1 99:59:59,999 --> 99:59:59,999 1 00:00:02,202 --> 00:00:07,274 [文字转视频] [用广角镜头呈现数字10形状的在轨运行空间站] 2 99:59:59,999 --> 99:59:59,999 2 00:00:10,677 --> 00:00:17,451 [使用Imagen和Veo生成] 3 99:59:59,999 --> 99:59:59,999 3 00:01:11,772 --> 00:01:13,607 这地方要大开眼界了 4 99:59:59,999 --> 99:59:59,999 4 00:01:13,707 --> 00:01:16,176 [使用Veo3生成] 5 99:59:59,999 --> 99:59:59,999 5 00:01:20,314 --> 00:01:27,254 [音乐] 6 99:59:59,999 --> 99:59:59,999 6 00:01:34,595 --> 00:01:38,665 醒醒吧孩子，我们都深陷白日梦 7 99:59:59,999 --> 99:59:59,999 7 00:01:38,732 --> 00:01:42,836 十四岁的年华，一直在向世界低头 8 99:59:59,999 --> 99:59:59,999 8 00:01:43,236 --> 00:01:48,308 却还是在做有礼貌的乖孩子 9 99:59:59,999 --> 99:59:59,999 9 00:01:49,142 --> 00:01:51,478 音乐在你的灵魂里激荡 10 99:59:59,999 --> 99:59:59,999 10 00:01:51,645 --> 00:01:55,482 别放弃，音乐在你的灵魂里激荡 11 99:59:59,999 --> 99:59:59,999 11 00:01:55,849 --> 00:01:57,884 最后一支舞马上开场 12 99:59:59,999 --> 99:59:59,999 12 00:01:57,951 --> 00:02:00,020 这世界终会变好 13 99:59:59,999 --> 99:59:59,999 13 00:02:00,087 --> 00:02:04,257 别放弃，你还有好好生活的理由 14 99:59:59,999 --> 99:59:59,999 14 00:02:04,324 --> 00:02:09,363 别忘记，付出几分，收获几何 15 99:59:59,999 --> 99:59:59,999 15 00:02:09,429 --> 00:02:12,532 [欢迎参加Google I/O大会] 16 99:59:59,999 --> 99:59:59,999 16 00:02:15,102 --> 00:02:16,503 我就要回家了 17 99:59:59,999 --> 99:59:59,999 17 00:02:16,737 --> 00:02:20,641 你是最棒的，现在就让我看到 18 99:59:59,999 --> 99:59:59,999 18 00:02:21,808 --> 00:02:25,178 凌晨四点，我们在奇迹一英里上狂奔 19 99:59:59,999 --> 99:59:59,999 19 00:02:25,379 --> 00:02:29,449 身无分文，但还是有型有格 20 99:59:59,999 --> 99:59:59,999 20 00:02:30,350 --> 00:02:33,854 好棒的视频，看来大家的制作过程应该很开心 21 99:59:59,999 --> 99:59:59,999 21 00:02:34,621 --> 00:02:36,223 大家上午好 22 99:59:59,999 --> 99:59:59,999 22 00:02:36,356 --> 00:02:37,624 欢迎参加Google I/O大会 23 99:59:59,999 --> 99:59:59,999 23 00:02:43,363 --> 00:02:45,832 欢迎亲临Shoreline会场的各位 24 99:59:59,999 --> 99:59:59,999 24 00:02:46,233 --> 00:02:49,736 以及世界各地在线上观看的观众 25 99:59:59,999 --> 99:59:59,999 25 00:02:50,337 --> 00:02:53,273 有人告诉我，Gemini季从今天起拉开序幕 26 99:59:59,999 --> 99:59:59,999 26 00:02:53,940 --> 00:02:55,776 我倒没觉得这有什么特别 27 99:59:59,999 --> 99:59:59,999 27 00:02:55,942 --> 00:02:58,478 因为对Google来说，每一天都是Gemini季 28 99:59:59,999 --> 99:59:59,999 28 00:03:00,180 --> 00:03:03,884 在I/O大会前的几周，我们通常不会有太多动作 29 99:59:59,999 --> 99:59:59,999 29 00:03:04,484 --> 00:03:08,188 因为我们想把最好的模型留到这个舞台上来展示 30 99:59:59,999 --> 99:59:59,999 30 00:03:08,822 --> 00:03:10,257 但在Gemini时代 31 99:59:59,999 --> 99:59:59,999 31 00:03:10,424 --> 00:03:12,626 我们也会在三月随便选个周二 32 99:59:59,999 --> 99:59:59,999 32 00:03:12,693 --> 00:03:15,862 发布我们最智能的模型 33 99:59:59,999 --> 99:59:59,999 33 00:03:16,329 --> 00:03:17,864 或是非凡的突破 34 99:59:59,999 --> 99:59:59,999 34 00:03:17,931 --> 00:03:20,300 就像在大会一周前发布AlphaEvolve那样 35 99:59:59,999 --> 99:59:59,999 35 00:03:21,201 --> 00:03:24,071 只想尽快将我们最好的模型交到你的手中 36 99:59:59,999 --> 99:59:59,999 36 00:03:24,404 --> 00:03:26,573 让你尽快体验到我们的产品 37 99:59:59,999 --> 99:59:59,999 37 00:03:27,074 --> 00:03:29,976 为此，我们的发布频率远超以往 38 99:59:59,999 --> 99:59:59,999 38 00:03:30,711 --> 00:03:31,878 自上届I/O大会至今 39 99:59:59,999 --> 99:59:59,999 39 00:03:31,945 --> 00:03:34,314 我们发布了十多种模型和研究突破 40 99:59:59,999 --> 99:59:59,999 40 00:03:34,548 --> 00:03:40,087 以及二十多种重要AI产品和功能 41 99:59:59,999 --> 99:59:59,999 41 00:03:40,987 --> 00:03:44,558 模型的快速进步让我振奋不已 42 99:59:59,999 --> 99:59:59,999 42 00:03:45,125 --> 00:03:47,427 从这张图表就能看出阶跃式的发展 43 99:59:59,999 --> 99:59:59,999 43 00:03:48,228 --> 00:03:50,697 Elo分数用来衡量进步的尺度 44 99:59:59,999 --> 99:59:59,999 44 00:03:50,964 --> 00:03:53,066 从第一代Gemini Pro至今 45 99:59:59,999 --> 99:59:59,999 45 00:03:53,133 --> 00:03:55,235 这项成绩提升了300多分 46 99:59:59,999 --> 99:59:59,999 46 00:03:55,769 --> 00:03:58,505 如今，Gemini 2.5 Pro 47 99:59:59,999 --> 99:59:59,999 47 00:03:58,572 --> 00:04:02,175 横扫LMArena排行榜的所有类别 48 99:59:59,999 --> 99:59:59,999 48 00:04:09,549 --> 00:04:12,452 在许多基准测试中都表现突出 49 99:59:59,999 --> 99:59:59,999 49 00:04:12,753 --> 00:04:15,589 在编程方面，我们也在快速进步 50 99:59:59,999 --> 99:59:59,999 50 00:04:16,056 --> 00:04:19,893 更新版的2.5 Pro在WebDev Arena中一马当先 51 99:59:59,999 --> 99:59:59,999 51 00:04:20,127 --> 00:04:24,631 Elo分数比上一个版本高142分 52 99:59:59,999 --> 99:59:59,999 52 00:04:25,432 --> 00:04:28,301 这款模型在各大主流编程平台上备受追捧 53 99:59:59,999 --> 99:59:59,999 53 00:04:28,368 --> 00:04:29,569 感谢所有人的支持 54 99:59:59,999 --> 99:59:59,999 54 00:04:30,203 --> 00:04:33,273 在领先AI代码编辑器cursor中 55 99:59:59,999 --> 99:59:59,999 55 00:04:33,673 --> 00:04:36,576 Gemini成为年度发展速度最快的模型 56 99:59:59,999 --> 99:59:59,999 56 00:04:36,877 --> 00:04:41,782 每分钟都有数十万行代码用它生成 57 99:59:59,999 --> 99:59:59,999 57 00:04:42,816 --> 00:04:46,019 最后这个里程碑或许最令人动容 58 99:59:59,999 --> 99:59:59,999 58 00:04:46,219 --> 00:04:47,788 至少在某些圈子里是如此 59 99:59:59,999 --> 99:59:59,999 59 00:04:48,288 --> 00:04:51,825 几周前，Gemini成功通关《宝可梦蓝》 60 99:59:59,999 --> 99:59:59,999 60 00:04:55,796 --> 00:04:59,432 集齐全部八枚徽章，挺进“冠军之路” 61 99:59:59,999 --> 99:59:59,999 61 00:04:59,499 --> 00:05:02,269 打败了四天王和联盟冠军 62 99:59:59,999 --> 99:59:59,999 62 00:05:02,636 --> 00:05:05,972 这让我们离实现API又近了一步 63 99:59:59,999 --> 99:59:59,999 63 00:05:06,406 --> 00:05:08,809 不过我们说的是“人工宝可梦智能” 64 99:59:59,999 --> 99:59:59,999 64 00:05:14,147 --> 00:05:17,818 所有这些进步都由我们卓越的基础设施提供支持 65 99:59:59,999 --> 99:59:59,999 65 00:05:17,918 --> 00:05:20,754 它是我们全栈式AI方法的基石 66 99:59:59,999 --> 99:59:59,999 66 00:05:21,454 --> 00:05:25,826 我们的第七代TPU Ironwood采用前沿设计 67 99:59:59,999 --> 99:59:59,999 67 00:05:25,926 --> 00:05:28,361 为大规模思考和推理赋能助力 68 99:59:59,999 --> 99:59:59,999 68 00:05:28,995 --> 00:05:32,332 其性能达到上一代的十倍 69 99:59:59,999 --> 99:59:59,999 69 00:05:32,666 --> 00:05:38,138 能迸发出每个pod 42.5 exaFLOPS的强大算力 70 99:59:59,999 --> 99:59:59,999 70 00:05:38,238 --> 00:05:39,072 令人赞叹 71 99:59:59,999 --> 99:59:59,999 71 00:05:39,573 --> 00:05:42,809 Google Cloud客户今年稍后就能切身体会它的实力 72 99:59:59,999 --> 99:59:59,999 72 00:05:46,880 --> 00:05:49,916 凭借始于底层TPU的基础设施实力 73 99:59:59,999 --> 99:59:59,999 73 00:05:49,983 --> 00:05:53,753 我们提供了速度更出众的模型 74 99:59:59,999 --> 99:59:59,999 74 00:05:54,287 --> 00:05:56,756 在LMArena排行榜的顶尖模型中 75 99:59:59,999 --> 99:59:59,999 75 00:05:56,823 --> 00:05:59,192 Gemini跻身前三 76 99:59:59,999 --> 99:59:59,999 76 00:05:59,392 --> 00:06:02,395 凭借最高的每秒生成token数 77 99:59:59,999 --> 99:59:59,999 77 00:06:02,896 --> 00:06:06,766 以及大幅下降的模型价格取得骄人成绩 78 99:59:59,999 --> 99:59:59,999 78 00:06:07,334 --> 00:06:10,170 在价格与性能之间权衡并不容易 79 99:59:59,999 --> 99:59:59,999 79 00:06:10,770 --> 00:06:13,006 然而，我们一次又一次地做到了 80 99:59:59,999 --> 99:59:59,999 80 00:06:13,073 --> 00:06:16,610 以最有效的价格点提供出色的模型 81 99:59:59,999 --> 99:59:59,999 81 00:06:17,244 --> 00:06:19,779 Google不仅在这个帕累托最优边界中先人一步 82 99:59:59,999 --> 99:59:59,999 82 00:06:20,046 --> 00:06:22,382 更从根本上改变了它 83 99:59:59,999 --> 99:59:59,999 83 00:06:23,316 --> 00:06:28,088 让每个人都能随时随地享受更好的智能体验 84 99:59:59,999 --> 99:59:59,999 84 00:06:28,688 --> 00:06:30,757 我们的努力收获了积极的回应 85 99:59:59,999 --> 99:59:59,999 85 00:06:30,857 --> 00:06:33,059 AI采用速度远超以往 86 99:59:59,999 --> 99:59:59,999 86 00:06:34,060 --> 00:06:37,664 作为进步的标志之一，在去年此时 87 99:59:59,999 --> 99:59:59,999 87 00:06:38,064 --> 00:06:44,571 我们所有产品和API 每个月的token处理总量是9.7万亿 88 99:59:59,999 --> 99:59:59,999 88 00:06:45,105 --> 00:06:49,376 而如今，每月token处理量达到了480万亿 89 99:59:59,999 --> 99:59:59,999 89 00:06:49,676 --> 00:06:53,046 短短一年，接近50倍的增幅 90 99:59:59,999 --> 99:59:59,999 90 00:06:58,985 --> 00:07:03,223 我们各类开发者AI工具的采用已成浪潮 91 99:59:59,999 --> 99:59:59,999 91 00:07:03,623 --> 00:07:06,226 在Google AI Studio和Vertex AI中 92 99:59:59,999 --> 99:59:59,999 92 00:07:06,293 --> 00:07:10,964 有700多万名开发者在使用Gemini API进行开发 93 99:59:59,999 --> 99:59:59,999 93 00:07:11,364 --> 00:07:13,767 达到去年召开I/O大会时的五倍多 94 99:59:59,999 --> 99:59:59,999 94 00:07:14,234 --> 00:07:15,635 自去年以来 95 99:59:59,999 --> 99:59:59,999 95 00:07:15,702 --> 00:07:19,839 Gemini在Vertex AI中的使用量增加到40多倍 96 99:59:59,999 --> 99:59:59,999 96 00:07:20,774 --> 00:07:23,443 我们的产品也在越来越多地采用AI 97 99:59:59,999 --> 99:59:59,999 97 00:07:23,677 --> 00:07:27,814 Gemini应用的月活用户数已超过4亿 98 99:59:59,999 --> 99:59:59,999 98 00:07:28,248 --> 00:07:30,850 增长幅度和参与度都在强势增长 99 99:59:59,999 --> 99:59:59,999 99 00:07:31,184 --> 00:07:33,253 尤其是2.5版的模型 100 99:59:59,999 --> 99:59:59,999 100 00:07:33,720 --> 00:07:36,756 对于在Gemini应用中使用2.5 Pro模型的用户 101 99:59:59,999 --> 99:59:59,999 101 00:07:36,990 --> 00:07:39,459 使用量增长了45% 102 99:59:59,999 --> 99:59:59,999 102 00:07:39,893 --> 00:07:42,996 稍后你会听到许多与Gemini应用有关的消息 103 99:59:59,999 --> 99:59:59,999 103 00:07:43,863 --> 00:07:46,266 在搜索领域，我们也看到了十分惊人的发展势头 104 99:59:59,999 --> 99:59:59,999 104 00:07:46,800 --> 00:07:52,372 AI概览的月用户量现已超过15亿 105 99:59:59,999 --> 99:59:59,999 105 00:07:53,106 --> 00:07:56,443 也就是说，通过Google搜索使用生成式AI的用户量 106 99:59:59,999 --> 99:59:59,999 106 00:07:56,943 --> 00:08:00,246 超过全球其他任何产品 107 99:59:59,999 --> 99:59:59,999 107 00:08:00,847 --> 00:08:06,186 除了AI概览之外，AI模式也是搜索领域的一大进步 108 99:59:59,999 --> 99:59:59,999 108 00:08:06,653 --> 00:08:08,621 稍后我们会分享更多相关信息 109 99:59:59,999 --> 99:59:59,999 109 00:08:09,789 --> 00:08:11,658 所有这些进步都指向一件事 110 99:59:59,999 --> 99:59:59,999 110 00:08:12,492 --> 00:08:15,528 我们身处AI平台变革的新阶段 111 99:59:59,999 --> 99:59:59,999 111 00:08:16,196 --> 00:08:19,265 数十载的研究成果转化为实用工具 112 99:59:59,999 --> 99:59:59,999 112 00:08:19,332 --> 00:08:21,301 惠及普罗大众 113 99:59:59,999 --> 99:59:59,999 113 00:08:21,968 --> 00:08:23,770 我想分享三个例子 114 99:59:59,999 --> 99:59:59,999 114 00:08:23,837 --> 00:08:27,407 证明研究如何彻底改变我们当今的产品 115 99:59:59,999 --> 99:59:59,999 115 00:08:27,841 --> 00:08:31,678 分别是Project Starline、Astra和Mariner 116 99:59:59,999 --> 99:59:59,999 116 00:08:32,712 --> 00:08:34,447 在几年前的I/O大会上 117 99:59:59,999 --> 99:59:59,999 117 00:08:34,514 --> 00:08:36,583 我们公布了Project Starline 118 99:59:59,999 --> 99:59:59,999 118 00:08:36,649 --> 00:08:38,184 这是我们突破性的3D视频技术 119 99:59:59,999 --> 99:59:59,999 119 00:08:38,752 --> 00:08:42,322 让你在与千里之外的人交流时 120 99:59:59,999 --> 99:59:59,999 120 00:08:42,722 --> 00:08:44,324 也能有共处一室的感觉 121 99:59:59,999 --> 99:59:59,999 121 00:08:44,824 --> 00:08:47,093 我们不断改进相关技术 122 99:59:59,999 --> 99:59:59,999 122 00:08:47,494 --> 00:08:50,330 今天，我们终于准备好开启它的下一篇章 123 99:59:59,999 --> 99:59:59,999 123 00:08:50,697 --> 00:08:52,298 隆重介绍Google Beam 124 99:59:59,999 --> 99:59:59,999 124 00:08:52,632 --> 00:08:56,703 全新的AI优先型视频沟通平台 125 99:59:59,999 --> 99:59:59,999 125 00:08:57,604 --> 00:09:00,874 Beam运用全新的先进视频模型 126 99:59:59,999 --> 99:59:59,999 126 00:09:01,374 --> 00:09:06,212 将2D视频流转化为栩栩如生的3D体验 127 99:59:59,999 --> 99:59:59,999 127 00:09:07,113 --> 00:09:12,252 在幕后，六摄像头阵列从不同角度拍摄你的影像 128 99:59:59,999 --> 99:59:59,999 128 00:09:12,819 --> 00:09:16,823 这些视频流随后会通过AI合并在一起 129 99:59:59,999 --> 99:59:59,999 129 00:09:16,990 --> 00:09:19,592 生成3D光场显示，展示你的立体形象 130 99:59:59,999 --> 99:59:59,999 130 00:09:20,126 --> 00:09:23,163 它能实现毫米级精度的头部跟踪 131 99:59:59,999 --> 99:59:59,999 131 00:09:23,530 --> 00:09:26,966 以及每秒60帧的实时渲染 132 99:59:59,999 --> 99:59:59,999 132 00:09:27,567 --> 00:09:33,239 成就高度自然流畅、沉浸感十足的对话体验 133 99:59:59,999 --> 99:59:59,999 133 00:09:33,807 --> 00:09:36,509 我们热切期待将这项技术带给大家 134 99:59:59,999 --> 99:59:59,999 134 00:09:37,143 --> 00:09:38,611 并为此与惠普开展了合作 135 99:59:59,999 --> 99:59:59,999 135 00:09:38,678 --> 00:09:40,580 首款Google Beam设备 136 99:59:59,999 --> 99:59:59,999 136 00:09:40,914 --> 00:09:44,884 将于今年晚些时候与首批客户见面 137 99:59:59,999 --> 99:59:59,999 137 00:09:45,985 --> 00:09:48,588 从现在起的几周内，惠普会分享大量相关资讯 138 99:59:59,999 --> 99:59:59,999 138 00:09:48,855 --> 00:09:49,823 敬请关注 139 99:59:59,999 --> 99:59:59,999 139 00:09:56,596 --> 00:09:59,099 多年来，我们一直在不懈努力 140 99:59:59,999 --> 99:59:59,999 140 00:09:59,165 --> 00:10:02,001 将Starline的底层技术整合到Google Meet中 141 99:59:59,999 --> 99:59:59,999 141 00:10:02,602 --> 00:10:07,107 其中包括帮助打破语言壁垒的实时语音翻译 142 99:59:59,999 --> 99:59:59,999 142 00:10:08,041 --> 00:10:11,144 这个例子展示了在南美洲预定度假屋 143 99:59:59,999 --> 99:59:59,999 143 00:10:11,411 --> 00:10:14,180 而你又不会讲当地语言时 144 99:59:59,999 --> 99:59:59,999 144 00:10:14,447 --> 00:10:15,949 这项功能的实用价值 145 99:59:59,999 --> 99:59:59,999 145 00:10:16,049 --> 00:10:16,883 我们一起来看看吧 146 99:59:59,999 --> 99:59:59,999 146 00:10:18,952 --> 00:10:20,153 嗨，Camilla 147 99:59:59,999 --> 99:59:59,999 147 00:10:20,487 --> 00:10:22,589 我来打开语音翻译功能 148 99:59:59,999 --> 99:59:59,999 148 00:10:25,024 --> 00:10:27,127 终于能跟你直接对话了，太棒了 149 99:59:59,999 --> 99:59:59,999 149 00:10:27,227 --> 00:10:31,364 [西班牙语翻译] 150 99:59:59,999 --> 99:59:59,999 150 00:10:32,398 --> 00:10:34,334 [西班牙语回答] 151 99:59:59,999 --> 99:59:59,999 151 00:10:34,400 --> 00:10:35,635 你会很开心的 152 99:59:59,999 --> 99:59:59,999 152 00:10:35,735 --> 00:10:38,238 我相信你也一定会爱上这座城市 153 99:59:59,999 --> 99:59:59,999 153 00:10:38,471 --> 00:10:42,475 这间房所在的社区环境很好，还能俯瞰山景 154 99:59:59,999 --> 99:59:59,999 154 00:10:42,709 --> 00:10:44,144 听起来真不错 155 99:59:59,999 --> 99:59:59,999 155 00:10:45,411 --> 00:10:50,450 [西班牙语翻译] 156 99:59:59,999 --> 99:59:59,999 156 00:10:51,050 --> 00:10:53,119 [西班牙语回答] 157 99:59:59,999 --> 99:59:59,999 157 00:10:53,253 --> 00:10:54,587 附近就有公交车 158 99:59:59,999 --> 99:59:59,999 158 00:10:54,921 --> 00:11:00,260 不过我推荐你租辆车，这样去野外玩更方便 159 99:59:59,999 --> 99:59:59,999 159 00:11:00,727 --> 00:11:02,529 听起来不错 160 99:59:59,999 --> 99:59:59,999 160 00:11:02,829 --> 00:11:04,264 [西班牙语翻译] 161 99:59:59,999 --> 99:59:59,999 161 00:11:11,905 --> 00:11:14,641 如你所见，语音翻译惟妙惟肖地模仿了 162 99:59:59,999 --> 99:59:59,999 162 00:11:14,707 --> 00:11:17,877 讲话人的语气、模式甚至是表达方式 163 99:59:59,999 --> 99:59:59,999 163 00:11:18,811 --> 00:11:24,117 我们距离自然流畅的跨语言沟通又近了一步 164 99:59:59,999 --> 99:59:59,999 164 00:11:24,684 --> 00:11:28,321 我们现已直接将这种实时语音翻译功能 165 99:59:59,999 --> 99:59:59,999 165 00:11:28,388 --> 00:11:29,956 直接整合到Google Meet中 166 99:59:59,999 --> 99:59:59,999 166 00:11:30,523 --> 00:11:34,561 订阅者从现在起即可使用英语与西班牙语互译功能 167 99:59:59,999 --> 99:59:59,999 167 00:11:34,928 --> 00:11:38,932 在未来几周内，我们还会推出更多语言的翻译功能 168 99:59:59,999 --> 99:59:59,999 168 00:11:39,499 --> 00:11:43,436 实时翻译功能将于今年晚些时候面向企业用户开放 169 99:59:59,999 --> 99:59:59,999 169 00:11:50,443 --> 00:11:52,378 另一个在往届I/O舞台上正式亮相的 170 99:59:59,999 --> 99:59:59,999 170 00:11:52,445 --> 00:11:54,781 早期研究项目是Project Astra 171 99:59:59,999 --> 99:59:59,999 171 00:11:55,381 --> 00:12:01,955 专门探索通用型AI助手未来理解你身边世界的能力 172 99:59:59,999 --> 99:59:59,999 172 00:12:02,488 --> 00:12:04,224 我们正在将其整合到Google产品中 173 99:59:59,999 --> 99:59:59,999 173 00:12:04,891 --> 00:12:06,426 Gemini Live现已整合 174 99:59:59,999 --> 99:59:59,999 174 00:12:06,492 --> 00:12:09,896 Project Astra的相机和屏幕共享功能 175 99:59:59,999 --> 99:59:59,999 175 00:12:10,263 --> 00:12:12,565 支持就你看到的任何事物展开对话 176 99:59:59,999 --> 99:59:59,999 176 00:12:13,032 --> 00:12:15,101 人们正在将它用于各种用途 177 99:59:59,999 --> 99:59:59,999 177 00:12:15,401 --> 00:12:17,470 无论是为工作面试做练习 178 99:59:59,999 --> 99:59:59,999 178 00:12:17,704 --> 00:12:19,339 还是为备战马拉松而训练 179 99:59:59,999 --> 99:59:59,999 179 00:12:19,806 --> 00:12:22,942 我们始终高度重视可信测试员的反馈 180 99:59:59,999 --> 99:59:59,999 180 00:12:23,309 --> 00:12:26,179 还有一些不那么“可信”的测试员 181 99:59:59,999 --> 99:59:59,999 181 00:12:26,246 --> 00:12:27,247 一起来看看吧 182 99:59:59,999 --> 99:59:59,999 182 00:12:30,350 --> 00:12:31,951 好漂亮的敞篷车 183 99:59:59,999 --> 99:59:59,999 183 00:12:32,585 --> 00:12:36,055 我觉得你把垃圾车误认成敞篷车了 184 99:59:59,999 --> 99:59:59,999 184 00:12:36,155 --> 00:12:37,890 我还能帮你些什么？ 185 99:59:59,999 --> 99:59:59,999 185 00:12:38,024 --> 00:12:40,326 这个细长的建筑物怎么杵在我家门口？ 186 99:59:59,999 --> 99:59:59,999 186 00:12:40,460 --> 00:12:43,029 这是路灯，不是建筑物 187 99:59:59,999 --> 99:59:59,999 187 00:12:43,496 --> 00:12:46,566 这些棕榈树怎么这么矮？看着让人有点担心 188 99:59:59,999 --> 99:59:59,999 188 00:12:47,100 --> 00:12:50,003 它们并不矮，其实很高的 189 99:59:59,999 --> 99:59:59,999 189 00:12:50,570 --> 00:12:51,938 帅气的敞篷车 190 99:59:59,999 --> 99:59:59,999 190 00:12:52,238 --> 00:12:53,740 还是垃圾车 191 99:59:59,999 --> 99:59:59,999 191 00:12:54,173 --> 00:12:55,408 还有别的事吗？ 192 99:59:59,999 --> 99:59:59,999 192 00:12:55,475 --> 00:12:58,444 为什么总有人把快递放到我的草坪上？ 193 99:59:59,999 --> 99:59:59,999 193 00:12:58,978 --> 00:13:01,581 那不是快递，是个变电箱 194 99:59:59,999 --> 99:59:59,999 194 00:13:02,015 --> 00:13:04,784 我走路的时候，这个人怎么一直跟踪我 195 99:59:59,999 --> 99:59:59,999 195 00:13:05,618 --> 00:13:08,521 没人在跟踪你，那只是你的影子 196 99:59:59,999 --> 99:59:59,999 196 00:13:17,196 --> 00:13:19,766 Gemini真的特别擅长给你挑错 197 99:59:59,999 --> 99:59:59,999 197 00:13:21,234 --> 00:13:26,572 即日起，此功能面向所有Android和iOS用户开放 198 99:59:59,999 --> 99:59:59,999 198 00:13:34,147 --> 00:13:37,784 再说说我们的研究原型Project Mariner 199 99:59:59,999 --> 99:59:59,999 199 00:13:38,484 --> 00:13:42,422 这种智能体可以与Web交互并完成各种任务 200 99:59:59,999 --> 99:59:59,999 200 00:13:43,256 --> 00:13:45,358 我们跳出原有思维模式，将智能体视为一种系统 201 99:59:59,999 --> 99:59:59,999 201 00:13:45,425 --> 00:13:50,530 集先进AI模型的智能和访问工具的能力于一身 202 99:59:59,999 --> 99:59:59,999 202 00:13:51,164 --> 00:13:54,634 它们可以自主帮你执行操作，而且完全由你掌控 203 99:59:59,999 --> 99:59:59,999 203 00:13:55,234 --> 00:13:58,538 使用计算机是智能体的一项重要能力 204 99:59:59,999 --> 99:59:59,999 204 00:13:59,072 --> 00:14:02,942 让智能体能够与浏览器和其他软件交互 205 99:59:59,999 --> 99:59:59,999 205 00:14:03,242 --> 00:14:04,577 并对其执行操作 206 99:59:59,999 --> 99:59:59,999 206 00:14:05,345 --> 00:14:09,782 Project Mariner是测试计算机使用能力的初步尝试 207 99:59:59,999 --> 99:59:59,999 207 00:14:10,383 --> 00:14:13,586 去年12月，我们将其作为早期研究原型发布 208 99:59:59,999 --> 99:59:59,999 208 00:14:13,886 --> 00:14:16,022 自那之后取得了长足进展 209 99:59:59,999 --> 99:59:59,999 209 00:14:16,556 --> 00:14:19,625 首先，我们引入了多任务处理能力 210 99:59:59,999 --> 99:59:59,999 210 00:14:19,859 --> 00:14:23,262 它如今能同时管理多达10个任务 211 99:59:59,999 --> 99:59:59,999 211 00:14:24,230 --> 00:14:27,333 其次，它使用了一项名为“示教与重复”的功能 212 99:59:59,999 --> 99:59:59,999 212 00:14:27,767 --> 00:14:30,370 你只要向它展示一次任务 213 99:59:59,999 --> 99:59:59,999 213 00:14:30,737 --> 00:14:34,107 它就能学会在将来为类似任务做计划 214 99:59:59,999 --> 99:59:59,999 214 00:14:34,707 --> 00:14:38,077 我们将通过Gemini API向开发者提供 215 99:59:59,999 --> 99:59:59,999 215 00:14:38,411 --> 00:14:40,947 Project Mariner的计算机使用能力 216 99:59:59,999 --> 99:59:59,999 216 00:14:41,748 --> 00:14:44,550 Automation Anywhere和UiPath等可信测试方 217 99:59:59,999 --> 99:59:59,999 217 00:14:44,617 --> 00:14:47,720 已经开始使用它构建产品 218 99:59:59,999 --> 99:59:59,999 218 00:14:48,187 --> 00:14:51,124 我们将于今年夏季将其提供给更广泛的用户 219 99:59:59,999 --> 99:59:59,999 219 00:14:51,991 --> 00:14:53,893 为了推动智能体生态系统繁荣发展 220 99:59:59,999 --> 99:59:59,999 220 00:14:53,960 --> 00:14:55,728 我们需要开发更广泛的工具组合 221 99:59:59,999 --> 99:59:59,999 221 00:14:55,795 --> 00:14:57,897 计算机使用能力就是这项工作的一环 222 99:59:59,999 --> 99:59:59,999 222 00:14:58,431 --> 00:15:00,032 这方面的例子有我们的开放智能体间协议 223 99:59:59,999 --> 99:59:59,999 223 00:15:00,099 --> 00:15:02,635 让智能体能够彼此对话 224 99:59:59,999 --> 99:59:59,999 224 00:15:03,269 --> 00:15:04,837 我们在Cloud Next大会上发布了这种协议 225 99:59:59,999 --> 99:59:59,999 225 00:15:04,904 --> 00:15:07,607 获得了60多家技术合作伙伴的支持 226 99:59:59,999 --> 99:59:59,999 226 00:15:07,907 --> 00:15:09,609 这个数字还有望继续增长 227 99:59:59,999 --> 99:59:59,999 227 00:15:10,343 --> 00:15:14,080 还有Anthropic推出的模型上下文协议 228 99:59:59,999 --> 99:59:59,999 228 00:15:14,213 --> 00:15:16,616 让智能体可以访问其他服务 229 99:59:59,999 --> 99:59:59,999 229 00:15:17,150 --> 00:15:19,118 今天，我们要激动万分地宣布 230 99:59:59,999 --> 99:59:59,999 230 00:15:19,185 --> 00:15:23,656 Gemini SDK现已兼容MCP工具 231 99:59:59,999 --> 99:59:59,999 231 00:15:29,462 --> 00:15:33,533 这些技术将协同运作，让智能体更实用 232 99:59:59,999 --> 99:59:59,999 232 00:15:33,900 --> 00:15:36,769 我们已经开始将代理式功能添加到 233 99:59:59,999 --> 99:59:59,999 233 00:15:36,869 --> 00:15:38,571 Chrome、Google搜索和Gemini应用中 234 99:59:59,999 --> 99:59:59,999 234 00:15:38,638 --> 00:15:42,975 我来展示一下Gemini应用中最让我们激动的功能 235 99:59:59,999 --> 99:59:59,999 235 00:15:43,142 --> 00:15:45,111 我们称之为智能体模式 236 99:59:59,999 --> 99:59:59,999 236 00:15:45,711 --> 00:15:49,715 假设你想在奥斯汀给自己和两名室友找间公寓 237 99:59:59,999 --> 99:59:59,999 237 00:15:50,116 --> 00:15:52,752 你们每个人每月的预算是1200美元 238 99:59:59,999 --> 99:59:59,999 238 00:15:53,352 --> 00:15:54,754 希望公寓里有洗衣机、干衣机 239 