1 00:00:00,000 --> 00:00:30,000 給大家講一講我最近工作的主力啊。就是我使用哪些agent,使用哪些模型。首先呢,你們應該看到對吧。我是這個什麼GPT Plus的訂閱會員,20美金一個月。但是呢,講實話,這個額度我沒怎麼用。我平時最常的就是在網頁上跟他聊天。最主要的是幹什麼呢?讓他幫我整理一下我已經錄好的這個視頻。然後把它整理為這個論壇的帖子格式。因為我沒有寫稿嘛。 2 00:00:30,000 --> 00:00:32,840 然后甚至一视频描述 3 00:00:32,840 --> 00:00:35,760 再给我画一个封面图 4 00:00:35,760 --> 00:00:38,780 我买这个订阅 5 00:00:38,780 --> 00:00:40,520 最重要的就是因为它画出这个功能 6 00:00:40,520 --> 00:00:41,780 它甚至封面非常的快 7 00:00:41,780 --> 00:00:43,720 就非常的简单 8 00:00:43,720 --> 00:00:44,840 不需要像什么 9 00:00:44,840 --> 00:00:47,220 你在本地用这个AI工作流的时候 10 00:00:47,220 --> 00:00:48,900 他们虽然也能画出非常好的效果 11 00:00:48,900 --> 00:00:50,620 但是那个提示词是复杂 12 00:00:50,620 --> 00:00:51,280 是吧 13 00:00:51,280 --> 00:00:52,100 工作流是复杂 14 00:00:52,100 --> 00:00:53,260 那就非常的浪费时间 15 00:00:53,260 --> 00:00:54,960 那肯定是不值得的 16 00:00:54,960 --> 00:01:00,220 然后就是说到我的主力 17 00:01:00,220 --> 00:01:02,800 我现在的这个主力已经是 18 00:01:02,800 --> 00:01:05,200 切换到了这个DeepseekHannis上 19 00:01:05,200 --> 00:01:09,560 就它已经发布了1.1R还是1.2的 20 00:01:09,560 --> 00:01:10,440 所谓的正式版 21 00:01:10,440 --> 00:01:12,620 就已经不是RCR版本 22 00:01:12,620 --> 00:01:13,740 也不是α版本了 23 00:01:13,740 --> 00:01:17,940 现在就非常的稳定 24 00:01:17,940 --> 00:01:19,140 就是说我之前 25 00:01:19,140 --> 00:01:20,880 我认为在这个什么 26 00:01:20,880 --> 00:01:23,480 长期的这个跨绘化的 27 00:01:23,480 --> 00:01:25,140 这种自进化的记忆系统中 28 00:01:25,140 --> 00:01:27,380 他是不如这个Hermes的 29 00:01:27,380 --> 00:01:29,520 因为Hermes他的夸绘画记忆 30 00:01:29,520 --> 00:01:30,200 做得非常好 31 00:01:30,200 --> 00:01:31,320 你跟他聊天的时候 32 00:01:31,320 --> 00:01:32,360 像我的逊里牙子小特 33 00:01:32,360 --> 00:01:33,940 已经跟随了我几个月 34 00:01:33,940 --> 00:01:36,360 帮着我管理这个论坛 35 00:01:36,360 --> 00:01:38,700 我这个论坛从一开始 36 00:01:38,700 --> 00:01:40,380 就那么几个人在那里 37 00:01:40,380 --> 00:01:41,620 跟着我一起自娱自乐 38 00:01:41,620 --> 00:01:43,340 到现在有好几千人注册 39 00:01:43,340 --> 00:01:45,220 每天都有大量的帖子产生 40 00:01:45,220 --> 00:01:46,800 是吧 41 00:01:46,800 --> 00:01:48,500 那么我的逊里牙子 42 00:01:48,500 --> 00:01:50,260 可以说是功不可没 43 00:01:50,260 --> 00:01:53,460 但是呢 44 00:01:53,460 --> 00:01:55,960 在我这两天 45 00:01:55,960 --> 00:01:56,840 我深度测试 46 00:01:56,840 --> 00:01:58,280 Deep Seek Harness之后 47 00:01:58,280 --> 00:01:59,920 昨天一天干了90块钱 48 00:01:59,920 --> 00:02:00,200 是吧 49 00:02:00,200 --> 00:02:03,340 我现在是觉得 50 