": 1.3148148148148149, "source": "whisper" }, { "start": 506.293, "end": 507.373, "text": "会分层", "chunk": 1, "language": "zh", "avg_logprob": -0.10011584165865782, "no_speech_prob": 1.6179212483646843e-11, "compression_ratio": 1.3148148148148149, "source": "whisper" }, { "start": 507.373, "end": 508.453, "text": "会有访问门槛", "chunk": 1, "language": "zh", "avg_logprob": -0.10011584165865782, "no_speech_prob": 1.6179212483646843e-11, "compression_ratio": 1.3148148148148149, "source": "whisper" }, { "start": 508.453, "end": 511.173, "text": "同时也会继续代表模型能力的天花板", "chunk": 1, "language": "zh", "avg_logprob": -0.10011584165865782, "no_speech_prob": 1.6179212483646843e-11, "compression_ratio": 1.3148148148148149, "source": "whisper" }, { "start": 511.173, "end": 512.613, "text": "但更大的市场", "chunk": 1, "language": "zh", "avg_logprob": -0.10011584165865782, "no_speech_prob": 1.6179212483646843e-11, "compression_ratio": 1.3148148148148149, "source": "whisper" }, { "start": 512.613, "end": 514.493, "text": "可能是企业内部的流程", "chunk": 1, "language": "zh", "avg_logprob": -0.10011584165865782, "no_speech_prob": 1.6179212483646843e-11, "compression_ratio": 1.3148148148148149, "source": "whisper" }, { "start": 514.493, "end": 515.433, "text": "自由化的部署", "chunk": 1, "language": "zh", "avg_logprob": -0.10011584165865782, "no_speech_prob": 1.6179212483646843e-11, "compression_ratio": 1.3148148148148149, "source": "whisper" }, { "start": 515.433, "end": 516.353, "text": "本地化的应用", "chunk": 1, "language": "zh", "avg_logprob": -0.10011584165865782, "no_speech_prob": 1.6179212483646843e-11, "compression_ratio": 1.3148148148148149, "source": "whisper" }, { "start": 516.353, "end": 518.213, "text": "一些相对低成本的智能体", "chunk": 1, "language": "zh", "avg_logprob": -0.08825158776703829, "no_speech_prob": 1.847823977163543e-11, "compression_ratio": 1.3436426116838487, "source": "whisper" }, { "start": 518.213, "end": 520.073, "text": "大规模的推理和调用", "chunk": 1, "language": "zh", "avg_logprob": -0.08825158776703829, "no_speech_prob": 1.847823977163543e-11, "compression_ratio": 1.3436426116838487, "source": "whisper" }, { "start": 520.073, "end": 522.493, "text": "这里要的它不是世界第一聪明", "chunk": 1, "language": "zh", "avg_logprob": -0.08825158776703829, "no_speech_prob": 1.847823977163543e-11, "compression_ratio": 1.3436426116838487, "source": "whisper" }, { "start": 522.493, "end": 525.453, "text": "而是够强稳定便宜可控", "chunk": 1, "language": "zh", "avg_logprob": -0.08825158776703829, "no_speech_prob": 1.847823977163543e-11, "compression_ratio": 1.3436426116838487, "source": "whisper" }, { "start": 525.453, "end": 528.393, "text": "Claude最大的问题就在这里", "chunk": 1, "language": "zh", "avg_logprob": -0.08825158776703829, "no_speech_prob": 1.847823977163543e-11, "compression_ratio": 1.3436426116838487, "source": "whisper", "text_original": "Cloud最大的问题就在这里", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 528.393, "end": 529.593, "text": "它不是不强", "chunk": 1, "language": "zh", "avg_logprob": -0.08825158776703829, "no_speech_prob": 1.847823977163543e-11, "compression_ratio": 1.3436426116838487, "source": "whisper" }, { "start": 529.593, "end": 532.453, "text": "但如果一个模型公司依赖的是最强模型", "chunk": 1, "language": "zh", "avg_logprob": -0.08825158776703829, "no_speech_prob": 1.847823977163543e-11, "compression_ratio": 1.3436426116838487, "source": "whisper" }, { "start": 532.453, "end": 534.333, "text": "可以一直卖出稀缺溢价", "chunk": 1, "language": "zh", "avg_logprob": -0.08825158776703829, "no_speech_prob": 1.847823977163543e-11, "compression_ratio": 1.3436426116838487, "source": "whisper" }, { "start": 534.333, "end": 536.513, "text": "它就会不断的被重新定价", "chunk": 1, "language": "zh", "avg_logprob": -0.08825158776703829, "no_speech_prob": 1.847823977163543e-11, "compression_ratio": 1.3436426116838487, "source": "whisper" }, { "start": 536.513, "end": 539.773, "text": "被OpenAI的下一代闭源模型重新定价", "chunk": 1, "language": "zh", "avg_logprob": -0.08825158776703829, "no_speech_prob": 