99:59:59,999 --> 99:59:59,999 239 00:15:54,821 --> 00:15:56,823 或者至少附近要有间洗衣店 240 99:59:59,999 --> 99:59:59,999 240 00:15:57,323 --> 00:16:01,394 通常情况下，你要花很多时间去浏览无数房源 241 99:59:59,999 --> 99:59:59,999 241 00:16:02,161 --> 00:16:05,364 但有了智能体模式，Gemini应用可以为你效力 242 99:59:59,999 --> 99:59:59,999 242 00:16:05,431 --> 00:16:07,800 它在后台访问Zillow等网站 243 99:59:59,999 --> 99:59:59,999 243 00:16:07,867 --> 00:16:10,803 查找符合你的条件的房源 244 99:59:59,999 --> 99:59:59,999 244 00:16:11,437 --> 00:16:13,673 并在需要时使用Project Mariner 245 99:59:59,999 --> 99:59:59,999 245 00:16:13,739 --> 00:16:16,008 调整极其细致的过滤条件 246 99:59:59,999 --> 99:59:59,999 246 00:16:16,542 --> 00:16:20,112 如果有你想具体了解的公寓，Gemini会使用MCP 247 99:59:59,999 --> 99:59:59,999 247 00:16:20,513 --> 00:16:24,750 访问房源，甚至可以帮你安排行程 248 99:59:59,999 --> 99:59:59,999 248 00:16:25,184 --> 00:16:28,888 只要你需要，它就会不断浏览查找新房源 249 99:59:59,999 --> 99:59:59,999 249 00:16:29,689 --> 00:16:32,225 让你可以腾出时间去做自己想做的事情 250 99:59:59,999 --> 99:59:59,999 250 00:16:32,291 --> 00:16:33,826 比如规划乔迁派对 251 99:59:59,999 --> 99:59:59,999 251 00:16:34,460 --> 00:16:36,095 这对Zillow这样的公司特别有用 252 99:59:59,999 --> 99:59:59,999 252 00:16:36,162 --> 00:16:39,632 能够带来新客源并提高转化率 253 99:59:59,999 --> 99:59:59,999 253 00:16:40,399 --> 00:16:43,536 Gemini应用中智能体模式的实验版本 254 99:59:59,999 --> 99:59:59,999 254 00:16:43,603 --> 00:16:46,105 即将面向订阅者开放 255 99:59:59,999 --> 99:59:59,999 255 00:16:46,706 --> 00:16:48,875 这是一个崭露头角的全新领域 256 99:59:59,999 --> 99:59:59,999 256 00:16:49,342 --> 00:16:50,977 我们很激动能亲身探索 257 99:59:59,999 --> 99:59:59,999 257 00:16:51,077 --> 00:16:56,482 如何将智能体的优势带给用户和更广泛的生态系统 258 99:59:59,999 --> 99:59:59,999 258 00:16:57,717 --> 00:17:00,453 要将研究成果落地，我们最好的方案就是 259 99:59:59,999 --> 99:59:59,999 259 00:17:00,586 --> 00:17:03,856 保证它在你的实际环境中拥有真正的实用价值 260 99:59:59,999 --> 99:59:59,999 260 00:17:04,423 --> 00:17:07,994 这正是个性定制的用武之地 261 99:59:59,999 --> 99:59:59,999 261 00:17:08,594 --> 00:17:12,999 为此，我们推出了个人使用场景功能 262 99:59:59,999 --> 99:59:59,999 262 00:17:13,866 --> 00:17:15,034 在征得你的许可之后 263 99:59:59,999 --> 99:59:59,999 263 00:17:15,167 --> 00:17:17,136 Gemini模型可在你的各Google应用之间 264 99:59:59,999 --> 99:59:59,999 264 00:17:17,203 --> 00:17:19,038 使用相关背景信息 265 99:59:59,999 --> 99:59:59,999 265 00:17:19,705 --> 00:17:21,974 并且采用一种保护隐私、高度透明 266 99:59:59,999 --> 99:59:59,999 266 00:17:22,041 --> 00:17:23,509 完全由你掌控的方式 267 99:59:59,999 --> 99:59:59,999 267 00:17:24,243 --> 00:17:26,479 我来展示一个Gmail中的例子 268 99:59:59,999 --> 99:59:59,999 268 00:17:27,146 --> 00:17:30,483 你可能很熟悉我们依托AI技术的智能回复功能 269 99:59:59,999 --> 99:59:59,999 269 00:17:30,650 --> 00:17:32,885 其受欢迎的程度甚至令人有几分意外 270 99:59:59,999 --> 99:59:59,999 270 00:17:33,119 --> 00:17:36,556 如果AI生成的回复能效仿你的口吻，那该有多好 271 99:59:59,999 --> 99:59:59,999 271 00:17:37,189 --> 00:17:40,293 这就是个性化智能回复背后的理念 272 99:59:59,999 --> 99:59:59,999 272 00:17:41,394 --> 00:17:43,996 假设朋友写信给我，向我寻求建议 273 99:59:59,999 --> 99:59:59,999 273 00:17:44,530 --> 00:17:49,835 他打算到犹他州去自驾游，记得我曾自驾去过那里 274 99:59:59,999 --> 99:59:59,999 274 00:17:50,570 --> 00:17:52,538 坦白说，如果我自己写回信 275 99:59:59,999 --> 99:59:59,999 275 00:17:52,605 --> 00:17:54,740 只会寥寥几句带过，帮不上什么忙 276 99:59:59,999 --> 99:59:59,999 276 00:17:55,041 --> 00:17:56,042 抱歉了，Felix 277 99:59:59,999 --> 99:59:59,999 277 00:17:57,009 --> 00:18:00,313 但使用个性化智能回复，我就能成为知心好友 278 99:59:59,999 --> 99:59:59,999 278 00:18:01,013 --> 00:18:05,685 因为Gemini会帮我做好几乎所有的“功课” 279 99:59:59,999 --> 99:59:59,999 279 00:18:06,285 --> 00:18:08,254 查看我云端硬盘中的笔记 280 99:59:59,999 --> 99:59:59,999 280 00:18:08,554 --> 00:18:11,023 查询旧电子邮件中的预订信息 281 99:59:59,999 --> 99:59:59,999 281 00:18:11,324 --> 00:18:13,926 查找Google文档中的行程安排 282 99:59:59,999 --> 99:59:59,999 282 00:18:14,260 --> 00:18:15,861 锡安国家公园之旅 283 99:59:59,999 --> 99:59:59,999 283 00:18:16,629 --> 00:18:20,099 Gemini效仿我既往邮件中的问候语 284 99:59:59,999 --> 99:59:59,999 284 00:18:20,566 --> 00:18:25,571 捕捉到了我的语气、风格和遣词用句的喜好 285 99:59:59,999 --> 99:59:59,999 285 00:18:25,805 --> 00:18:28,641 随后自动生成回复 286 99:59:59,999 --> 99:59:59,999 286 00:18:29,375 --> 00:18:31,143 我最欣赏的是它添加了贴心的细节 287 99:59:59,999 --> 99:59:59,999 287 00:18:31,210 --> 00:18:33,879 比如将驾车时间控制在每天五小时以下 288 99:59:59,999 --> 99:59:59,999 288 00:18:34,213 --> 00:18:37,350 还使用了我最喜欢的形容词：激动人心 289 99:59:59,999 --> 99:59:59,999 289 00:18:39,018 --> 00:18:40,252 看起来不错 290 99:59:59,999 --> 99:59:59,999 290 00:18:40,319 --> 00:18:43,689 或许简单修改几处，就可以点击“发送”了 291 99:59:59,999 --> 99:59:59,999 291 00:18:44,323 --> 00:18:48,027 今年夏天，这项功能就会在Gmail中提供给订阅者 292 99:59:59,999 --> 99:59:59,999 292 00:18:53,933 --> 00:18:56,569 可想而知，覆盖搜索、文档、Gemini 293 99:59:59,999 --> 99:59:59,999 293 00:18:56,636 --> 00:19:00,306 及其他产品的个人使用场景功能会有多实用 294 99:59:59,999 --> 99:59:59,999 294 00:19:01,474 --> 00:19:06,245 今天我谈到了智能、智能体和个性化 295 99:59:59,999 --> 99:59:59,999 295 00:19:06,846 --> 00:19:10,082 这是我们将努力推进的几个前沿领域 296 99:59:59,999 --> 99:59:59,999 296 00:19:10,149 --> 00:19:13,185 在整场主旨演讲中，你会听到更多例子 297 99:59:59,999 --> 99:59:59,999 297 00:19:13,919 --> 00:19:15,454 但首先，我想请一个人 298 99:59:59,999 --> 99:59:59,999 298 00:19:15,521 --> 00:19:19,091 详细分享有关推动我们未来创新的智能 299 99:59:59,999 --> 99:59:59,999 299 00:19:19,792 --> 00:19:22,828 去年在介绍他的时候，我称他为Demis爵士 300 99:59:59,999 --> 99:59:59,999 300 00:19:23,129 --> 00:19:27,233 今年，我们还可以在他的头衔里 加上“诺贝尔奖获得者” 301 99:59:59,999 --> 99:59:59,999 301 00:19:27,299 --> 00:19:28,834 Demis，请上台来 302 99:59:59,999 --> 99:59:59,999 302 00:19:43,482 --> 00:19:46,485 大家好！很高兴能再次参加Google I/O大会 303 99:59:59,999 --> 99:59:59,999 303 00:19:47,586 --> 00:19:50,189 我们身处一个重要的历史节点 304 99:59:59,999 --> 99:59:59,999 304 00:19:50,690 --> 00:19:54,026 AI正在开启令人向往的全新未来 305 99:59:59,999 --> 99:59:59,999 305 00:19:54,760 --> 00:19:56,395 这一年，我们在不懈的探索中前行 306 99:59:59,999 --> 99:59:59,999 306 00:19:56,462 --> 00:19:57,596 收获了累累硕果 307 99:59:59,999 --> 99:59:59,999 307 00:19:58,264 --> 00:20:02,268 正如Sundar所说，人们喜欢与Gemini 2.5交互 308 99:59:59,999 --> 99:59:59,999 308 00:20:03,002 --> 00:20:04,770 我们见证了你们用它进行氛围编程 309 99:59:59,999 --> 99:59:59,999 309 00:20:04,870 --> 00:20:07,139 一气呵成开发应用与游戏 310 99:59:59,999 --> 99:59:59,999 310 00:20:07,840 --> 00:20:10,242 见证了你们用它强大的推理能力完成各种任务 311 99:59:59,999 --> 99:59:59,999 311 00:20:10,309 --> 00:20:13,679 从拆解科学论文，到解析YouTube视频 312 99:59:59,999 --> 99:59:59,999 312 00:20:14,146 --> 00:20:16,382 你们告诉我们，Gemini协作性强、洞见深刻， 313 99:59:59,999 --> 99:59:59,999 313 00:20:16,482 --> 00:20:19,452 是真正的得力助手 314 99:59:59,999 --> 99:59:59,999 314 00:20:20,319 --> 00:20:24,724 看到大家用它创造出的惊人成果，我们倍感振奋 315 99:59:59,999 --> 99:59:59,999 315 00:20:26,525 --> 00:20:29,829 Gemini 2.5是我们迄今为止最智能的模型 316 99:59:59,999 --> 99:59:59,999 316 00:20:30,096 --> 00:20:32,264 也是全球先进基础模型 317 99:59:59,999 --> 99:59:59,999 317 00:20:32,998 --> 00:20:37,002 就在两周前，我们发布了升级后的2.5 Pro预览版 318 99:59:59,999 --> 99:59:59,999 318 00:20:37,236 --> 00:20:40,239 让大家可以开始亲手尝试用它将创想化为现实 319 99:59:59,999 --> 99:59:59,999 319 00:20:41,440 --> 00:20:43,576 大家的成果让我们大开眼界 320 99:59:59,999 --> 99:59:59,999 320 00:20:43,876 --> 00:20:49,315 从将草图转化为交互式应用到模拟3D城市 321 99:59:59,999 --> 99:59:59,999 321 00:20:50,349 --> 00:20:52,418 全新的2.5 Pro模型在热门编程排行榜 322 99:59:59,999 --> 99:59:59,999 322 00:20:52,485 --> 00:20:54,620 WebDev Arena上一马当先 323 99:59:59,999 --> 99:59:59,999 323 00:20:55,421 --> 00:20:57,590 现在，它又整合了LearnLM 324 99:59:59,999 --> 99:59:59,999 324 00:20:57,790 --> 00:21:00,493 该系列模型的开发得到了教育专家的支持 325 99:59:59,999 --> 99:59:59,999 325 00:21:00,693 --> 00:21:03,829 这让2.5 Pro也成为全球出色的学习辅助模型 326 99:59:59,999 --> 99:59:59,999 326 00:21:04,563 --> 00:21:08,434 在LMArena的所有排行榜上独占鳌头 327 99:59:59,999 --> 99:59:59,999 327 00:21:10,603 --> 00:21:14,006 Gemini Flash是我们极其高效的王牌模型 328 99:59:59,999 --> 99:59:59,999 328 00:21:14,507 --> 00:21:18,944 因出色的速度和较低的成本备受开发者青睐 329 99:59:59,999 --> 99:59:59,999 329 00:21:20,012 --> 00:21:24,617 今天我要激动地宣布推出更新版的2.5 Flash 330 99:59:59,999 --> 99:59:59,999 330 00:21:25,284 --> 00:21:28,387 新版Flash模型几乎在每一个方面都有所改进 331 99:59:59,999 --> 99:59:59,999 331 00:21:28,587 --> 00:21:30,222 在各类重要基准测试中都取得了更好的成绩 332 99:59:59,999 --> 99:59:59,999 332 00:21:30,289 --> 00:21:33,025 包括推理、编程和长上下文 333 99:59:59,999 --> 99:59:59,999 333 00:21:33,659 --> 00:21:38,931 实际上，它在LMArena排行榜上仅次于2.5 Pro 334 99:59:59,999 --> 99:59:59,999 334 00:21:40,232 --> 00:21:44,303 我要很高兴地告诉大家，Flash将于六月初正式上线 335 99:59:59,999 --> 99:59:59,999 335 00:21:44,470 --> 00:21:45,938 Pro紧随其后 336 99:59:59,999 --> 99:59:59,999 336 00:21:52,645 --> 00:21:55,681 我们正在根据大家的反馈做出最终调整 337 99:59:59,999 --> 99:59:59,999 337 00:21:55,815 --> 00:21:58,651 不过你现在可以在AI Studio 338 99:59:59,999 --> 99:59:59,999 338 00:21:58,717 --> 00:22:01,287 Vertex AI和Gemini应用中试用预览版 339 99:59:59,999 --> 99:59:59,999 339 00:22:02,855 --> 00:22:06,625 我们将一如既往，继续不懈前行，推动技术向前发展 340 99:59:59,999 --> 99:59:59,999 340 00:22:06,826 --> 00:22:09,528 并对我们近期的研究突破感到十分欣喜 341 99:59:59,999 --> 99:59:59,999 341 00:22:10,095 --> 00:22:12,898 稍后我会请Tulsee来与大家进一步分享 342 99:59:59,999 --> 99:59:59,999 342 00:22:13,065 --> 00:22:14,767 不过首先，我们来看一下 343 99:59:59,999 --> 99:59:59,999 343 00:22:14,834 --> 00:22:16,902 目前你可以用Gemini来做些什么 344 99:59:59,999 --> 99:59:59,999 344 00:22:20,139 --> 00:22:22,208 [用Gemini可以做的30件事] 345 99:59:59,999 --> 99:59:59,999 345 00:22:22,775 --> 00:22:23,876 [物理模拟] 346 99:59:59,999 --> 99:59:59,999 346 00:22:24,043 --> 00:22:25,344 [自然模拟] 347 99:59:59,999 --> 99:59:59,999 347 00:22:25,444 --> 00:22:26,645 [光学模拟] 348 99:59:59,999 --> 99:59:59,999 348 00:22:27,046 --> 00:22:28,214 [谜题模拟] 349 99:59:59,999 --> 99:59:59,999 349 00:22:28,981 --> 00:22:30,149 [将图片转变成代码] 350 99:59:59,999 --> 99:59:59,999 350 00:22:31,650 --> 00:22:33,219 [将想法转变成3D模型] 351 99:59:59,999 --> 99:59:59,999 351 00:22:36,155 --> 00:22:37,656 [为画作加入3D效果] 352 99:59:59,999 --> 99:59:59,999 352 00:22:39,758 --> 00:22:41,126 [与AI共同创作] 353 99:59:59,999 --> 99:59:59,999 353 00:22:41,360 --> 00:22:42,228 [发型设计] 354 99:59:59,999 --> 99:59:59,999 354 00:22:43,495 --> 00:22:45,698 [给这只猫穿上礼服] 355 99:59:59,999 --> 99:59:59,999 355 00:22:47,933 --> 00:22:49,201 [在积攒了几十年的文件中搜索内容] 356 99:59:59,999 --> 99:59:59,999 356 00:22:49,268 --> 00:22:50,102 [菜谱] 357 99:59:59,999 --> 99:59:59,999 357 00:22:50,970 --> 00:22:52,504 [在数小时长的视频中搜索内容] 358 99:59:59,999 --> 99:59:59,999 358 00:22:54,073 --> 00:22:55,241 [建造机器人] 359 99:59:59,999 --> 99:59:59,999 359 00:22:55,608 --> 00:23:00,446 [直接说话就能编程] 你能做个从我嘴里爆出一团圆形彩纸的特效吗？ 360 99:59:59,999 --> 99:59:59,999 360 00:23:01,447 --> 00:23:02,715 我更新了草图 361 99:59:59,999 --> 99:59:59,999 361 00:23:03,482 --> 00:23:05,784 [万物皆可AI：娱乐、学习、动画、可视化] [创造、转录、练习、翻译、查询、表情符号] 362 99:59:59,999 --> 99:59:59,999 362 00:23:05,851 --> 00:23:07,152 [混音、计算、探索、理解、地图、解释] 363 99:59:59,999 --> 99:59:59,999 363 00:23:07,253 --> 00:23:09,688 [生成千般内容] [构想万种可能] 364 99:59:59,999 --> 99:59:59,999 364 00:23:10,789 --> 00:23:12,658 接下来我们做点什么？ 365 99:59:59,999 --> 99:59:59,999 365 00:23:13,058 --> 00:23:15,060 这里还有30个创意供你参考 366 99:59:59,999 --> 99:59:59,999 366 00:23:16,028 --> 00:23:18,731 [使用Gemini开始构建] [AI.Studio] 367 99:59:59,999 --> 99:59:59,999 367 00:23:34,647 --> 00:23:35,748 Demis，谢谢你！ 368 99:59:59,999 --> 99:59:59,999 368 00:23:36,282 --> 00:23:39,318 很高兴能借此机会告诉大家 369 99:59:59,999 --> 99:59:59,999 369 00:23:39,385 --> 00:23:41,053 为了帮助像你一样的开发者 370 99:59:59,999 --> 99:59:59,999 370 00:23:41,153 --> 00:23:44,256 更轻松地使用Gemini 2.5进行构建 371 99:59:59,999 --> 99:59:59,999 371 00:23:44,890 --> 00:23:46,625 我们根据大家的反馈意见做出了多项改进 372 99:59:59,999 --> 99:59:59,999 372 00:23:46,926 --> 00:23:51,530 其中包括：功能改进、安全性和透明度加强 373 99:59:59,999 --> 99:59:59,999 373 00:23:51,964 --> 00:23:55,167 成本效益提升和可控性提升 374 99:59:59,999 --> 99:59:59,999 374 00:23:56,368 --> 00:24:00,572 首先，除了Demis提到的新版2.5 Flash之外 375 99:59:59,999 --> 99:59:59,999 375 00:24:01,106 --> 00:24:05,010 我们还推出了文字转语音功能的全新预览版 376 99:59:59,999 --> 99:59:59,999 376 00:24:05,778 --> 00:24:11,684 基于原生音频输出提供 同类产品中首创的多讲话人支持 377 99:59:59,999 --> 99:59:59,999 377 00:24:11,884 --> 00:24:13,886 可以支持两个人的语音 378 99:59:59,999 --> 99:59:59,999 378 00:24:16,088 --> 00:24:20,192 这意味着模型可以通过 更富于表现力的方式进行对话交流 379 99:59:59,999 --> 99:59:59,999 379 00:24:20,392 --> 00:24:23,529 能够捕捉我们言语表述之中的微妙之处 380 99:59:59,999 --> 99:59:59,999 380 00:24:25,331 --> 00:24:29,034 甚至能无缝切换到低声耳语，就像这样 381 99:59:59,999 --> 99:59:59,999 381 00:24:30,803 --> 00:24:33,405 它支持超过24种语言 382 99:59:59,999 --> 99:59:59,999 382 00:24:33,639 --> 00:24:36,675 甚至能流畅自如地在不同语言之间切换 383 99:59:59,999 --> 99:59:59,999 383 00:24:38,410 --> 00:24:42,448 模型可以在最初讲英语，然后 384 99:59:59,999 --> 99:59:59,999 384 00:24:43,949 --> 00:24:47,086 [印地语] 385 99:59:59,999 --> 99:59:59,999 385 00:24:48,687 --> 00:24:49,888 然后再切换回来 386 99:59:59,999 --> 99:59:59,999 386 00:24:50,222 --> 00:24:51,857 而且声音丝毫不变 387 99:59:59,999 --> 99:59:59,999 387 00:24:52,925 --> 00:24:54,326 不错吧？ 388 99:59:59,999 --> 99:59:59,999 388 00:25:00,933 --> 00:25:03,202 从今天开始，你就可以开始在Gemini API里 389 99:59:59,999 --> 99:59:59,999 389 00:25:03,268 --> 00:25:05,938 使用这种文字转语音功能了 390 99:59:59,999 --> 99:59:59,999 390 00:25:11,176 --> 00:25:13,545 今天稍后，Live API还将提供 391 99:59:59,999 --> 99:59:59,999 391 00:25:13,645 --> 00:25:18,717 2.5 Flash原生音频对话功能的预览版 392 99:59:59,999 --> 99:59:59,999 392 00:25:19,351 --> 00:25:24,390 让你能通过更自然流畅的方式直接与Gemini会话 393 99:59:59,999 --> 99:59:59,999 393 00:25:24,890 --> 00:25:28,694 它甚至能区分讲话人与背景噪音 394 99:59:59,999 --> 99:59:59,999 394 00:25:28,927 --> 00:25:30,496 从而判断合适的回复时机 395 99:59:59,999 --> 99:59:59,999 395 00:25:31,363 --> 00:25:33,832 Logan和Paige将在开发者主旨演讲中分享更多内容 396 99:59:59,999 --> 99:59:59,999 396 00:25:35,734 --> 00:25:36,602 其次 397 99:59:59,999 --> 99:59:59,999 397 00:25:37,202 --> 00:25:41,073 我们加强了针对间接提示注入等安全威胁的防护 398 99:59:59,999 --> 99:59:59,999 398 00:25:41,607 --> 00:25:45,711 让Gemini 2.5成为我们迄今为止最安全的模型 399 99:59:59,999 --> 99:59:59,999 399 00:25:46,779 --> 00:25:49,515 在2.5 Pro和Flash模型中 400 99:59:59,999 --> 99:59:59,999 400 00:25:49,782 --> 00:25:51,984 我们都添加了思考总结功能 401 99:59:59,999 --> 99:59:59,999 401 00:25:52,051 --> 00:25:54,286 可以通过Gemini API和Vertex AI来使用 402 99:59:59,999 --> 99:59:59,999 402 00:25:55,287 --> 00:25:58,290 思考总结功能会提取模型的原始思考过程 403 99:59:59,999 --> 99:59:59,999 403 00:25:58,457 --> 00:26:02,861 将其整理为清晰有序的格式，包含标题、重要细节 404 99:59:59,999 --> 99:59:59,999 404 00:26:03,128 --> 00:26:05,531 以及有关模型操作的信息，比如工具调用 405 99:59:59,999 --> 99:59:59,999 405 00:26:06,398 --> 00:26:09,768 这能让你更清晰地了解模型的思考过程 406 99:59:59,999 --> 99:59:59,999 406 00:26:10,169 --> 00:26:16,241 不但对高延迟任务特别有用，同时方便调试和使用 407 99:59:59,999 --> 99:59:59,999 407 00:26:17,943 --> 00:26:20,012 大家也呼吁提高模型处理效率 408 99:59:59,999 --> 99:59:59,999 408 00:26:20,379 --> 00:26:23,715 而现在，2.5 Flash的效率更胜以往 409 99:59:59,999 --> 99:59:59,999 409 00:26:24,083 --> 00:26:28,487 同等性能的情况下，使用的token更少 410 99:59:59,999 --> 99:59:59,999 410 00:26:28,654 --> 00:26:32,491 在实际测评中效率提升达22% 411 99:59:59,999 --> 99:59:59,999 411 00:26:34,159 --> 00:26:37,796 最后，我们为2.5 Flash引入了思考预算功能 412 99:59:59,999 --> 99:59:59,999 412 00:26:38,063 --> 00:26:41,567 让你能够在成本、延迟和输出质量之间取得平衡 413 99:59:59,999 --> 99:59:59,999 413 00:26:41,800 --> 00:26:43,635 由于大家的反响很好 414 99:59:59,999 --> 99:59:59,999 414 00:26:44,169 --> 00:26:47,406 我们将思考预算功能添加到了2.5 Pro之中 415 99:59:59,999 --> 99:59:59,999 415 00:26:47,706 --> 00:26:51,643 并将在未来几周内随同我们的正式版模型一同发布 416 99:59:59,999 --> 99:59:59,999 416 00:26:52,644 --> 00:26:53,779 借助思考预算功能 417 99:59:59,999 --> 99:59:59,999 417 00:26:53,846 --> 00:26:57,182 你可以更好地掌控模型使用多少token进行思考 418 99:59:59,999 --> 99:59:59,999 418 00:26:57,249 --> 00:26:58,217 然后再给出回答 419 99:59:59,999 --> 99:59:59,999 419 00:26:58,817 --> 00:27:00,819 也可以直接关闭思考功能 420 99:59:59,999 --> 99:59:59,999 420 00:27:02,821 --> 00:27:04,590 之前Demis提到过 421 99:59:59,999 --> 99:59:59,999 421 00:27:04,957 --> 00:27:07,893 Gemini 2.5 Pro有着超强的编码能力 422 99:59:59,999 --> 99:59:59,999 422 00:27:08,393 --> 00:27:13,165 所以接下来，我想展示一下如何将构思落地 423 99:59:59,999 --> 99:59:59,999 423 00:27:14,733 --> 00:27:19,004 如果你参观过纽约市的美国自然历史博物馆 424 99:59:59,999 --> 99:59:59,999 424 00:27:19,371 --> 00:27:21,673 一定记得那里精彩的展品 425 99:59:59,999 --> 99:59:59,999 425 00:27:22,207 --> 00:27:23,775 为了让大家能够看到这些展品 426 99:59:59,999 --> 99:59:59,999 426 00:27:23,842 --> 00:27:26,678 我让2.5 Pro在Google AI Studio中 427 99:59:59,999 --> 99:59:59,999 427 00:27:26,778 --> 00:27:28,447 编写了一个简单的Web应用 428 99:59:59,999 --> 99:59:59,999 428 00:27:28,614 --> 00:27:30,382 用于分享一些照片以及详细信息 429 99:59:59,999 --> 99:59:59,999 429 00:27:31,416 --> 00:27:32,985 这是该应用目前为止的模样 430 99:59:59,999 --> 99:59:59,999 430 00:27:34,119 --> 00:27:36,855 不过我还想进一步加强其互动性 431 99:59:59,999 --> 99:59:59,999 431 00:27:37,256 --> 00:27:39,258 我还在就设计思路集思广益 432 99:59:59,999 --> 99:59:59,999 432 00:27:39,625 --> 00:27:41,059 不过已经有了一些想法 433 99:59:59,999 --> 99:59:59,999 433 00:27:41,960 --> 00:27:43,929 大家都看到过像这样的情况吧？ 434 99:59:59,999 --> 99:59:59,999 434 00:27:44,463 --> 00:27:45,430 某人找到你 435 99:59:59,999 --> 99:59:59,999 435 00:27:45,497 --> 00:27:48,300 给你看在餐巾纸上勾勒出的“绝妙”点子 436 99:59:59,999 --> 99:59:59,999 436 00:27:49,368 --> 00:27:51,803 作为PM，我常常就是拿着餐巾纸的这个人 437 99:59:59,999 --> 99:59:59,999 437 00:27:53,872 --> 00:27:56,375 常规的二维网页设计已经不简单 438 99:59:59,999 --> 99:59:59,999 438 00:27:56,742 --> 00:27:58,644 但我想进一步实现三维设计 439 99:59:59,999 --> 99:59:59,999 439 00:27:59,878 --> 00:28:02,514 我知道转为三维并不容易 440 99:59:59,999 --> 99:59:59,999 440 00:28:03,015 --> 00:28:05,918 需要学习许许多多新知识 441 99:59:59,999 --> 99:59:59,999 441 00:28:06,185 --> 00:28:10,656 比如设置场景、相机、光线等 442 99:59:59,999 --> 99:59:59,999 442 00:28:11,757 --> 00:28:14,493 但幸运的是，2.5 Pro可以帮到我 443 99:59:59,999 --> 99:59:59,999 443 00:28:15,194 --> 00:28:17,029 这里我要做的是 444 99:59:59,999 --> 99:59:59,999 444 00:28:17,663 --> 00:28:22,534 把刚才展示的球体图片添加进来 445 99:59:59,999 --> 99:59:59,999 445 00:28:25,037 --> 00:28:30,309 输入提示，让2.5 Pro根据此图片更新我的代码 446 99:59:59,999 --> 99:59:59,999 446 00:28:31,543 --> 00:28:33,779 我们让2.5 Pro开始工作 447 99:59:59,999 --> 99:59:59,999 447 00:28:33,845 --> 00:28:36,248 可以看到，它开始思考 448 99:59:59,999 --> 99:59:59,999 448 00:28:36,615 --> 00:28:40,285 并根据我的要求制定了一份计划 449 99:59:59,999 --> 99:59:59,999 449 00:28:40,652 --> 00:28:43,488 稍后会将其应用于我的原有代码库 450 99:59:59,999 --> 99:59:59,999 450 00:28:44,489 --> 00:28:46,391 Gemini是一种多模态模型 451 99:59:59,999 --> 99:59:59,999 451 00:28:46,592 --> 00:28:48,994 因此能理解这张抽象的球体草图 452 99:59:59,999 --> 99:59:59,999 452 00:28:49,061 --> 00:28:51,597 通过编程生成精美的三维动画 453 99:59:59,999 --> 99:59:59,999 453 00:28:51,797 --> 00:28:54,166 再将其应用于我的原有应用 454 99:59:59,999 --> 99:59:59,999 454 00:28:55,067 --> 00:28:56,668 此过程大概需要两分钟 455 99:59:59,999 --> 99:59:59,999 455 00:28:56,735 --> 00:29:00,572 为节省时间，我们将以类似烘焙节目的形式来展示 456 99:59:59,999 --> 99:59:59,999 456 00:29:00,639 --> 00:29:02,507 我直接切换到另外一个标签页 457 99:59:59,999 --> 99:59:59,999 457 00:29:02,574 --> 00:29:05,744 在这场主旨演讲开始之前，我运行了相同的提示 458 99:59:59,999 --> 99:59:59,999 458 00:29:05,811 --> 00:29:08,614 这就是Gemini生成的结果 459 99:59:59,999 --> 99:59:59,999 459 00:29:17,022 --> 00:29:20,826 我们把一张粗略的草图直接变成了代码 460 99:59:59,999 --> 99:59:59,999 460 00:29:21,260 --> 00:29:23,295 更新了多个文件 461 99:59:59,999 --> 99:59:59,999 461 00:29:23,762 --> 00:29:26,665 而且可以看到，它思考了37秒时间 462 99:59:59,999 --> 99:59:59,999 462 00:29:26,832 --> 00:29:28,834 你也可以看到它的思考变化过程 463 99:59:59,999 --> 99:59:59,999 463 00:29:29,368 --> 00:29:30,936 以及它更新了哪些文件 464 99:59:59,999 --> 99:59:59,999 464 00:29:31,737 --> 00:29:33,705 这一切都是在AI Studio中完成的 465 99:59:59,999 --> 99:59:59,999 465 00:29:33,805 --> 00:29:35,574 所以在完成原型设计之后 466 99:59:59,999 --> 99:59:59,999 466 00:29:35,841 --> 00:29:39,611 我就可以直接用Gemini API密钥部署代码 467 99:59:59,999 --> 99:59:59,999 467 00:29:40,879 --> 00:29:42,714 这是我们的最终版应用在Chrome中的效果 468 99:59:59,999 --> 99:59:59,999 468 00:29:44,816 --> 00:29:46,852 看看这些动画 469 99:59:59,999 --> 99:59:59,999 469 00:29:48,186 --> 00:29:52,958 我不需要有多精深的Three.js库知识 470 99:59:59,999 --> 99:59:59,999 470 00:29:53,158 --> 00:29:56,128 也不必钻研复杂的三维算法，就做成了这样的作品 471 99:59:59,999 --> 99:59:59,999 471 00:29:56,728 --> 00:29:59,464 如果要亲自动手来做，我可能永远也没法实现 472 99:59:59,999 --> 99:59:59,999 472 00:29:59,798 --> 00:30:03,969 但这里，我只用一张草图就做出来了 473 99:59:59,999 --> 99:59:59,999 473 00:30:05,437 --> 00:30:08,573 我可以利用多模态，让整个体验更加丰富 474 99:59:59,999 --> 99:59:59,999 474 00:30:09,007 --> 00:30:13,011 我用2.5 Flash给每张照片添加了一个问题 475 99:59:59,999 --> 99:59:59,999 475 00:30:13,211 --> 00:30:14,680 邀请你进一步了解相关信息 476 99:59:59,999 --> 99:59:59,999 476 00:30:17,382 --> 00:30:18,984 但如果它能说话会怎样呢？ 