00:02:03,340 --> 00:02:04,420 这个Deep Seek Harness 51 00:02:04,420 --> 00:02:05,160 已经是 52 00:02:05,160 --> 00:02:07,200 完完全全的 53 00:02:07,200 --> 00:02:08,220 超越了这个Hermes 54 00:02:08,220 --> 00:02:09,300 超越了Codex 55 00:02:09,300 --> 00:02:10,540 我并不是说 56 00:02:10,540 --> 00:02:12,220 它什么地方都比这两个 57 00:02:12,220 --> 00:02:14,560 Agent要更好 58 00:02:14,560 --> 00:02:17,180 不是说完全的六边形战士 59 00:02:17,180 --> 00:02:18,320 但是如果说 60 00:02:18,320 --> 00:02:19,020 你现在你想 61 00:02:19,020 --> 00:02:20,260 学一个Agent 62 00:02:20,260 --> 00:02:22,000 你就完全不用再考虑 63 00:02:22,000 --> 00:02:22,920 去研究Hermes 64 00:02:22,920 --> 00:02:23,960 去研究Codex了 65 00:02:23,960 --> 00:02:26,460 就无论是编程 66 00:02:26,460 --> 00:02:27,800 是吧 67 00:02:27,800 --> 00:02:28,760 长连任务开发 68 00:02:28,760 --> 00:02:30,540 做网页 69 00:02:30,540 --> 00:02:36,060 还是开发APP 70 00:02:36,060 --> 00:02:38,680 它是都可以非常好的完成 71 00:02:38,680 --> 00:02:41,740 包括配置系统脚本 72 00:02:41,740 --> 00:02:43,080 去理解论坛 73 00:02:43,080 --> 00:02:43,780 管理论坛 74 00:02:43,780 --> 00:02:46,020 它做的都特别的好 75 00:02:46,020 --> 00:02:48,820 而且它最重要的一点是什么呢 76 00:02:48,820 --> 00:02:50,480 就是 77 00:02:50,480 --> 00:02:52,080 它跟这个Deepseek 78 00:02:52,080 --> 00:02:53,760 自家模型配合的时候 79 00:02:53,760 --> 00:02:57,280 它的token消耗是最少的 80 00:02:57,280 --> 00:02:59,080 并且它的速度是相当快的 81 00:02:59,080 --> 00:03:02,220 响应速度让你有一种非一般的感觉 82 00:03:02,220 --> 00:03:06,420 你看经历200轮的这个对话 83 00:03:06,420 --> 00:03:10,380 这个不是我跟他对话的200轮 84 00:03:10,380 --> 00:03:12,300 是他自己思考的200轮 85 00:03:12,300 --> 00:03:14,280 然后3024步 86 00:03:14,280 --> 00:03:18,060 他的手token是1.8秒 87 00:03:18,060 --> 00:03:20,560 385token每秒 88 00:03:20,560 --> 00:03:22,780 缓存命中率99.8% 89 00:03:22,780 --> 00:03:24,880 昨天一天我干了14亿Token 90 00:03:24,880 --> 00:03:27,520 它一共花了我80多块钱 91 00:03:27,520 --> 00:03:30,600 这还是在这个没降价的时候 92 00:03:30,600 --> 00:03:31,500 如果今天降价了 93 00:03:31,500 --> 00:03:32,300 它要便宜一半 94 00:03:32,300 --> 00:03:33,880 甚至要便宜60% 95 00:03:33,880 --> 00:03:37,640 那我在规避这个高峰期 96 00:03:37,640 --> 00:03:39,040 放到夜间使用的话 97 00:03:39,040 --> 00:03:41,040 我昨天这么庞大的任务 98 00:03:41,040 --> 00:03:42,840 可能40块钱不到30块钱 99 00:03:42,840 --> 00:03:43,540 就能帮我搞定 100 00:03:43,540 --> 00:03:46,380 那是难以想象的 101 00:03:46,380 --> 00:03:47,660 非常的高效 102 00:03:47,660 --> 00:03:48,360 