1
00:00:00,000 --> 00:00:07,200
這是我剛發布的Cloud Ops 5.5做的3D程式生成器。你覺得怎麼樣?

2
00:00:07,200 --> 00:00:24,200
官方號稱在大多數任務上都能打平自家最強的Fable 5.1。成本卻比上一代Ops 5還低40%。不愧是A2,呼籲AI前沿發展放慢節奏,自己卻偷偷卷。

3
00:00:24,200 --> 00:00:27,660
紧接着 OpenAI也发布了GPT-6 SAUP

4
00:00:27,660 --> 00:00:31,220
号称继承了GPT-6 AUSTRA大部分优势

5
00:00:31,220 --> 00:00:35,460
API的价格却比GPT-5.6直接砍了一半

6
00:00:35,460 --> 00:00:38,800
这个历史似曾相识啊

7
00:00:38,800 --> 00:00:42,800
两大顶级AI公司再次上演中门对局

8
00:00:42,800 --> 00:00:46,060
这次我不想再花时间科普跑分了

9
00:00:46,060 --> 00:00:48,360
新模型太多已经麻了

10
00:00:48,360 --> 00:00:51,880
感兴趣的朋友直接看我的AI大模型世界网站

11
00:00:51,880 --> 00:00:53,300
上的对比就好

12
00:00:53,300 --> 00:00:56,700
本期我准备了五大案例实战测评

13
00:00:56,700 --> 00:00:58,520
点着三连 咱们开始

14
00:00:58,520 --> 00:01:03,440
首先是AI圈的经典测试

15
00:01:03,440 --> 00:01:05,400
提湖骑自行车

16
00:01:05,400 --> 00:01:07,600
那为了防止AI直接备答案

17
00:01:07,600 --> 00:01:09,420
我把题目改成了

18
00:01:09,420 --> 00:01:11,320
熊猫骑自行车送外卖

19
00:01:11,320 --> 00:01:13,380
直接让AI生成代码

20
00:01:13,380 --> 00:01:16,100
两种模型都在三分钟内完成了

21
00:01:16,100 --> 00:01:18,520
你觉得哪边的效果正好呢

22
00:01:18,520 --> 00:01:21,280
显然是左边的OPS5.5吧

23
00:01:21,280 --> 00:01:22,540
熊猫快送

24
00:01:22,540 --> 00:01:23,940
很接地气

25
00:01:23,940 --> 00:01:25,600
还有对话气泡

26
00:01:25,600 --> 00:01:26,560
奶茶

27
00:01:26,560 --> 00:01:27,640
城市背景

28
00:01:27,640 --> 00:01:29,700
完整度很高啊

29
00:01:29,700 --> 00:01:31,100
再看看右边

30
00:01:31,100 --> 00:01:33,800
什么玩意儿

31
00:01:33,800 --> 00:01:35,120
我真的绷不住了

32
00:01:35,120 --> 00:01:37,940
这一轮OPS 5.5完胜

33
00:01:37,940 --> 00:01:41,900
第二个测试是让AI写300字的科普文案

34
00:01:41,900 --> 00:01:43,500
比一比写作能力

35
00:01:43,500 --> 00:01:46,600
因为这次Anthoropi的官方特别强调

36
00:01:46,600 --> 00:01:49,180
OPS 5.5写东西正向人画

37
00:01:49,180 --> 00:01:51,020
我还是有点期待的

38
00:01:51,020 --> 00:01:53,280
这是两个AI生成的文案

39
00:01:53,280 --> 00:01:56,380
你可以猜猜哪篇是OPS 5.5写的

40
00:01:56,380 --> 00:01:57,720
更喜欢哪一篇呢

41
00:01:57,720 --> 00:01:59,440
可以打在弹幕上

42
00:01:59,440 --> 00:02:02,280
揭晓答案

43
00:02:02,280 --> 00:02:05,380
我个人更喜欢OPS 5.5写的这一篇

44
00:02:05,380 --> 00:02:07,440
他先从你天天在用的

45
00:02:07,440 --> 00:02:09,020
拆的GPT豆包说起

46
00:02:09,020 --> 00:02:11,960
再拿一个特别爱读书的学生打比方

47
00:02:11,960 --> 00:02:15,240
然后用一句什么移着字移着字往下接

48
00:02:15,240 --> 00:02:17,560
就把大模型的原理讲明白了

49
00:02:17,560 --> 00:02:19,420
中间还穿插了一句

50
00:02:19,420 --> 00:02:20,800
听起来挺简单

51
00:02:20,800 --> 00:02:21,660
对吧

52
00:02:21,660 --> 00:02:23,960
读起来真的很像朋友在跟你聊天

53
00:02:23,960 --> 00:02:26,800
GPT6Soul这篇也还行

54