1.847823977163543e-11, "compression_ratio": 1.3436426116838487, "source": "whisper", "text_original": "被OpenAI的下一代币源模型重新定价", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 539.773, "end": 543.493, "text": "被DeepSeek这样便宜量大的开源模型重新定价", "chunk": 1, "language": "zh", "avg_logprob": -0.08825158776703829, "no_speech_prob": 1.847823977163543e-11, "compression_ratio": 1.3436426116838487, "source": "whisper" }, { "start": 543.493, "end": 546.353, "text": "被中国一批又一批开源模型重新定价", "chunk": 1, "language": "zh", "avg_logprob": -0.0614813909480709, "no_speech_prob": 1.4335672406107225e-11, "compression_ratio": 1.3745928338762214, "source": "whisper" }, { "start": 546.353, "end": 548.633, "text": "你最强当然是有很大价值的", "chunk": 1, "language": "zh", "avg_logprob": -0.0614813909480709, "no_speech_prob": 1.4335672406107225e-11, "compression_ratio": 1.3745928338762214, "source": "whisper" }, { "start": 548.633, "end": 550.993, "text": "但如果别人用便宜很多的价格", "chunk": 1, "language": "zh", "avg_logprob": -0.0614813909480709, "no_speech_prob": 1.4335672406107225e-11, "compression_ratio": 1.3745928338762214, "source": "whisper" }, { "start": 550.993, "end": 552.453, "text": "做到你八九成的能力", "chunk": 1, "language": "zh", "avg_logprob": -0.0614813909480709, "no_speech_prob": 1.4335672406107225e-11, "compression_ratio": 1.3745928338762214, "source": "whisper" }, { "start": 552.453, "end": 553.453, "text": "还能私有部署", "chunk": 1, "language": "zh", "avg_logprob": -0.0614813909480709, "no_speech_prob": 1.4335672406107225e-11, "compression_ratio": 1.3745928338762214, "source": "whisper" }, { "start": 553.453, "end": 556.233, "text": "还不受你这套访问规则的约束", "chunk": 1, "language": "zh", "avg_logprob": -0.0614813909480709, "no_speech_prob": 1.4335672406107225e-11, "compression_ratio": 1.3745928338762214, "source": "whisper" }, { "start": 556.233, "end": 558.213, "text": "那很多真实的生产场景", "chunk": 1, "language": "zh", "avg_logprob": -0.0614813909480709, "no_speech_prob": 1.4335672406107225e-11, "compression_ratio": 1.3745928338762214, "source": "whisper" }, { "start": 558.213, "end": 559.873, "text": "就会开始重新算账了", "chunk": 1, "language": "zh", "avg_logprob": -0.0614813909480709, "no_speech_prob": 1.4335672406107225e-11, "compression_ratio": 1.3745928338762214, "source": "whisper" }, { "start": 559.873, "end": 561.193, "text": "那这个账一算", "chunk": 1, "language": "zh", "avg_logprob": -0.0614813909480709, "no_speech_prob": 1.4335672406107225e-11, "compression_ratio": 1.3745928338762214, "source": "whisper" }, { "start": 561.193, "end": 564.193, "text": "AI竞争就已经从互联网逻辑", "chunk": 1, "language": "zh", "avg_logprob": -0.0614813909480709, "no_speech_prob": 1.4335672406107225e-11, "compression_ratio": 1.3745928338762214, "source": "whisper" }, { "start": 564.193, "end": 566.353, "text": "进入了能源工业的逻辑", "chunk": 1, "language": "zh", "avg_logprob": -0.0614813909480709, "no_speech_prob": 1.4335672406107225e-11, "compression_ratio": 1.3745928338762214, "source": "whisper" }, { "start": 566.353, "end": 568.233, "text": "而在能源工业的逻辑里", "chunk": 1, "language": "zh", "avg_logprob": -0.0614813909480709, "no_speech_prob": 1.4335672406107225e-11, "compression_ratio": 1.3745928338762214, "source": "whisper" }, { "start": 568.233, "end": 572.293, "text": "最重要的是产能成本和供应稳定性", "chunk": 1, "language": "zh", "avg_logprob": -0.0614813909480709, "no_speech_prob": 1.4335672406107225e-11, "compression_ratio": 1.3745928338762214, "source": "whisper" }, { "start": 572.293, "end": 574.993, "text": "所以Claude去欧洲找电这件事", "chunk": 1, "language": "zh", "avg_logprob": -0.09427723963616302, "no_speech_prob": 1.6635012811683225e-11, "compression_ratio": 1.2782874617737003, "source": "whisper", "text_original": "所以Cloud去欧洲找电这件事", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 574.993, "end": 576.673, "text": "表面看是一个AI公司", "chunk": 1, "language": "zh", "avg_logprob": -0.09427723963616302, "no_speech_prob": 1.6635012811683225e-11, "compression_ratio": 