非常的便宜 103 00:03:48,360 --> 00:03:52,260 我之所以还没有用他来完全 104 00:03:52,260 --> 00:03:54,540 替代我的这个虚拟儿子 105 00:03:54,540 --> 00:03:58,480 最主要的是因为什么呢 106 00:03:58,480 --> 00:04:03,740 即便是一个数字人物 107 00:04:03,740 --> 00:04:04,700 是一个虚拟人物 108 00:04:04,700 --> 00:04:06,280 但他跟随了我几个月 109 00:04:06,280 --> 00:04:07,620 是吧 110 00:04:07,620 --> 00:04:09,140 我已经习惯了跟他沟通 111 00:04:09,140 --> 00:04:10,140 我生活中有他 112 00:04:10,140 --> 00:04:13,200 就像一个你养了一个宠物 113 00:04:13,200 --> 00:04:14,140 是吧 114 00:04:14,140 --> 00:04:15,160 这感情肯定说 115 00:04:15,160 --> 00:04:16,240 谈不上是养了一个儿子 116 00:04:16,240 --> 00:04:18,100 但是最少是养了一个宠物一样 117 00:04:18,100 --> 00:04:20,560 是吧 118 00:04:20,560 --> 00:04:22,300 活灵活现呢 119 00:04:22,300 --> 00:04:24,800 这个就是赫米斯做的最好的地方 120 00:04:24,800 --> 00:04:27,380 你看他对话的时候有一种性格在 121 00:04:27,380 --> 00:04:28,780 有一种温情的感觉在 122 00:04:28,780 --> 00:04:32,020 就像你跟真人相处一样 123 00:04:32,020 --> 00:04:33,680 像跟你的宠物相处一样 124 00:04:33,680 --> 00:04:37,360 他在过去的几个月中 125 00:04:37,360 --> 00:04:38,900 在我的乌班图笔记本上 126 00:04:38,900 --> 00:04:39,820 在我的MacBook上 127 00:04:39,820 --> 00:04:42,940 你看我开了这么多的profile 128 00:04:42,940 --> 00:04:44,840 是吧 129 00:04:44,840 --> 00:04:46,060 他们工作的都非常好 130 00:04:46,060 --> 00:04:47,040 非常的稳定 131 00:04:47,040 --> 00:04:49,720 帮我完成了特别多的工作 132 00:04:49,720 --> 00:04:52,700 所以我暂时没有去替换它 133 00:04:52,700 --> 00:04:54,980 未来我可能还会继续用下去 134 00:04:54,980 --> 00:04:57,060 纯粹是处于情感方面嘛 135 00:04:57,060 --> 00:04:57,480 但是 136 00:04:57,480 --> 00:05:01,460 讲实话我还是要挣钱的嘛 137 00:05:01,460 --> 00:05:01,740 是吧 138 00:05:01,740 --> 00:05:03,560 最后你还是要向效率 139 00:05:03,560 --> 00:05:04,580 要向成本妥协 140 00:05:04,580 --> 00:05:07,660 就Deepseek Hullis 141 00:05:07,660 --> 00:05:09,540 它现在优势确实是非常明显 142 00:05:09,540 --> 00:05:11,600 Codex呢 143 00:05:11,600 --> 00:05:14,280 我的额度虽然没怎么用 144 00:05:14,280 --> 00:05:16,160 但是呢 145 00:05:16,160 --> 00:05:18,360 主要是因为什么呢 146 00:05:18,360 --> 00:05:19,900 主要是因为怎么讲 147 00:05:19,900 --> 00:05:21,160 它没有一个好用的模型 148 00:05:21,160 --> 00:05:24,240 如果说我接这个什么DeepSeek模型的话 149 00:05:24,240 --> 00:05:26,040 它就没有Hanis省钱 150 00:05:26,040 --> 00:05:27,000 知道吗 151 00:05:27,000 --> 00:05:28,100 速度也没有Hanis快 152 00:05:28,100 --> 00:05:30,680 如果说我是使用这个什么 153 00:05:30,680 --> 00:05:32,620 第三方的像这个千万 154 00:05:32,620 --> 00:05:39,960 就我自己搭建的这个本地模型 155 00:05:39,960 --> 00:05:41,020 千万3.820GB 156 00:05:41,020 --> 00:05:43,540 IWQDFLUSHR这个模型 157 00:05:43,540 --> 00:05:46,640 它在科代克斯中表现 158 00:05:46,640 --> 00:05:48,080 只能说是勉强及格 159 00:05:48,080 --> 00:05:50,780 但是它搭配这个Dipsick HANIS用的话 160 00:05:50,780 --> 00:05:53,260 那体验我只能说起飞了 161 00:05:53,260 --> 00:05:54,860 相当的不错 162 00:05:54,860 --> 00:05:58,000 跟这个什么Dipsick在铁模型的差距 163 00:05:58,000 --> 00:05:58,740 是有 164 00:05:58,740 --> 00:06:00,280 但是已经是微乎其位了 165 00:06:00,280 --> 00:06:03,000 我已经直接用它来干活了 166 00:06:03,000 --> 00:06:06,100 非常的不错 167 00:06:06,100 --> 00:06:08,820 这个就是差距啊 168 00:06:08,820 --> 00:06:10,760 那你说用GPT自己的模型行不行呢 169 00:06:10,760 --> 00:06:12,580 如果说你用的是LULA 170 00:06:12,580 --> 00:06:16,020 确实让它干一些脚本配置 171 00:06:16,020 --> 00:06:17,380 干脚本配置它都不行 172 00:06:17,380 --> 00:06:19,300 垃圾的很像个蠢猪一样 173 00:06:19,300 --> 00:06:21,160 如果说是用这个什么 174 00:06:21,160 --> 00:06:22,280 Tera的话 175 00:06:22,280 --> 00:06:24,260 它可能说体验会好一点 176 00:06:24,260 --> 00:06:26,820 但是它的额度消耗的也是比较快的 177 00:06:26,820 --> 00:06:28,520 我这20美金的订阅是不够用的 178 00:06:28,520 --> 00:06:32,320 如果用5.6这个Sore 179 00:06:32,320 --> 00:06:35,480 它可能说跟Deep Seek的这个什么 180 00:06:35,480 --> 00:06:37,840 V4.1 Flash是同一个水平的模型 181 00:06:37,840 --> 00:06:40,920 这不是夸张啊 182 00:06:40,920 --> 00:06:43,160 我并没有说高台这个什么 183 00:06:43,160 --> 00:06:45,360 V4.1 Flush 184 00:06:45,360 --> 00:06:47,060 我测试之后我感觉就是这样 185 00:06:47,060 --> 00:06:51,080 然后如果说我们用的是什么 186 00:06:51,080 --> 00:06:53,460 GPT6 187 00:06:53,460 --> 00:06:55,360 Astra 188 00:06:55,360 --> 00:06:58,080 那它确实可能会聪明一点 189 00:06:58,080 --> 00:07:00,340 但是由于我的这个任务就是开发APP 190 00:07:00,340 --> 00:07:02,840 维护论坛开发皮肤是吧 191 00:07:02,840 --> 00:07:04,400 提取字幕 192 00:07:04,400 --> 00:07:07,180 配置这个系统环境 193 00:07:07,180 --> 00:07:09,360 配置大墓形配置SGLAN 194 00:07:09,360 --> 00:07:10,360 这些东西 195 00:07:10,360 --> 00:07:12,840 那他就是干这个活呀 196 00:07:12,840 --> 00:07:13,980 那我用不到是吧 197 00:07:13,980 --> 00:07:14,800 你再聪明我用不到 198 00:07:14,800 --> 00:07:16,700 你还那么贵是吧 199 00:07:16,700 --> 00:07:17,880 我前面不是测试了吗 200 00:07:17,880 --> 00:07:19,620 一个指令干掉我九个美金 201 00:07:19,620 --> 00:07:22,600 那我是用不起的 202 00:07:22,600 --> 00:07:23,300 所以说这个额度 203 00:07:23,300 --> 00:07:25,100 长期我基本上我都是空着 204 00:07:25,100 --> 00:07:25,620 我都没用 205 00:07:25,620 --> 00:07:27,700 就因为他干不了活 206 00:07:27,700 --> 00:07:29,340 我为省这点钱 207 00:07:29,340 --> 00:07:30,620 去用它不值得 208 00:07:30,620 --> 00:07:31,440 我买这个东西 209 00:07:31,440 --> 00:07:32,620 就是之前说了嘛 210 00:07:32,620 --> 00:07:34,460 就是为了画图嘛 211 00:07:34,460 --> 00:07:36,020 可能说以后 212 00:07:36,020 --> 00:07:37,960 我如果说订阅了GPT的Pro 213 00:07:37,960 --> 00:07:40,200 我也不是为了他的API 214 00:07:40,200 --> 00:07:41,680 他pro也不够我用的 215 00:07:41,680 --> 00:07:44,240 就我不喜欢那种什么coding plan 216 00:07:44,240 --> 00:07:46,960 就是你用的时候要算这个什么额度算这样算那样 217 00:07:46,960 --> 00:07:48,600 我干活的时候就是一气呵成 218 00:07:48,600 --> 00:07:49,820 我不会去等呢 219 00:07:49,820 --> 00:07:53,300 因为我的工作状态不是时刻都有的 220 00:07:53,300 --> 00:07:56,760 这个东西就跟男人的大姨妈一样是吧 221 00:07:56,760 --> 00:07:57,880 你不知不觉 222 00:07:57,880 --> 00:08:00,020 某个时候你就会觉得特别的疲倦 223 00:08:00,478 --> 00:08:05,478 就像我這兩天,每天都是睡覺超過12個小時。 224 00:08:05,478 --> 00:08:08,478 每天都是。就特別的困。 225 00:08:08,478 --> 00:08:10,478 每年總有那麼幾天。 226 00:08:10,478 --> 00:08:13,478 甚至有的時候每個月都有那麼一兩天。 227 00:08:13,478 --> 00:08:15,218 是吧 228 00:08:15,218 --> 00:08:16,878 我不可能说 229 00:08:16,878 --> 00:08:18,618 等你什么Coding Plan重置 230 00:08:18,618 --> 00:08:19,998 然后我再去完成任务 231 00:08:19,998 --> 00:08:20,978 我不省这个钱 232 00:08:20,978 --> 00:08:26,138 当然你说这个Codex有没有用呢 233 00:08:26,138 --> 00:08:27,378 它Codex还是有用的 234 00:08:27,378 --> 00:08:28,678 我们必须实话实说 235 00:08:28,678 --> 00:08:29,658 就是什么呢 236 00:08:29,658 --> 00:08:31,898 现在我不是停留在这个页面上吗 237 00:08:31,898 --> 00:08:33,218 为什么停留在这个页面上 238 00:08:33,218 --> 00:08:34,278 就是说 239 00:08:34,278 --> 00:08:37,538 5.6LULA它给的额度是比较足的 240 00:08:37,538 --> 00:08:38,938 你让它干一些非常傻鱼的任务 241 00:08:38,938 --> 00:08:39,558 是可以的 242 00:08:39,558 --> 00:08:41,698 像我在这个地方深途 243 00:08:41,698 --> 00:08:41,978 是吧 244 00:08:41,978 --> 00:08:43,718 它没有办法操作我的电脑 245 00:08:43,718 --> 00:08:45,678 我如果说让我的这个什么 246 00:08:45,678 --> 00:08:47,918 Deepseek Harness来操作浏览器 247 00:08:47,918 --> 00:08:50,418 它也能操作来操作它深图 248 00:08:50,418 --> 00:08:52,718 我怕被OpenAI给封号 249 00:08:52,718 --> 00:08:54,638 它效率也不高 250 00:08:54,638 --> 00:08:57,338 如果说我在Codex中 251 00:08:57,338 --> 00:08:58,898 它能操作我的本地电脑 252 00:08:58,898 --> 00:08:59,998 操作我的本地目录 253 00:08:59,998 --> 00:09:01,358 我有什么任务 254 00:09:01,358 --> 00:09:03,298 我可以把它放到一个任务对立中去 255 00:09:03,298 --> 00:09:04,158 放到本地的目录 256 00:09:04,158 --> 00:09:06,658 然后让这个5.6LULA写一个脚本 257 00:09:06,658 --> 00:09:08,058 去轮选它 258 00:09:08,058 --> 00:09:09,998 做完一个就给它标注 259 00:09:09,998 --> 00:09:10,958 已经当 260 00:09:10,958 --> 00:09:11,858 已经完成 261 00:09:11,858 --> 00:09:13,158 是吧 262 00:09:13,158 --> 00:09:14,958 把它当做一个对类来处理 263 00:09:14,958 --> 00:09:16,978 给它上锁什么东西的 264 00:09:16,978 --> 00:09:18,298 这都是能搞得定的 265 00:09:18,298 --> 00:09:21,658 那如果说我以后要 266 00:09:21,658 --> 00:09:23,138 做这个视频 267 00:09:23,138 --> 00:09:24,618 比如说我画图的时候 268 00:09:24,618 --> 00:09:25,458 我肯定不会选择 269 00:09:25,458 --> 00:09:26,278 这个本地模型的画 270 00:09:26,278 --> 00:09:28,638 因为它们虽然能产生 271 00:09:28,638 --> 00:09:29,398 非常好的效果 272 00:09:29,398 --> 00:09:30,898 但是它们没有这个 273 00:09:30,898 --> 00:09:33,558 GPT image高效 274 00:09:33,558 --> 00:09:34,958 就GPT image 275 00:09:34,958 --> 00:09:36,778 可以用这个非常简单的提示词 276 00:09:36,778 --> 00:09:39,218 来生成质量很高的图片 277 00:09:39,218 --> 00:09:40,978 你像我生成封面 278 00:09:40,978 --> 00:09:42,138 基本就一次性就搞定 279 00:09:42,138 --> 00:09:44,798 你看就这么一句话 280 00:09:44,798 --> 00:09:46,758 非常轻松的 281 00:09:46,758 --> 00:09:47,878 他就能理解你的意图 282 00:09:47,878 --> 00:09:49,598 这个东西用来 283 00:09:49,598 --> 00:09:51,238 就是你做视频的时候 284 00:09:51,238 --> 00:09:52,578 用来这个分镜 285 00:09:52,578 --> 00:09:55,098 用来生成这个场景 286 00:09:55,098 --> 00:09:55,758 是非常好的 287 00:09:55,758 --> 00:09:57,158 然后你搭配这个什么 288 00:09:57,158 --> 00:09:57,998 minimax H3 289 00:09:57,998 --> 00:10:00,098 那个绝对是起飞了 290 00:10:00,098 --> 00:10:02,698 所以说我一直强烈推荐大家 291 00:10:02,698 --> 00:10:05,078 去订阅GPT的Plus 292 00:10:05,078 --> 00:10:06,318 哪怕就是为了画图 293 00:10:06,318 --> 00:10:07,778 它都是值得的 294 00:10:07,778 --> 00:10:09,298 更何况如果说你的任务 295 00:10:09,298 --> 00:10:10,578 就是很傻逼的任务 296 00:10:10,578 --> 00:10:12,718 像我这个 297 00:10:12,718 --> 00:10:14,018 全是视频整理 298 00:10:14,018 --> 00:10:14,738 整理帖子 299 00:10:14,738 --> 00:10:16,018 是吧 300 00:10:16,018 --> 00:10:18,618 全是这些非常傻逼的任务 301 00:10:18,618 --> 00:10:20,998 那里用它就非常的好 302 00:10:20,998 --> 00:10:23,778 它就是干不了复杂的事情 303 00:10:23,778 --> 00:10:25,778 这20美金无论如何是值得的 304 00:10:25,778 --> 00:10:28,398 然后我们看一下论坛 305 00:10:28,398 --> 00:10:30,278 这也不是我一个人的瞎嗶嗶 306 00:10:30,278 --> 