00:02:26,800 --> 00:02:29,400
但是两段话都写得太长了

55
00:02:29,400 --> 00:02:32,860
而且它只说大模型会一点点组织出回复

56
00:02:32,860 --> 00:02:35,340
却没有把怎么组织的讲清楚

57
00:02:35,340 --> 00:02:38,400
那这一轮OPS 5.5略胜一筹

58
00:02:38,400 --> 00:02:40,540
第三轮上点强度

59
00:02:40,540 --> 00:02:43,440
让AI开发3D程式生成器

60
00:02:43,440 --> 00:02:46,360
考察AI的前端编码和工程能力

61
00:02:46,360 --> 00:02:48,540
我还要求AI做完之后

62
00:02:48,540 --> 00:02:50,560
自己打开网页验证效果

63
00:02:50,560 --> 00:02:52,340
不满意就自己接着改

64
00:02:52,340 --> 00:02:54,040
直到满意了再交付

65
00:02:54,040 --> 00:02:57,940
OPS 5.5反复打磨了将近一个半小时

66
00:02:57,940 --> 00:03:00,660
严重影响了我这期视频的发布时间

67
00:03:00,660 --> 00:03:02,580
本来我很生气

68
00:03:02,580 --> 00:03:04,840
但是当我打开主页时

69
00:03:04,840 --> 00:03:07,260
我直接大呼卧槽了

70
00:03:07,260 --> 00:03:08,360
这这细腻度

71
00:03:08,360 --> 00:03:09,540
这这质感

72
00:03:09,540 --> 00:03:12,620
我感觉一切的等待都是值得的

73
00:03:12,620 --> 00:03:15,700
鼠标拖一拖就能旋转平移缩放

74
00:03:15,700 --> 00:03:17,160
还能切换视角

75
00:03:17,160 --> 00:03:19,560
最让我惊喜的是细节

76
00:03:19,560 --> 00:03:21,620
明明有几百栋大楼

77
00:03:21,620 --> 00:03:24,620
放大之后你竟然能看到中文站牌

78
00:03:24,620 --> 00:03:27,960
还有一列轻轨正从楼中间穿过去

79
00:03:27,960 --> 00:03:32,000
整体来说山水桥都刻画得非常细腻

80
00:03:32,000 --> 00:03:33,720
可以说是栩栩如生

81
00:03:33,720 --> 00:03:37,240
至少让我感受到重清立交的复杂度了

82
00:03:37,240 --> 00:03:41,160
还能在右侧面板灵活调整生成的尝试效果

83
00:03:41,160 --> 00:03:42,780
你感觉怎么样呢

84
00:03:42,780 --> 00:03:45,400
这个效果我可以直接给到夯

85
00:03:45,400 --> 00:03:47,760
再来看看GPT6Soul

86
00:03:47,760 --> 00:03:50,500
虽然它只用了12分钟就完成了

87
00:03:50,500 --> 00:03:53,960
但是做出来的东西就很一般了

88
00:03:53,960 --> 00:03:58,280
我相信大家一眼就能感受到和OPS 5.5之间的差距了

89
00:03:58,280 --> 00:04:00,100
而且问题还不少

90
00:04:00,100 --> 00:04:03,400
这是玄空城啊

91
00:04:03,400 --> 00:04:06,160
而且像轻轨穿楼这种场景

92
00:04:06,160 --> 00:04:07,820
穿模问题也很明显

93
00:04:07,820 --> 00:04:10,220
车厢一头扎到墙里去了

94
00:04:10,220 --> 00:04:12,540
虽然功能也是完整的

95
00:04:12,540 --> 00:04:13,920
但无论是UI效果

96
00:04:13,920 --> 00:04:15,800
还是右侧菜单的丰富程度

97
00:04:15,800 --> 00:04:18,320
都被OPS 5.5降为打击

98
00:04:18,320 --> 00:04:19,760
第四个测试

99
00:04:19,760 --> 00:04:21,680
我让AI操作电脑画图

100
00:04:21,680 --> 00:04:24,120
我准备了一个简洁的画板网站

101
00:04:24,120 --> 00:04:25,260
和一张参考图

102
00:04:25,260 --> 00:04:27,320
让它们只能通过鼠标

103
00:04:27,320 --> 00:04:29,360
一笔一笔画出参考图

104
00:04:29,360 --> 00:04:31,500
不能靠代码和脚本生成

105
00:04:31,500 --> 00:04:35,040
OPS 5.5是在Cursor自带的浏览器里画的

106
00:04:35,040 --> 00:04:36,720
前后花了一个多小时

107
00:04:36,720 --> 00:04:39,040
画出来的成品效果是这样的

108
00:04:39,040 --> 00:04:40,060