477 99:59:59,999 --> 99:59:59,999 477 00:30:20,485 --> 00:30:22,688 这时我们就可以用到Gemini的原生音频功能 478 99:59:59,999 --> 99:59:59,999 478 00:30:24,556 --> 00:30:25,957 这是一只穿山甲 479 99:59:59,999 --> 99:59:59,999 479 00:30:26,024 --> 00:30:29,828 它的鳞片由角蛋白构成，就像你的指甲一样 480 99:59:59,999 --> 99:59:59,999 480 00:30:30,962 --> 00:30:33,065 哇！现在它支持语音功能了 481 99:59:59,999 --> 99:59:59,999 481 00:30:33,799 --> 00:30:37,736 你可以听到，如何在应用中添加富有表现力的音频 482 99:59:59,999 --> 99:59:59,999 482 00:30:38,470 --> 00:30:40,405 在分享更多内容之前 483 99:59:59,999 --> 99:59:59,999 483 00:30:40,605 --> 00:30:43,008 我想以另一个好玩的设计来结束演示 484 99:59:59,999 --> 99:59:59,999 484 00:30:43,075 --> 00:30:45,510 这也是通过2.5 Pro编码实现的 485 99:59:59,999 --> 99:59:59,999 485 00:30:56,054 --> 00:30:57,889 我们在Google AI Studio提供了起始代码 486 99:59:59,999 --> 99:59:59,999 486 00:30:57,956 --> 00:31:00,392 让大家都可以在本演示的基础上进行二次开发 487 99:59:59,999 --> 99:59:59,999 487 00:31:00,492 --> 00:31:01,593 欢迎大家去尝试 488 99:59:59,999 --> 99:59:59,999 488 00:31:02,728 --> 00:31:04,496 这个演示仅仅是一个例子 489 99:59:59,999 --> 99:59:59,999 489 00:31:04,796 --> 00:31:07,999 证明Gemini 2.5必将大幅改变我们的构建方式 490 99:59:59,999 --> 99:59:59,999 490 00:31:08,066 --> 00:31:10,635 改变你利用非凡的功能 491 99:59:59,999 --> 99:59:59,999 491 00:31:10,969 --> 00:31:15,674 在自己的应用中集成视觉、语音和编码的方式 492 99:59:59,999 --> 99:59:59,999 492 00:31:16,675 --> 00:31:20,345 你可以在自己喜爱的IDE平台 493 99:59:59,999 --> 99:59:59,999 493 00:31:20,979 --> 00:31:22,814 以及各种Google产品中使用2.5 Pro 494 99:59:59,999 --> 99:59:59,999 494 00:31:22,981 --> 00:31:27,819 例如Android Studio、Firebase Studio Gemini Code Assist 495 99:59:59,999 --> 99:59:59,999 495 00:31:28,153 --> 00:31:31,423 以及我们最新的异步编码智能体Jules 496 99:59:59,999 --> 99:59:59,999 496 00:31:38,563 --> 00:31:41,600 你只需提交任务，剩下的就交给Jules来处理 497 99:59:59,999 --> 99:59:59,999 497 00:31:41,833 --> 00:31:44,236 修复bug，执行更新 498 99:59:59,999 --> 99:59:59,999 498 00:31:44,603 --> 00:31:47,606 它与GitHub集成，可以独立运行 499 99:59:59,999 --> 99:59:59,999 499 00:31:48,707 --> 00:31:51,476 Jules能够处理大型代码库中的复杂任务 500 99:59:59,999 --> 99:59:59,999 500 00:31:51,543 --> 00:31:52,978 这些任务过去需要数小时才能完成 501 99:59:59,999 --> 99:59:59,999 501 00:31:53,245 --> 00:31:55,747 比如更新较早版本的Node.js 502 99:59:59,999 --> 99:59:59,999 502 00:31:56,348 --> 00:32:00,886 它能在几分钟内完成规划步骤、修改文件等任务 503 99:59:59,999 --> 99:59:59,999 503 00:32:01,586 --> 00:32:03,555 今天，我很高兴地宣布 504 99:59:59,999 --> 99:59:59,999 504 00:32:03,622 --> 00:32:05,357 Jules将开启公开Beta版测试 505 99:59:59,999 --> 99:59:59,999 505 00:32:05,624 --> 00:32:08,260 所有人都可以通过jules.google注册 506 99:59:59,999 --> 99:59:59,999 506 00:32:15,233 --> 00:32:19,571 正如Demis所说，我们一直在创新优化模型的方法 507 99:59:59,999 --> 99:59:59,999 507 00:32:20,138 --> 00:32:22,908 包括提高模型的效率和性能 508 99:59:59,999 --> 99:59:59,999 508 00:32:23,675 --> 00:32:28,880 我们率先采用扩散技术，革新了图像和视频的生成 509 99:59:59,999 --> 99:59:59,999 509 00:32:29,548 --> 00:32:34,286 扩散模型可通过逐步优化噪声来生成输出 510 99:59:59,999 --> 99:59:59,999 510 00:32:35,153 --> 00:32:40,025 现在，我们正借助最新研究模型将扩散应用于文本 511 99:59:59,999 --> 99:59:59,999 511 00:32:41,259 --> 00:32:43,595 这样有助于提升处理编辑等任务的能力 512 99:59:59,999 --> 99:59:59,999 512 00:32:43,762 --> 00:32:46,097 包括在数学和编码等应用场景中 513 99:59:59,999 --> 99:59:59,999 513 00:32:46,798 --> 00:32:48,934 因为它不只是从左到右线性生成内容 514 99:59:59,999 --> 99:59:59,999 514 00:32:49,334 --> 00:32:51,269 还能够快速反复试验解决方案 515 99:59:59,999 --> 99:59:59,999 515 00:32:51,636 --> 00:32:54,239 并在生成过程中不断纠错 516 99:59:59,999 --> 99:59:59,999 516 00:32:56,107 --> 00:33:00,579 Gemini Diffusion是一种领先的文本扩散实验模型 517 99:59:59,999 --> 99:59:59,999 517 00:33:01,079 --> 00:33:05,817 利用并行生成技术，实现超低延迟 518 99:59:59,999 --> 99:59:59,999 518 00:33:06,618 --> 00:33:10,422 例如，我们今天发布的Gemini Diffusion版本 519 99:59:59,999 --> 99:59:59,999 519 00:33:10,956 --> 00:33:12,891 生成速度是我们目前速度最快的模型 520 99:59:59,999 --> 99:59:59,999 520 00:33:13,191 --> 00:33:16,995 2.0 Flash-Lite的五倍 521 99:59:59,999 --> 99:59:59,999 521 00:33:17,395 --> 00:33:19,464 而编码能力却旗鼓相当 522 99:59:59,999 --> 99:59:59,999 522 00:33:20,165 --> 00:33:21,700 以这个数学问题为例 523 99:59:59,999 --> 99:59:59,999 523 00:33:22,033 --> 00:33:23,335 准备，开始！ 524 99:59:59,999 --> 99:59:59,999 524 00:33:25,537 --> 00:33:26,805 眨眼间整个过程就完成了 525 99:59:59,999 --> 99:59:59,999 525 00:33:33,645 --> 00:33:38,917 之前我们加快了速度，但这次我们会放慢一点 526 99:59:59,999 --> 99:59:59,999 526 00:33:43,955 --> 00:33:47,425 看到模型如何一步步计算出39这个答案，感觉很棒 527 99:59:59,999 --> 99:59:59,999 527 00:33:48,994 --> 00:33:51,263 此模型目前正在小范围测试 528 99:59:59,999 --> 99:59:59,999 528 00:33:52,330 --> 00:33:54,799 我们会继续探索各种方法 529 99:59:59,999 --> 99:59:59,999 529 00:33:54,900 --> 00:33:57,269 降低所有Gemini模型的延迟 530 99:59:59,999 --> 99:59:59,999 530 00:33:57,636 --> 00:34:00,739 并即将推出速度更快的2.5 Flash-Lite模型 531 99:59:59,999 --> 99:59:59,999 531 00:34:02,007 --> 00:34:04,409 更多有关Gemini未来发展的信息 532 99:59:59,999 --> 99:59:59,999 532 00:34:04,709 --> 00:34:05,877 请Demis与大家进一步分享 533 99:59:59,999 --> 99:59:59,999 533 00:34:14,152 --> 00:34:15,186 谢谢你，Tulsi 534 99:59:59,999 --> 99:59:59,999 534 00:34:15,854 --> 00:34:20,859 我们一直在不懈探索Gemini 2.5思考功能的边界 535 99:59:59,999 --> 99:59:59,999 535 00:34:21,726 --> 00:34:24,062 使用AlphaGo的经验告诉我们 536 99:59:59,999 --> 99:59:59,999 536 00:34:24,195 --> 00:34:27,732 给模型更多思考时间，其回答质量会得到改善 537 99:59:59,999 --> 99:59:59,999 537 00:34:29,100 --> 00:34:32,270 如今，我们进一步优化了2.5 Pro 538 99:59:59,999 --> 99:59:59,999 538 00:34:32,337 --> 00:34:35,807 引入了名为Deep Think的新模式 539 99:59:59,999 --> 99:59:59,999 539 00:34:36,875 --> 00:34:41,146 它能将模型性能发挥到极致，带来突破性的成果 540 99:59:59,999 --> 99:59:59,999 540 00:34:41,613 --> 00:34:46,184 Deep Think采用了思考与推理领域的前沿研究 541 99:59:59,999 --> 99:59:59,999 541 00:34:46,384 --> 00:34:48,253 包括平行技术 542 99:59:59,999 --> 99:59:59,999 542 00:34:49,287 --> 00:34:51,990 目前为止，我们看到的性能堪称惊艳 543 99:59:59,999 --> 99:59:59,999 543 00:34:52,524 --> 00:34:55,860 在目前难度最高的数学基准测试USAMO 2025中 544 99:59:59,999 --> 99:59:59,999 544 00:34:56,294 --> 00:34:58,897 它的评分十分亮眼 545 99:59:59,999 --> 99:59:59,999 545 00:34:59,297 --> 00:35:00,799 在LiveCodeBench中也处于领先水平 546 99:59:59,999 --> 99:59:59,999 546 00:35:00,865 --> 00:35:04,169 这是一项针对竞赛级编码的高难度基准测试 547 99:59:59,999 --> 99:59:59,999 547 00:35:04,636 --> 00:35:07,739 Gemini从一开始就具备原生多模态能力 548 99:59:59,999 --> 99:59:59,999 548 00:35:08,173 --> 00:35:12,243 所以自然在衡量此能力的主要基准中表现同样出色 549 99:59:59,999 --> 99:59:59,999 549 00:35:12,577 --> 00:35:13,812 这一基准是：MMMU 550 99:59:59,999 --> 99:59:59,999 550 00:35:15,246 --> 00:35:18,683 因为我们正在通过2.5 Pro Deep Think来引领前沿技术 551 99:59:59,999 --> 99:59:59,999 551 00:35:19,150 --> 00:35:23,121 我们需要额外花一些时间来进行更多前沿安全评估 552 99:59:59,999 --> 99:59:59,999 552 00:35:23,455 --> 00:35:25,657 并从安全专家那里获取更多意见 553 99:59:59,999 --> 99:59:59,999 553 00:35:26,324 --> 00:35:28,526 因此，我们将通过Gemini API 554 99:59:59,999 --> 99:59:59,999 554 00:35:28,593 --> 00:35:31,096 向可信测试员开放此功能 555 99:59:59,999 --> 99:59:59,999 555 00:35:31,596 --> 00:35:34,299 以便在大规模推出此功能前获取反馈意见 556 99:59:59,999 --> 99:59:59,999 556 00:35:35,233 --> 00:35:38,136 今天稍后，Josh会更详细地给大家介绍 557 99:59:59,999 --> 99:59:59,999 557 00:35:38,603 --> 00:35:40,105 我们针对2.5 Pro Deep Think的计划 558 99:59:59,999 --> 99:59:59,999 558 00:35:42,107 --> 00:35:43,341 过去十年 559 99:59:59,999 --> 99:59:59,999 559 00:35:43,541 --> 00:35:46,745 我们为现代AI纪元的到来做了很多基础工作 560 99:59:59,999 --> 99:59:59,999 560 00:35:48,046 --> 00:35:51,082 从我们开创性地发明了Transformer架构 561 99:59:59,999 --> 99:59:59,999 561 00:35:51,383 --> 00:35:54,486 它如今已成为所有大语言模型的基础 562 99:59:59,999 --> 99:59:59,999 562 00:35:54,786 --> 00:35:57,656 到打造AlphaGo和AlphaZero等智能体系统 563 99:59:59,999 --> 99:59:59,999 563 00:35:57,956 --> 00:36:02,927 我们将继续拓展基础研究的深度和广度 564 99:59:59,999 --> 99:59:59,999 564 00:36:03,395 --> 00:36:05,196 探索下一个重大突破 565 99:59:59,999 --> 99:59:59,999 565 00:36:05,263 --> 00:36:08,466 满足人们对通用人工智能的需要 566 99:59:59,999 --> 99:59:59,999 566 00:36:09,734 --> 00:36:13,505 Gemini如今已是卓尔不凡的多模态基础模型 567 99:59:59,999 --> 99:59:59,999 567 00:36:13,905 --> 00:36:15,607 但我们依然脚步未停 568 99:59:59,999 --> 99:59:59,999 568 00:36:16,141 --> 00:36:18,376 竭力将其打造成所谓的“世界模型” 569 99:59:59,999 --> 99:59:59,999 569 00:36:19,144 --> 00:36:22,380 待到那时，它将能够通过模拟世界的方方面面 570 99:59:59,999 --> 99:59:59,999 570 00:36:22,447 --> 00:36:26,418 制定计划和构想全新体验，就像人类的大脑一样 571 99:59:59,999 --> 99:59:59,999 571 00:36:27,252 --> 00:36:29,721 这一直是我个人的热情所在 572 99:59:59,999 --> 99:59:59,999 572 00:36:29,821 --> 00:36:31,690 早在我为《主题公园》等模拟游戏 573 99:59:59,999 --> 99:59:59,999 573 00:36:31,756 --> 00:36:33,692 开发AI的早期职业生涯中 574 99:59:59,999 --> 99:59:59,999 574 00:36:33,825 --> 00:36:35,760 我就已经开始探索虚拟世界 575 99:59:59,999 --> 99:59:59,999 575 00:36:37,095 --> 00:36:39,798 多年来，我们一直向这个方向迈进 576 99:59:59,999 --> 99:59:59,999 576 00:36:39,864 --> 00:36:44,636 从训练智能体掌握围棋和《星际争霸》等复杂游戏 577 99:59:59,999 --> 99:59:59,999 577 00:36:45,403 --> 00:36:47,672 到最近推出的Genie 2模型 578 99:59:59,999 --> 99:59:59,999 578 00:36:47,906 --> 00:36:50,742 它仅通过一张图片作为提示 579 99:59:59,999 --> 99:59:59,999 579 00:36:50,975 --> 00:36:54,245 就能生成可交互的三维模拟环境 580 99:59:59,999 --> 99:59:59,999 580 00:36:55,847 --> 00:36:58,083 大家已经可以看到这些功能正在逐渐显现 581 99:59:59,999 --> 99:59:59,999 581 00:36:58,216 --> 00:37:00,852 例如Gemini能够运用对世界的认知和推理能力 582 99:59:59,999 --> 99:59:59,999 582 00:37:01,219 --> 00:37:02,687 来呈现自然万物 583 99:59:59,999 --> 99:59:59,999 583 00:37:03,555 --> 00:37:06,424 还有我们先进的视频模型Veo 584 99:59:59,999 --> 99:59:59,999 584 00:37:06,491 --> 00:37:08,960 它对基本物理规律有着深刻理解 585 99:59:59,999 --> 99:59:59,999 585 00:37:09,127 --> 00:37:12,097 比如重力、光线和材料的属性 586 99:59:59,999 --> 99:59:59,999 586 00:37:12,864 --> 00:37:14,733 Veo令人惊叹的地方在于 587 99:59:59,999 --> 99:59:59,999 587 00:37:14,799 --> 00:37:18,503 它能在所有帧中保持高度的准确性和连贯性 588 99:59:59,999 --> 99:59:59,999 588 00:37:19,137 --> 00:37:22,107 即使提示脑洞大开，它也知道该怎么做 589 99:59:59,999 --> 99:59:59,999 589 00:37:22,474 --> 00:37:25,343 比如这个用救生筏生成的人 590 99:59:59,999 --> 99:59:59,999 590 00:37:27,479 --> 00:37:31,449 理解物理环境对于机器人技术也至关重要 591 99:59:59,999 --> 99:59:59,999 591 00:37:31,916 --> 00:37:34,085 AI系统需要具备世界模型 592 99:59:59,999 --> 99:59:59,999 592 00:37:34,219 --> 00:37:36,421 才能在现实世界中高效地运作 593 99:59:59,999 --> 99:59:59,999 593 00:37:37,689 --> 00:37:40,592 我们微调了一个专用模型，Gemini Robotics 594 99:59:59,999 --> 99:59:59,999 594 00:37:40,658 --> 00:37:42,727 它可以教机器人完成各种实用任务 595 99:59:59,999 --> 99:59:59,999 595 00:37:43,094 --> 00:37:47,298 例如抓取物体、执行指令和动态适应新的任务 596 99:59:59,999 --> 99:59:59,999 596 00:37:48,233 --> 00:37:52,337 在场的各位，可以在AI展示区中体验这些机器人 597 99:59:59,999 --> 99:59:59,999 597 00:37:59,244 --> 00:38:01,446 将Gemini打造成完善的世界模型 598 99:59:59,999 --> 99:59:59,999 598 00:38:01,513 --> 00:38:04,549 对于解锁新型AI至关重要 599 99:59:59,999 --> 99:59:59,999 599 00:38:05,617 --> 00:38:07,786 比如能够为日常生活提供帮助的AI 600 99:59:59,999 --> 99:59:59,999 600 00:38:08,052 --> 00:38:10,922 智能且理解你所处情境的AI 601 99:59:59,999 --> 99:59:59,999 601 00:38:11,189 --> 00:38:14,926 以及能够跨设备为你制定计划并采取行动的AI 602 99:59:59,999 --> 99:59:59,999 602 00:38:15,593 --> 00:38:18,129 我们对Gemini应用的终极愿景 603 99:59:59,999 --> 99:59:59,999 603 00:38:18,797 --> 00:38:21,966 就是将其打造成一个通用型的AI助理 604 99:59:59,999 --> 99:59:59,999 604 00:38:22,600 --> 00:38:26,070 让它成为富有个性、主动思考、功能强大的AI 605 99:59:59,999 --> 99:59:59,999 605 00:38:26,471 --> 00:38:29,607 以及我们迈向通用人工智能的重要里程碑 606 99:59:59,999 --> 99:59:59,999 606 00:38:30,341 --> 00:38:34,479 这要从我们去年在Project Astra中 首次探索的功能开始 607 99:59:59,999 --> 99:59:59,999 607 00:38:34,846 --> 00:38:37,649 包括视频理解、屏幕共享和记忆等功能 608 99:59:59,999 --> 99:59:59,999 608 00:38:38,750 --> 00:38:43,254 过去一年，我们一直在将这些功能引入Gemini Live 609 99:59:59,999 --> 99:59:59,999 609 00:38:43,788 --> 00:38:45,523 以便更多人今天就能体验到 610 99:59:59,999 --> 99:59:59,999 610 00:38:46,558 --> 00:38:51,129 当然，我们也会不断改进，并探索新的前沿创新 611 99:59:59,999 --> 99:59:59,999 611 00:38:52,297 --> 00:38:56,367 比如通过原生音频升级语音输出，使其变得更自然 612 99:59:59,999 --> 99:59:59,999 612 00:38:57,001 --> 00:38:59,871 我们改进了记忆功能，并增加了计算机控件 613 99:59:59,999 --> 99:59:59,999 613 00:39:00,238 --> 00:39:01,105 我们来看一下 614 99:59:59,999 --> 99:59:59,999 614 00:39:02,273 --> 00:39:05,043 我现在向大家展示的是 Project Astra的最新研究原型 615 99:59:59,999 --> 99:59:59,999 615 00:39:05,210 --> 00:39:07,145 下面来看看它如何帮我修自行车 616 99:59:59,999 --> 99:59:59,999 616 00:39:08,146 --> 00:39:10,849 [Project Astra] [行动智能+Gemini] 617 99:59:59,999 --> 99:59:59,999 617 00:39:11,049 --> 00:39:14,853 上网查查Huffy山地自行车的用户手册 618 99:59:59,999 --> 99:59:59,999 618 00:39:16,187 --> 00:39:18,056 已找到一份用户手册 619 99:59:59,999 --> 99:59:59,999 619 00:39:18,256 --> 00:39:19,524 你想要什么信息？ 620 99:59:59,999 --> 99:59:59,999 620 00:39:19,624 --> 00:39:23,261 好，现在向下滚动页面，找到刹车系统的相关章节 621 99:59:59,999 --> 99:59:59,999 621 00:39:23,528 --> 00:39:25,597 好的，正在为你查找相应页面 622 99:59:59,999 --> 99:59:59,999 622 00:39:26,264 --> 00:39:28,132 这颗螺丝可能滑牙了 623 99:59:59,999 --> 99:59:59,999 623 00:39:28,399 --> 00:39:31,069 能否上YouTube找个修复教程视频？ 624 99:59:59,999 --> 99:59:59,999 624 00:39:31,536 --> 00:39:33,705 没问题，正在打开YouTube 625 99:59:59,999 --> 99:59:59,999 625 00:39:36,875 --> 00:39:38,343 这个视频看起来不错 626 99:59:59,999 --> 99:59:59,999 626 00:39:38,409 --> 00:39:40,478 将橡皮筋套在螺丝头上 627 99:59:59,999 --> 99:59:59,999 627 00:39:40,545 --> 00:39:41,813 噢，不错，谢谢 628 99:59:59,999 --> 99:59:59,999 628 00:39:41,880 --> 00:39:43,915 查下我跟自行车店的往来邮件 629 99:59:59,999 --> 99:59:59,999 629 00:39:43,982 --> 00:39:46,351 看看我需要哪种规格的六角螺母？ 630 99:59:59,999 --> 99:59:59,999 630 00:39:46,551 --> 00:39:49,654 好的，Bicycle Habitat的邮件显示 631 99:59:59,999 --> 99:59:59,999 631 00:39:49,721 --> 00:39:52,523 你的链轮需要一个3/8英寸的内六角螺母 632 99:59:59,999 --> 99:59:59,999 632 00:39:52,590 --> 00:39:54,392 我已经帮你标出了相应的工具盒 633 99:59:59,999 --> 99:59:59,999 633 00:39:54,926 --> 00:39:57,562 我好像需要一个备用的张力螺丝 634 99:59:59,999 --> 99:59:59,999 634 00:39:58,096 --> 00:40:00,498 能否联系最近的五金店确认是否有货？ 635 99:59:59,999 --> 99:59:59,999 635 00:40:00,698 --> 00:40:02,367 好的，正在联系对方 636 99:59:59,999 --> 99:59:59,999 636 00:40:02,634 --> 00:40:04,802 稍后我会告诉你具体库存信息 637 99:59:59,999 --> 99:59:59,999 637 00:40:04,869 --> 00:40:06,771 我的刹车片有磨损了 638 99:59:59,999 --> 99:59:59,999 638 00:40:06,905 --> 00:40:08,172 麻烦再查一下用户手册 639 99:59:59,999 --> 99:59:59,999 639 00:40:08,239 --> 00:40:09,674 看看上面有没有提到刹车片？ 640 99:59:59,999 --> 99:59:59,999 640 00:40:09,807 --> 00:40:12,844 用户手册的第24页显示 641 99:59:59,999 --> 99:59:59,999 641 00:40:13,311 --> 00:40:14,545 要去吃午饭吗？ 642 99:59:59,999 --> 99:59:59,999 642 00:40:15,280 --> 00:40:17,048 好，再等我五分钟 643 99:59:59,999 --> 99:59:59,999 643 00:40:17,682 --> 00:40:19,217 你刚才说到哪了？ 644 99:59:59,999 --> 99:59:59,999 644 00:40:19,484 --> 00:40:23,554 刚才说到，第24页第2节有相关说明 645 99:59:59,999 --> 99:59:59,999 645 00:40:23,788 --> 00:40:25,623 介绍了如何更换刹车片 646 99:59:59,999 --> 99:59:59,999 646 00:40:25,723 --> 00:40:27,492 之前说的那个电话有进展了吗？ 647 99:59:59,999 --> 99:59:59,999 647 00:40:28,159 --> 00:40:30,328 嗯，我刚和五金店通完电话 648 99:59:59,999 --> 99:59:59,999 648 00:40:30,495 --> 00:40:32,630 他们确认张力螺丝还有库存 649 99:59:59,999 --> 99:59:59,999 649 00:40:32,764 --> 00:40:34,465 要我下一个自取订单吗？ 650 99:59:59,999 --> 99:59:59,999 650 00:40:34,966 --> 00:40:38,403 能否展示几款适合我的自行车用的狗篮？ 651 99:59:59,999 --> 99:59:59,999 651 00:40:38,469 --> 00:40:41,272 没问题，乐意效劳，稍等片刻 652 99:59:59,999 --> 99:59:59,999 652 00:40:42,473 --> 00:40:43,675 看看这几款怎么样 653 99:59:59,999 --> 99:59:59,999 653 00:40:43,741 --> 00:40:45,810 我觉得特别适合Zuco 654 99:59:59,999 --> 99:59:59,999 654 00:40:47,478 --> 00:40:48,513 [行动智能+Gemini] 655 99:59:59,999 --> 99:59:59,999 655 00:40:48,579 --> 00:40:53,418 [电脑控制/重点标示/通话助理/知识接地 原生音频/主动思考/聚焦讲话人/个性化购物] 656 99:59:59,999 --> 99:59:59,999 656 00:40:56,654 --> 00:41:00,625 通用AI助理将帮助我们处理各种任务 657 99:59:59,999 --> 99:59:59,999 657 00:41:00,959 --> 00:41:05,863 它会处理琐碎的行政工作 并给出令人愉快的新推荐 658 99:59:59,999 --> 99:59:59,999 658 00:41:06,164 --> 00:41:08,900 从而提高我们的效率，丰富我们的生活 659 99:59:59,999 --> 99:59:59,999 659 00:41:09,534 --> 00:41:13,371 我们正向可信测试员收集这些新功能的反馈意见 660 99:59:59,999 --> 99:59:59,999 660 00:41:13,805 --> 00:41:15,907 并致力于将它们引入Gemini Live 661 99:59:59,999 --> 99:59:59,999 661 00:41:16,140 --> 00:41:17,575 提供新的搜索体验 662 99:59:59,999 --> 99:59:59,999 662 00:41:17,642 --> 00:41:19,510 以及面向开发者的Live API 663 99:59:59,999 --> 99:59:59,999 663 00:41:19,677 --> 00:41:23,348 还有Android XR眼镜等全新形态的设备 664 99:59:59,999 --> 99:59:59,999 664 00:41:23,948 --> 00:41:25,817 今天后面还会有更详细的介绍 665 99:59:59,999 --> 99:59:59,999 665 00:41:27,618 --> 00:41:28,820 在我整个职业生涯中 666 99:59:59,999 --> 99:59:59,999 666 00:41:28,886 --> 00:41:31,656 核心都是关于如何利用AI来推动知识进步 667 99:59:59,999 --> 99:59:59,999 667 00:41:31,723 --> 00:41:33,458 以及加速科学发现 668 99:59:59,999 --> 99:59:59,999 668 00:41:34,392 --> 00:41:36,094 Google DeepMind长期以来都在致力于 669 99:59:59,999 --> 99:59:59,999 669 00:41:36,160 --> 00:41:39,797 将AI应用于近乎每个科学领域 670 99:59:59,999 --> 99:59:59,999 670 00:41:40,331 --> 00:41:41,599 仅在过去一年 671 99:59:59,999 --> 99:59:59,999 671 00:41:41,766 --> 00:41:44,602 我们就已经在多个领域取得了巨大突破 672 99:59:59,999 --> 99:59:59,999 672 00:41:44,802 --> 00:41:47,105 包括数学、生命科学等领域 673 99:59:59,999 --> 99:59:59,999 673 00:41:48,039 --> 00:41:49,207 我们构建了AlphaProof 674 99:59:59,999 --> 99:59:59,999 674 00:41:49,440 --> 00:41:52,477 它可以解决奥林匹克数学竞赛中银牌水平的问题 675 99:59:59,999 --> 99:59:59,999 675 00:41:53,244 --> 00:41:55,880 Co-Scientist，它可以与研究人员协作 676 99:59:59,999 --> 99:59:59,999 676 00:41:55,980 --> 00:41:58,549 帮助他们开发和测试新的假设 677 99:59:59,999 --> 99:59:59,999 677 00:41:59,017 --> 00:42:01,352 我们还刚发布了AlphaEvolve 678 99:59:59,999 --> 99:59:59,999 678 00:42:01,519 --> 00:42:05,423 它可以发现新的科学知识，并加速AI本身的训练 679 99:59:59,999 --> 99:59:59,999 679 00:42:07,258 --> 00:42:09,427 在生命科学领域，我们构建了AMIE 680 99:59:59,999 --> 99:59:59,999 680 00:42:09,594 --> 00:42:13,064 一个可以帮助临床医生进行医疗诊断的研究系统 681 99:59:59,999 --> 99:59:59,999 681 00:42:13,631 --> 00:42:14,699 AlphaFold 3 682 99:59:59,999 --> 99:59:59,999 682 00:42:14,966 --> 00:42:18,870 能够预测所有生命体的分子结构和相互作用 683 99:59:59,999 --> 99:59:59,999 683 00:42:19,470 --> 00:42:22,273 以及Isomorphic Labs，它以AlphaFold为基础 684 99:59:59,999 --> 99:59:59,999 684 00:42:22,340 --> 00:42:26,511 致力于用AI来革新药物研发流程 685 99:59:59,999 --> 99:59:59,999 685 00:42:27,345 --> 00:42:30,148 未来有望帮助攻克多种全球性疾病 686 99:59:59,999 --> 99:59:59,999 686 00:42:31,482 --> 00:42:32,917 短短几年时间 687 99:59:59,999 --> 99:59:59,999 687 00:42:33,117 --> 00:42:36,554 AlphaFold就已经在科学界产生了深远影响 688 99:59:59,999 --> 99:59:59,999 688 00:42:37,055 --> 00:42:40,024 它已成为生物学及医学研究的标配工具 689 99:59:59,999 --> 99:59:59,999 689 00:42:40,458 --> 00:42:42,593 全球超过250万的研究人员 690 99:59:59,999 --> 99:59:59,999 690 00:42:42,660 --> 00:42:44,862 正在其关键工作中使用这一工具 691 99:59:59,999 --> 99:59:59,999 691 00:42:46,030 --> 00:42:48,433 在发展通用人工智能的道路上 692 99:59:59,999 --> 99:59:59,999 692 00:42:48,833 --> 00:42:52,370 我始终坚信，如果以安全和负责任的方式进行 693 99:59:59,999 --> 99:59:59,999 693 00:42:52,437 --> 00:42:55,106 AI有望加速科学发现进程 694 99:59:59,999 --> 99:59:59,999 694 00:42:55,373 --> 00:42:58,709 并成为有史以来最有益的技术 695 99:59:59,999 --> 99:59:59,999 695 00:43:06,284 --> 00:43:11,022 回想起来，真是不可思议，就在几年前 696 99:59:59,999 --> 99:59:59,999 696 00:43:11,222 --> 00:43:15,593 大家今天看到的这些尖端技术还跟魔法一样 697 99:59:59,999 --> 99:59:59,999 697 00:43:16,727 --> 00:43:22,066 看到这些技术为搜索和Gemini等产品的新体验赋能 698 99:59:59,999 --> 99:59:59,999 698 00:43:22,166 --> 00:43:25,136 并且共同在日常生活中帮助人们，真是令人兴奋 699 99:59:59,999 --> 99:59:59,999 699 00:43:25,903 --> 00:43:28,573 例如，我们最近与Aira建立了合作关系 700 99:59:59,999 --> 99:59:59,999 700 00:43:28,873 --> 00:43:33,711 这家公司帮助盲人和低视力人群探索世界 701 99:59:59,999 --> 99:59:59,999 701 00:43:33,945 --> 00:43:35,546 方法是将该人群通过视频 702 99:59:59,999 --> 99:59:59,999 702 00:43:35,613 --> 00:43:37,548 与人工“视觉翻译”连接起来 703 99:59:59,999 --> 99:59:59,999 703 00:43:38,349 --> 00:43:39,617 利用Astra技术 704 99:59:59,999 --> 99:59:59,999 704 00:43:39,784 --> 00:43:43,888 我们设计了一个帮助更多人获得此类协助的原型 705 99:59:59,999 --> 99:59:59,999 705 00:43:44,956 --> 00:43:46,757 我们正在持续收集用户反馈 706 99:59:59,999 --> 99:59:59,999 706 00:43:46,924 --> 00:43:51,729 同时Aira的翻译人员也在积极监督安全性及可靠性 707 99:59:59,999 --> 99:59:59,999 707 00:43:58,102 --> 00:44:00,938 凭借这项技术以及我们所有的开创性工作 708 99:59:59,999 --> 99:59:59,999 708 00:44:01,139 --> 00:44:05,076 我们正在打造更具个性、更主动、更强大的AI 709 99:59:59,999 --> 99:59:59,999 709 00:44:05,676 --> 00:44:09,280 从而丰富我们的生活、加速科学进程 710 99:59:59,999 --> 99:59:59,999 710 00:44:09,514 --> 00:44:13,751 并开启一个充满发现和奇迹的黄金时代 711 99:59:59,999 --> 99:59:59,999 711 00:44:18,222 --> 00:44:21,959 我记得我小时候经常弹吉他到天黑 712 99:59:59,999 --> 99:59:59,999 712 00:44:22,026 --> 00:44:25,596 来证明即使我看不见，也可以弹吉他 713 99:59:59,999 --> 99:59:59,999 713 00:44:27,765 --> 00:44:32,136 我在大概四岁左右被诊断出患有视网膜色素变性症 714 99:59:59,999 --> 99:59:59,999 714 00:44:32,236 --> 00:44:35,173 我的视力水平不断恶化 715 99:59:59,999 --> 99:59:59,999 715 00:44:36,240 --> 00:44:41,345 但面对视力逐渐模糊的现实 716 99:59:59,999 --> 99:59:59,999 716 00:44:41,646 --> 00:44:45,416 音乐便是我的解药 717 99:59:59,999 --> 99:59:59,999 717 00:44:45,816 --> 00:44:48,386 玩音乐让我倍感自由 718 99:59:59,999 --> 99:59:59,999 718 00:44:49,620 --> 00:44:52,790 如果Project Astra能助我更独立地生活 719 99:59:59,999 --> 99:59:59,999 719 00:44:52,857 --> 00:44:55,359 那真是再好不过了 720 99:59:59,999 --> 99:59:59,999 720 00:44:58,963 --> 00:45:00,998 就拿旅游来说，我非常喜欢旅游 721 99:59:59,999 --> 99:59:59,999 721 00:45:01,065 --> 00:45:05,236 因为可以走出去，去从未到过的地方，结识新朋友 722 99:59:59,999 --> 99:59:59,999 722 00:45:05,303 --> 00:45:08,039 但这确实变得越来越困难 723 99:59:59,999 --> 99:59:59,999 723 00:45:08,139 --> 00:45:09,173 你好 724 99:59:59,999 --> 99:59:59,999 724 00:45:09,240 --> 00:45:12,643 嗨，我来拍摄周围环境 725 99:59:59,999 --> 99:59:59,999 725 00:45:12,710 --> 00:45:16,147 你能告诉我这个绿色房间里有什么吗？ 