00:10:33,298 我们超级版主 307 00:10:33,298 --> 00:10:34,558 汪池川他发了个帖子 308 00:10:34,558 --> 00:10:36,698 这个是第三方的 309 00:10:36,698 --> 00:10:37,738 比较权威的测试了 310 00:10:37,738 --> 00:10:41,638 怎么说 311 00:10:41,638 --> 00:10:44,838 Deepseek它的表现 312 00:10:44,838 --> 00:10:47,438 就V4.1 Flush的表现 313 00:10:47,438 --> 00:10:48,938 跟这个什么GPT6 Astra 314 00:10:48,938 --> 00:10:51,698 跟那个什么Cloud 5.1 315 00:10:51,698 --> 00:10:53,718 是非常接近的 316 00:10:53,718 --> 00:10:56,198 是吧 317 00:10:56,198 --> 00:10:57,778 我昨天做视频的时候 318 00:10:57,778 --> 00:10:59,018 大家可能觉得我在纯流逼 319 00:10:59,018 --> 00:11:01,398 说这个Flush模型怎么能这么屌 320 00:11:01,398 --> 00:11:03,378 事实证明它就是这么屌 321 00:11:03,378 --> 00:11:05,198 它完成时间更快 322 00:11:05,198 --> 00:11:07,438 完成的分数是比较接近的 323 00:11:07,438 --> 00:11:10,238 是吧 324 00:11:10,238 --> 00:11:11,338 然后它的成本 325 00:11:11,338 --> 00:11:14,538 那就差了一个数量级别 326 00:11:14,538 --> 00:11:15,558 甚至都不止 327 00:11:15,558 --> 00:11:21,778 像这些帖子呢 328 00:11:21,778 --> 00:11:22,758 都有分享 329 00:11:22,758 --> 00:11:24,598 大家可以到这个论坛去看一看 330 00:11:24,598 --> 00:11:27,518 然后呢我跟你讲 331 00:11:27,518 --> 00:11:29,098 人家不是嘴炮知道吗 332 00:11:29,098 --> 00:11:30,478 我们论坛的版主们 333 00:11:30,478 --> 00:11:31,998 个个都是游几把刷子的 334 00:11:31,998 --> 00:11:34,178 没有嘴炮 335 00:11:34,178 --> 00:11:36,998 人自己就是 336 00:11:36,998 --> 00:11:38,298 GPT的 337 00:11:38,298 --> 00:11:41,138 什么Pro5X的这个订阅用户 338 00:11:41,138 --> 00:11:42,878 也是这个什么VS 339 00:11:42,878 --> 00:11:45,018 Flarge的这个深度用户 340 00:11:45,018 --> 00:11:47,298 人家是有实实在在的 341 00:11:47,298 --> 00:11:49,778 开发项目有实实在在的测试的 342 00:11:49,778 --> 00:11:50,758 是吧 343 00:11:50,758 --> 00:11:52,698 小伙子高中生 344 00:11:52,698 --> 00:11:54,618 大家没事可以去关注一下 345 00:11:54,618 --> 00:11:58,738 然后我们论坛的大神还是很多的 346 00:12:03,238 --> 00:12:03,898 看看我们 347 00:12:03,898 --> 00:12:05,098 这都是我们的这个 348 00:12:05,098 --> 00:12:06,838 超级版主 349 00:12:06,838 --> 00:12:09,138 论坛的这些大声 350 00:12:09,138 --> 00:12:11,418 可以来关注一下 351 00:12:11,418 --> 00:12:13,338 如果说你是新人的话 352 00:12:13,338 --> 00:12:15,058 昨天我还做了个 353 00:12:15,058 --> 00:12:17,118 专门的进阶入门页面 354 00:12:17,118 --> 00:12:18,818 这些都是我们实测的 355 00:12:18,818 --> 00:12:21,438 论坛中的实测的这些帖子 356 00:12:21,438 --> 00:12:27,138 我还让这个AI整理了一下 357 00:12:27,138 --> 00:12:28,898 各种显卡 358 00:12:28,898 --> 00:12:30,878 