好家伙

109
00:04:40,060 --> 00:04:41,500
怎么画成卷发了

110
00:04:41,500 --> 00:04:43,280
我第一眼看成了美洋洋

111
00:04:43,280 --> 00:04:46,800
应该是因为Cursor内置浏览器的拖拽功能

112
00:04:46,800 --> 00:04:48,100
只能拖动网页元素

113
00:04:48,100 --> 00:04:51,020
没办法按住鼠标画出连续的线条

114
00:04:51,020 --> 00:04:53,440
于是AI只能用大圆笔

115
00:04:53,440 --> 00:04:56,680
一个一个把画面的元素点出来

116
00:04:56,680 --> 00:04:58,200
不过即使是这样

117
00:04:58,200 --> 00:05:01,520
它的构图和配色还原的挺像的

118
00:05:01,520 --> 00:05:02,240
对话气泡

119
00:05:02,240 --> 00:05:02,840
DEMO

120
00:05:02,840 --> 00:05:04,340
白色蕾丝头饰

121
00:05:04,340 --> 00:05:05,580
蓝色大眼睛

122
00:05:05,580 --> 00:05:06,200
腮红

123
00:05:06,200 --> 00:05:06,920
蝴蝶结

124
00:05:06,920 --> 00:05:07,540
都在

125
00:05:07,540 --> 00:05:11,660
GPT-6 Saw用的是Codex的电脑操作能力

126
00:05:11,660 --> 00:05:14,380
直接控制真实的Tram浏览器来画

127
00:05:14,380 --> 00:05:16,700
它可以自由拖动鼠标画线

128
00:05:16,700 --> 00:05:18,420
十几分钟就找定了

129
00:05:18,420 --> 00:05:20,560
不过这些成品

130
00:05:20,560 --> 00:05:22,320
怎么说呢

131
00:05:22,320 --> 00:05:24,500
你能看出来这是啥字吗

132
00:05:24,500 --> 00:05:26,100
崩的也太明显了

133
00:05:26,100 --> 00:05:27,740
其他的不多说了

134
00:05:27,740 --> 00:05:29,920
你觉得哪张正向参考图呢

135
00:05:29,920 --> 00:05:32,300
我个人投OPS5.5一票

136
00:05:32,300 --> 00:05:34,360
如果没被AI工具限制

137
00:05:34,360 --> 00:05:36,180
还原度应该会更高

138
00:05:36,180 --> 00:05:38,900
最后我让AI来复彻Cursor

139
00:05:38,900 --> 00:05:40,100
这些AI编程工具

140
00:05:40,100 --> 00:05:42,500
那这个案例我搞过很多次了

141
00:05:42,500 --> 00:05:46,640
这是一个非常考验长城任务能力的全战项目

142
00:05:46,640 --> 00:05:47,800
CloudOps 5.5

143
00:05:47,800 --> 00:05:49,480
花了50多分钟完成

144
00:05:49,480 --> 00:05:51,200
来看看效果

145
00:05:51,200 --> 00:05:54,900
这个Agent面板的布局

146
00:05:54,900 --> 00:05:57,260
是不是跟主流的AI工具神色

147
00:05:57,260 --> 00:05:59,380
AI的输出也很清晰

148
00:05:59,380 --> 00:06:01,400
每次的改动一目了然

149
00:06:01,400 --> 00:06:02,940
光看到这儿

150
00:06:02,940 --> 00:06:05,280
你可能还没意识到问题的严重性

151
00:06:05,280 --> 00:06:08,200
那如果我可以打开资源管理器

152
00:06:08,200 --> 00:06:09,000
查看文件

153
00:06:09,000 --> 00:06:10,500
可以打开终端

154
00:06:10,500 --> 00:06:11,980
输入和运行命令

155
00:06:11,980 --> 00:06:13,980
可以打开Agent侧边栏

156
00:06:13,980 --> 00:06:15,820
管理所有的对话呢

157
00:06:15,820 --> 00:06:18,800
这些功能全都能正常使用

158
00:06:18,800 --> 00:06:20,420
切换到Editor Window

159
00:06:20,420 --> 00:06:21,340
编辑器模式

160
00:06:21,340 --> 00:06:22,440
文件招量

161
00:06:22,440 --> 00:06:23,240
代码招量

162
00:06:23,240 --> 00:06:23,960
代码补权

163
00:06:23,960 --> 00:06:26,520
甚至是Gate版本管理

164