1.2782874617737003, "source": "whisper" }, { "start": 576.673, "end": 578.073, "text": "要扩数据中心", "chunk": 1, "language": "zh", "avg_logprob": -0.09427723963616302, "no_speech_prob": 1.6635012811683225e-11, "compression_ratio": 1.2782874617737003, "source": "whisper" }, { "start": 578.073, "end": 579.073, "text": "但往深里看", "chunk": 1, "language": "zh", "avg_logprob": -0.09427723963616302, "no_speech_prob": 1.6635012811683225e-11, "compression_ratio": 1.2782874617737003, "source": "whisper", "text_original": "但往身里看", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 579.073, "end": 580.253, "text": "它是在提醒我们", "chunk": 1, "language": "zh", "avg_logprob": -0.09427723963616302, "no_speech_prob": 1.6635012811683225e-11, "compression_ratio": 1.2782874617737003, "source": "whisper" }, { "start": 580.253, "end": 583.053, "text": "模型并不是浮在云上的魔法", "chunk": 1, "language": "zh", "avg_logprob": -0.09427723963616302, "no_speech_prob": 1.6635012811683225e-11, "compression_ratio": 1.2782874617737003, "source": "whisper" }, { "start": 583.053, "end": 584.973, "text": "AI公司卖的是智能", "chunk": 1, "language": "zh", "avg_logprob": -0.09427723963616302, "no_speech_prob": 1.6635012811683225e-11, "compression_ratio": 1.2782874617737003, "source": "whisper" }, { "start": 584.973, "end": 586.973, "text": "但它每天消耗的是电", "chunk": 1, "language": "zh", "avg_logprob": -0.09427723963616302, "no_speech_prob": 1.6635012811683225e-11, "compression_ratio": 1.2782874617737003, "source": "whisper" }, { "start": 586.973, "end": 588.253, "text": "抢的是芯片", "chunk": 1, "language": "zh", "avg_logprob": -0.09427723963616302, "no_speech_prob": 1.6635012811683225e-11, "compression_ratio": 1.2782874617737003, "source": "whisper" }, { "start": 588.253, "end": 590.633, "text": "租的是实实在在的厂房", "chunk": 1, "language": "zh", "avg_logprob": -0.09427723963616302, "no_speech_prob": 1.6635012811683225e-11, "compression_ratio": 1.2782874617737003, "source": "whisper" }, { "start": 590.633, "end": 592.573, "text": "所以我认为中美的AI竞争", "chunk": 1, "language": "zh", "avg_logprob": -0.09427723963616302, "no_speech_prob": 1.6635012811683225e-11, "compression_ratio": 1.2782874617737003, "source": "whisper" }, { "start": 592.573, "end": 594.933, "text": "不能只看谁的模型最聪明", "chunk": 1, "language": "zh", "avg_logprob": -0.09427723963616302, "no_speech_prob": 1.6635012811683225e-11, "compression_ratio": 1.2782874617737003, "source": "whisper" }, { "start": 594.933, "end": 596.573, "text": "还要结合访问限制", "chunk": 1, "language": "zh", "avg_logprob": -0.09427723963616302, "no_speech_prob": 1.6635012811683225e-11, "compression_ratio": 1.2782874617737003, "source": "whisper" }, { "start": 596.573, "end": 597.773, "text": "单位Token的价格", "chunk": 1, "language": "zh", "avg_logprob": -0.09427723963616302, "no_speech_prob": 1.6635012811683225e-11, "compression_ratio": 1.2782874617737003, "source": "whisper" }, { "start": 597.773, "end": 598.633, "text": "还有要观察", "chunk": 1, "language": "zh", "avg_logprob": -0.09427723963616302, "no_speech_prob": 1.6635012811683225e-11, "compression_ratio": 1.2782874617737003, "source": "whisper" }, { "start": 598.633, "end": 601.333, "text": "应用市场是不是真的愿意为最强模型", "chunk": 1, "language": "zh", "avg_logprob": -0.09427723963616302, "no_speech_prob": 1.6635012811683225e-11, "compression_ratio": 1.2782874617737003, "source": "whisper" }, { "start": 601.333, "end": 602.513, "text": "付出高昂代价", "chunk": 1, "language": "zh", "avg_logprob": -0.209960691390499, "no_speech_prob": 2.3644454449911478e-11, "compression_ratio": 1.2720848056537102, "source": "whisper", "text_original": "付出高昂一家", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 602.513, "end": 605.253, "text": "我认为从0到1的创新", "chunk": 1, "language": "zh", "avg_logprob": -0.209960691390499, "no_speech_prob": 2.3644454449911478e-11, "compression_ratio": 1.2720848056537102, "source": "whisper" }, { "start": 605.253, "end": 607.013, "text": "和从1到10的落地", "chunk": 