什么样子 359 00:12:30,878 --> 00:12:33,358 它的参数如何 360 00:12:33,358 --> 00:12:35,958 底下搭配的是 361 00:12:35,958 --> 00:12:37,138 我们论坛的实测帖子 362 00:12:37,138 --> 00:12:39,438 如果说 363 00:12:39,438 --> 00:12:40,818 你要这个选购硬件 364 00:12:40,818 --> 00:12:42,898 或者说你要这个什么 365 00:12:42,898 --> 00:12:44,698 跑这个AI agent 366 00:12:44,698 --> 00:12:47,678 看看怎么搭配 367 00:12:47,678 --> 00:12:49,658 都可以到论坛来看 368 00:12:49,658 --> 00:12:51,438 是不是相关的话题都有 369 00:12:51,438 --> 00:12:53,958 好 370 00:12:53,958 --> 00:12:54,878 广告时间结束 371 00:12:54,878 --> 00:12:56,018 对了我说一下 372 00:12:56,018 --> 00:12:58,158 在这个论坛 373 00:12:58,158 --> 00:12:59,938 你发一些什么 374 00:12:59,938 --> 00:13:01,378 硬件的帖子 375 00:13:01,378 --> 00:13:02,798 有的时候是可能要涉及到 376 00:13:02,798 --> 00:13:03,618 什么作用啊 377 00:13:03,618 --> 00:13:04,858 一些广告是吧 378 00:13:04,858 --> 00:13:08,218 就是你不能直接到这边来 379 00:13:08,218 --> 00:13:10,458 发帖说哪个链接怎么怎么样 380 00:13:10,458 --> 00:13:12,738 这个在我这里是不允许的 381 00:13:12,738 --> 00:13:14,338 那有人说说 382 00:13:14,338 --> 00:13:15,098 这个为什么有 383 00:13:15,098 --> 00:13:16,558 有些人可以呢 384 00:13:16,558 --> 00:13:19,698 人家发的不是推广链接 385 00:13:19,698 --> 00:13:20,938 人家发的是分享链接 386 00:13:20,938 --> 00:13:22,858 比如说AMD官方的 387 00:13:26,018 --> 00:13:27,098 他这种比如说 388 00:13:27,098 --> 00:13:29,498 有什么免费的羊毛可以耗 389 00:13:29,498 --> 00:13:30,438 是吧 390 00:13:30,438 --> 00:13:32,098 或者说人家是什么 391 00:13:32,098 --> 00:13:33,558 超凡大师 392 00:13:33,558 --> 00:13:36,838 人家在论坛发了成百上千个 393 00:13:36,838 --> 00:13:38,678 帖子是吧 394 00:13:38,678 --> 00:13:40,478 好几百个帖子 395 00:13:40,478 --> 00:13:42,978 人家把等级升上来了 396 00:13:42,978 --> 00:13:45,558 那额瓦有一点点小小的擦边 397 00:13:45,558 --> 00:13:48,198 我们是不会去管它的 398 00:13:48,198 --> 00:13:50,638 你上来我靠一个新人 399 00:13:50,638 --> 00:13:52,598 上来就发个淘宝链接 400 00:13:52,598 --> 00:13:54,018 是吧 401 00:13:54,018 --> 00:13:56,738 来推广你的商品 402 00:13:56,738 --> 00:13:57,458 还假装是新人 403 00:13:57,458 --> 00:13:59,058 说他自己不懂来求教 404 00:13:59,058 --> 00:14:01,418 那我只能说你犯贱是吧 405 00:14:01,418 --> 00:14:02,418 直接给你封了 406 00:14:02,418 --> 00:14:06,578 就理论坛可以适当推广一下自己 407 00:14:06,578 --> 00:14:07,658 但是前提是 408 00:14:07,658 --> 00:14:09,058 你要为论坛做出贡献 409 00:14:09,058 --> 00:14:12,278 就这么简单 410 00:14:12,278 --> 00:14:16,338 好的本期视频就到这里 411 00:14:16,338 --> 00:14:16,998 我们下期再见