00:06:26,520 --> 00:06:27,800
一应俱全

165
00:06:27,800 --> 00:06:30,720
如果我把这个产品直接发布了

166
00:06:30,720 --> 00:06:31,980
你绝对想不到

167
00:06:31,980 --> 00:06:33,880
这是AI一把缩的

168
00:06:33,880 --> 00:06:35,220
Ops 5.5

169
00:06:35,220 --> 00:06:36,560
行报了

170
00:06:36,560 --> 00:06:39,300
再来看看GPT-6Soul的效果

171
00:06:39,300 --> 00:06:42,940
虽然核心功能和UI布局是合理的

172
00:06:42,940 --> 00:06:45,180
但从第一眼的界面上来看

173
00:06:45,180 --> 00:06:46,300
就已经输了

174
00:06:46,300 --> 00:06:47,660
配色有点怪

175
00:06:47,660 --> 00:06:49,180
整个任务的执行界面

176
00:06:49,180 --> 00:06:50,760
也是浓浓的AI味

177
00:06:50,760 --> 00:06:54,000
我感觉GPT的风格一贯如此

178
00:06:54,000 --> 00:06:57,040
功能丰富度也远远不如OPS 5.5

179
00:06:57,040 --> 00:06:58,180
不支持搜索

180
00:06:58,180 --> 00:06:59,280
没有Jet能力

181
00:06:59,280 --> 00:07:00,160
没有终端

182
00:07:00,160 --> 00:07:02,400
也没有那么多布局调整和面板

183
00:07:02,400 --> 00:07:04,720
页面上很多按钮都是死的

184
00:07:04,720 --> 00:07:05,820
不好意思

185
00:07:05,820 --> 00:07:07,940
在看完了OPS5.5之后

186
00:07:07,940 --> 00:07:09,780
我只能这刀拉完了

187
00:07:09,780 --> 00:07:12,780
最后花一分钟说说感受吧

188
00:07:12,780 --> 00:07:15,740
高傲的OPS5.5真的赢麻了

189
00:07:15,740 --> 00:07:17,260
不仅编程能力强

190
00:07:17,260 --> 00:07:19,120
写作也有人味了

191
00:07:19,120 --> 00:07:22,700
我感觉甚至比Fable5.1还要强

192
00:07:23,120 --> 00:07:29,120
那A2嘴上建議大家放緩AI發展,自己在那嘎喳進步是吧?

193
00:07:29,120 --> 00:07:31,280
不过强的代价就是速度慢

194
00:07:31,280 --> 00:07:32,120
价格贵

195
00:07:32,120 --> 00:07:34,080
本期这5个案例加起来

196
00:07:34,080 --> 00:07:36,720
花了我600多块钱

197
00:07:36,720 --> 00:07:38,680
成本巨大

198
00:07:38,680 --> 00:07:41,120
GPT6SAL就友好多了

199
00:07:41,120 --> 00:07:44,160
虽然能力无法跟GPT6OSTRA媲美

200
00:07:44,160 --> 00:07:46,080
但消耗低了很多

201
00:07:46,080 --> 00:07:48,360
本期我跑完一整套测试

202
00:07:48,360 --> 00:07:51,280
才刚好把PLUS会员的额度用完

203
00:07:51,280 --> 00:07:53,880
这才是给大家日常用的模型

204
00:07:53,880 --> 00:07:57,080
我感觉跟国产旗舰模型的能力差不多

205
00:07:57,080 --> 00:07:58,920
仅个人体验

206
00:07:58,920 --> 00:08:01,440
那看到这里你已经了解了新模型

207
00:08:01,440 --> 00:08:04,220
相信你也有自己的使用选择了

208
00:08:04,220 --> 00:08:06,220
最后我再送大家一段

209
00:08:06,220 --> 00:08:08,080
由Cloud Ops 5.5

210
00:08:08,080 --> 00:08:11,120
自主生成的3D网页动画大作

211
00:08:11,120 --> 00:08:12,380
牛来骑车

212
00:08:12,380 --> 00:08:15,260
这一段大作花了我几十块钱

213
00:08:15,260 --> 00:08:17,120
能三连回回血吗

214
00:08:28,920 --> 00:08:58,900
中文字幕志愿者 李宗盛

215
00:08:58,920 --> 00:09:28,900
中文字幕志愿者 李宗盛

216
00:09:28,920 --> 00:09:58,900
中文字幕志愿者 李宗盛

217
00:09:58,920 --> 00:10:00,920
我可以看到你

218
00:10:28,920 --> 00:10:30,920
好