726 99:59:59,999 --> 99:59:59,999 726 00:45:16,214 --> 00:45:18,849 你面向的墙壁上贴有一个指示牌 727 99:59:59,999 --> 99:59:59,999 727 00:45:18,916 --> 00:45:22,286 写着“Wi-Fi network: TheGreyEagle” 728 99:59:59,999 --> 99:59:59,999 728 00:45:22,353 --> 00:45:24,422 密码是“livemusic” 729 99:59:59,999 --> 99:59:59,999 729 00:45:24,488 --> 00:45:28,259 非常感谢。你能告诉我麦克风架在哪儿吗？ 730 99:59:59,999 --> 99:59:59,999 730 00:45:28,326 --> 00:45:29,560 没问题，非常乐意为你提供帮助 731 99:59:59,999 --> 99:59:59,999 731 00:45:29,627 --> 00:45:31,429 我可以打开你的手电筒吗？ 732 99:59:59,999 --> 99:59:59,999 732 00:45:31,529 --> 00:45:32,563 当然可以 733 99:59:59,999 --> 99:59:59,999 733 00:45:34,265 --> 00:45:36,567 这面墙上有一些盘绕的电缆 734 99:59:59,999 --> 99:59:59,999 734 00:45:36,634 --> 00:45:39,870 视频右上方位置应该就是麦克风架 735 99:59:59,999 --> 99:59:59,999 735 00:45:42,640 --> 00:45:45,843 我能做的最有力量的事就是登上舞台 736 99:59:59,999 --> 99:59:59,999 736 00:45:46,143 --> 00:45:47,745 拿起吉他，尽情演奏 737 99:59:59,999 --> 99:59:59,999 737 00:45:48,412 --> 00:45:53,017 让人们意识到，看不看得见 738 99:59:59,999 --> 99:59:59,999 738 00:45:53,251 --> 00:45:54,685 残疾与否，并不能决定一切 739 99:59:59,999 --> 99:59:59,999 739 00:45:54,919 --> 00:45:57,588 如果Project Astra能助我一臂之力 740 99:59:59,999 --> 99:59:59,999 740 00:45:57,655 --> 00:45:58,956 我将全力支持它 741 99:59:59,999 --> 99:59:59,999 741 00:46:00,157 --> 00:46:01,559 祝你演出愉快 742 99:59:59,999 --> 99:59:59,999 742 00:46:01,626 --> 00:46:03,361 [前往goo.gle/AstraVI注册加入视觉翻译等候名单] 743 99:59:59,999 --> 99:59:59,999 743 00:46:11,035 --> 00:46:12,136 谢谢你，Demis 744 99:59:59,999 --> 99:59:59,999 744 00:46:12,203 --> 00:46:16,941 Project Astra在无障碍功能方面的潜力让人惊艳 745 99:59:59,999 --> 99:59:59,999 745 00:46:17,842 --> 00:46:19,610 这又是一个激动人心的例子 746 99:59:59,999 --> 99:59:59,999 746 00:46:19,777 --> 00:46:22,613 说明AI可以推动我们一直以来的使命 747 99:59:59,999 --> 99:59:59,999 747 00:46:23,381 --> 00:46:28,185 即：整理全球信息，让所有人都能访问并从中受益 748 99:59:59,999 --> 99:59:59,999 748 00:46:28,853 --> 00:46:32,023 要说最能体现我们使命的产品，非Google搜索莫属 749 99:59:59,999 --> 99:59:59,999 749 00:46:32,556 --> 00:46:36,594 正是因为搜索，我们几十年前就开始投资AI 750 99:59:59,999 --> 99:59:59,999 750 00:46:37,094 --> 00:46:39,497 也正是通过搜索，我们得以将AI的益处惠及大众 751 99:59:59,999 --> 99:59:59,999 751 00:46:39,597 --> 00:46:41,666 满足人类好奇心 752 99:59:59,999 --> 99:59:59,999 752 00:46:42,466 --> 00:46:45,236 Gemini模型正在让Google搜索 753 99:59:59,999 --> 99:59:59,999 753 00:46:45,303 --> 00:46:48,306 变得更智能、更真实、更个性化 754 99:59:59,999 --> 99:59:59,999 754 00:46:49,340 --> 00:46:52,877 我们取得了不俗的进步 AI概览就是一个绝佳的例证 755 99:59:59,999 --> 99:59:59,999 755 00:46:53,411 --> 00:46:55,446 自去年I/O大会推出以来 756 99:59:59,999 --> 99:59:59,999 756 00:46:55,780 --> 00:46:59,050 该功能已覆盖200多个国家/地区 757 99:59:59,999 --> 99:59:59,999 757 00:46:59,116 --> 00:47:01,719 每月服务超15亿用户 758 99:59:59,999 --> 99:59:59,999 758 00:47:02,453 --> 00:47:04,288 随着人们使用AI概览 759 99:59:59,999 --> 99:59:59,999 759 00:47:04,488 --> 00:47:08,492 我们发现他们对搜索结果更满意 并且搜索频率更高 760 99:59:59,999 --> 99:59:59,999 760 00:47:09,160 --> 00:47:11,962 在我们的主要市场，如美国、印度 761 99:59:59,999 --> 99:59:59,999 761 00:47:12,396 --> 00:47:16,434 在AI概览的推动下，查询类型 762 99:59:59,999 --> 99:59:59,999 762 00:47:16,500 --> 00:47:18,002 实现了超过10%的增长 763 99:59:59,999 --> 99:59:59,999 763 00:47:18,703 --> 00:47:23,607 最令人振奋的是，这种增长势头与日俱增 764 99:59:59,999 --> 99:59:59,999 764 00:47:24,141 --> 00:47:28,412 这是搜索领域过去十年最成功的产品之一 765 99:59:59,999 --> 99:59:59,999 765 00:47:36,120 --> 00:47:41,025 AI概览也是推动Google智能镜头 视觉搜索增长的最强劲动力之一 766 99:59:59,999 --> 99:59:59,999 766 00:47:41,859 --> 00:47:44,762 智能镜头的搜索量同比增长65% 767 99:59:59,999 --> 99:59:59,999 767 00:47:45,129 --> 00:47:49,033 今年已有超千亿次视觉搜索 768 99:59:59,999 --> 99:59:59,999 768 00:47:49,667 --> 00:47:51,569 在我们最新的Gemini模型发布后 不仅用户查询量显著增加 769 99:59:59,999 --> 99:59:59,999 769 00:47:52,069 --> 00:47:56,974 而且他们还在尝试更为复杂的查询 770 99:59:59,999 --> 99:59:59,999 770 00:47:57,041 --> 00:47:59,477 AI概览达到了用户期待的 771 99:59:59,999 --> 99:59:59,999 771 00:47:59,543 --> 00:48:02,346 搜索质量与准确程度 772 99:59:59,999 --> 99:59:59,999 772 00:48:02,646 --> 00:48:04,882 并拥有行业领先的响应速度 773 99:59:59,999 --> 99:59:59,999 773 00:48:05,516 --> 00:48:08,552 为满足端到端AI搜索体验需求 774 99:59:59,999 --> 99:59:59,999 774 00:48:08,986 --> 00:48:12,056 我们正推出全新AI模式 775 99:59:59,999 --> 99:59:59,999 775 00:48:12,556 --> 00:48:17,128 这是对搜索的全面重构 同时搭配更先进的推理能力 776 99:59:59,999 --> 99:59:59,999 776 00:48:17,528 --> 00:48:22,666 你可以向AI模式提出更长、更复杂的查询 就像这样 777 99:59:59,999 --> 99:59:59,999 777 00:48:23,334 --> 00:48:30,074 实际上，用户查询长度已是传统搜索的2到3倍 778 99:59:59,999 --> 99:59:59,999 778 00:48:30,441 --> 00:48:33,511 而且我们支持多轮追问 779 99:59:59,999 --> 99:59:59,999 779 00:48:34,111 --> 00:48:38,649 所有这些今天正式上线 将作为新标签页整合到搜索中 780 99:59:59,999 --> 99:59:59,999 780 00:48:39,250 --> 00:48:43,154 我个人高频使用后发现 它彻底改变了我的搜索方式 781 99:59:59,999 --> 99:59:59,999 781 00:48:43,487 --> 00:48:46,891 很高兴宣布AI模式今日起 782 99:59:59,999 --> 99:59:59,999 782 00:48:46,957 --> 00:48:49,093 面向全美用户开放 783 99:59:59,999 --> 99:59:59,999 783 00:48:55,666 --> 00:48:59,837 AI模式将率先为Google搜索 引入我们的前沿技术 784 99:59:59,999 --> 99:59:59,999 784 00:49:00,204 --> 00:49:06,877 本周起，我们最智能的模型系列Gemini 2.5 将接入Google搜索 785 99:59:59,999 --> 99:59:59,999 785 00:49:07,211 --> 00:49:08,612 带来更丰富的功能 786 99:59:59,999 --> 99:59:59,999 786 00:49:08,913 --> 00:49:09,980 接下来有请Liz 787 99:59:59,999 --> 99:59:59,999 787 00:49:21,859 --> 00:49:23,027 Sundar，谢谢 788 99:59:59,999 --> 99:59:59,999 788 00:49:23,527 --> 00:49:28,432 短短一年时间 用户使用Google搜索的方式就发生了巨大变化 789 99:59:59,999 --> 99:59:59,999 789 00:49:29,166 --> 00:49:33,504 以前，用户在Google上搜索时 常常会问这样的问题 790 99:59:59,999 --> 99:59:59,999 790 00:49:34,505 --> 00:49:38,108 但现在，用户的问题变成了这样 791 99:59:59,999 --> 99:59:59,999 791 00:49:39,310 --> 00:49:42,346 用户的问题更长，也更复杂了 792 99:59:59,999 --> 99:59:59,999 792 00:49:42,980 --> 00:49:45,916 正如Sundar刚才所说，问题数量也大大增加了 793 99:59:59,999 --> 99:59:59,999 793 00:49:46,851 --> 00:49:49,553 用户正在亲身体验AI赋能的搜索 794 99:59:59,999 --> 99:59:59,999 794 00:49:50,087 --> 00:49:53,624 我们将Gemini模型的先进功能 795 99:59:59,999 --> 99:59:59,999 795 00:49:53,824 --> 00:49:57,995 与Google搜索对网络和世界信息的深刻理解 结合在一起 796 99:59:59,999 --> 99:59:59,999 796 00:49:59,129 --> 00:50:01,532 现在，你会发现你可以提出任何问题 797 99:59:59,999 --> 99:59:59,999 797 00:50:01,932 --> 00:50:08,105 更智能、更个性化的代理式搜索功能 将可以处理你最棘手的问题 798 99:59:59,999 --> 99:59:59,999 798 00:50:08,405 --> 00:50:09,907 并帮助你完成任务 799 99:59:59,999 --> 99:59:59,999 799 00:50:11,041 --> 00:50:13,110 这就是Google搜索的未来 800 99:59:59,999 --> 99:59:59,999 800 00:50:13,511 --> 00:50:16,747 它将超越信息本身，变得越来越智能 801 99:59:59,999 --> 99:59:59,999 801 00:50:17,948 --> 00:50:21,685 大家已经开始看到AI概览将这一切变为现实 802 99:59:59,999 --> 99:59:59,999 802 00:50:21,919 --> 00:50:24,188 而AI模式更是将它们提升到了新高度 803 99:59:59,999 --> 99:59:59,999 803 00:50:25,689 --> 00:50:29,894 AI模式是以Gemini 2.5为核心的变革性搜索体验 804 99:59:59,999 --> 99:59:59,999 804 00:50:30,794 --> 00:50:35,266 它是我们功能最强大的AI搜索 能够应对任何问题 805 99:59:59,999 --> 99:59:59,999 805 00:50:36,467 --> 00:50:37,668 正如Sundar所说 806 99:59:59,999 --> 99:59:59,999 806 00:50:37,735 --> 00:50:42,907 从今天开始 我们将在美国全面推出AI模式 807 99:59:59,999 --> 99:59:59,999 807 00:50:49,947 --> 00:50:53,851 它会以新标签页的形式直接出现在搜索页面中 大家也可以从搜索栏进入AI模式 808 99:59:59,999 --> 99:59:59,999 808 00:50:54,818 --> 00:50:59,723 AI模式将带来我们最精华的AI功能 809 99:59:59,999 --> 99:59:59,999 809 00:51:00,524 --> 00:51:02,092 但还不止于此 810 99:59:59,999 --> 99:59:59,999 810 00:51:02,159 --> 00:51:04,228 从它可窥见未来的广阔前景 811 99:59:59,999 --> 99:59:59,999 811 00:51:05,396 --> 00:51:09,967 未来，我们将逐步推出大量AI模型、先进的功能 812 99:59:59,999 --> 99:59:59,999 812 00:51:10,034 --> 00:51:12,670 并会直接整合到核心搜索体验中 813 99:59:59,999 --> 99:59:59,999 813 00:51:13,470 --> 00:51:19,443 今天就是起点 我们带来了为AI模式和AI概览提供支持的模型 814 99:59:59,999 --> 99:59:59,999 814 00:51:19,977 --> 00:51:23,714 大家在搜索框中就可以直接提出最难的问题 815 99:59:59,999 --> 99:59:59,999 815 00:51:24,848 --> 00:51:26,850 今天，我们将带大家全面了解AI模式 816 99:59:59,999 --> 99:59:59,999 816 00:51:26,917 --> 00:51:29,653 看看它如何运作，如何变得越来越强大 817 99:59:59,999 --> 99:59:59,999 817 00:51:30,554 --> 00:51:31,722 尤其是在个人使用场景 818 99:59:59,999 --> 99:59:59,999 818 00:51:32,122 --> 00:51:35,626 深度研究、复杂分析、可视化 819 99:59:59,999 --> 99:59:59,999 819 00:51:35,826 --> 00:51:38,662 实时多模态以及购物新方式等方面 820 99:59:59,999 --> 99:59:59,999 820 00:51:39,430 --> 00:51:42,232 亮点很多，因为AI模式的功能很强大 821 99:59:59,999 --> 99:59:59,999 821 00:51:42,766 --> 00:51:48,606 下面，我们具体来说一下 首先，使用AI模式，你可以随心所欲地提问 822 99:59:59,999 --> 99:59:59,999 822 00:51:48,939 --> 00:51:51,709 大家可以看到，搜索功能会立即开始工作 823 99:59:59,999 --> 99:59:59,999 823 00:51:52,443 --> 00:51:56,180 生成回答，将所有信息整合在一起 824 99:59:59,999 --> 99:59:59,999 824 00:51:56,714 --> 00:52:00,851 包括你可能从未发现的内容和创作者的链接 825 99:59:59,999 --> 99:59:59,999 825 00:52:00,918 --> 00:52:05,022 以及包含评分等实用信息的商家和企业的链接 826 99:59:59,999 --> 99:59:59,999 826 00:52:05,956 --> 00:52:09,393 Google搜索使用AI动态地调整整个界面 827 99:59:59,999 --> 99:59:59,999 827 00:52:10,094 --> 00:52:14,832 灵活组合文字、图片、链接，甚至是地图 828 99:59:59,999 --> 99:59:59,999 828 00:52:15,032 --> 00:52:16,433 只为更好地回答你的问题 829 99:59:59,999 --> 99:59:59,999 829 00:52:17,034 --> 00:52:19,003 你可以追问，就像跟人对话一样 830 99:59:59,999 --> 99:59:59,999 830 00:52:20,070 --> 00:52:22,473 现在，AI模式不仅仅是提供信息 831 99:59:59,999 --> 99:59:59,999 831 00:52:22,906 --> 00:52:26,010 更将搜索带到了全新的智能水平 832 99:59:59,999 --> 99:59:59,999 832 00:52:27,444 --> 00:52:31,148 这一切之所以能实现 得益于称之为查询扇出的技术 833 99:59:59,999 --> 99:59:59,999 833 00:52:32,182 --> 00:52:36,320 其原理是 Google搜索会识别需要高级推理的问题 834 99:59:59,999 --> 99:59:59,999 834 00:52:37,254 --> 00:52:39,256 并调用定制版的Gemini 835 99:59:59,999 --> 99:59:59,999 835 00:52:39,423 --> 00:52:41,925 将问题分解为若干子主题 836 99:59:59,999 --> 99:59:59,999 836 00:52:42,226 --> 00:52:46,230 然后替你同时发起大量查询 837 99:59:59,999 --> 99:59:59,999 837 00:52:46,830 --> 00:52:51,902 它会在整个网络进行搜索，深度远超传统搜索 838 99:59:59,999 --> 99:59:59,999 838 00:52:52,703 --> 00:52:55,806 它还会利用我们所有的实时信息数据集 839 99:59:59,999 --> 99:59:59,999 839 00:52:56,173 --> 00:53:01,211 例如知识图谱、购物信息图谱 以及本例中涉及的本地数据 840 99:59:59,999 --> 99:59:59,999 840 00:53:01,345 --> 00:53:06,116 这其中包含地图社区中，5亿多贡献者的数据洞见 841 99:59:59,999 --> 99:59:59,999 841 00:53:07,217 --> 00:53:09,987 Google搜索会整合所有信息给出回答 并进行自我检查 842 99:59:59,999 --> 99:59:59,999 842 00:53:10,487 --> 00:53:13,424 确保符合我们对信息质量的高标准要求 843 99:59:59,999 --> 99:59:59,999 843 00:53:14,058 --> 00:53:18,262 如果发现任何不足，它会发起更多搜索来弥补 844 99:59:59,999 --> 99:59:59,999 844 00:53:19,463 --> 00:53:20,631 这意味着，在AI模式下 845 99:59:59,999 --> 99:59:59,999 845 00:53:20,698 --> 00:53:25,369 你只需搜索一次，就能快速获得所有这些信息 846 99:59:59,999 --> 99:59:59,999 846 00:53:26,603 --> 00:53:30,374 现在来看看AI模式即将增加的新功能 这些功能会先在实验室中进行测试 847 99:59:59,999 --> 99:59:59,999 847 00:53:31,875 --> 00:53:35,579 不久之后，AI模式将能根据你以往的搜索记录 848 99:59:59,999 --> 99:59:59,999 848 00:53:35,946 --> 00:53:39,316 提供个性化的建议，从而使回答更具针对性 849 99:59:59,999 --> 99:59:59,999 849 00:53:40,150 --> 00:53:43,887 你还可以选择关联其他的Google应用 首先是Gmail 850 99:59:59,999 --> 99:59:59,999 850 00:53:44,988 --> 00:53:47,024 我们称之为个人使用场景 851 99:59:59,999 --> 99:59:59,999 851 00:53:47,224 --> 00:53:50,294 关联之后 AI模式可以引入你的个人使用场景来提供帮助 852 99:59:59,999 --> 99:59:59,999 852 00:53:51,195 --> 00:53:54,264 现在，根据你最近的酒店预订和搜索记录 853 99:59:59,999 --> 99:59:59,999 853 00:53:54,631 --> 00:53:56,467 它能推断出你更喜欢户外的座位 854 99:59:59,999 --> 99:59:59,999 854 00:53:57,201 --> 00:53:59,703 由于你订阅了这些画廊简报 855 99:59:59,999 --> 99:59:59,999 855 00:53:59,970 --> 00:54:03,107 它就会推荐一些精彩的艺术展 方便你在当地停留期间参观 856 99:59:59,999 --> 99:59:59,999 856 00:54:04,274 --> 00:54:05,409 但还不止这些 857 99:59:59,999 --> 99:59:59,999 857 00:54:05,876 --> 00:54:08,946 如果你的航班和酒店确认邮件都在收件箱里 858 99:59:59,999 --> 99:59:59,999 858 00:54:09,313 --> 00:54:12,783 它还会推荐你在纳什维尔时刚好能参加的活动 859 99:59:59,999 --> 99:59:59,999 859 00:54:13,016 --> 00:54:14,785 推荐你在停留期间的周边好去处 860 99:59:59,999 --> 99:59:59,999 860 00:54:16,153 --> 00:54:19,923 你会发现，AI模式中的个人使用场景功能 可以将Google搜索变成为你量身定制的工具 861 99:59:59,999 --> 99:59:59,999 861 00:54:20,491 --> 00:54:22,960 专门为你提供推荐内容 862 99:59:59,999 --> 99:59:59,999 862 00:54:23,827 --> 00:54:25,529 这一切始终都由你掌控 863 99:59:59,999 --> 99:59:59,999 863 00:54:25,996 --> 00:54:28,732 你可以随时选择关联或解除关联 864 99:59:59,999 --> 99:59:59,999 864 00:54:29,833 --> 00:54:32,536 个人使用场景将于今年夏天在AI模式中上线 865 99:59:59,999 --> 99:59:59,999 865 00:54:40,811 --> 00:54:44,081 另外，我们将针对需要更深入解答的问题 866 99:59:59,999 --> 99:59:59,999 866 00:54:44,648 --> 00:54:47,851 为AI模式引入Deep Research功能 867 99:59:59,999 --> 99:59:59,999 867 00:54:48,652 --> 00:54:51,789 大家已经在使用Google搜索来剖析某个主题 868 99:59:59,999 --> 99:59:59,999 868 00:54:52,156 --> 00:54:54,391 而这项功能将研究的深度和广度又提升了一个级别 869 99:59:59,999 --> 99:59:59,999 869 00:54:54,992 --> 00:54:57,861 所以我们称之为Deep Search 870 99:59:59,999 --> 99:59:59,999 870 00:54:58,796 --> 00:55:02,032 Deep Search同样采用前面提到的查询扇出技术 871 99:59:59,999 --> 99:59:59,999 871 00:55:02,399 --> 00:55:03,567 但功能更强 872 99:59:59,999 --> 99:59:59,999 872 00:55:03,867 --> 00:55:07,805 它可以替你发起数十次甚至数百次搜索 873 99:59:59,999 --> 99:59:59,999 873 00:55:08,539 --> 00:55:11,708 它会对所有分散的信息进行分析推理 874 99:59:59,999 --> 99:59:59,999 874 00:55:11,975 --> 00:55:15,746 并在几分钟内生成一份 带有完整引用信息的专家级报告 875 99:59:59,999 --> 99:59:59,999 875 00:55:16,747 --> 00:55:18,448 报告中会包含网页链接 876 99:59:59,999 --> 99:59:59,999 876 00:55:18,715 --> 00:55:21,151 方便你进一步探索和采取行动 877 99:59:59,999 --> 99:59:59,999 877 00:55:22,419 --> 00:55:25,589 这是我们构建AI模式的核心理念 878 99:59:59,999 --> 99:59:59,999 878 00:55:26,223 --> 00:55:28,826 以及我们对搜索领域中AI运用的一贯思考 879 99:59:59,999 --> 99:59:59,999 879 00:55:29,693 --> 00:55:32,629 因为我们坚信，AI将成为Web有史以来 880 99:59:59,999 --> 99:59:59,999 880 00:55:32,696 --> 00:55:34,598 最强大的发现引擎 881 99:59:59,999 --> 99:59:59,999 881 00:55:35,566 --> 00:55:38,602 帮助人们发现网络中更多有价值的信息 882 99:59:59,999 --> 99:59:59,999 882 00:55:39,236 --> 00:55:41,772 找到令人惊叹且高度相关的内容 883 99:59:59,999 --> 99:59:59,999 883 00:55:43,240 --> 00:55:46,109 大家已经开始感受到搜索变得越来越智能 884 99:59:59,999 --> 99:59:59,999 884 00:55:46,376 --> 00:55:47,845 但精彩远不止于此 885 99:59:59,999 --> 99:59:59,999 885 00:55:48,545 --> 00:55:50,013 接下来，我把时间交给Rajan 886 99:59:59,999 --> 99:59:59,999 886 00:55:50,080 --> 00:55:52,983 他将介绍AI模式如何帮助他分析 887 99:59:59,999 --> 99:59:59,999 887 00:55:53,584 --> 00:55:55,853 他热爱的一个体育领域里的复杂数据 888 99:59:59,999 --> 99:59:59,999 888 00:56:06,964 --> 00:56:11,301 Liz，谢谢你！我是一个狂热的棒球迷 889 99:59:59,999 --> 99:59:59,999 889 00:56:11,368 --> 00:56:14,972 最近关于这种叫做“鱼雷球棒”的新型球棒 引起了很多讨论 890 99:59:59,999 --> 99:59:59,999 890 00:56:15,439 --> 00:56:16,940 为不太关注棒球的朋友解释一下 891 99:59:59,999 --> 99:59:59,999 891 00:56:17,140 --> 00:56:21,612 这是一种新型球棒 它的特点是将更多重量集中在最佳击球点 892 99:59:59,999 --> 99:59:59,999 892 00:56:22,246 --> 00:56:26,650 正如大家所见 我一直在研究这种球棒是否真的对比赛产生了影响 893 99:59:59,999 --> 99:59:59,999 893 00:56:27,117 --> 00:56:29,786 现在，我想知道数据会怎么说 894 99:59:59,999 --> 99:59:59,999 894 00:56:30,220 --> 00:56:31,521 于是，我向Google搜索提出了一个问题 895 99:59:59,999 --> 99:59:59,999 895 00:56:31,588 --> 00:56:35,425 请显示目前使用鱼雷球棒的知名球员 896 99:59:59,999 --> 99:59:59,999 896 00:56:35,726 --> 00:56:38,795 在本赛季和上赛季的打击率和上垒率 897 99:59:59,999 --> 99:59:59,999 897 00:56:39,563 --> 00:56:43,000 这个问题并不简单，包含了多重维度 898 99:59:59,999 --> 99:59:59,999 898 00:56:43,400 --> 00:56:46,370 Google搜索需要识别哪些是知名球员 899 99:59:59,999 --> 99:59:59,999 899 00:56:46,637 --> 00:56:49,840 谁在使用鱼雷球棒，并整合他们的赛季数据 900 99:59:59,999 --> 99:59:59,999 900 00:56:50,908 --> 00:56:54,745 但它马上给出了非常实用的回答 其中还有容易理解的表格 901 99:59:59,999 --> 99:59:59,999 901 00:56:55,746 --> 00:56:58,215 而且，数据新鲜、精准 902 99:59:59,999 --> 99:59:59,999 902 00:56:58,415 --> 00:57:02,286 因为它使用的是我们持续更新的体育数据 903 99:59:59,999 --> 99:59:59,999 903 00:57:02,386 --> 00:57:03,654 甚至精确到最近一次好球 904 99:59:59,999 --> 99:59:59,999 904 00:57:04,655 --> 00:57:06,990 Google搜索甚至还提供了重要的背景信息 905 99:59:59,999 --> 99:59:59,999 905 00:57:07,157 --> 00:57:09,559 比如“目前赛季仍处于早期阶段” 906 99:59:59,999 --> 99:59:59,999 906 00:57:10,460 --> 00:57:14,965 我继续追问：这些球员本赛季击出多少支本垒打？ 