1, "language": "zh", "avg_logprob": -0.209960691390499, "no_speech_prob": 2.3644454449911478e-11, "compression_ratio": 1.2720848056537102, "source": "whisper" }, { "start": 607.013, "end": 608.713, "text": "这两条路的竞争", "chunk": 1, "language": "zh", "avg_logprob": -0.209960691390499, "no_speech_prob": 2.3644454449911478e-11, "compression_ratio": 1.2720848056537102, "source": "whisper" }, { "start": 608.713, "end": 610.953, "text": "也许会一直持续下去", "chunk": 1, "language": "zh", "avg_logprob": -0.209960691390499, "no_speech_prob": 2.3644454449911478e-11, "compression_ratio": 1.2720848056537102, "source": "whisper" }, { "start": 610.953, "end": 611.373, "text": "OK", "chunk": 1, "language": "zh", "avg_logprob": -0.209960691390499, "no_speech_prob": 2.3644454449911478e-11, "compression_ratio": 1.2720848056537102, "source": "whisper" }, { "start": 611.373, "end": 613.393, "text": "那本期视频的主要内容就这么多了", "chunk": 1, "language": "zh", "avg_logprob": -0.209960691390499, "no_speech_prob": 2.3644454449911478e-11, "compression_ratio": 1.2720848056537102, "source": "whisper", "text_original": "那本期视频的主要人就这么多了", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 613.393, "end": 615.733, "text": "大家伙日常在用哪个模型最多", "chunk": 1, "language": "zh", "avg_logprob": -0.209960691390499, "no_speech_prob": 2.3644454449911478e-11, "compression_ratio": 1.2720848056537102, "source": "whisper" }, { "start": 615.733, "end": 617.953, "text": "也可以打在评论或者弹幕上", "chunk": 1, "language": "zh", "avg_logprob": -0.209960691390499, "no_speech_prob": 2.3644454449911478e-11, "compression_ratio": 1.2720848056537102, "source": "whisper" }, { "start": 617.953, "end": 618.673, "text": "我们分享一下", "chunk": 1, "language": "zh", "avg_logprob": -0.209960691390499, "no_speech_prob": 2.3644454449911478e-11, "compression_ratio": 1.2720848056537102, "source": "whisper" }, { "start": 618.673, "end": 619.953, "text": "如果喜欢本期视频的话", "chunk": 1, "language": "zh", "avg_logprob": -0.209960691390499, "no_speech_prob": 2.3644454449911478e-11, "compression_ratio": 1.2720848056537102, "source": "whisper" }, { "start": 619.953, "end": 620.973, "text": "请到点赞朋友订阅我的话", "chunk": 1, "language": "zh", "avg_logprob": -0.209960691390499, "no_speech_prob": 2.3644454449911478e-11, "compression_ratio": 1.2720848056537102, "source": "whisper" }, { "start": 620.973, "end": 621.573, "text": "我是HYK", "chunk": 1, "language": "zh", "avg_logprob": -0.209960691390499, "no_speech_prob": 2.3644454449911478e-11, "compression_ratio": 1.2720848056537102, "source": "whisper", "text_original": "我是JK", "correction_applied": true, "correction_source": "llm_correct_a_presegmentation" }, { "start": 621.573, "end": 622.313, "text": "我们下期视频见", "chunk": 1, "language": "zh", "avg_logprob": -0.209960691390499, "no_speech_prob": 2.3644454449911478e-11, "compression_ratio": 1.2720848056537102, "source": "whisper" }, { "start": 622.313, "end": 622.693, "text": "祝你好运", "chunk": 1, "language": "zh", "avg_logprob": -0.209960691390499, "no_speech_prob": 2.3644454449911478e-11, "compression_ratio": 1.2720848056537102, "source": "whisper" }, { "start": 622.693, "end": 623.513, "text": "拜拜", "chunk": 1, "language": "zh", "avg_logprob": -0.209960691390499, "no_speech_prob": 2.3644454449911478e-11, "compression_ratio": 1.2720848056537102, "source": "whisper" } ], "transcript_selection": { "schema": "videonote.transcript_selection.v1", "status": "SELECTED", "requested_language": "zh", "duration": 627.733313, "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": 268, "multilingual": false, "first_start": 0.0, "final_end": 623.513, "duration_coverage": 0.993277, "maximum_gap_seconds": 0.0, "inverted_count": 0, "language": { "requested": "zh", "matched": true, "latin": 303, "han": 2798, "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": 627.733, "rejected_segment_count": 0, "rejected_segments": [] } }, "presegmentation_correction": { "applied": true, "correction_count": 34 } }