907 99:59:59,999 --> 99:59:59,999 907 00:57:15,465 --> 00:57:17,935 瞬间就获得了一张图表 908 99:59:59,999 --> 99:59:59,999 908 00:57:19,002 --> 00:57:20,804 这正好印证了Liz刚才提到的 909 99:59:59,999 --> 99:59:59,999 909 00:57:20,871 --> 00:57:25,776 AI模式能够为每个回答动态生成合适的界面 910 99:59:59,999 --> 99:59:59,999 910 00:57:26,576 --> 00:57:30,681 Google搜索确定 最适合呈现这些信息的方式是图表 911 99:59:59,999 --> 99:59:59,999 911 00:57:30,914 --> 00:57:36,553 因此生成了图表 就像Google搜索里内置了一个专属体育分析师 912 99:59:59,999 --> 99:59:59,999 912 00:57:37,487 --> 00:57:42,225 针对体育和金融问题的复杂数据分析和可视化功能 913 99:59:59,999 --> 99:59:59,999 913 00:57:42,292 --> 00:57:43,760 将在今年夏天上线 914 99:59:59,999 --> 99:59:59,999 914 00:57:50,334 --> 00:57:55,072 说了这么多棒球，我都想去现场看比赛了 915 99:59:59,999 --> 99:59:59,999 915 00:57:55,439 --> 00:57:57,708 比如说，下一场就去看 916 99:59:59,999 --> 99:59:59,999 916 00:57:58,308 --> 00:58:01,278 但是，购买好票总是让人头大 917 99:59:59,999 --> 99:59:59,999 917 00:58:01,945 --> 00:58:03,981 这里很高兴地告诉大家 918 99:59:59,999 --> 99:59:59,999 918 00:58:04,214 --> 00:58:08,885 我们正将Project Mariner的代理式功能 整合到AI模式中 919 99:59:59,999 --> 99:59:59,999 919 00:58:09,653 --> 00:58:13,690 大家已经看到AI模式正变得更加智能和个性化 920 99:59:59,999 --> 99:59:59,999 920 00:58:14,191 --> 00:58:17,527 现在可以开始看到Google搜索正变得更加主动 921 99:59:59,999 --> 99:59:59,999 921 00:58:18,161 --> 00:58:22,399 它能帮我完成任务，但仍由我保持掌控权 922 99:59:59,999 --> 99:59:59,999 922 00:58:22,933 --> 00:58:23,767 我会告诉它 923 99:59:59,999 --> 99:59:59,999 923 00:58:24,134 --> 00:58:28,572 “找两张这周六红人队比赛下层看台的平价门票” 924 99:59:59,999 --> 99:59:59,999 924 00:58:29,639 --> 00:58:31,274 Google搜索启动查询扇出 925 99:59:59,999 --> 99:59:59,999 925 00:58:31,608 --> 00:58:35,712 同时扫描多个网站，分析上百个潜在的门票选项 926 99:59:59,999 --> 99:59:59,999 926 00:58:36,246 --> 00:58:40,951 按我提出的所有条件填写表单 不用我去做这些繁琐的工作 927 99:59:59,999 --> 99:59:59,999 927 00:58:42,052 --> 00:58:43,687 然后将所有信息汇总在一起 928 99:59:59,999 --> 99:59:59,999 928 00:58:43,987 --> 00:58:48,392 基于搜索结果进行推理，分析实时票价和库存 929 99:59:59,999 --> 99:59:59,999 929 00:58:48,792 --> 00:58:52,529 然后，像这样，任务就完成了 930 99:59:59,999 --> 99:59:59,999 930 00:58:53,263 --> 00:58:56,400 它给出了不错的门票选项，还有实用的背景信息 931 99:59:59,999 --> 99:59:59,999 931 00:58:56,600 --> 00:58:58,568 我可以在此基础上做出明智的决定 932 99:59:59,999 --> 99:59:59,999 932 00:58:59,636 --> 00:59:02,906 这些座位看起来视野良好，价格合理 933 99:59:59,999 --> 99:59:59,999 933 00:59:03,740 --> 00:59:05,742 Google搜索已代我完成了大量步骤填写这些信息 934 99:59:59,999 --> 99:59:59,999 934 00:59:06,076 --> 00:59:08,211 并提供了链接，引导我完成结账 935 99:59:59,999 --> 99:59:59,999 935 00:59:09,279 --> 00:59:10,414 定下这些门票 936 99:59:59,999 --> 99:59:59,999 936 00:59:18,455 --> 00:59:20,490 Google搜索很快将能够帮助处理这类任务 937 99:59:59,999 --> 99:59:59,999 937 00:59:20,557 --> 00:59:25,896 首先覆盖的是活动门票、餐厅预订和本地生活服务预约 938 99:59:59,999 --> 99:59:59,999 938 00:59:28,198 --> 00:59:31,168 接下来，让我们聊聊多模态技术 939 99:59:59,999 --> 99:59:59,999 939 00:59:31,768 --> 00:59:34,571 从多模态搜索这个概念还不怎么流行的时候 940 99:59:59,999 --> 99:59:59,999 940 00:59:34,638 --> 00:59:36,373 我们就一直是这个领域的方向引领者 941 99:59:59,999 --> 99:59:59,999 941 00:59:36,907 --> 00:59:40,844 回到2017年，我们就是在这个讲台上 宣布推出了Google智能镜头 942 99:59:59,999 --> 99:59:59,999 942 00:59:41,044 --> 00:59:45,315 此后我们不断优化，让所见即所搜更加便捷 943 99:59:59,999 --> 99:59:59,999 943 00:59:46,016 --> 00:59:49,286 只需用Google智能镜头拍照或使用圈定即搜 944 99:59:59,999 --> 99:59:59,999 944 00:59:49,352 --> 00:59:52,055 就能立即获得AI概览 945 99:59:59,999 --> 99:59:59,999 945 00:59:52,522 --> 00:59:53,723 Sundar刚才提到过 946 99:59:59,999 --> 99:59:59,999 946 00:59:53,990 --> 00:59:55,926 视觉搜索目前正呈现出燎原之势 947 99:59:59,999 --> 99:59:59,999 947 00:59:55,992 --> 00:59:58,261 今天我很激动地宣布 948 99:59:59,999 --> 99:59:59,999 948 00:59:58,328 --> 01:00:02,332 每月有超过15亿用户使用Google智能镜头 949 99:59:59,999 --> 99:59:59,999 949 01:00:08,105 --> 01:00:11,408 现在，我们正围绕多模态迈出重大的一步 950 99:59:59,999 --> 99:59:59,999 950 01:00:11,675 --> 01:00:16,279 将Project Astra的实时功能引入AI模式 951 99:59:59,999 --> 99:59:59,999 951 01:00:16,947 --> 01:00:20,884 想想那些对话交流一下，就会变得超级简单的问题 952 99:59:59,999 --> 99:59:59,999 952 01:00:21,151 --> 01:00:22,953 而且自己的意图还能得到准确的展示 953 99:59:59,999 --> 99:59:59,999 953 01:00:23,487 --> 01:00:25,455 比如家庭DIY维修 954 99:59:59,999 --> 99:59:59,999 954 01:00:25,655 --> 01:00:28,391 棘手的学校作业，或是学习新技能 955 99:59:59,999 --> 99:59:59,999 955 01:00:28,959 --> 01:00:30,694 我们称之为实时搜索 956 99:59:59,999 --> 99:59:59,999 956 01:00:30,927 --> 01:00:32,963 现在利用摄像头 957 99:59:59,999 --> 99:59:59,999 957 01:00:33,130 --> 01:00:36,299 Google搜索能随着用户来回走动 958 99:59:59,999 --> 99:59:59,999 958 01:00:36,366 --> 01:00:39,035 实时感知他们看到的画面，并提供有用信息 959 99:59:59,999 --> 99:59:59,999 959 01:00:39,402 --> 01:00:42,072 就像在与Google搜索进行视频通话 960 99:59:59,999 --> 99:59:59,999 960 01:00:43,073 --> 01:00:46,943 我有三个孩子，他们每分钟能问上百万个问题 961 99:59:59,999 --> 99:59:59,999 961 01:00:47,310 --> 01:00:49,346 随着暑假临近 962 99:59:59,999 --> 99:59:59,999 962 01:00:49,513 --> 01:00:53,350 我和团队决定对实时搜索进行终极测试 963 99:59:59,999 --> 99:59:59,999 963 01:00:53,483 --> 01:00:57,320 帮助我们和孩子们解决些新问题 964 99:59:59,999 --> 99:59:59,999 964 01:00:58,054 --> 01:01:01,958 上周末，我们与我的家人在家里录制了演示 965 99:59:59,999 --> 99:59:59,999 965 01:01:02,058 --> 01:01:02,893 一起看下吧 966 99:59:59,999 --> 99:59:59,999 966 01:01:04,161 --> 01:01:06,696 看起来你们准备做有趣的科学实验 967 99:59:59,999 --> 99:59:59,999 967 01:01:06,763 --> 01:01:09,232 好的。准备好了吗？ 968 99:59:59,999 --> 99:59:59,999 968 01:01:09,299 --> 01:01:10,467 -好了 -好的 969 99:59:59,999 --> 99:59:59,999 969 01:01:10,534 --> 01:01:13,003 -你准备好开始实验了吗？ -好了 970 99:59:59,999 --> 99:59:59,999 970 01:01:13,103 --> 01:01:17,040 能猜出我要做什么实验吗？我有双氧水 971 99:59:59,999 --> 99:59:59,999 971 01:01:17,107 --> 01:01:18,375 洗洁精和酵母 972 99:59:59,999 --> 99:59:59,999 972 01:01:18,441 --> 01:01:20,443 你可能要做大象牙膏实验 973 99:59:59,999 --> 99:59:59,999 973 01:01:20,510 --> 01:01:22,979 我知道哪颗草莓可以吃了 974 99:59:59,999 --> 99:59:59,999 974 01:01:23,046 --> 01:01:25,282 我在想办法让遥控器工作 975 99:59:59,999 --> 99:59:59,999 975 01:01:25,382 --> 01:01:29,252 看来有人准备种下青豆幼苗 976 99:59:59,999 --> 99:59:59,999 976 01:01:29,319 --> 01:01:31,321 把压力调到每平方英寸50磅左右 977 99:59:59,999 --> 99:59:59,999 977 01:01:31,388 --> 01:01:32,989 但不要超过每平方英寸90磅 978 99:59:59,999 --> 99:59:59,999 978 01:01:33,156 --> 01:01:35,725 我混合了小苏打，接下来做什么？ 979 99:59:59,999 --> 99:59:59,999 979 01:01:35,792 --> 01:01:38,728 看起来有人把手指伸进肉桂水里了 980 99:59:59,999 --> 99:59:59,999 980 01:01:38,795 --> 01:01:40,931 如何让实验更壮观？ 981 99:59:59,999 --> 99:59:59,999 981 01:01:40,997 --> 01:01:42,933 可以尝试不同的催化剂 982 99:59:59,999 --> 99:59:59,999 982 01:01:42,999 --> 01:01:45,202 碘化钾是个好选择 983 99:59:59,999 --> 99:59:59,999 983 01:01:45,569 --> 01:01:47,003 哇！为什么会这样？ 984 99:59:59,999 --> 99:59:59,999 984 01:01:47,204 --> 01:01:49,472 成熟的草莓也会有甜香味 985 99:59:59,999 --> 99:59:59,999 985 01:01:49,539 --> 01:01:52,309 顶部绿叶应新鲜、有活力 986 99:59:59,999 --> 99:59:59,999 986 01:01:52,375 --> 01:01:54,177 化学反应进行得很顺利 987 99:59:59,999 --> 99:59:59,999 987 01:01:54,377 --> 01:01:55,946 -太棒了！谢谢！ 988 99:59:59,999 --> 99:59:59,999 988 01:01:57,681 --> 01:01:58,615 -再来一次可以吗？ 989 99:59:59,999 --> 99:59:59,999 989 01:01:58,682 --> 01:02:04,221 小伙子，怎么洗掉衣服上的草莓渍？ 990 99:59:59,999 --> 99:59:59,999 990 01:02:04,654 --> 01:02:07,290 尝试用一汤匙白醋 991 99:59:59,999 --> 99:59:59,999 991 01:02:07,357 --> 01:02:09,259 加半茶匙洗衣液 992 99:59:59,999 --> 99:59:59,999 992 01:02:23,907 --> 01:02:29,846 通过AI模式 我们正在带来全新Google智能购物体验 993 99:59:59,999 --> 99:59:59,999 993 01:02:30,914 --> 01:02:32,148 前面提到过 994 99:59:59,999 --> 99:59:59,999 994 01:02:32,549 --> 01:02:37,554 AI模式整合了网络信息和实时数据 995 99:59:59,999 --> 99:59:59,999 995 01:02:37,954 --> 01:02:42,492 在这种情况下 这意味着既可通过Google图片获得视觉灵感 996 99:59:59,999 --> 99:59:59,999 996 01:02:43,026 --> 01:02:48,932 又能获得购物信息图谱提供的 海量商品和零售商数据 997 99:59:59,999 --> 99:59:59,999 997 01:02:49,366 --> 01:02:54,671 其中包含超过500亿条持续更新的商品详情 998 99:59:59,999 --> 99:59:59,999 998 01:02:55,538 --> 01:02:57,974 我来演示一下这是如何实现的 999 99:59:59,999 --> 99:59:59,999 999 01:02:58,575 --> 01:03:02,212 我一直在为我的客厅寻找一块新地毯 1000 99:59:59,999 --> 99:59:59,999 1000 01:03:02,646 --> 01:03:06,416 因此，我给出提示：我家有一个浅灰色的沙发 1001 99:59:59,999 --> 99:59:59,999 1001 01:03:06,549 --> 01:03:09,352 想找一块能让房间看起来更明亮的地毯 1002 99:59:59,999 --> 99:59:59,999 1002 01:03:10,353 --> 01:03:12,889 我首先需要的是视觉灵感 1003 99:59:59,999 --> 99:59:59,999 1003 01:03:13,857 --> 01:03:18,428 Google搜索根据我的喜好 动态生成了一个可浏览的图片拼接图 1004 99:59:59,999 --> 99:59:59,999 1004 01:03:18,828 --> 01:03:22,565 并显示了一些可购买的商品 1005 99:59:59,999 --> 99:59:59,999 1005 01:03:23,967 --> 01:03:28,138 其中就有我常用品牌的地毯 以及风格更现代的选项 1006 99:59:59,999 --> 99:59:59,999 1006 01:03:28,204 --> 01:03:30,740 因为我经常搜索这类风格的物品 1007 99:59:59,999 --> 99:59:59,999 1007 01:03:31,708 --> 01:03:37,781 这些推荐虽好，但大多数在我家恐怕连一天都撑不过 1008 99:59:59,999 --> 99:59:59,999 1008 01:03:39,249 --> 01:03:40,350 于是我补充道 1009 99:59:59,999 --> 99:59:59,999 1009 01:03:43,086 --> 01:03:46,990 家里有四个活泼的孩子，他们经常带朋友来玩 1010 99:59:59,999 --> 99:59:59,999 1010 01:03:48,291 --> 01:03:51,695 Google搜索立刻意识到了那种杂乱的场面 1011 99:59:59,999 --> 99:59:59,999 1011 01:03:52,329 --> 01:03:57,400 于是推荐了采用耐用材质的短绒地毯或可水洗地毯 1012 99:59:59,999 --> 99:59:59,999 1012 01:03:57,467 --> 01:03:59,436 还附上了一些可直接购买的产品 1013 99:59:59,999 --> 99:59:59,999 1013 01:04:00,704 --> 01:04:02,539 不过我对地毯并不专业 1014 99:59:59,999 --> 99:59:59,999 1014 01:04:02,605 --> 01:04:06,309 好在Google搜索贴心地指出了需要考虑的事项 1015 99:59:59,999 --> 99:59:59,999 1015 01:04:06,443 --> 01:04:09,145 并提供了后续实用建议 1016 99:59:59,999 --> 99:59:59,999 1016 01:04:10,180 --> 01:04:15,352 原本需要几周的研究，现在只需几分钟 1017 99:59:59,999 --> 99:59:59,999 1017 01:04:16,286 --> 01:04:20,557 可以想象，这种全新的AI模式体验 1018 99:59:59,999 --> 99:59:59,999 1018 01:04:20,890 --> 01:04:23,059 对购物需求会有多实用 1019 99:59:59,999 --> 99:59:59,999 1019 01:04:23,593 --> 01:04:27,263 说到购物，我需要几件夏天穿的新裙子 1020 99:59:59,999 --> 99:59:59,999 1020 01:04:28,098 --> 01:04:30,567 Google搜索给了我一些不错的选择 1021 99:59:59,999 --> 99:59:59,999 1021 01:04:30,967 --> 01:04:36,139 但现在我遇到了经典网购难题 1022 99:59:59,999 --> 99:59:59,999 1022 01:04:37,006 --> 01:04:41,144 那就是完全不知道这些款式穿在我身上会是什么样 1023 99:59:59,999 --> 99:59:59,999 1023 01:04:42,045 --> 01:04:44,881 于是，我们将推出一项新的试穿功能 1024 99:59:59,999 --> 99:59:59,999 1024 01:04:45,315 --> 01:04:48,251 帮用户虚拟试穿 1025 99:59:59,999 --> 99:59:59,999 1025 01:04:48,385 --> 01:04:51,955 感受款式在身上的效果 1026 99:59:59,999 --> 99:59:59,999 1026 01:04:52,989 --> 01:04:54,491 我来实时演示一下 1027 99:59:59,999 --> 99:59:59,999 1027 01:04:56,393 --> 01:05:02,365 在一通搜索之后，我找到一条喜欢的蓝裙子 1028 99:59:59,999 --> 99:59:59,999 1028 01:05:03,099 --> 01:05:05,635 然后点击这个按钮试穿 1029 99:59:59,999 --> 99:59:59,999 1029 01:05:05,702 --> 01:05:10,306 系统提示我上传照片，并引导我进入相册 1030 99:59:59,999 --> 99:59:59,999 1030 01:05:10,840 --> 01:05:12,175 我有很多照片 1031 99:59:59,999 --> 99:59:59,999 1031 01:05:12,242 --> 01:05:16,312 于是打算选一张清晰的全身照 1032 99:59:59,999 --> 99:59:59,999 1032 01:05:16,846 --> 01:05:18,515 处理就开始了 1033 99:59:59,999 --> 99:59:59,999 1033 01:05:19,783 --> 01:05:23,319 在等待系统处理的时候 我想介绍下背后的技术原理 1034 99:59:59,999 --> 99:59:59,999 1034 01:05:24,220 --> 01:05:27,323 为了打造这种规模化运作的试穿功能 1035 99:59:59,999 --> 99:59:59,999 1035 01:05:27,624 --> 01:05:30,393 我们需要深度理解人体结构 1036 99:59:59,999 --> 99:59:59,999 1036 01:05:30,794 --> 01:05:32,862 及服装在人体上的呈现效果 1037 99:59:59,999 --> 99:59:59,999 1037 01:05:32,929 --> 01:05:36,399 为此，我们构建了一个专门针对时尚行业训练的 1038 99:59:59,999 --> 99:59:59,999 1038 01:05:36,599 --> 01:05:38,968 定制图片生成模型 1039 99:59:59,999 --> 99:59:59,999 1039 01:05:40,036 --> 01:05:41,237 哇，处理完成了！ 1040 99:59:59,999 --> 99:59:59,999 1040 01:05:50,079 --> 01:05:53,616 必须说实时演示成功的感觉太棒了 1041 99:59:59,999 --> 99:59:59,999 1041 01:05:56,486 --> 01:05:58,221 我们来继续聊背后的原理 1042 99:59:59,999 --> 99:59:59,999 1042 01:05:58,421 --> 01:06:01,391 该技术整合了先进的3D形体理解能力 1043 99:59:59,999 --> 99:59:59,999 1043 01:06:01,925 --> 01:06:05,862 能更精准地感知形状与深度 1044 99:59:59,999 --> 99:59:59,999 1044 01:06:06,029 --> 01:06:08,765 从而更好地呈现人体轮廓 1045 99:59:59,999 --> 99:59:59,999 1045 01:06:09,332 --> 01:06:12,101 这种体验完全基于你的照片 1046 99:59:59,999 --> 99:59:59,999 1046 01:06:12,168 --> 01:06:16,039 而非预先拍摄的图片或跟你不像的模特 1047 99:59:59,999 --> 99:59:59,999 1047 01:06:16,573 --> 01:06:19,943 对于你心仪的服饰 1048 99:59:59,999 --> 99:59:59,999 1048 01:06:20,143 --> 01:06:26,950 AI模型能够呈现面料褶皱、延展和垂坠感 1049 99:59:59,999 --> 99:59:59,999 1049 01:06:28,284 --> 01:06:32,188 这是一项业内领先的技术，可大规模应用 1050 99:59:59,999 --> 99:59:59,999 1050 01:06:32,255 --> 01:06:34,190 借助它，我们能以直观的方式 1051 99:59:59,999 --> 99:59:59,999 1051 01:06:34,457 --> 01:06:38,895 呈现数十亿服装产品 在不同体型的人群身上的穿着效果 1052 99:59:59,999 --> 99:59:59,999 1052 01:06:39,262 --> 01:06:40,530 大家可以看到 1053 99:59:59,999 --> 99:59:59,999 1053 01:06:40,663 --> 01:06:44,334 这个功能让我真实感受到了裙子上身的效果 1054 99:59:59,999 --> 99:59:59,999 1054 01:06:45,235 --> 01:06:47,604 选好裙子后 1055 99:59:59,999 --> 99:59:59,999 1055 01:06:47,904 --> 01:06:51,207 Google搜索还能通过全新代理式结账功能 1056 99:59:59,999 --> 99:59:59,999 1056 01:06:51,574 --> 01:06:55,144 按目标价位帮我购买 1057 99:59:59,999 --> 99:59:59,999 1057 01:06:55,411 --> 01:06:57,914 现在回到这件裙子的页面 1058 99:59:59,999 --> 99:59:59,999 1058 01:06:58,481 --> 01:07:02,252 我想点击按钮，跟踪价格 1059 99:59:59,999 --> 99:59:59,999 1059 01:07:02,952 --> 01:07:04,754 选择尺码后 1060 99:59:59,999 --> 99:59:59,999 1060 01:07:05,555 --> 01:07:07,090 设定目标价 格 1061 99:59:59,999 --> 99:59:59,999 1061 01:07:07,156 --> 01:07:09,826 我打算设成50美元左右 1062 99:59:59,999 --> 99:59:59,999 1062 01:07:12,462 --> 01:07:14,063 跟踪功能马上启动 1063 99:59:59,999 --> 99:59:59,999 1063 01:07:14,197 --> 01:07:18,368 Google搜索现在将持续监测销售该裙子的网站 1064 99:59:59,999 --> 99:59:59,999 1064 01:07:18,434 --> 01:07:21,104 在价格降到位时通知我 1065 99:59:59,999 --> 99:59:59,999 1065 01:07:21,204 --> 01:07:24,641 现在我们退出实时演示 1066 99:59:59,999 --> 99:59:59,999 1066 01:07:25,108 --> 01:07:27,477 这里我施展一点魔法 1067 99:59:59,999 --> 99:59:59,999 1067 01:07:27,744 --> 01:07:29,913 假设价格现在降到位了 1068 99:59:59,999 --> 99:59:59,999 1068 01:07:30,847 --> 01:07:35,351 此时系统会推送类似画面中的通知 1069 99:59:59,999 --> 99:59:59,999 1069 01:07:35,718 --> 01:07:36,920 我确认购买后 1070 99:59:59,999 --> 99:59:59,999 1070 01:07:37,287 --> 01:07:41,791 结账智能体会将尺码和颜色合适的裙子 添加到购物车中 1071 99:59:59,999 --> 99:59:59,999 1071 01:07:42,392 --> 01:07:45,828 我可以选择核对所有支付与配送信息 1072 99:59:59,999 --> 99:59:59,999 1072 01:07:46,229 --> 01:07:48,965 或直接授权智能体完成购买 1073 99:59:59,999 --> 99:59:59,999 1073 01:07:58,408 --> 01:08:03,646 只需点按一下，Google搜索就能 使用Google Pay安全地代我完成购买 1074 99:59:59,999 --> 99:59:59,999 1074 01:08:03,713 --> 01:08:06,382 当然，从灵感激发到最终购买 1075 99:59:59,999 --> 99:59:59,999 1075 01:08:07,383 --> 01:08:09,919 整个流程全程由我主导 1076 99:59:59,999 --> 99:59:59,999 1076 01:08:10,887 --> 01:08:13,089 我们的全新智能功能 1077 99:59:59,999 --> 99:59:59,999 1077 01:08:13,389 --> 01:08:18,027 将最好的购物体验与顶尖AI技术 融合在了Google搜索中 1078 99:59:59,999 --> 99:59:59,999 1078 01:08:18,761 --> 01:08:21,898 全新可视化购物及结账功能 1079 99:59:59,999 --> 99:59:59,999 1079 01:08:21,965 --> 01:08:23,766 将在未来几个月内逐步推出 1080 99:59:59,999 --> 99:59:59,999 1080 01:08:23,967 --> 01:08:28,571 虚拟试穿功能即日起可在实验室版本中体验 1081 99:59:59,999 --> 99:59:59,999 1081 01:08:36,245 --> 01:08:37,647 现在把时间交还给Liz 1082 99:59:59,999 --> 99:59:59,999 1082 01:08:45,955 --> 01:08:48,758 感谢Vidhya！正如Sundar所说 1083 99:59:59,999 --> 99:59:59,999 1083 01:08:48,925 --> 01:08:52,028 本周晚些时候即将迎来重大升级 1084 99:59:59,999 --> 99:59:59,999 1084 01:08:52,428 --> 01:08:58,501 Gemini 2.5定制版将集成到AI模式和AI概览中 1085 99:59:59,999 --> 99:59:59,999 1085 01:08:59,435 --> 01:09:02,105 这意味着，Google搜索迎来了新篇章 1086 99:59:59,999 --> 99:59:59,999 1086 01:09:02,772 --> 01:09:05,441 受益于此，AI概览将会把AI的实用性 1087 99:59:59,999 --> 99:59:59,999 1087 01:09:05,608 --> 01:09:08,311 带给核心Google搜索体验的每一个用户 1088 99:59:59,999 --> 99:59:59,999 1088 01:09:09,078 --> 01:09:10,380 正如你今天所见 1089 99:59:59,999 --> 99:59:59,999 1089 01:09:10,613 --> 01:09:13,716 AI模式是一种彻底重构的AI搜索 1090 99:59:59,999 --> 99:59:59,999 1090 01:09:14,350 --> 01:09:17,654 集成了我们最前沿的AI特性和功能 1091 99:59:59,999 --> 99:59:59,999 1091 01:09:18,488 --> 01:09:21,624 我们很激动，Google搜索迎来了新篇章 1092 99:59:59,999 --> 99:59:59,999 1092 01:09:21,991 --> 01:09:26,996 毫不夸张地说，用户可以真正地问它任何事 1093 99:59:59,999 --> 99:59:59,999 1093 01:09:27,764 --> 01:09:30,066 从最简单到最棘手的问题 1094 99:59:59,999 --> 99:59:59,999 1094 01:09:30,466 --> 01:09:33,936 最深的研究、个性化购物需求 1095 99:59:59,999 --> 99:59:59,999 1095 01:09:34,137 --> 01:09:36,305 希望代劳的琐事 1096 99:59:59,999 --> 99:59:59,999 1096 01:09:37,106 --> 01:09:39,275 只需提问，即可获得响应 1097 99:59:59,999 --> 99:59:59,999 1097 01:09:42,311 --> 01:09:47,417 好戏开场！我的魅力自然流露 1098 99:59:59,999 --> 99:59:59,999 1098 01:09:47,583 --> 01:09:50,653 我的风格如此惊艳，让所有人都为之震撼 1099 99:59:59,999 --> 99:59:59,999 1099 01:09:50,720 --> 01:09:53,122 等等！我感觉妙不可言 1100 99:59:59,999 --> 99:59:59,999 1100 01:09:54,323 --> 01:09:57,193 来吧，展示你如何舞动 舞动、舞动，释放自我 1101 99:59:59,999 --> 99:59:59,999 1101 01:09:57,260 --> 01:09:59,095 我用崭新视角迎接挑战 1102 99:59:59,999 --> 99:59:59,999 1102 01:09:59,295 --> 01:10:01,130 将热情点燃，让全场沸腾 1103 99:59:59,999 --> 99:59:59,999 1103 01:10:01,197 --> 01:10:03,499 我们将激情调到最高点，就是这样，来吧！ 1104 99:59:59,999 --> 99:59:59,999 1104 01:10:03,566 --> 01:10:05,068 舞动、舞动，释放精彩 1105 99:59:59,999 --> 99:59:59,999 1105 01:10:05,201 --> 01:10:06,602 来吧，尽情展现你的独特魅力 1106 99:59:59,999 --> 99:59:59,999 1106 01:10:06,669 --> 01:10:08,638 我还应该做些什么来让它更坚固？ 1107 99:59:59,999 --> 99:59:59,999 1107 01:10:08,705 --> 01:10:12,275 为了让它更坚固 建议在设计中运用更多三角形结构 1108 99:59:59,999 --> 99:59:59,999 1108 01:10:13,876 --> 01:10:15,511 已经足够完美了 1109 99:59:59,999 --> 99:59:59,999 1109 01:10:15,645 --> 01:10:19,115 尽情沐浴在聚光灯下，因为你值得闪耀 1110 99:59:59,999 --> 99:59:59,999 1110 01:10:19,615 --> 01:10:23,553 F-A-B-U-L-O-U-S，每一天都光芒四射 1111 99:59:59,999 --> 99:59:59,999 1111 01:10:23,619 --> 01:10:25,321 想要就来追求，勇敢出击 1112 99:59:59,999 --> 99:59:59,999 1112 01:10:25,388 --> 01:10:29,058 无意冒犯，只为精彩舞动 舞动、舞动，释放精彩 1113 99:59:59,999 --> 99:59:59,999 1113 01:10:29,225 --> 01:10:33,129 我带来崭新的创意风格 将热情点燃，让全场沸腾 1114 99:59:59,999 --> 99:59:59,999 1114 01:10:33,196 --> 01:10:34,831 我们把热情推向极致 1115 99:59:59,999 --> 99:59:59,999 1115 01:10:34,897 --> 01:10:37,066 就是这样，舞动、舞动，释放精彩 1116 99:59:59,999 --> 99:59:59,999 1116 01:10:37,300 --> 01:10:39,869 来吧，尽情展现你的独特魅力，让我大声喊出 1117 99:59:59,999 --> 99:59:59,999 1117 01:10:39,936 --> 01:10:41,904 就是这样，太棒了！ 1118 99:59:59,999 --> 99:59:59,999 1118 01:10:43,139 --> 01:10:45,174 舞动、舞动，释放精彩 1119 99:59:59,999 --> 99:59:59,999 1119 01:10:45,241 --> 01:10:47,276 我带来崭新的创意风格 1120 99:59:59,999 --> 99:59:59,999 1120 01:10:47,343 --> 01:10:49,112 将热情点燃，让全场沸腾 1121 99:59:59,999 --> 99:59:59,999 1121 01:10:49,178 --> 01:10:53,082 我们把热情推向极致 来吧，舞动、舞动，释放精彩 1122 99:59:59,999 --> 99:59:59,999 1122 01:10:53,316 --> 01:10:57,687 来吧，尽情展现你的独特魅力，光芒四射！ 1123 99:59:59,999 --> 99:59:59,999 1123 01:10:57,754 --> 01:10:59,555 每一天都耀眼夺目 1124 99:59:59,999 --> 99:59:59,999 1124 01:10:59,622 --> 01:11:01,290 想要就来追求，勇敢出击 1125 99:59:59,999 --> 99:59:59,999 1125 01:11:01,357 --> 01:11:03,226 无意冒犯，只为精彩——等等！ 1126 99:59:59,999 --> 99:59:59,999 1126 01:11:17,807 --> 01:11:20,510 多年来，人们一直致力于构建 1127 99:59:59,999 --> 99:59:59,999 1127 01:11:20,576 --> 01:11:24,680 一种不仅能给出回应，而且具有理解能力的AI助理 1128 99:59:59,999 --> 99:59:59,999 1128 01:11:25,348 --> 01:11:28,351 一个不仅仅等待用户提问，而是能够提前预判需求的助理 1129 99:59:59,999 --> 99:59:59,999 1129 01:11:28,818 --> 01:11:32,321 一个真正能为用户赋能的通用助理 1130 99:59:59,999 --> 99:59:59,999 1130 01:11:33,022 --> 01:11:36,425 今天，我们正朝着这个方向迈出新步伐 1131 99:59:59,999 --> 99:59:59,999 1131 01:11:36,659 --> 01:11:39,061 而Gemini应用正是我们实现这一目标的工具 1132 99:59:59,999 --> 99:59:59,999 1132 01:11:40,029 --> 01:11:46,936 我们的目标是将Gemini打造成 最个性化、最主动、最强大的AI助理 1133 99:59:59,999 --> 99:59:59,999 1133 01:11:47,336 --> 01:11:49,672 而这一切始于个性化 1134 99:59:59,999 --> 99:59:59,999 1134 01:11:51,040 --> 01:11:55,011 如果你的AI助理真正属于你 完全为你定制，会怎样？ 1135 99:59:59,999 --> 99:59:59,999 1135 01:11:55,411 --> 01:12:00,583 它能学习你的特点、你的偏好 你的项目、你的生活场景 1136 99:59:59,999 --> 99:59:59,999 1136 01:12:01,050 --> 01:12:02,985 而你始终处于主导地位 1137 99:59:59,999 --> 99:59:59,999 1137 01:12:03,586 --> 01:12:05,555 在你授权的情况下 1138 99:59:59,999 --> 99:59:59,999 1138 01:12:05,621 --> 01:12:08,424 你可以与它分享你所有的Google信息 1139 99:59:59,999 --> 99:59:59,999 1139 01:12:08,658 --> 01:12:10,459 使其成为你的延伸 1140 99:59:59,999 --> 99:59:59,999 1140 01:12:11,127 --> 01:12:13,529 我们将这项功能称为“个人使用场景” 1141 99:59:59,999 --> 99:59:59,999 1141 01:12:13,896 --> 01:12:18,000 我们已经实现了让用户与Gemini分享搜索记录 1142 99:59:59,999 --> 99:59:59,999 1142 01:12:18,134 --> 01:12:21,504 这样它就能知道用户曾经搜索过哪些食谱 1143 99:59:59,999 --> 99:59:59,999 1143 01:12:21,604 --> 01:12:23,706 并据此生成这样的回答 1144 99:59:59,999 --> 99:59:59,999 1144 01:12:24,574 --> 01:12:26,342 很快，你还可以添加 1145 99:59:59,999 --> 99:59:59,999 1145 01:12:26,409 --> 01:12:29,579 你在各种Google服务中的更多个人使用场景 1146 99:59:59,999 --> 99:59:59,999 1146 01:12:29,645 --> 01:12:33,015 让Gemini为你提供专属帮助 1147 99:59:59,999 --> 99:59:59,999 1147 01:12:33,749 --> 01:12:35,618 如果你选择启用这个功能 1148 99:59:59,999 --> 99:59:59,999 1148 01:12:35,685 --> 01:12:37,687 将始终掌控自己的体验 1149 99:59:59,999 --> 99:59:59,999 1149 01:12:38,221 --> 01:12:40,223 启用后，你可以随时查看和管理自己的信息 1150 99:59:59,999 --> 99:59:59,999 1150 01:12:40,289 --> 01:12:43,092 关联或解除关联各种Google应用 1151 99:59:59,999 --> 99:59:59,999 1151 01:12:44,060 --> 01:12:49,098 这种个性化将孕育主动型AI助理 1152 99:59:59,999 --> 99:59:59,999 1152 01:12:49,498 --> 01:12:51,133 带来革命性改变 1153 99:59:59,999 --> 99:59:59,999 1153 01:12:51,934 --> 01:12:54,604 要知道，目前大多数AI都是被动反应式的 1154 99:59:59,999 --> 99:59:59,999 1154 01:12:54,904 --> 01:12:56,706 你问一句，它答一句 1155 99:59:59,999 --> 99:59:59,999 1155 01:12:57,139 --> 01:12:59,609 但如果它能预见即将发生的事情 1156 99:59:59,999 --> 99:59:59,999 1156 01:12:59,675 --> 01:13:02,745 在你问它之前就帮你做好准备呢？ 1157 99:59:59,999 --> 99:59:59,999 1157 01:13:03,546 --> 01:13:04,947 假设你是一名学生 1158 99:59:59,999 --> 99:59:59,999 1158 01:13:05,114 --> 01:13:06,983 你的物理考试即将来临 1159 99:59:59,999 --> 99:59:59,999 1159 01:13:07,049 --> 01:13:08,651 Gemini能提前一周在你的日历上圈出这场考试 1160 99:59:59,999 --> 99:59:59,999 1160 01:13:08,718 --> 01:13:11,587 而不是让你手忙脚乱地准备 1161 99:59:59,999 --> 99:59:59,999 1161 01:13:11,988 --> 01:13:13,556 它并不只是提醒你而已 1162 99:59:59,999 --> 99:59:59,999 1162 01:13:13,856 --> 01:13:16,192 它会为你提供个性化测验 1163 99:59:59,999 --> 99:59:59,999 1163 01:13:16,592 --> 01:13:18,227 这些测验整合了你的课堂材料 1164 99:59:59,999 --> 99:59:59,999 1164 01:13:18,294 --> 01:13:21,530 教授的讲义，甚至是你手写笔记的照片 1165 99:59:59,999 --> 99:59:59,999 1165 01:13:22,098 --> 01:13:23,466 这不仅仅是有用 1166 99:59:59,999 --> 99:59:59,999 1166 01:13:23,666 --> 01:13:25,268 这会让人感觉就像魔法一样神奇 1167 99:59:59,999 --> 99:59:59,999 1167 01:13:26,068 --> 01:13:27,904 再更进一步 1168 99:59:59,999 --> 99:59:59,999 1168 01:13:28,170 --> 01:13:30,072 设想Gemini能够基于你的兴趣 1169 99:59:59,999 --> 99:59:59,999 1169 01:13:30,172 --> 01:13:34,143 制作个性化讲解视频 1170 99:59:59,999 --> 99:59:59,999 1170 01:13:34,410 --> 01:13:36,579 帮助你更好地理解学习主题 1171 99:59:59,999 --> 99:59:59,999 1171 01:13:37,513 --> 01:13:40,383 例如，Gemini了解到我喜欢骑行 1172 99:59:59,999 --> 99:59:59,999 1172 01:13:40,549 --> 01:13:45,554 它就会用自行车相关的类比来为我讲解热力学原理 1173 99:59:59,999 --> 99:59:59,999 1173 01:13:46,455 --> 01:13:48,357 这就是Gemini的发展方向 1174 99:59:59,999 --> 99:59:59,999 1174 01:13:48,557 --> 01:13:50,393 而这一切之所以成为可能 1175 99:59:59,999 --> 99:59:59,999 1175 01:13:50,626 --> 01:13:53,763 都要归功于底层模型的强大能力 1176 99:59:59,999 --> 99:59:59,999 1176 01:13:54,730 --> 01:13:58,534 上个月，我们发布了最新的Gemini 2.5 Pro模型 1177 99:59:59,999 --> 99:59:59,999 1177 01:13:58,668 --> 01:14:01,370 这不只是一次升级，而是一次飞跃 1178 99:59:59,999 --> 99:59:59,999 1178 01:14:01,871 --> 01:14:04,573 人们正在用它创造出非凡的成果 1179 99:59:59,999 --> 99:59:59,999 1179 01:14:05,007 --> 01:14:06,409 他们不仅仅是进行氛围编程 1180 99:59:59,999 --> 99:59:59,999 1180 01:14:06,475 --> 01:14:10,846 而是能在几分钟内创造出 完整的游戏和功能齐全的网站 1181 99:59:59,999 --> 99:59:59,999 1181 01:14:11,414 --> 01:14:13,716 他们只需要在脑海中构思，就能让创意变为现实 1182 99:59:59,999 --> 99:59:59,999 1182 01:14:14,250 --> 01:14:17,086 而且这很有趣，无论是第一次编写软件 1183 99:59:59,999 --> 99:59:59,999 1183 01:14:17,153 --> 01:14:21,324 还是使用Veo模型创建视频，或者生成和编辑图像 1184 99:59:59,999 --> 99:59:59,999 1184 01:14:21,490 --> 01:14:23,459 这些技能现在成了每个人都能掌握的“超能力” 1185 99:59:59,999 --> 99:59:59,999 1185 01:14:23,659 --> 01:14:29,332 因此，个性化、主动性、强大功能 1186 99:59:59,999 --> 99:59:59,999 1186 01:14:29,665 --> 01:14:31,968 这三个方面是AI辅助的前沿领域 1187 99:59:59,999 --> 99:59:59,999 1187 01:14:32,034 --> 01:14:33,836 而借助Gemini 2.5 Pro 1188 99:59:59,999 --> 99:59:59,999 1188 01:14:33,903 --> 01:14:36,339 我们在这三个方面都取得了重大进展 1189 99:59:59,999 --> 99:59:59,999 1189 01:14:37,540 --> 01:14:39,308 接下来，我们来详细了解一下 1190 99:59:59,999 --> 99:59:59,999 1190 01:14:39,375 --> 01:14:42,244 这些功能如何在Gemini应用中实现 1191 99:59:59,999 --> 99:59:59,999 1191 01:14:42,545 --> 01:14:44,246 我们今天将推出五项新功能 1192 99:59:59,999 --> 99:59:59,999 1192 01:14:44,814 --> 01:14:47,083 首先，我们来看看Gemini Live 1193 99:59:59,999 --> 99:59:59,999 1193 01:14:47,817 --> 01:14:52,421 许多用户对其自然流畅的交互和对话方式赞不绝口 1194 99:59:59,999 --> 99:59:59,999 1194 01:14:52,555 --> 01:14:56,692 它支持150多个国家/地区的45种以上的语言 1195 99:59:59,999 --> 99:59:59,999 1195 01:14:56,959 --> 01:14:59,395 它是如此直观、如此富有互动性 1196 99:59:59,999 --> 99:59:59,999 1196 01:14:59,562 --> 01:15:00,997 以至于用户与它的对话时长 1197 99:59:59,999 --> 99:59:59,999 1197 01:15:01,063 --> 01:15:04,600 是Gemini应用中文本对话的五倍 1198 99:59:59,999 --> 99:59:59,999 1198 01:15:05,001 --> 01:15:07,036 从我的个人经验来说 1199 99:59:59,999 --> 99:59:59,999 1199 01:15:07,103 --> 01:15:11,040 它非常适合在早上开车上班的路上 交流讨论、梳理思路 1200 99:59:59,999 --> 99:59:59,999 1200 01:15:11,907 --> 01:15:13,209 正如Sundar提到的 1201 99:59:59,999 --> 99:59:59,999 1201 01:15:13,275 --> 01:15:14,643 Gemini Live现在增加了 1202 99:59:59,999 --> 99:59:59,999 1202 01:15:14,710 --> 01:15:18,080 出色的相机和屏幕共享功能 1203 99:59:59,999 --> 99:59:59,999 1203 01:15:18,414 --> 01:15:19,548 而从今天起 1204 99:59:59,999 --> 99:59:59,999 1204 01:15:19,615 --> 01:15:21,584 你可以在Gemini移动应用中免费使用这些功能 1205 99:59:59,999 --> 99:59:59,999 1205 01:15:21,650 --> 01:15:23,319 并同时支持Android和iOS设备 1206 99:59:59,999 --> 99:59:59,999 1206 01:15:29,625 --> 01:15:30,726 在接下来的几周内 1207 99:59:59,999 --> 99:59:59,999 1207 01:15:30,793 --> 01:15:34,096 你将能够把Gemini Live连接到你喜爱的部分应用中 1208 99:59:59,999 --> 99:59:59,999 1208 01:15:34,230 --> 01:15:36,932 如日历、地图、Keep和Tasks 1209 99:59:59,999 --> 99:59:59,999 1209 01:15:37,133 --> 01:15:39,568 这样一来，你只需将摄像头对准邀请内容 1210 99:59:59,999 --> 99:59:59,999 1210 01:15:39,635 --> 01:15:43,339 就能让Gemini立即将其添加到日历中 1211 99:59:59,999 --> 99:59:59,999 1211 01:15:43,906 --> 01:15:48,010 或者，在面对室友手写的潦草购物清单时 1212 99:59:59,999 --> 99:59:59,999 1212 01:15:48,244 --> 01:15:52,448 Gemini Live能够将那些潦草的笔记 转换成Google Keep中的整洁列表 1213 99:59:59,999 --> 99:59:59,999 1213 01:15:53,115 --> 01:15:57,219 我们为Gemini Live规划了众多令人期待的新功能 1214 99:59:59,999 --> 99:59:59,999 1214 01:15:57,486 --> 01:16:01,424 这些功能正在刚才展示的Project Astra中进行原型设计 1215 99:59:59,999 --> 99:59:59,999 1215 01:16:01,657 --> 01:16:03,325 当这些功能成熟后 1216 99:59:59,999 --> 99:59:59,999 1216 01:16:03,392 --> 01:16:06,195 我们会在Gemini Live中推出，让所有人都能使用 1217 99:59:59,999 --> 99:59:59,999 1217 01:16:07,329 --> 01:16:10,399 由于Gemini和Android合作紧密 1218 99:59:59,999 --> 99:59:59,999 1218 01:16:10,466 --> 01:16:12,168 这些创新功能中的大部分 1219 99:59:59,999 --> 99:59:59,999 1219 01:16:12,234 --> 01:16:15,838 会在整个生态系统中的Android设备上大放异彩 1220 99:59:59,999 --> 99:59:59,999 1220 01:16:15,971 --> 01:16:17,139 所以，敬请期待未来的更多更新！ 1221 99:59:59,999 --> 99:59:59,999 1221 01:16:18,340 --> 01:16:23,179 实时交互确实很棒，但有时候，你需要深入挖掘 1222 99:59:59,999 --> 99:59:59,999 1222 01:16:23,446 --> 01:16:25,381 梳理一些复杂的内容 1223 99:59:59,999 --> 99:59:59,999 1223 01:16:25,714 --> 01:16:27,817 这就是Deep Research的用武之地 1224 99:59:59,999 --> 99:59:59,999 1224 01:16:28,084 --> 01:16:30,853 从今天起，你可以在使用Deep Research时 1225 99:59:59,999 --> 99:59:59,999 1225 01:16:30,920 --> 01:16:34,156 上传自己的文件来指导研究智能体 1226 99:59:59,999 --> 99:59:59,999 1226 01:16:34,423 --> 01:16:36,926 这也是用户呼声最高的一项功能 1227 99:59:59,999 --> 99:59:59,999 1227 01:16:37,426 --> 01:16:41,497 很快，你还能够将Deep Research 与Google云端硬盘和Gmail连接 1228 99:59:59,999 --> 99:59:59,999 1228 01:16:41,730 --> 01:16:44,433 以便轻松地从中导入信息 1229 99:59:59,999 --> 99:59:59,999 1229 01:16:45,234 --> 01:16:48,304 假如你获得了一份精彩、翔实的研究报告 1230 99:59:59,999 --> 99:59:59,999 1230 01:16:48,471 --> 01:16:52,174 比如这份关于彗星在太空中运行轨迹的科学分析 1231 99:59:59,999 --> 99:59:59,999 1231 01:16:53,109 --> 01:16:57,446 但要如何才能让它变得易于理解又能吸引读者呢？ 1232 99:59:59,999 --> 99:59:59,999 1232 01:16:57,513 --> 01:16:59,748 以及如何分享这些精彩内容呢？ 1233 99:59:59,999 --> 99:59:59,999 1233 01:17:00,516 --> 01:17:02,251 这就是Canvas大显身手的时刻 1234 99:59:59,999 --> 99:59:59,999 1234 01:17:02,384 --> 01:17:05,521 Canvas是Gemini提供的交互式协同创作空间 1235 99:59:59,999 --> 99:59:59,999 1235 01:17:06,055 --> 01:17:07,490 现在借助Canvas 1236 99:59:59,999 --> 99:59:59,999 1236 01:17:07,556 --> 01:17:12,027 只需轻轻一点就能转换报告格式 1237 99:59:59,999 --> 99:59:59,999 1237 01:17:12,328 --> 01:17:16,132 你可以将它转换成动态网页、信息图、实用测验 1238 99:59:59,999 --> 99:59:59,999 1238 01:17:16,398 --> 01:17:19,034 甚至是支持45种语言的自定义播客 1239 99:59:59,999 --> 99:59:59,999 1239 01:17:19,635 --> 01:17:22,404 要想更进一步，你可以在Canvas中通过氛围编程 1240 99:59:59,999 --> 99:59:59,999 1240 01:17:22,471 --> 01:17:27,109 创建各种精彩内容，让它按照你的想法不断完善作品 1241 99:59:59,999 --> 99:59:59,999 1241 01:17:27,643 --> 01:17:30,179 直到创造出你脑海中的构想 1242 99:59:59,999 --> 99:59:59,999 1242 01:17:30,713 --> 01:17:33,349 例如，我们的一位Google同事 1243 99:59:59,999 --> 99:59:59,999 1243 01:17:33,415 --> 01:17:37,119 仅仅是描述了自己的构想，然后与Gemini展开协作 1244 99:59:59,999 --> 99:59:59,999 1244 01:17:37,286 --> 01:17:40,256 便创造出了这个精确的交互式彗星运行模拟 1245 99:59:59,999 --> 99:59:59,999 1245 01:17:41,056 --> 01:17:42,958 现在，你也可以分享这样的应用 1246 99:59:59,999 --> 99:59:59,999 1246 01:17:43,025 --> 01:17:46,929 让其他人也能一起查看、修改甚至二次创作 1247 99:59:59,999 --> 99:59:59,999 1247 01:17:47,696 --> 01:17:50,599 这种能够转化任何内容的能力 1248 99:59:59,999 --> 99:59:59,999 1248 01:17:50,866 --> 01:17:53,402 开创了使用Gemini的全新方式 1249 99:59:59,999 --> 99:59:59,999 1249 01:17:54,870 --> 01:17:57,673 今天，我们还将推出Gemini的另一种全新使用方式 1250 99:59:59,999 --> 99:59:59,999 1250 01:17:58,207 --> 01:18:00,509 Chrome中的Gemini 1251 99:59:59,999 --> 99:59:59,999 1251 01:18:01,944 --> 01:18:07,216 它将成为你在桌面端浏览网页时的AI助理 1252 99:59:59,999 --> 99:59:59,999 1252 01:18:07,750 --> 01:18:10,085 最棒的是 1253 99:59:59,999 --> 99:59:59,999 1253 01:18:10,152 --> 01:18:14,523 它能直接理解你当前浏览的页面内容 1254 99:59:59,999 --> 99:59:59,999 1254 01:18:14,924 --> 01:18:17,459 只要你想到任何问题，都能获得针对性的解答 1255 99:59:59,999 --> 99:59:59,999 1255 01:18:17,993 --> 01:18:21,330 我个人最喜欢用它来对比一长串的评价内容 1256 99:59:59,999 --> 99:59:59,999 1256 01:18:21,397 --> 01:18:23,032 比如这个露营网站 1257 99:59:59,999 --> 99:59:59,999 1257 01:18:23,599 --> 01:18:24,900 从本周开始 1258 99:59:59,999 --> 99:59:59,999 1258 01:18:24,967 --> 01:18:27,136 我们将向美国地区的Gemini订阅者 1259 99:59:59,999 --> 99:59:59,999 1259 01:18:27,203 --> 01:18:29,004 推出Chrome中的Gemini 1260 99:59:59,999 --> 99:59:59,999 1260 01:18:36,478 --> 01:18:39,415 至此，我们已经体验了测验、Gemini Live对话 1261 99:59:59,999 --> 99:59:59,999 1261 01:18:39,481 --> 01:18:41,584 学习了彗星和露营地的知识 1262 99:59:59,999 --> 99:59:59,999 1262 01:18:41,750 --> 01:18:43,419 现在，我们来创造一些内容 1263 99:59:59,999 --> 99:59:59,999 1263 01:18:43,919 --> 01:18:45,321 从今天起 1264 99:59:59,999 --> 99:59:59,999 1264 01:18:45,487 --> 01:18:49,558 Google旗下最先进、功能最强大的图像生成模型 1265 99:59:59,999 --> 99:59:59,999 1265 01:18:49,892 --> 01:18:51,393 正式在Gemini应用上推出 1266 99:59:59,999 --> 99:59:59,999 1266 01:18:51,961 --> 01:18:55,664 它叫做Imagen 4，这是又一次重大飞跃 1267 99:59:59,999 --> 99:59:59,999 1267 01:19:00,302 --> 01:19:06,075 生成的图像更加饱满，色彩更加细腻，细节更为精细 1268 99:59:59,999 --> 99:59:59,999 1268 01:19:06,442 --> 01:19:08,444 例如各种场景中的阴影层次 1269 99:59:59,999 --> 99:59:59,999 1269 01:19:08,510 --> 01:19:11,113 以及照片中栩栩如生的水滴效果 1270 99:59:59,999 --> 99:59:59,999 1270 01:19:11,480 --> 01:19:14,016 经过长期的研究 1271 99:59:59,999 --> 99:59:59,999 1271 01:19:14,083 --> 01:19:16,318 我可以肯定地说，这些模型生成的图像质量 1272 99:59:59,999 --> 99:59:59,999 1272 01:19:16,552 --> 01:19:21,090 已经实现了从“不错”到“出色”再到“惊艳”的飞跃 1273 99:59:59,999 --> 99:59:59,999 1273 01:19:21,757 --> 01:19:25,894 Imagen 4在文字和排版方面也有很大提升 1274 99:59:59,999 --> 99:59:59,999 1274 01:19:26,328 --> 01:19:29,431 之前的模型虽然能生成不错的图像 1275 99:59:59,999 --> 99:59:59,999 1275 01:19:29,632 --> 01:19:33,035 但在图像中加入文字时总会出现瑕疵 1276 99:59:59,999 --> 99:59:59,999 1276 01:19:33,469 --> 01:19:37,439 举个例子，假设我想创建一张音乐节海报 1277 99:59:59,999 --> 99:59:59,999 1277 01:19:37,673 --> 01:19:39,908 将那只Chrome恐龙作为主角 1278 99:59:59,999 --> 99:59:59,999 1278 01:19:40,242 --> 01:19:43,879 Imagen 4不仅能够准确呈现文本内容和正确拼写 1279 99:59:59,999 --> 99:59:59,999 1279 01:19:43,946 --> 01:19:46,548 它还能做出富有创意的设计决策 1280 99:59:59,999 --> 99:59:59,999 1280 01:19:46,715 --> 01:19:49,318 比如在字体中融入恐龙骨骼元素 1281 99:59:59,999 --> 99:59:59,999 1281 01:19:49,385 --> 01:19:53,222 调整文字间距、字号大小和整体布局 1282 99:59:59,999 --> 99:59:59,999 1282 01:19:53,289 --> 01:19:54,723 创造出这张很棒的海报 1283 99:59:59,999 --> 99:59:59,999 1283 01:19:55,457 --> 01:19:59,995 总之，图像质量更高、生成速度更快、文字效果更好 1284 99:59:59,999 --> 99:59:59,999 1284 01:20:00,429 --> 01:20:04,767 这样一来，你就可以制作海报、派对邀请函等内容 1285 99:59:59,999 --> 99:59:59,999 1285 01:20:05,567 --> 01:20:08,037 借助原生图像生成功能 1286 99:59:59,999 --> 99:59:59,999 1286 01:20:08,103 --> 01:20:11,006 你还可以直接在应用中轻松编辑这些图像 1287 99:59:59,999 --> 99:59:59,999 1287 01:20:11,974 --> 01:20:15,911 我们还开发了一个快速版本的Imagen 4 1288 99:59:59,999 --> 99:59:59,999 1288 01:20:16,345 --> 01:20:18,314 我们非常期待你能亲自体验它 1289 99:59:59,999 --> 99:59:59,999 1289 01:20:18,380 --> 01:20:22,518 它的速度是我们先前模型的十倍 1290 99:59:59,999 --> 99:59:59,999 1290 01:20:22,584 --> 01:20:25,187 让你能够快速尝试和完善各种创意 1291 99:59:59,999 --> 99:59:59,999 1291 01:20:26,288 --> 01:20:28,357 最后，我想向大家展示一项新内容 1292 99:59:59,999 --> 99:59:59,999 1292 01:20:28,857 --> 01:20:30,225 这些生成的图像的确很棒 1293 99:59:59,999 --> 99:59:59,999 1293 01:20:30,559 --> 01:20:34,997 但要讲述一个完整的故事有时需要动作和声音的配合 1294 99:59:59,999 --> 99:59:59,999 1294 01:20:35,698 --> 01:20:38,367 去年12月，我们发布的Veo 2模型 1295 99:59:59,999 --> 99:59:59,999 1295 01:20:38,434 --> 01:20:41,904 重新定义了整个行业的视频生成标准 1296 99:59:59,999 --> 99:59:59,999 1296 01:20:42,304 --> 01:20:45,741 如果你注意到了Demis昨天发布的煎炒洋葱视频 1297 99:59:59,999 --> 99:59:59,999 1297 01:20:45,941 --> 01:20:47,843 就能猜到我们还有更多新鲜出炉的惊喜 1298 99:59:59,999 --> 99:59:59,999 1298 01:20:48,711 --> 01:20:53,215 今天，我很高兴地宣布推出我们最先进的模型 1299 99:59:59,999 --> 99:59:59,999 1299 01:20:53,615 --> 01:20:54,750 Veo 3 1300 99:59:59,999 --> 99:59:59,999 1300 01:21:01,890 --> 01:21:05,127 与今天发布的其他内容一样 1301 99:59:59,999 --> 99:59:59,999 1301 01:21:05,194 --> 01:21:06,995 它从今天起就已开放使用 1302 99:59:59,999 --> 99:59:59,999 1302 01:21:11,934 --> 01:21:14,370 视觉质量更加出色 1303 99:59:59,999 --> 99:59:59,999 1303 01:21:14,436 --> 01:21:17,639 对物理规律的理解更加强大 1304 99:59:59,999 --> 99:59:59,999 1304 01:21:17,840 --> 01:21:23,512 但这次质的飞跃在于，Veo 3实现了原生音频生成 1305 99:59:59,999 --> 99:59:59,999 1305 01:21:24,146 --> 01:21:30,552 这意味着Veo 3能够生成音效 1306 99:59:59,999 --> 99:59:59,999 1306 01:21:30,953 --> 01:21:33,756 背景音，甚至对话 1307 99:59:59,999 --> 99:59:59,999 1307 01:21:34,590 --> 01:21:37,726 只需输入提示，你的角色就会开口说话 1308 99:59:59,999 --> 99:59:59,999 1308 01:21:38,594 --> 01:21:42,698 这是森林中一只睿智的老猫头鹰与一只紧张的小獾 1309 99:59:59,999 --> 99:59:59,999 1309 01:21:43,098 --> 01:21:43,966 敬请聆听 1310 99:59:59,999 --> 99:59:59,999 1310 01:21:53,942 --> 01:21:56,445 他们今天留下了一个球 1311 99:59:59,999 --> 99:59:59,999 1311 01:21:56,512 --> 01:21:58,580 它能弹得比我跳得还高！ 1312 99:59:59,999 --> 99:59:59,999 1312 01:21:59,581 --> 01:22:01,683 这是什么神奇的魔法？ 1313 99:59:59,999 --> 99:59:59,999 1313 01:22:15,297 --> 01:22:18,967 很酷吧？Veo不仅添加了森林的声音 1314 99:59:59,999 --> 99:59:59,999 1314 01:22:19,034 --> 01:22:20,369 还添加了对话 1315 99:59:59,999 --> 99:59:59,999 1315 01:22:20,869 --> 01:22:26,708 内容创作正在进入结合音频和视频生成的新时代 1316 99:59:59,999 --> 99:59:59,999 1316 01:22:26,775 --> 01:22:28,610 无论是画面还是声音都栩栩如生 1317 99:59:59,999 --> 99:59:59,999 1317 01:22:28,977 --> 01:22:31,213 让人感觉身临其境 1318 99:59:59,999 --> 99:59:59,999 1318 01:22:31,280 --> 01:22:35,384 在这个视频中，就像真的坐在这艘船上 1319 99:59:59,999 --> 99:59:59,999 1319 01:22:37,219 --> 01:22:38,954 这片海洋，是一股无与伦比的力量 1320 99:59:59,999 --> 99:59:59,999 1320 01:22:39,021 --> 01:22:41,190 一种未被驯服的野性力量 1321 99:59:59,999 --> 99:59:59,999 1321 01:22:41,890 --> 01:22:43,292 她指挥着你的每一次划桨 1322 99:59:59,999 --> 99:59:59,999 1322 01:22:43,592 --> 01:22:45,127 伴随着每一道破浪的闪光 1323 99:59:59,999 --> 99:59:59,999 1323 01:22:52,734 --> 01:22:56,271 如此逼真的画面、丰富的情感表达 1324 99:59:59,999 --> 99:59:59,999 1324 01:22:56,338 --> 01:23:00,008 匹配的口型，以及背景中的海洋 1325 99:59:59,999 --> 99:59:59,999 1325 01:23:00,342 --> 01:23:02,811 作为一款强大的创意工具 1326 99:59:59,999 --> 99:59:59,999 1326 01:23:02,878 --> 01:23:05,981 Veo的发展速度令人叹为观止 1327 99:59:59,999 --> 99:59:59,999 1327 01:23:06,582 --> 01:23:09,485 我们一直在与电影行业密切合作 1328 99:59:59,999 --> 99:59:59,999 1328 01:23:09,551 --> 01:23:12,554 共同探索Veo的可能性，直至做好发布的准备 1329 99:59:59,999 --> 99:59:59,999 1329 01:23:12,621 --> 01:23:14,890 稍后也将为大家分享更多具体内容 1330 99:59:59,999 --> 99:59:59,999 1330 01:23:15,757 --> 01:23:17,893 以上就是Gemini的更新内容 1331 99:59:59,999 --> 99:59:59,999 1331 01:23:18,527 --> 01:23:21,964 从今天起，Gemini Live的各项功能 1332 99:59:59,999 --> 99:59:59,999 1332 01:23:22,097 --> 01:23:24,533 在Android和iOS上完全免费提供 1333 99:59:59,999 --> 99:59:59,999 1333 01:23:25,100 --> 01:23:28,737 Deep Research和Canvas迎来了迄今为止最大的更新 1334 99:59:59,999 --> 99:59:59,999 1334 01:23:29,404 --> 01:23:30,706 正如之前介绍的 1335 99:59:59,999 --> 99:59:59,999 1335 01:23:30,772 --> 01:23:33,675 支持多步骤操作的Agent Mode也即将推出 1336 99:59:59,999 --> 99:59:59,999 1336 01:23:34,176 --> 01:23:37,646 我们推出了Chrome中的Gemini，帮助你浏览网页 1337 99:59:59,999 --> 99:59:59,999 1337 01:23:37,946 --> 01:23:40,182 你可以使用全新的Imagen 4和Veo 3模型 1338 99:59:59,999 --> 99:59:59,999 1338 01:23:40,382 --> 01:23:44,887 创建令人惊叹的图像和音画结合的视频 1339 99:59:59,999 --> 99:59:59,999 1339 01:23:45,420 --> 01:23:48,023 所有这些功能都将在Gemini应用中整合 1340 99:59:59,999 --> 99:59:59,999 1340 01:23:48,090 --> 01:23:54,429 助力我们打造最个性化、最主动、最强大的AI助理 1341 99:59:59,999 --> 99:59:59,999 1341 01:23:55,063 --> 01:23:57,132 接下来，Jason将为大家详细介绍 1342 99:59:59,999 --> 99:59:59,999 1342 01:23:57,199 --> 01:24:00,602 这些生成模型如何为创意表达开启全新可能 1343 99:59:59,999 --> 99:59:59,999 1343 01:24:00,702 --> 01:24:01,703 有请Jason 1344 99:59:59,999 --> 99:59:59,999 1344 01:24:12,247 --> 01:24:13,148 谢谢，Josh 1345 99:59:59,999 --> 99:59:59,999 1345 01:24:14,349 --> 01:24:18,153 无论你是创作者、音乐人还是电影制作人 1346 99:59:59,999 --> 99:59:59,999 1346 01:24:18,587 --> 01:24:21,657 生成式媒体技术正在拓展创意表达的可能性 1347 99:59:59,999 --> 99:59:59,999 1347 01:24:22,424 --> 01:24:25,727 通过与艺术家和创作者一直以来的密切合作 1348 99:59:59,999 --> 99:59:59,999 1348 01:24:25,861 --> 01:24:30,032 我们开发了能够赋能其创作过程的技术 1349 99:59:59,999 --> 99:59:59,999 1349 01:24:30,699 --> 01:24:35,170 例如，我们与音乐人合作开发了Music AI Sandbox 1350 99:59:59,999 --> 99:59:59,999 1350 01:24:35,437 --> 01:24:37,172 这是一款专为音乐专业人士设计的工具 1351 99:59:59,999 --> 99:59:59,999 1351 01:24:37,239 --> 01:24:38,774 让他们能够在创作过程中 1352 99:59:59,999 --> 99:59:59,999 1352 01:24:38,840 --> 01:24:41,643 探索我们的音乐生成模型Lyria所带来的可能性 1353 99:59:59,999 --> 99:59:59,999 1353 01:24:42,444 --> 01:24:46,014 例如这段视频记录了音乐传奇人物、格莱美获奖歌手 1354 99:59:59,999 --> 99:59:59,999 1354 01:24:46,081 --> 01:24:47,816 作曲家Shankar Mahadevan 1355 99:59:59,999 --> 99:59:59,999 1355 01:24:47,883 --> 01:24:50,919 使用Music AI Sandbox和Lyria的场景 1356 99:59:59,999 --> 99:59:59,999 1356 01:24:52,487 --> 01:24:56,858 我是Shankar Mahadevan 一名音乐作曲家、歌手、制作人 1357 99:59:59,999 --> 99:59:59,999 1357 01:24:56,925 --> 01:24:58,393 大家都说我笑口常开 1358 99:59:59,999 --> 99:59:59,999 1358 01:25:01,163 --> 01:25:02,731 使用Sandbox的体验非常棒 1359 99:59:59,999 --> 99:59:59,999 1359 01:25:03,198 --> 01:25:07,169 我们输入需求后，它生成了一段背景音乐 1360 99:59:59,999 --> 99:59:59,999 1360 01:25:07,636 --> 01:25:10,305 我们在此基础上创作出了完整的歌曲 1361 99:59:59,999 --> 99:59:59,999 1361 01:25:14,610 --> 01:25:17,245 对一名音乐人而言，这是一个极具启发性的工具 1362 99:59:59,999 --> 99:59:59,999 1362 01:25:17,779 --> 01:25:20,382 就像你打开一扇门，发现了另一个空间 1363 99:59:59,999 --> 99:59:59,999 1363 01:25:20,449 --> 01:25:23,085 然后再开一扇门，发现更多可能性 1364 99:59:59,999 --> 99:59:59,999 1364 01:25:23,151 --> 01:25:25,020 AI就是这样，不断开启新视野 1365 99:59:59,999 --> 99:59:59,999 1365 01:25:51,279 --> 01:25:52,114 太棒了 1366 99:59:59,999 --> 99:59:59,999 1366 01:25:56,385 --> 01:25:58,253 我们最近推出的Lyria 2模型 1367 99:59:59,999 --> 99:59:59,999 1367 01:25:58,320 --> 01:26:01,690 可以生成高保真音乐和专业级音频 1368 99:59:59,999 --> 99:59:59,999 1368 01:26:02,257 --> 01:26:03,759 它生成的音乐旋律优美动听 1369 99:59:59,999 --> 99:59:59,999 1369 01:26:03,825 --> 01:26:06,395 还能编入独唱和合唱人声部分 1370 99:59:59,999 --> 99:59:59,999 1370 01:26:07,162 --> 01:26:09,965 它能够创作出富有表现力且层次丰富的内容 例如接下来的这段音乐 1371 99:59:59,999 --> 99:59:59,999 1371 01:26:27,449 --> 01:26:29,718 Lyria 2从今天起正式向企业用户 1372 99:59:59,999 --> 99:59:59,999 1372 01:26:29,785 --> 01:26:32,120 和YouTube创作者和专业音乐人开放 1373 99:59:59,999 --> 99:59:59,999 1373 01:26:33,355 --> 01:26:37,426 这种协作理念不仅体现在我们构建的产品中 也体现在我们的构建方式上 1374 99:59:59,999 --> 99:59:59,999 1374 01:26:38,260 --> 01:26:40,228 目前，无论是个人还是组织 1375 99:59:59,999 --> 99:59:59,999 1375 01:26:40,295 --> 01:26:43,465 都难以有效识别AI生成的图像 1376 99:59:59,999 --> 99:59:59,999 1376 01:26:44,332 --> 01:26:49,771 随着技术的进步和生成的音乐 音频、图像、视频日益逼真 1377 99:59:59,999 --> 99:59:59,999 1377 01:26:49,838 --> 01:26:51,573 这一情况只会更加泛滥 1378 99:59:59,999 --> 99:59:59,999 1378 01:26:52,374 --> 01:26:55,110 因此，我们在这一领域持续创新 1379 99:59:59,999 --> 99:59:59,999 1379 01:26:56,278 --> 01:26:58,547 两年前，我们开创了SynthID技术 1380 99:59:59,999 --> 99:59:59,999 1380 01:26:58,714 --> 01:27:01,917 它能在生成的媒体内容中嵌入不可见的水印 1381 99:59:59,999 --> 99:59:59,999 1381 01:27:02,484 --> 01:27:07,022 到目前为止，已有超过100亿生成的内容带有水印 1382 99:59:59,999 --> 99:59:59,999 1382 01:27:07,956 --> 01:27:09,925 另外，我们也在进一步扩大合作伙伴关系 1383 99:59:59,999 --> 99:59:59,999 1383 01:27:09,991 --> 01:27:13,028 确保更多内容带有SynthID水印 1384 99:59:59,999 --> 99:59:59,999 1384 01:27:13,095 --> 01:27:15,597 同时让更多组织能够检测到水印 1385 99:59:59,999 --> 99:59:59,999 1385 01:27:16,398 --> 01:27:20,068 最后，我们还简化了水印的检测机制 1386 99:59:59,999 --> 99:59:59,999 1386 01:27:20,435 --> 01:27:22,771 我们新推出的SynthID检测器能够识别 1387 99:59:59,999 --> 99:59:59,999 1387 01:27:22,838 --> 01:27:27,175 图像、音频、文本或视频是否含有SynthID水印 1388 99:59:59,999 --> 99:59:59,999 1388 01:27:27,242 --> 01:27:29,878 即使只是部分内容被标记也能识别 1389 99:59:59,999 --> 99:59:59,999 1389 01:27:31,012 --> 01:27:33,715 这一功能从今天起开始向早期测试员逐步开放 1390 99:59:59,999 --> 99:59:59,999 1390 01:27:40,288 --> 01:27:43,058 凭借广泛的合作，我们还开始探索Veo 1391 99:59:59,999 --> 99:59:59,999 1391 01:27:43,125 --> 01:27:44,426 在电影制作领域的应用 1392 99:59:59,999 --> 99:59:59,999 1392 01:27:45,160 --> 01:27:48,897 最近，我们与极富远见的导演Darren Aronofsky 1393 99:59:59,999 --> 99:59:59,999 1393 01:27:48,964 --> 01:27:51,666 及其创新故事创作公司Primordial Soup达成了合作 1394 99:59:59,999 --> 99:59:59,999 1394 01:27:51,933 --> 01:27:54,870 共同探索如何塑造Veo的能力以满足创作需求 1395 99:59:59,999 --> 99:59:59,999 1395 01:27:54,936 --> 01:27:57,773 让艺术家引领技术创新方向 1396 99:59:59,999 --> 99:59:59,999 1396 01:27:58,573 --> 01:28:01,943 通过这一合作，我们将世界顶尖的视频生成模型 1397 99:59:59,999 --> 99:59:59,999 1397 01:28:02,010 --> 01:28:05,480 交付给最杰出的电影制作人，推动技术突破 1398 99:59:59,999 --> 99:59:59,999 1398 01:28:05,547 --> 01:28:08,850 打造更具创意性和情感表现力的创作工具 1399 99:59:59,999 --> 99:59:59,999 1399 01:28:09,718 --> 01:28:12,120 在计划制作的三部短片中 1400 99:59:59,999 --> 99:59:59,999 1400 01:28:12,187 --> 01:28:15,190 第一部作品是导演Eliza McNitt的《ANCESTRA》 1401 99:59:59,999 --> 99:59:59,999 1401 01:28:15,624 --> 01:28:16,591 现在就让我们来一睹为快 1402 99:59:59,999 --> 99:59:59,999 1402 01:28:19,227 --> 01:28:22,230 电影本质上是一种深刻的人类行为 1403 99:59:59,999 --> 99:59:59,999 1403 01:28:22,297 --> 01:28:24,399 它能将人们与彼此的故事联系起来 1404 99:59:59,999 --> 99:59:59,999 1404 01:28:25,033 --> 01:28:29,104 它能够将我们从自己的经历中抽离出来 1405 99:59:59,999 --> 99:59:59,999 1405 01:28:29,171 --> 01:28:31,072 带我们踏上另一段旅程 1406 99:59:59,999 --> 99:59:59,999 1406 01:28:31,807 --> 01:28:33,642 我认为这一点永远不会改变 1407 99:59:59,999 --> 99:59:59,999 1407 01:28:34,242 --> 01:28:38,513 这个故事虽然跨越整个宇宙的历史长河 1408 99:59:59,999 --> 99:59:59,999 1408 01:28:39,147 --> 01:28:41,249 但本质上讲述的是一位母亲的经历 1409 99:59:59,999 --> 99:59:59,999 1409 01:28:41,316 --> 01:28:44,820 以及当她那有心脏缺陷的孩子刚出生时所发生的一切 1410 99:59:59,999 --> 99:59:59,999 1410 01:28:45,587 --> 01:28:48,323 我们拍摄了演员真实的情感表演 1411 99:59:59,999 --> 99:59:59,999 1411 01:28:48,390 --> 01:28:53,061 但同时也通过AI生成了 那些用传统方式根本无法拍摄的画面 1412 99:59:59,999 --> 99:59:59,999 1412 01:28:54,396 --> 01:28:57,465 我想要婴儿紧握着母亲的手指那样的画面 1413 99:59:59,999 --> 99:59:59,999 1413 01:28:57,966 --> 01:28:59,401 展现他们之间那种幸福的时刻 1414 99:59:59,999 --> 99:59:59,999 1414 01:28:59,467 --> 01:29:00,302 是的 1415 99:59:59,999 --> 99:59:59,999 1415 01:29:01,603 --> 01:29:03,972 Veo是一个生成视频模型 1416 99:59:59,999 --> 99:59:59,999 1416 01:29:04,039 --> 01:29:06,041 但在我看来，它更像是一个全新的镜头 1417 99:59:59,999 --> 99:59:59,999 1417 01:29:06,107 --> 01:29:08,443 让我能够以不同方式想象身边的一切 1418 99:59:59,999 --> 99:59:59,999 1418 01:29:09,511 --> 01:29:11,379 探索这些模型的优势和局限 1419 99:59:59,999 --> 99:59:59,999 1419 01:29:11,446 --> 01:29:17,819 并尝试将它们打造成创作工具，这个过程非常有趣 1420 99:59:59,999 --> 99:59:59,999 1420 01:29:19,654 --> 01:29:23,592 老实说，我从未想过要讲述我出生那天的故事 1421 99:59:59,999 --> 99:59:59,999 1421 01:29:23,859 --> 01:29:25,260 但现在我们拍出来了 1422 99:59:59,999 --> 99:59:59,999 1422 01:29:31,800 --> 01:29:33,068 这原本只是一次普通的产检而已 1423 99:59:59,999 --> 99:59:59,999 1423 01:29:33,134 --> 01:29:34,436 紧急剖腹产！ 1424 99:59:59,999 --> 99:59:59,999 1424 01:29:35,270 --> 01:29:36,371 胎儿有危险 1425 99:59:59,999 --> 99:59:59,999 1425 01:29:36,438 --> 01:29:37,339 我们这里需要帮助！ 1426 99:59:59,999 --> 99:59:59,999 1426 01:29:37,405 --> 01:29:38,240 我们必须立刻手术 1427 99:59:59,999 --> 99:59:59,999 1427 01:29:38,306 --> 01:29:39,641 我的孩子能保住吗？ 1428 99:59:59,999 --> 99:59:59,999 1428 01:29:39,708 --> 01:29:41,109 我们会尽一切努力 1429 99:59:59,999 --> 99:59:59,999 1429 01:29:41,176 --> 01:29:42,043 那我也会平安无事吗？ 1430 99:59:59,999 --> 99:59:59,999 1430 01:29:42,544 --> 01:29:45,180 来自在你之前存在的每一个生命 1431 99:59:59,999 --> 99:59:59,999 1431 01:29:46,481 --> 01:29:51,453 来自为你的诞生而消亡的每一颗星辰 1432 99:59:59,999 --> 99:59:59,999 1432 01:30:09,070 --> 01:30:10,138 真是令人惊叹的杰作 1433 99:59:59,999 --> 99:59:59,999 1433 01:30:17,279 --> 01:30:19,681 Eliza巧妙地将真人实景表演 1434 99:59:59,999 --> 99:59:59,999 1434 01:30:19,881 --> 01:30:24,619 与Veo生成的内容，包括微观世界、宇宙景象 1435 99:59:59,999 --> 99:59:59,999 1435 01:30:24,886 --> 01:30:26,988 以及她自己作为新生儿的画面融为一体 1436 99:59:59,999 --> 99:59:59,999 1436 01:30:27,822 --> 01:30:31,393 这种方法开辟了全新的叙事可能性 1437 99:59:59,999 --> 99:59:59,999 1437 01:30:31,459 --> 01:30:33,795 使Eliza能够将宏大的电影视觉效果 1438 99:59:59,999 --> 99:59:59,999 1438 01:30:33,862 --> 01:30:36,665 和深厚的情感表达带入一个极为个人化的故事 1439 99:59:59,999 --> 99:59:59,999 1439 01:30:37,766 --> 01:30:40,702 这次合作也帮助我们塑造了Veo的功能 1440 99:59:59,999 --> 99:59:59,999 1440 01:30:41,002 --> 01:30:43,705 以满足像Eliza这样的故事创作者的需求 1441 99:59:59,999 --> 99:59:59,999 1441 01:30:44,272 --> 01:30:46,608 我们为电影制作人开发了多项新功能 1442 99:59:59,999 --> 99:59:59,999 1442 01:30:46,675 --> 01:30:48,176 所以当你创作视频时 1443 99:59:59,999 --> 99:59:59,999 1443 01:30:48,243 --> 01:30:49,911 无论你提供什么创作元素 1444 99:59:59,999 --> 99:59:59,999 1444 01:30:50,111 --> 01:30:54,282 角色、场景还是艺术风格 Veo都能确保它们在整个创作过程中保持一致性 1445 99:59:59,999 --> 99:59:59,999 1445 01:30:54,983 --> 01:30:56,718 你还可以像真正的导演一样指挥Veo 1446 99:59:59,999 --> 99:59:59,999 1446 01:30:56,785 --> 01:30:58,787 提供精确的相机指令 1447 99:59:59,999 --> 99:59:59,999 1447 01:30:58,853 --> 01:31:01,022 让它沿着特定路径拍摄 1448 99:59:59,999 --> 99:59:59,999 1448 01:31:02,157 --> 01:31:04,159 这些功能将帮助电影制作人 1449 99:59:59,999 --> 99:59:59,999 1449 01:31:04,225 --> 01:31:06,995 用Veo顺畅地表达自己的创意 1450 99:59:59,999 --> 99:59:59,999 1450 01:31:07,963 --> 01:31:10,065 而我们的创新并未止步于此 1451 99:59:59,999 --> 99:59:59,999 1451 01:31:10,732 --> 01:31:11,700 下面请Josh重返舞台 1452 99:59:59,999 --> 99:59:59,999 1452 01:31:11,766 --> 01:31:12,867 为大家介绍更多内容 1453 99:59:59,999 --> 99:59:59,999 1453 01:31:21,009 --> 01:31:21,977 谢谢，Jason 1454 99:59:59,999 --> 99:59:59,999 1454 01:31:22,777 --> 01:31:25,246 在与创意社区合作的基础上 1455 99:59:59,999 --> 99:59:59,999 1455 01:31:25,413 --> 01:31:29,684 我们为创作者打造了一款全新的AI电影制作工具 1456 99:59:59,999 --> 99:59:59,999 1456 01:31:30,251 --> 01:31:34,589 它结合了Veo、Imagen和Gemini三大技术的精髓 1457 99:59:59,999 --> 99:59:59,999 1457 01:31:35,090 --> 01:31:38,927 一款源自创作者、为了创作者打造的工具 1458 99:59:59,999 --> 99:59:59,999 1458 01:31:39,661 --> 01:31:42,530 它的设计灵感源自创作者那种 1459 99:59:59,999 --> 99:59:59,999 1459 01:31:42,630 --> 01:31:45,834 完全沉浸在创作中思如泉涌的奇妙境界 1460 99:59:59,999 --> 99:59:59,999 1460 01:31:46,468 --> 01:31:49,704 我们将这款工具命名为Flow，从今天起正式推出 1461 99:59:59,999 --> 99:59:59,999 1461 01:31:50,405 --> 01:31:51,673 我来给大家展示一下它的运作方式 1462 99:59:59,999 --> 99:59:59,999 1462 01:31:54,142 --> 01:31:56,244 我们来看看我正在创作的一个项目 1463 99:59:59,999 --> 99:59:59,999 1463 01:31:56,444 --> 01:31:58,380 故事的主角是一位老爷爷 1464 99:59:59,999 --> 99:59:59,999 1464 01:31:58,446 --> 01:32:01,649 他正在一位鸟类朋友的帮助下制造一辆会飞的汽车 1465 99:59:59,999 --> 99:59:59,999 1465 01:32:02,183 --> 01:32:03,451 这些是我的素材 1466 99:59:59,999 --> 99:59:59,999 1466 01:32:03,618 --> 01:32:05,620 这位老人和他的汽车 1467 99:59:59,999 --> 99:59:59,999 1467 01:32:06,254 --> 01:32:07,355 借助简便的界面 1468 99:59:59,999 --> 99:59:59,999 1468 01:32:07,422 --> 01:32:09,391 你可以轻松将自己的图像上传到工具中 1469 99:59:59,999 --> 99:59:59,999 1469 01:32:09,591 --> 01:32:12,293 或者，你可以使用内置的Imagen 1470 99:59:59,999 --> 99:59:59,999 1470 01:32:12,360 --> 01:32:13,695 即时生成所需的图像 1471 99:59:59,999 --> 99:59:59,999 1471 01:32:14,362 --> 01:32:18,833 只需简单描述就能生成一张定制金色变速杆图像 1472 99:59:59,999 --> 99:59:59,999 1472 01:32:19,467 --> 01:32:20,935 看，相当不错 1473 99:59:59,999 --> 99:59:59,999 1473 01:32:21,836 --> 01:32:25,740 接下来，你可以开始把所有这些素材组合起来 1474 99:59:59,999 --> 99:59:59,999 1474 01:32:26,274 --> 01:32:27,742 只需输入一条提示 1475 99:59:59,999 --> 99:59:59,999 1475 01:32:27,809 --> 01:32:31,746 你可以描述想要的效果，包括非常精确的镜头控制 1476 99:59:59,999 --> 99:59:59,999 1476 01:32:32,380 --> 01:32:35,050 Flow能将所有素材安排到恰当的位置 1477 99:59:59,999 --> 99:59:59,999 1477 01:32:35,450 --> 01:32:37,819 我可以在场景构建器中继续进行调整迭代 1478 99:59:59,999 --> 99:59:59,999 1478 01:32:38,353 --> 01:32:40,055 接下来是最精彩的部分 1479 99:59:59,999 --> 99:59:59,999 1479 01:32:40,555 --> 01:32:42,924 如果我想继续拍摄场景的下一个镜头 1480 99:59:59,999 --> 99:59:59,999 1480 01:32:43,258 --> 01:32:46,528 只要点一下这个加号，新镜头就自动生成了 1481 99:59:59,999 --> 99:59:59,999 1481 01:32:47,095 --> 01:32:49,664 我只需用文字描述接下来想要呈现的场景内容 1482 99:59:59,999 --> 99:59:59,999 1482 01:32:49,931 --> 01:32:52,600 比如在后座加上一只3米高的鸡 1483 99:59:59,999 --> 99:59:59,999 1483 01:32:53,034 --> 01:32:54,502 Flow将完成剩下的工作 1484 99:59:59,999 --> 99:59:59,999 1484 01:32:55,103 --> 01:32:58,239 角色将保持一致，场景也将保持一致 1485 99:59:59,999 --> 99:59:59,999 1485 01:32:58,406 --> 01:32:59,541 整个过程行云流水 1486 99:59:59,999 --> 99:59:59,999 1486 01:32:59,741 --> 01:33:03,211 如果对有些地方不太满意，也没关系 1487 99:59:59,999 --> 99:59:59,999 1487 01:33:03,445 --> 01:33:06,414 你可以像使用其他视频剪辑工具一样 1488 99:59:59,999 --> 99:59:59,999 1488 01:33:06,581 --> 01:33:09,050 随时返回剪掉那些不合适的片段 1489 99:59:59,999 --> 99:59:59,999 1489 01:33:09,551 --> 01:33:12,554 但Flow还支持反向操作 1490 99:59:59,999 --> 99:59:59,999 1490 01:33:12,754 --> 01:33:15,290 你还可以用它来延长素材片段 1491 99:59:59,999 --> 99:59:59,999 1491 01:33:15,924 --> 01:33:19,160 这样，我就得到了我一直追求的理想收尾镜头 1492 99:59:59,999 --> 99:59:59,999 1492 01:33:20,028 --> 01:33:23,431 所有片段制作完毕后，我可以把它们下载下来 1493 99:59:59,999 --> 99:59:59,999 1493 01:33:23,631 --> 01:33:26,401 然后把它们导入我常用的视频剪辑软件 1494 99:59:59,999 --> 99:59:59,999 1494 01:33:26,468 --> 01:33:28,436 再加上一小段Lyria生成的配乐 1495 99:59:59,999 --> 99:59:59,999 1495 01:33:28,636 --> 01:33:32,874 就这样，老爷爷终于开上了自己的“飞车” 1496 99:59:59,999 --> 99:59:59,999 1496 01:33:56,164 --> 01:33:57,732 简直太棒了 1497 99:59:59,999 --> 99:59:59,999 1497 01:33:57,932 --> 01:34:01,636 接下来，让我们听听几位新锐AI电影制作人的心声 1498 99:59:59,999 --> 99:59:59,999 1498 01:34:01,703 --> 01:34:04,305 他们在Flow的开发过程中为我们提供了宝贵意见 1499 99:59:59,999 --> 99:59:59,999 1499 01:34:04,372 --> 01:34:06,608 他们分享了自己使用Flow的感受 1500 99:59:59,999 --> 99:59:59,999 1500 01:34:06,774 --> 01:34:08,543 以及他们如何让这一工具变得更出色 1501 99:59:59,999 --> 99:59:59,999 1501 01:34:08,776 --> 01:34:09,744 一起来看看吧 1502 99:59:59,999 --> 99:59:59,999 1502 01:34:13,281 --> 01:34:15,416 我不知道自己是否走在正确的道路上 1503 99:59:59,999 --> 99:59:59,999 1503 01:34:15,483 --> 01:34:17,485 但一直在努力探索 1504 99:59:59,999 --> 99:59:59,999 1504 01:34:17,752 --> 01:34:21,656 我不断探索，不断质疑 1505 99:59:59,999 --> 99:59:59,999 1505 01:34:21,723 --> 01:34:24,225 直到某一刻，有些东西悄然改变 1506 99:59:59,999 --> 99:59:59,999 1506 01:34:24,626 --> 01:34:27,362 我不再纠结于正确与否，只是专注去做 1507 99:59:59,999 --> 99:59:59,999 1507 01:34:29,898 --> 01:34:35,403 然后，那些散落的拼图开始慢慢归位 1508 99:59:59,999 --> 99:59:59,999 1508 01:34:39,474 --> 01:34:41,709 我的思维变得非常清晰 1509 99:59:59,999 --> 99:59:59,999 1509 01:34:42,410 --> 01:34:46,748 无数种可能在我脑海中闪现 1510 99:59:59,999 --> 99:59:59,999 1510 01:34:47,081 --> 01:34:49,751 就像穿越时空般奇妙 1511 99:59:59,999 --> 99:59:59,999 1511 01:34:49,817 --> 01:34:51,719 当我以上帝视角审视身为创作者的自己 1512 99:59:59,999 --> 99:59:59,999 1512 01:34:51,786 --> 01:34:53,721 以及我所创作的不同世界中的角色 1513 99:59:59,999 --> 99:59:59,999 1513 01:34:53,788 --> 01:34:56,024 我感觉这些角色仿佛有着自己的生命 1514 99:59:59,999 --> 99:59:59,999 1514 01:34:56,090 --> 01:34:59,194 尽管我知道故事情节的走向由我掌控 1515 99:59:59,999 --> 99:59:59,999 1515 01:34:59,861 --> 01:35:01,696 似乎在某个时刻 1516 99:59:59,999 --> 99:59:59,999 1516 01:35:01,763 --> 01:35:04,032 这些故事开始沿着自己的脉络逐渐展开 1517 99:59:59,999 --> 99:59:59,999 1517 01:35:04,799 --> 01:35:09,370 你知道，你的故事可以有无数种结局 1518 99:59:59,999 --> 99:59:59,999 1518 01:35:11,206 --> 01:35:15,176 用Flow创作的作品不是简单的素材堆砌 1519 99:59:59,999 --> 99:59:59,999 1519 01:35:16,044 --> 01:35:19,581 它就像花园里的花，肆意绽放 1520 99:59:59,999 --> 99:59:59,999 1520 01:35:20,648 --> 01:35:25,153 它自然生长，充满生命力，浑然天成 1521 99:59:59,999 --> 99:59:59,999 1521 01:35:26,254 --> 01:35:27,855 我没有刻意去安排什么 1522 99:59:59,999 --> 99:59:59,999 1522 01:35:27,922 --> 01:35:31,092 那些画面只是在等着我找到它们 1523 99:59:59,999 --> 99:59:59,999 1523 01:35:31,326 --> 01:35:34,195 当我有这种感觉时，我就知道自己找对了方向 1524 99:59:59,999 --> 99:59:59,999 1524 01:35:54,582 --> 01:35:57,185 真是叹为观止，听到这些电影制作人讲述 1525 99:59:59,999 --> 99:59:59,999 1525 01:35:57,252 --> 01:35:59,354 如何将不可思议的想象变为现实 1526 99:59:59,999 --> 99:59:59,999 1526 01:35:59,854 --> 01:36:01,522 我们希望通过我们的工具 1527 99:59:59,999 --> 99:59:59,999 1527 01:36:01,589 --> 01:36:05,260 让大家都能享受这种自由构建、挥洒创意的感觉 1528 99:59:59,999 --> 99:59:59,999 1528 01:36:05,927 --> 01:36:07,562 所以我很激动地宣布 1529 99:59:59,999 --> 99:59:59,999 1529 01:36:07,629 --> 01:36:11,266 今天，我们将升级Google的两大AI订阅方案 1530 99:59:59,999 --> 99:59:59,999 1530 01:36:11,599 --> 01:36:13,668 我们将推出Google AI Pro 1531 99:59:59,999 --> 99:59:59,999 1531 01:36:13,735 --> 01:36:16,571 以及全新的Google AI Ultra 1532 99:59:59,999 --> 99:59:59,999 1532 01:36:17,605 --> 01:36:20,842 我们将面向全球推出Google AI Pro 1533 99:59:59,999 --> 99:59:59,999 1533 01:36:20,942 --> 01:36:22,710 相比Google AI的免费版本 1534 99:59:59,999 --> 99:59:59,999 1534 01:36:22,777 --> 01:36:24,679 你将拥有全套Google AI产品的使用权 1535 99:59:59,999 --> 99:59:59,999 1535 01:36:24,746 --> 01:36:27,615 并获享更高的用量限额和特殊功能 1536 99:59:59,999 --> 99:59:59,999 1536 01:36:28,016 --> 01:36:30,785 其中包括Gemini应用的Pro版本 1537 99:59:59,999 --> 99:59:59,999 1537 01:36:30,885 --> 01:36:33,054 也就是先前的Gemini Advanced 1538 99:59:59,999 --> 99:59:59,999 1538 01:36:34,155 --> 01:36:36,124 然后让我来介绍一下Google AI Ultra 1539 99:59:59,999 --> 99:59:59,999 1539 01:36:36,591 --> 01:36:39,594 它专门面向行业开拓者和先行者 1540 99:59:59,999 --> 99:59:59,999 1540 01:36:39,827 --> 01:36:43,464 以及Google前沿AI技术的尝鲜者 1541 99:59:59,999 --> 99:59:59,999 1541 01:36:43,865 --> 01:36:46,834 该方案将为用户提供最高用量限额 1542 99:59:59,999 --> 99:59:59,999 1542 01:36:47,068 --> 01:36:48,870 订阅者还可以抢先体验 1543 99:59:59,999 --> 99:59:59,999 1543 01:36:48,936 --> 01:36:51,239 Google发布的各项新产品和功能 1544 99:59:59,999 --> 99:59:59,999 1544 01:36:51,673 --> 01:36:53,841 Google AI Ultra今天仅面向美国推出 1545 99:59:59,999 --> 99:59:59,999 1545 01:36:54,042 --> 01:36:56,411 并将很快推广至全球 1546 99:59:59,999 --> 99:59:59,999 1546 01:36:56,878 --> 01:36:58,913 你可以把这个Ultra方案当成是 1547 99:59:59,999 --> 99:59:59,999 1547 01:36:58,980 --> 01:37:01,215 你专属的Google AI VIP通行证 1548 99:59:59,999 --> 99:59:59,999 1548 01:37:02,016 --> 01:37:03,685 订阅Google AI Ultra 1549 99:59:59,999 --> 99:59:59,999 1549 01:37:03,918 --> 01:37:05,653 你就能获得超高用量限额 1550 99:59:59,999 --> 99:59:59,999 1550 01:37:05,753 --> 01:37:07,555 等到2.5 Pro Deep Think模式推出后 1551 99:59:59,999 --> 99:59:59,999 1551 01:37:07,622 --> 01:37:10,725 还可以在Gemini应用中使用这种模式 1552 99:59:59,999 --> 99:59:59,999 1552 01:37:11,426 --> 01:37:15,997 你还可以抢先体验我们今天发布的Veo 3和Flow 1553 99:59:59,999 --> 99:59:59,999 1553 01:37:16,597 --> 01:37:17,799 该方案还附赠 1554 99:59:59,999 --> 99:59:59,999 1554 01:37:17,865 --> 01:37:20,635 YouTube Premium会员和海量存储空间 1555 99:59:59,999 --> 99:59:59,999 1555 01:37:21,336 --> 01:37:24,572 我们热切期待看到，借助这些全新订阅方案 1556 99:59:59,999 --> 99:59:59,999 1556 01:37:24,639 --> 01:37:27,575 你将构建、创造和挖掘出怎样的精彩 1557 99:59:59,999 --> 99:59:59,999 1557 01:37:28,209 --> 01:37:30,378 下面，让我们一同展望未来 1558 99:59:59,999 --> 99:59:59,999 1558 01:37:30,712 --> 01:37:33,981 探索我们将如何使用AI与现实世界互动 1559 99:59:59,999 --> 99:59:59,999 1559 01:37:34,248 --> 01:37:36,084 有请Shahram 1560 99:59:59,999 --> 99:59:59,999 1560 01:37:50,365 --> 01:37:51,432 大家好 1561 99:59:59,999 --> 99:59:59,999 1561 01:37:51,666 --> 01:37:55,970 此刻，Android平台正在发生许多激动人心的变化 1562 99:59:59,999 --> 99:59:59,999 1562 01:37:56,671 --> 01:37:59,874 在这个平台中，你将率先预见未来 1563 99:59:59,999 --> 99:59:59,999 1563 01:38:00,708 --> 01:38:03,277 在上周的Android Show上 1564 99:59:59,999 --> 99:59:59,999 1564 01:38:03,478 --> 01:38:05,780 我们公布了Android 16和Wear OS 6 1565 99:59:59,999 --> 99:59:59,999 1565 01:38:06,013 --> 01:38:10,184 全新的大胆设计和重大更新 1566 99:59:59,999 --> 99:59:59,999 1566 01:38:11,185 --> 01:38:15,089 当然，Android也是体验Google AI的绝佳平台 1567 99:59:59,999 --> 99:59:59,999 1567 01:38:15,957 --> 01:38:18,793 你今天看到的许多Gemini突破性创新 1568 99:59:59,999 --> 99:59:59,999 1568 01:38:18,860 --> 01:38:21,129 都会在不久之后登陆Android 1569 99:59:59,999 --> 99:59:59,999 1569 01:38:21,763 --> 01:38:25,433 你现已可以通过电源按钮立即访问Gemini 1570 99:59:59,999 --> 99:59:59,999 1570 01:38:25,833 --> 01:38:29,070 它能理解你的意图和上下文，并随时为你提供帮助 1571 99:59:59,999 --> 99:59:59,999 1571 01:38:30,138 --> 01:38:32,540 Android不止是手机系统 1572 99:59:59,999 --> 99:59:59,999 1572 01:38:33,040 --> 01:38:35,810 更是一个完整的设备生态系统 1573 99:59:59,999 --> 99:59:59,999 1573 01:38:36,377 --> 01:38:37,745 在接下来的几个月 1574 99:59:59,999 --> 99:59:59,999 1574 01:38:38,079 --> 01:38:42,550 我们将把Gemini引入你的手表、汽车仪表盘 1575 99:59:59,999 --> 99:59:59,999 1575 01:38:42,984 --> 01:38:44,352 甚至电视 1576 99:59:59,999 --> 99:59:59,999 1576 01:38:45,086 --> 01:38:46,421 无论你身在何处 1577 99:59:59,999 --> 99:59:59,999 1577 01:38:46,521 --> 01:38:50,558 都会有一位有用的AI助理伴你左右，让你的生活更轻松 1578 99:59:59,999 --> 99:59:59,999 1578 01:38:51,559 --> 01:38:54,162 那么，要不要尝试一些新兴的设备形态 1579 99:59:59,999 --> 99:59:59,999 1579 01:38:54,262 --> 01:38:58,733 让我们以全新方式体验AI助理？ 1580 99:59:59,999 --> 99:59:59,999 1580 01:38:59,267 --> 01:39:02,870 这正是我们打造Android XR的原因 1581 99:59:59,999 --> 99:59:59,999 1581 01:39:10,077 --> 01:39:14,115 它是首个在Gemini时代构建的Android平台 1582 99:59:59,999 --> 99:59:59,999 1582 01:39:14,348 --> 01:39:18,820 支持不同应用场景的各种设备 1583 99:59:59,999 --> 99:59:59,999 1583 01:39:19,220 --> 01:39:23,424 从头戴设备到智能眼镜，以及其他各种智能设备 1584 99:59:59,999 --> 99:59:59,999 1584 01:39:24,192 --> 01:39:27,228 我们认为，没有哪种扩展现实设备是万能的 1585 99:59:59,999 --> 99:59:59,999 1585 01:39:27,395 --> 01:39:30,264 你在一天当中会用到不同的设备 1586 99:59:59,999 --> 99:59:59,999 1586 01:39:30,898 --> 01:39:36,170 比如，在看电影、玩游戏或处理工作时 1587 99:59:59,999 --> 99:59:59,999 1587 01:39:36,337 --> 01:39:38,606 你想要的是沉浸式头戴设备 1588 99:59:59,999 --> 99:59:59,999 1588 01:39:39,307 --> 01:39:43,044 但外出时，你会想要一副轻便的智能眼镜 1589 99:59:59,999 --> 99:59:59,999 1589 01:39:43,110 --> 01:39:46,848 让你不用掏出手机就能获得及时的信息 1590 99:59:59,999 --> 99:59:59,999 1590 01:39:47,949 --> 01:39:51,719 我们携手三星共同打造了Android XR 1591 99:59:59,999 --> 99:59:59,999 1591 01:39:52,019 --> 01:39:55,056 并与高通合作，针对骁龙处理器对其进行了优化 1592 99:59:59,999 --> 99:59:59,999 1592 01:39:55,957 --> 01:39:59,460 自从去年Android XR开发者预览版发布以来 1593 99:59:59,999 --> 99:59:59,999 1593 01:39:59,861 --> 01:40:02,730 已有数百名开发者在为该平台构建应用 1594 99:59:59,999 --> 99:59:59,999 1594 01:40:03,431 --> 01:40:05,733 我们也在为Android XR重新构建 1595 99:59:59,999 --> 99:59:59,999 1595 01:40:05,800 --> 01:40:07,902 你喜爱的各种Google应用 1596 99:59:59,999 --> 99:59:59,999 1596 01:40:08,469 --> 01:40:10,204 由于它是Android系统 1597 99:59:59,999 --> 99:59:59,999 1597 01:40:10,438 --> 01:40:13,207 所以你的手机和平板应用也与它兼容 1598 99:59:59,999 --> 99:59:59,999 1598 01:40:14,041 --> 01:40:16,110 今天，我想跟大家分享 1599 99:59:59,999 --> 99:59:59,999 1599 01:40:16,177 --> 01:40:21,816 Gemini如何颠覆头戴设备和智能眼镜用户的体验 1600 99:59:59,999 --> 99:59:59,999 1600 01:40:22,650 --> 01:40:24,151 在这些设备上 1601 99:59:59,999 --> 99:59:59,999 1601 01:40:24,218 --> 01:40:28,789 AI助理会以更丰富的方式理解你的上下文和意图 1602 99:59:59,999 --> 99:59:59,999 1602 01:40:28,990 --> 01:40:30,791 全天候为你提供支持 1603 99:59:59,999 --> 99:59:59,999 1603 01:40:31,726 --> 01:40:34,228 让我们从头戴设备中的Gemini说起 1604 99:59:59,999 --> 99:59:59,999 1604 01:40:34,762 --> 01:40:40,134 这是三星的Project Moohan，首款Android XR设备 1605 99:59:59,999 --> 99:59:59,999 1605 01:40:46,974 --> 01:40:49,544 Moohan提供无限延展的屏幕空间 1606 99:59:59,999 --> 99:59:59,999 1606 01:40:49,610 --> 01:40:52,747 让你在Gemini的陪伴下尽情探索各种应用 1607 99:59:59,999 --> 99:59:59,999 1607 01:40:53,180 --> 01:40:55,016 Android XR还集成了Google地图 1608 99:59:59,999 --> 99:59:59,999 1608 01:40:55,182 --> 01:40:57,418 只要告诉Gemini你想去哪里 1609 99:59:59,999 --> 99:59:59,999 1609 01:40:57,618 --> 01:41:00,555 它就能带你“瞬间传送”到世界任何角落 1610 99:59:59,999 --> 99:59:59,999 1610 01:41:01,322 --> 01:41:05,426 你可以和AI助理谈论你看到的一切 1611 99:59:59,999 --> 99:59:59,999 1611 01:41:05,726 --> 01:41:09,897 并让它调出你想要探索的视频和网站 1612 99:59:59,999 --> 99:59:59,999 1612 01:41:11,032 --> 01:41:15,336 我们都希望能买到自己支持球队的季票 1613 99:59:59,999 --> 99:59:59,999 1613 01:41:15,836 --> 01:41:19,941 想象一下，当你使用Moohan在MLB应用上观看球赛 1614 99:59:59,999 --> 99:59:59,999 1614 01:41:20,007 --> 01:41:21,375 你将仿佛置身现场 1615 99:59:59,999 --> 99:59:59,999 1615 01:41:21,943 --> 01:41:25,446 还可以一边观赛一边和Gemini讨论球员和比赛数据 1616 99:59:59,999 --> 99:59:59,999 1616 01:41:26,614 --> 01:41:31,552 三星的Project Moohan将于今年晚些时候发售 1617 99:59:59,999 --> 99:59:59,999 1617 01:41:32,453 --> 01:41:35,590 我们迫切期待你们能亲身体验这款设备 1618 99:59:59,999 --> 99:59:59,999 1618 01:41:36,724 --> 01:41:40,661 现在，让我们把注意力转向智能眼镜 1619 99:59:59,999 --> 99:59:59,999 1619 01:41:41,429 --> 01:41:45,700 众所周知，我们在智能眼镜领域耕耘超过十年 1620 99:59:59,999 --> 99:59:59,999 1620 01:41:45,866 --> 01:41:47,435 从未停下脚步 1621 99:59:59,999 --> 99:59:59,999 1621 01:41:48,235 --> 01:41:53,240 搭载Android XR的智能眼镜满载先进科技 1622 99:59:59,999 --> 99:59:59,999 1622 01:41:53,307 --> 01:41:56,110 但无比轻盈，适合全天佩戴 1623 99:59:59,999 --> 99:59:59,999 1623 01:41:57,011 --> 01:41:58,879 眼镜内置镜头和麦克风 1624 99:59:59,999 --> 99:59:59,999 1624 01:41:59,213 --> 01:42:02,783 让Gemini能够“看到”和“听到”这个世界 1625 99:59:59,999 --> 99:59:59,999 1625 01:42:03,384 --> 01:42:07,655 扬声器则让你可以与AI对话、播放音乐或接听电话 1626 99:59:59,999 --> 99:59:59,999 1626 01:42:08,189 --> 01:42:10,191 你也可以通过镜片内置显示屏 1627 99:59:59,999 --> 99:59:59,999 1627 01:42:10,524 --> 01:42:14,395 以私密的方式在需要时查看有用的信息 1628 99:59:59,999 --> 99:59:59,999 1628 01:42:15,529 --> 01:42:17,398 这款眼镜可以与你的手机连接 1629 99:59:59,999 --> 99:59:59,999 1629 01:42:17,498 --> 01:42:20,901 让你不用动手也能访问应用 1630 99:59:59,999 --> 99:59:59,999 1630 01:42:21,936 --> 01:42:25,873 这一切都让智能眼镜成为天然适合AI的设备形态 1631 99:59:59,999 --> 99:59:59,999 1631 01:42:26,307 --> 01:42:29,577 让Gemini的强大功能随时伴你左右 1632 99:59:59,999 --> 99:59:59,999 1632 01:42:30,478 --> 01:42:32,580 所以，与摘下眼镜才能变身的超人不同 1633 99:59:59,999 --> 99:59:59,999 1633 01:42:32,813 --> 01:42:36,017 戴上这款智能眼镜，你就能获得“超能力” 1634 99:59:59,999 --> 99:59:59,999 1634 01:42:37,885 --> 01:42:42,256 好了，谁想看Android XR眼镜的最新演示？ 1635 99:59:59,999 --> 99:59:59,999 1635 01:42:47,862 --> 01:42:53,300 让我们看看它在最忙乱的环境中表现如何 1636 99:59:59,999 --> 99:59:59,999 1636 01:42:53,567 --> 01:42:55,169 比如…I/O大会的后台 1637 99:59:59,999 --> 99:59:59,999 1637 01:42:55,970 --> 01:42:59,373 我们的“超级女侠”Nishtha就在后台 1638 99:59:59,999 --> 99:59:59,999 1638 01:42:59,473 --> 01:43:01,575 她将向我们展示这款眼镜在真实环境中的使用效果 1639 99:59:59,999 --> 99:59:59,999 1639 01:43:02,243 --> 01:43:04,845 我现在就给她发个消息，然后我们马上开始 1640 99:59:59,999 --> 99:59:59,999 1640 01:43:08,849 --> 01:43:10,051 大家好！ 1641 99:59:59,999 --> 99:59:59,999 1641 01:43:10,518 --> 01:43:13,120 现在，你们应该能在大屏幕上看到 1642 99:59:59,999 --> 99:59:59,999 1642 01:43:13,187 --> 01:43:16,090 我通过Android XR智能眼镜镜片看到的一切 1643 99:59:59,999 --> 99:59:59,999 1643 01:43:16,323 --> 01:43:18,292 比如这杯美味的咖啡 1644 99:59:59,999 --> 99:59:59,999 1644 01:43:18,526 --> 01:43:20,428 还有Shahram刚刚发来的这条消息 1645 99:59:59,999 --> 99:59:59,999 1645 01:43:20,494 --> 01:43:21,696 让我们看看他说了什么 1646 99:59:59,999 --> 99:59:59,999 1646 01:43:23,764 --> 01:43:25,900 很好，现在就是见证奇迹的时刻 1647 99:59:59,999 --> 99:59:59,999 1647 01:43:25,966 --> 01:43:28,402 我要启动Gemini，开始演示 1648 99:59:59,999 --> 99:59:59,999 1648 01:43:29,804 --> 01:43:31,906 给Shahram发一条消息，告诉他我已经开始了 1649 99:59:59,999 --> 99:59:59,999 1649 01:43:31,972 --> 01:43:34,108 然后把我的通知静音 1650 99:59:59,999 --> 99:59:59,999 1650 01:43:36,644 --> 01:43:38,579 好的，我已经给他发了消息 1651 99:59:59,999 --> 99:59:59,999 1651 01:43:38,713 --> 01:43:40,881 并且把你所有的通知静音了 1652 99:59:59,999 --> 99:59:59,999 1652 01:43:41,215 --> 01:43:42,349 很好 1653 99:59:59,999 --> 99:59:59,999 1653 01:43:43,317 --> 01:43:44,151 嘿，Nishtha！ 1654 99:59:59,999 --> 99:59:59,999 1654 01:43:44,218 --> 01:43:45,086 嘿，Dieter！ 1655 99:59:59,999 --> 99:59:59,999 1655 01:43:45,152 --> 01:43:48,856 你的灯亮着，看来我们已经开始直播了？ 1656 99:59:59,999 --> 99:59:59,999 1656 01:43:48,923 --> 01:43:51,492 是的，我们已经和I/O大会的工作人员连上线了 1657 99:59:59,999 --> 99:59:59,999 1657 01:43:51,659 --> 01:43:54,528 大家好。从这个视角看I/O大会，真是太棒了！ 1658 99:59:59,999 --> 99:59:59,999 1658 01:43:54,595 --> 01:43:55,830 Nishtha，你答应过我 1659 99:59:59,999 --> 99:59:59,999 1659 01:43:55,896 --> 01:43:59,233 只要我来帮忙，就送我一副Android XR眼镜 1660 99:59:59,999 --> 99:59:59,999 1660 01:43:59,667 --> 01:44:00,668 说话算数吗？ 1661 99:59:59,999 --> 99:59:59,999 1661 01:44:00,735 --> 01:44:01,669 当然！ 1662 99:59:59,999 --> 99:59:59,999 1662 01:44:01,736 --> 01:44:02,903 待会结束后我们去喝杯咖啡 1663 99:59:59,999 --> 99:59:59,999 1663 01:44:02,970 --> 01:44:04,171 我会把你的智能眼镜带过去！ 1664 99:59:59,999 --> 99:59:59,999 1664 01:44:04,238 --> 01:44:05,139 太棒了！待会见！祝你演示顺利 1665 99:59:59,999 --> 99:59:59,999 1665 01:44:05,206 --> 01:44:06,173 谢谢 1666 99:59:59,999 --> 99:59:59,999 1666 01:44:06,440 --> 01:44:09,877 大家可以看到，我们在后台，这里真是热闹非凡 1667 99:59:59,999 --> 99:59:59,999 1667 01:44:10,244 --> 01:44:13,881 那是职业篮球运动员Giannis戴着我们的眼镜吗？ 1668 99:59:59,999 --> 99:59:59,999 1668 01:44:14,515 --> 01:44:15,716 我太喜欢这眼镜了！ 1669 99:59:59,999 --> 99:59:59,999 1669 01:44:15,883 --> 01:44:18,652 它解放了我的双手，让我随时能来个双手击掌 1670 99:59:59,999 --> 99:59:59,999 1670 01:44:19,053 --> 01:44:23,057 太棒了，我继续来展示这副眼镜的功能 1671 99:59:59,999 --> 99:59:59,999 1671 01:44:23,524 --> 01:44:26,360 今天一整天，我都对这面照片墙很好奇 1672 99:59:59,999 --> 99:59:59,999 1672 01:44:26,794 --> 01:44:31,265 比如这是什么乐队，他们和这个地方有什么联系？ 1673 99:59:59,999 --> 99:59:59,999 1673 01:44:36,871 --> 01:44:38,672 这是Counting Crows 1674 99:59:59,999 --> 99:59:59,999 1674 01:44:38,739 --> 01:44:41,175 一支来自旧金山湾区的传奇摇滚乐队 1675 99:59:59,999 --> 99:59:59,999 1675 01:44:41,242 --> 01:44:43,477 他们在Shoreline Amphitheatre演出过很多次 1676 99:59:59,999 --> 99:59:59,999 1676 01:44:43,544 --> 01:44:45,746 这里通常被视为是他们的返乡主场 1677 99:59:59,999 --> 99:59:59,999 1677 01:44:45,980 --> 01:44:49,650 天哪！你能给我看看他们在这里演出时的照片吗？ 1678 99:59:59,999 --> 99:59:59,999 1678 01:44:52,520 --> 01:44:55,556 当然，这有一张，想让我播放一首他们的歌曲吗？ 1679 99:59:59,999 --> 99:59:59,999 1679 01:44:55,990 --> 01:44:58,759 好啊，我可以一边听一边走向舞台 1680 99:59:59,999 --> 99:59:59,999 1680 01:45:02,530 --> 01:45:06,167 太好了！请听Counting Crows的《Under the Aurora》 1681 99:59:59,999 --> 99:59:59,999 1681 01:45:07,201 --> 01:45:09,804 好了，大家准备好一睹这副眼镜的风采了吗？ 1682 99:59:59,999 --> 99:59:59,999 1682 01:45:10,204 --> 01:45:12,439 让我们欢迎Nishtha！ 1683 99:59:59,999 --> 99:59:59,999 1683 01:45:13,741 --> 01:45:15,042 Nishtha，欢迎欢迎！ 1684 99:59:59,999 --> 99:59:59,999 1684 01:45:15,109 --> 01:45:16,443 大家好！ 1685 99:59:59,999 --> 99:59:59,999 1685 01:45:19,246 --> 01:45:21,682 谢谢你为我们带来这段星光熠熠的幕后之旅 1686 99:59:59,999 --> 99:59:59,999 1686 01:45:21,949 --> 01:45:24,385 对了，你现在要约Dieter去喝咖啡吗？ 1687 99:59:59,999 --> 99:59:59,999 1687 01:45:24,451 --> 01:45:27,955 当然…刚刚工作人员在后台给了我一杯超棒的咖啡 1688 99:59:59,999 --> 99:59:59,999 1688 01:45:28,055 --> 01:45:29,523 所以让来我试试一些好玩的功能 1689 99:59:59,999 --> 99:59:59,999 1689 01:45:31,058 --> 01:45:35,196 Gemini，刚刚那杯咖啡上的咖啡店名字是什么？ 1690 99:59:59,999 --> 99:59:59,999 1690 01:45:37,865 --> 01:45:39,600 嗯…可能是Bloomsgiving 1691 99:59:59,999 --> 99:59:59,999 1691 01:45:39,667 --> 01:45:40,734 据我所知 1692 99:59:59,999 --> 99:59:59,999 1692 01:45:40,801 --> 01:45:42,870 那是卡斯特罗街上一家充满活力的咖啡馆 1693 99:59:59,999 --> 99:59:59,999 1693 01:45:43,204 --> 01:45:44,538 记性真好 1694 99:59:59,999 --> 99:59:59,999 1694 01:45:50,678 --> 01:45:52,913 你能给我看看那家咖啡馆的照片吗？ 1695 99:59:59,999 --> 99:59:59,999 1695 01:45:53,047 --> 01:45:54,381 我想看看那里的氛围 1696 99:59:59,999 --> 99:59:59,999 1696 01:45:57,117 --> 01:45:57,952 没问题 1697 99:59:59,999 --> 99:59:59,999 1697 01:45:58,018 --> 01:45:59,653 这些是来自Google地图的照片，请问有帮助吗？ 1698 99:59:59,999 --> 99:59:59,999 1698 01:46:00,387 --> 01:46:01,422 哦，我知道那个地方 1699 99:59:59,999 --> 99:59:59,999 1699 01:46:01,488 --> 01:46:04,692 它是一家花店，同时也卖咖啡，但它在市中心 1700 99:59:59,999 --> 99:59:59,999 1700 01:46:05,426 --> 01:46:08,462 好吧，Gemini，我走过去需要多久？ 1701 99:59:59,999 --> 99:59:59,999 1701 01:46:11,799 --> 01:46:13,267 正在获取路线 1702 99:59:59,999 --> 99:59:59,999 1702 01:46:14,168 --> 01:46:15,769 步行过去大概需要1小时 1703 99:59:59,999 --> 99:59:59,999 1703 01:46:16,136 --> 01:46:17,104 好吧 1704 99:59:59,999 --> 99:59:59,999 1704 01:46:17,805 --> 01:46:19,540 也许我可以走过去，顺便刷点步数 1705 99:59:59,999 --> 99:59:59,999 1705 01:46:19,607 --> 01:46:24,144 眼镜上会直接显示完整3D地图和导航，会很轻松 1706 99:59:59,999 --> 99:59:59,999 1706 01:46:33,587 --> 01:46:39,059 帮我约Dieter今天下午3点在那家咖啡馆喝咖啡 1707 99:59:59,999 --> 99:59:59,999 1707 01:46:42,763 --> 01:46:44,131 我马上发出邀请 1708 99:59:59,999 --> 99:59:59,999 1708 01:46:44,231 --> 01:46:45,366 希望你们享受愉快的咖啡时光 1709 99:59:59,999 --> 99:59:59,999 1709 01:46:46,233 --> 01:46:49,737 如你们所见，Gemini帮助Nishtha搜索她看到的东西 1710 99:59:59,999 --> 99:59:59,999 1710 01:46:50,037 --> 01:46:53,140 记住咖啡杯这种细节 1711 99:59:59,999 --> 99:59:59,999 1711 01:46:53,741 --> 01:46:55,943 为她安排日程、甚至为她导航 1712 99:59:59,999 --> 99:59:59,999 1712 01:46:56,277 --> 01:46:59,313 而Nishtha全程都不用掏出手机 1713 99:59:59,999 --> 99:59:59,999 1713 01:46:59,947 --> 01:47:02,116 其实我现在也戴着这副眼镜 1714 99:59:59,999 --> 99:59:59,999 1714 01:47:02,549 --> 01:47:04,718 它相当于是我的个人提词器 1715 99:59:59,999 --> 99:59:59,999 1715 01:47:04,985 --> 01:47:07,421 而且它的镜片还带度数，所以我能看清你们 1716 99:59:59,999 --> 99:59:59,999 1716 01:47:08,689 --> 01:47:11,025 Nishtha，这是智能眼镜的高光时刻 1717 99:59:59,999 --> 99:59:59,999 1717 01:47:11,091 --> 01:47:11,926 我们把这一刻记录下来吧 1718 99:59:59,999 --> 99:59:59,999 1718 01:47:11,992 --> 01:47:12,826 好的 1719 99:59:59,999 --> 99:59:59,999 1719 01:47:12,927 --> 01:47:14,728 大家准备好，我们快速合个影 1720 99:59:59,999 --> 99:59:59,999 1720 01:47:14,795 --> 01:47:16,797 让我们把后台的明星也请上来！ 1721 99:59:59,999 --> 99:59:59,999 1721 01:47:18,666 --> 01:47:20,134 这是Dieter 1722 99:59:59,999 --> 99:59:59,999 1722 01:47:23,304 --> 01:47:24,505 Janice来了 1723 99:59:59,999 --> 99:59:59,999 1723 01:47:25,272 --> 01:47:26,907 Gemini，帮我拍张照 1724 99:59:59,999 --> 99:59:59,999 1724 01:47:35,816 --> 01:47:39,286 Gemini，帮我拍张照，保存到我的收藏夹里 1725 99:59:59,999 --> 99:59:59,999 1725 01:47:47,194 --> 01:47:48,729 拍得真好！ 1726 99:59:59,999 --> 99:59:59,999 1726 01:47:56,437 --> 01:47:58,205 我是个彻头彻尾的追星族 1727 99:59:59,999 --> 99:59:59,999 1727 01:47:58,639 --> 01:48:00,507 好的，还有最后一件事 1728 99:59:59,999 --> 99:59:59,999 1728 01:48:01,175 --> 01:48:05,245 之前，Sundar展示了实时翻译已经发展到什么程度 1729 99:59:59,999 --> 99:59:59,999 1729 01:48:05,612 --> 01:48:08,282 现在让我们看看它在智能眼镜上的表现 1730 99:59:59,999 --> 99:59:59,999 1730 01:48:08,849 --> 01:48:12,286 这个演示很有挑战，但我们试试吧！ 1731 99:59:59,999 --> 99:59:59,999 1731 01:48:14,288 --> 01:48:18,092 Nishtha和我要用各自的母语交谈 1732 99:59:59,999 --> 99:59:59,999 1732 01:48:18,225 --> 01:48:19,760 Nishtha说印地语 1733 99:59:59,999 --> 99:59:59,999 1733 01:48:20,094 --> 01:48:22,796 我说波斯语，虽然我说得很烂 1734 99:59:59,999 --> 99:59:59,999 1734 01:48:23,230 --> 01:48:26,967 你们会在大屏幕上看到我俩眼镜上的实时画面 1735 99:59:59,999 --> 99:59:59,999 1735 01:48:27,368 --> 01:48:29,036 这样你们都可以明白我们在说什么 1736 99:59:59,999 --> 99:59:59,999 1736 01:48:29,470 --> 01:48:32,439 我们会实时显示英语译文 1737 99:59:59,999 --> 99:59:59,999 1737 01:48:33,007 --> 01:48:36,810 好吧，我们试试，但愿一切顺利 1738 99:59:59,999 --> 99:59:59,999 1738 01:48:38,078 --> 01:48:43,283 [波斯语] 1739 99:59:59,999 --> 99:59:59,999 1739 01:48:43,350 --> 01:48:45,652 [波斯语] 1740 99:59:59,999 --> 99:59:59,999 1740 01:48:50,657 --> 01:48:57,398 [印地语] 1741 99:59:59,999 --> 99:59:59,999 1741 01:49:00,367 --> 01:49:04,438 [波斯语] 1742 99:59:59,999 --> 99:59:59,999 1742 01:49:04,938 --> 01:49:09,910 [波斯语] 1743 99:59:59,999 --> 99:59:59,999 1743 01:49:16,683 --> 01:49:18,318 我们之前没说错，确实演示有风险 1744 99:59:59,999 --> 99:59:59,999 1744 01:49:20,821 --> 01:49:21,722 让我们… 1745 99:59:59,999 --> 99:59:59,999 1745 01:49:38,505 --> 01:49:39,440 感谢Nishtha 1746 99:59:59,999 --> 99:59:59,999 1746 01:49:39,506 --> 01:49:42,376 带我们领略了Android XR智能眼镜的魅力 1747 99:59:59,999 --> 99:59:59,999 1747 01:49:47,347 --> 01:49:49,416 有了这些Android XR设备 1748 99:59:59,999 --> 99:59:59,999 1748 01:49:49,483 --> 01:49:52,252 就好像身边有一位非常得力的助理 1749 99:59:59,999 --> 99:59:59,999 1749 01:49:52,319 --> 01:49:56,023 它将为我们带来怎样的精彩？让我们拭目以待 1750 99:59:59,999 --> 99:59:59,999 1750 01:49:56,523 --> 01:49:57,524 这还不是全部 1751 99:59:59,999 --> 99:59:59,999 1751 01:49:57,958 --> 01:50:01,662 我们正在与三星深化合作 1752 99:59:59,999 --> 99:59:59,999 1752 01:50:01,962 --> 01:50:06,467 将Android XR从头戴设备扩展到智能眼镜 1753 99:59:59,999 --> 99:59:59,999 1753 01:50:07,334 --> 01:50:10,637 我们正在创建软件和参考硬件平台 1754 99:59:59,999 --> 99:59:59,999 1754 01:50:10,771 --> 01:50:12,706 以便生态系统中的伙伴们能够与我们一起 1755 99:59:59,999 --> 99:59:59,999 1755 01:50:12,773 --> 01:50:15,142 打造出强大的智能眼镜 1756 99:59:59,999 --> 99:59:59,999 1756 01:50:16,443 --> 01:50:20,314 我们的眼镜原型目前正在由可信测试员试用 1757 99:59:59,999 --> 99:59:59,999 1757 01:50:20,514 --> 01:50:24,918 今年晚些时候，你们就可以开始为这款眼镜开发应用了 1758 99:59:59,999 --> 99:59:59,999 1758 01:50:26,720 --> 01:50:29,089 我们知道，智能眼镜必须足够时尚 1759 99:59:59,999 --> 99:59:59,999 1759 01:50:29,156 --> 01:50:31,391 你才会愿意全天佩戴 1760 99:59:59,999 --> 99:59:59,999 1760 01:50:32,159 --> 01:50:35,229 因此，今天，我们很高兴地宣布 1761 99:59:59,999 --> 99:59:59,999 1761 01:50:35,762 --> 01:50:38,899 Gentle Monster和Warby Parker 1762 99:59:59,999 --> 99:59:59,999 1762 01:50:38,966 --> 01:50:41,435 将成为我们的首批眼镜制造商合作伙伴 1763 99:59:59,999 --> 99:59:59,999 1763 01:50:41,502 --> 01:50:43,670 与我们一同打造搭载Android XR的时尚眼镜 1764 99:59:59,999 --> 99:59:59,999 1764 01:50:51,678 --> 01:50:55,249 我们希望你能戴上符合你个人品味的眼镜 1765 99:59:59,999 --> 99:59:59,999 1765 01:50:56,316 --> 01:50:57,518 这些还仅仅是开始 1766 99:59:59,999 --> 99:59:59,999 1766 01:50:58,051 --> 01:51:01,255 我热切期待你们亲自体验Android XR 1767 99:59:59,999 --> 99:59:59,999 1767 01:51:01,388 --> 01:51:03,991 在接下来的几个月，我们还会有更多精彩与大家分享 1768 99:59:59,999 --> 99:59:59,999 1768 01:51:04,491 --> 01:51:05,559 谢谢大家 1769 99:59:59,999 --> 99:59:59,999 1769 01:51:21,041 --> 01:51:23,343 这些XR眼镜实在太棒了！ 1770 99:59:59,999 --> 99:59:59,999 1770 01:51:24,077 --> 01:51:27,748 就在我们眼前，研究成果转化成了现实应用 1771 99:59:59,999 --> 99:59:59,999 1771 01:51:28,415 --> 01:51:30,584 好的，我们的演讲已经接近尾声 1772 99:59:59,999 --> 99:59:59,999 1772 01:51:31,251 --> 01:51:33,420 今天你们听到了很多关于Elo评分 1773 99:59:59,999 --> 99:59:59,999 1773 01:51:34,087 --> 01:51:36,523 基准测试和先进性能的内容 1774 99:59:59,999 --> 99:59:59,999 1774 01:51:37,224 --> 01:51:40,060 但我知道大家还在等着看另外一项数据 1775 99:59:59,999 --> 99:59:59,999 1775 01:51:40,527 --> 01:51:42,162 那就是AI统计的词频数据 1776 99:59:59,999 --> 99:59:59,999 1776 01:51:42,829 --> 01:51:45,699 那么，让我们来看这最后一个排行榜 1777 99:59:59,999 --> 99:59:59,999 1777 01:51:47,167 --> 01:51:50,971 看起来，又有新选手入围了 1778 99:59:59,999 --> 99:59:59,999 1778 01:51:51,572 --> 01:51:53,207 Gemini领先 1779 99:59:59,999 --> 99:59:59,999 1779 01:51:53,574 --> 01:51:56,076 以95次的成绩胜出 1780 99:59:59,999 --> 99:59:59,999 1780 01:51:57,711 --> 01:51:59,179 非常令人振奋 1781 99:59:59,999 --> 99:59:59,999 1781 01:52:02,916 --> 01:52:06,220 言归正传，以上就是我们今天分享的所有内容 1782 99:59:59,999 --> 99:59:59,999 1782 01:52:06,787 --> 01:52:12,059 从新产品发布、产品扩展，到未来展望 1783 99:59:59,999 --> 99:59:59,999 1783 01:52:12,593 --> 01:52:16,063 AI带来了前所未有的机遇 1784 99:59:59,999 --> 99:59:59,999 1784 01:52:16,897 --> 01:52:20,968 未来的关键就在你们这些开发者和技术创新者手中 1785 99:59:59,999 --> 99:59:59,999 1785 01:52:21,101 --> 01:52:25,505 你们将推动这项技术的普及，让更多人能从中受益 1786 99:59:59,999 --> 99:59:59,999 1786 01:52:26,173 --> 01:52:29,676 最后，我想跟大家分享几个给我带来激励的真实案例 1787 99:59:59,999 --> 99:59:59,999 1787 01:52:29,776 --> 01:52:33,113 首先是加州居民以及世界各地的人们 1788 99:59:59,999 --> 99:59:59,999 1788 01:52:33,513 --> 01:52:35,449 最关心的问题——野火 1789 99:59:59,999 --> 99:59:59,999 1789 01:52:36,183 --> 01:52:39,353 想必有许多人都认识野火的受灾者 1790 99:59:59,999 --> 99:59:59,999 1790 01:52:39,886 --> 01:52:43,657 野火的发生往往很突然，并在几分钟内失控蔓延 1791 99:59:59,999 --> 99:59:59,999 1791 01:52:44,391 --> 01:52:47,194 因此，野火检测的速度和精度至关重要 1792 99:59:59,999 --> 99:59:59,999 1792 01:52:47,828 --> 01:52:50,163 我们正在与一群了不起的合作伙伴一起 1793 99:59:59,999 --> 99:59:59,999 1793 01:52:50,631 --> 01:52:52,866 构建一个名为FireSat的项目 1794 99:59:59,999 --> 99:59:59,999 1794 01:52:52,966 --> 01:52:58,372 它是一个卫星星座，利用多光谱卫星图像和AI 1795 99:59:59,999 --> 99:59:59,999 1795 01:52:58,739 --> 01:53:01,742 旨在提供近乎实时的野火观测数据 1796 99:59:59,999 --> 99:59:59,999 1796 01:53:02,175 --> 01:53:03,710 看看这个分辨率 1797 99:59:59,999 --> 99:59:59,999 1797 01:53:04,111 --> 01:53:07,781 它可以探测到面积小至25平方米的火灾 1798 99:59:59,999 --> 99:59:59,999 1798 01:53:08,448 --> 01:53:10,617 大约相当于一个单车位车库的大小 1799 99:59:59,999 --> 99:59:59,999 1799 01:53:11,318 --> 01:53:13,353 我们的第一颗卫星现已入轨 1800 99:59:59,999 --> 99:59:59,999 1800 01:53:13,720 --> 01:53:15,289 项目全面投入运行后 1801 99:59:59,999 --> 99:59:59,999 1801 01:53:15,689 --> 01:53:18,759 图像更新频率将大幅提高 1802 99:59:59,999 --> 99:59:59,999 1802 01:53:19,192 --> 01:53:21,094 从目前的每12小时更新一次 1803 99:59:59,999 --> 99:59:59,999 1803 01:53:21,161 --> 01:53:23,030 缩短到每20分钟更新一次 1804 99:59:59,999 --> 99:59:59,999 1804 01:53:31,772 --> 01:53:34,741 在其他类型的应急响应中，速度同样至关重要 1805 99:59:59,999 --> 99:59:59,999 1805 01:53:35,375 --> 01:53:36,910 例如，在海伦飓风期间 1806 99:59:59,999 --> 99:59:59,999 1806 01:53:37,144 --> 01:53:40,047 Wing与沃尔玛、红十字会合作 1807 99:59:59,999 --> 99:59:59,999 1807 01:53:40,514 --> 01:53:43,317 用无人机投送物资，为救援工作提供了支持 1808 99:59:59,999 --> 99:59:59,999 1808 01:53:43,850 --> 01:53:45,118 在AI的帮助下 1809 99:59:59,999 --> 99:59:59,999 1809 01:53:45,252 --> 01:53:48,288 我们能够根据实时需求将食物、药品等关键物资 1810 99:59:59,999 --> 99:59:59,999 1810 01:53:48,755 --> 01:53:52,693 投放到北卡罗来纳州的一个基督教青年会避难所 1811 99:59:59,999 --> 99:59:59,999 1811 01:53:53,293 --> 01:53:57,331 可以想象这在其他社区的救灾行动中会有多大帮助 1812 99:59:59,999 --> 99:59:59,999 1812 01:53:57,998 --> 01:54:00,200 我们也正在积极努力扩大这一技术的应用范围 1813 99:59:59,999 --> 99:59:59,999 1813 01:54:00,967 --> 01:54:04,571 这些都是AI目前造福社会的例子 1814 99:59:59,999 --> 99:59:59,999 1814 01:54:05,339 --> 01:54:07,007 尤其令人鼓舞的是 1815 99:59:59,999 --> 99:59:59,999 1815 01:54:07,074 --> 01:54:11,511 今天的研究成果将在短短几年内成为现实 1816 99:59:59,999 --> 99:59:59,999 1816 01:54:12,279 --> 01:54:15,716 无论是构建下一代实用机器人 1817 99:59:59,999 --> 99:59:59,999 1817 01:54:16,583 --> 01:54:19,386 为世界致死率最高的疾病寻找治疗方法 1818 99:59:59,999 --> 99:59:59,999 1818 01:54:20,187 --> 01:54:22,889 还是推进纠错量子计算机的发展 1819 99:59:59,999 --> 99:59:59,999 1819 01:54:23,523 --> 01:54:28,695 或是打造能将乘客安全送达 任何地点的全自动驾驶汽车 1820 99:59:59,999 --> 99:59:59,999 1820 01:54:29,496 --> 01:54:33,533 所有这些都极有可能在几年内实现，而不是几十年 1821 99:59:59,999 --> 99:59:59,999 1821 01:54:33,867 --> 01:54:34,768 真是令人向往 1822 99:59:59,999 --> 99:59:59,999 1822 01:54:35,469 --> 01:54:39,239 这种利用AI改善生活的机会并不是我在想当然 1823 99:59:59,999 --> 99:59:59,999 1823 01:54:40,006 --> 01:54:42,409 最近的一次经历让我深有体会 1824 99:59:59,999 --> 99:59:59,999 1824 01:54:42,843 --> 01:54:45,278 当时我和父母一起到旧金山旅游 1825 99:59:59,999 --> 99:59:59,999 1825 01:54:45,712 --> 01:54:47,214 与许多其他游客一样 1826 99:59:59,999 --> 99:59:59,999 1826 01:54:47,280 --> 01:54:49,783 他们想做的第一件事就是体验Waymo的自动驾驶 1827 99:59:59,999 --> 99:59:59,999 1827 01:54:50,884 --> 01:54:52,386 我以前也坐过Waymo 1828 99:59:59,999 --> 99:59:59,999 1828 01:54:52,452 --> 01:54:57,591 但看着我八十多岁的父亲坐在前排那种惊讶的表情 1829 99:59:59,999 --> 99:59:59,999 1829 01:54:57,824 --> 01:55:00,727 我从全新的视角看到了科技进步的影响 1830 99:59:59,999 --> 99:59:59,999 1830 01:55:01,328 --> 01:55:04,931 它让我看到，科技拥有多么不可思议的力量 1831 99:59:59,999 --> 99:59:59,999 1831 01:55:04,998 --> 01:55:08,368 它能激发灵感、造福所有人、并推动我们前进 1832 99:59:59,999 --> 99:59:59,999 1832 01:55:08,769 --> 01:55:10,137 我迫不及待想看看 1833 99:59:59,999 --> 99:59:59,999 1833 01:55:10,203 --> 01:55:12,572 接下来我们会共同创造出怎样的精彩成果 1834 99:59:59,999 --> 99:59:59,999 1834 01:55:12,706 --> 01:55:13,540 谢谢大家 1835 99:59:59,999 --> 99:59:59,999 1835 01:55:26,019 --> 01:55:28,889 Flow为我开启了无限可能 1836 99:59:59,999 --> 99:59:59,999 1836 01:55:29,055 --> 01:55:30,957 那是什么样的魔法？ 1837 99:59:59,999 --> 99:59:59,999 1837 01:55:31,491 --> 01:55:34,928 AI模式是自搜索引擎发明以来最伟大的革命 1838 99:59:59,999 --> 99:59:59,999 1838 01:55:35,162 --> 01:55:37,831 只需准确地告诉Gemini你想要构建什么 1839 99:59:59,999 --> 99:59:59,999 1839 01:55:37,898 --> 01:55:39,332 它就会为你搞定 1840 99:59:59,999 --> 99:59:59,999 1840 01:55:39,399 --> 01:55:40,233 天哪 1841 99:59:59,999 --> 99:59:59,999 1841 01:55:40,300 --> 01:55:42,536 你可以用Gemini 2.5 Pro编程 1842 99:59:59,999 --> 99:59:59,999 1842 01:55:42,936 --> 01:55:44,671 实在太好用了 1843 99:59:59,999 --> 99:59:59,999 1843 01:55:44,738 --> 01:55:47,040 Project Mariner将改变我们使用浏览器的方式 1844 99:59:59,999 --> 99:59:59,999 1844 01:55:47,107 --> 01:55:50,177 它是一个能帮你处理事务的AI智能体 1845 99:59:59,999 --> 99:59:59,999 1845 01:55:50,243 --> 01:55:53,346 Notebook LMS思维导图颠覆了我们学习的方式 1846 99:59:59,999 --> 99:59:59,999 1846 01:55:53,413 --> 01:55:56,216 为什么无论我走到哪，这个人都在跟踪我？ 1847 99:59:59,999 --> 99:59:59,999 1847 01:55:56,450 --> 01:55:57,784 那是你的影子 1848 99:59:59,999 --> 99:59:59,999 1848 01:55:58,118 --> 01:55:58,985 等等，你要右转 1849 99:59:59,999 --> 99:59:59,999 1849 01:56:01,354 --> 01:56:03,957 AI正在彻底改变我们对抗野火的方式 1850 99:59:59,999 --> 99:59:59,999 1850 01:56:04,224 --> 01:56:07,427 生成式媒体正在拓展创造力的边界 1851 99:59:59,999 --> 99:59:59,999 1851 01:56:07,794 --> 01:56:09,362 我第一次看到这个的时候，感觉就像魔法一样 1852 99:59:59,999 --> 99:59:59,999 1852 01:56:09,563 --> 01:56:10,597 它不仅仅是又一项工具 1853 99:59:59,999 --> 99:59:59,999 1853 01:56:10,664 --> 01:56:11,631 它颠覆了一切 1854 99:59:59,999 --> 99:59:59,999 1854 01:56:11,698 --> 01:56:13,400 简直让我大开眼界 1855 99:59:59,999 --> 99:59:59,999 1855 01:56:13,467 --> 01:56:15,435 AI的潜力近乎无限