1
00:00:00,000 --> 00:00:07,000
论坛的超级版主王池川發帖說Deepseek vc.1flash 0910開始累測了。

2
00:00:07,000 --> 00:00:11,000
實際上就給了兩天時間啊。他說手特根1.1秒。

3
00:00:11,000 --> 00:00:17,000
生成速度最高329特根每秒。經過我的長期實測啊。就是長時間的工作的實測。

4
00:00:17,000 --> 00:00:21,000
這個數字是差不多的啊。是能夠達到的。

5
00:00:21,000 --> 00:00:25,000
他就說這個模型非常的好用。超級屌。

6
00:00:25,000 --> 00:00:28,020
问我要不要发一个测试视频

7
00:00:28,020 --> 00:00:30,380
其实他当时发这个帖子的时候

8
00:00:30,380 --> 00:00:31,560
我已经在干活了

9
00:00:31,560 --> 00:00:33,000
而且给他规划了个大任务

10
00:00:33,000 --> 00:00:35,940
这个Sky Knight说

11
00:00:35,940 --> 00:00:39,860
好消息就是Flash明天中午降价

12
00:00:39,860 --> 00:00:41,000
就是今天中午降价了

13
00:00:41,000 --> 00:00:42,960
但是当时呢

14
00:00:42,960 --> 00:00:45,260
我是想早一点把这个任务给跑完

15
00:00:45,260 --> 00:00:46,260
早一点测试

16
00:00:46,260 --> 00:00:47,860
我也不想省那一点钱

17
00:00:47,860 --> 00:00:49,380
我当时估计我这个任务

18
00:00:49,380 --> 00:00:51,080
可能花个50块钱左右就差不多了

19
00:00:51,080 --> 00:00:51,860
是吧

20
00:00:51,860 --> 00:00:52,600
你说我省那个

21
00:00:52,600 --> 00:00:53,780
就是降价一半

22
00:00:53,780 --> 00:00:54,580
我省个25块钱

23
00:00:54,580 --> 00:00:55,520
有什么意义呢

24
00:00:55,520 --> 00:00:58,100
但是呢最终

25
00:00:58,100 --> 00:00:59,800
还是有点超出我的预料

26
00:00:59,800 --> 00:01:01,540
他一共花了我多少钱

27
00:01:01,540 --> 00:01:03,960
他这个任务花了我83块4

28
00:01:03,960 --> 00:01:06,280
就比我想象的要贵不少

29
00:01:06,280 --> 00:01:08,780
他消耗这个tokens

30
00:01:08,780 --> 00:01:10,880
一共是

31
00:01:10,880 --> 00:01:14,180
13.6亿

32
00:01:14,180 --> 00:01:15,380
tokens输入

33
00:01:15,380 --> 00:01:17,600
输出是980k

34
00:01:17,600 --> 00:01:19,580
那基本上就相当于是

35
00:01:19,580 --> 00:01:21,500
14亿的这个总tokens消耗

36
00:01:21,500 --> 00:01:23,980
那我相信在这样的价格下

37
00:01:23,980 --> 00:01:25,640
尤其是在这个什么

38
00:01:25,640 --> 00:01:28,520
降价之前

39
00:01:28,520 --> 00:01:29,860
它能有这样的表现

40
00:01:29,860 --> 00:01:31,120
我觉得非常的便宜了

41
00:01:31,120 --> 00:01:32,140
因为降价之后

42
00:01:32,140 --> 00:01:33,620
它可能说最高能降60%

43
00:01:33,620 --> 00:01:35,260
我们按照降一半来算的话

44
00:01:35,260 --> 00:01:36,640
做这个庞大的任务

45
00:01:36,640 --> 00:01:38,180
也达到了我的预期

46
00:01:38,180 --> 00:01:39,960
我的心理承受价格是50块钱

47
00:01:39,960 --> 00:01:41,560
它如果40块钱能完成的话

48
00:01:41,560 --> 00:01:43,240
那我觉得就非常的划算了

49
00:01:43,240 --> 00:01:45,700
那这个任务是什么呢

50
00:01:45,700 --> 00:01:47,460
就我这个论坛

51
00:01:47,460 --> 00:01:50,140
不是有这个硬件页面吗

52
00:01:50,140 --> 00:01:52,200
我就想啊

53
00:01:52,200 --> 00:01:55,640
老是发这个用户的帖子

54
00:01:55,640 --> 00:01:56,680
就这样展现给大家

55
00:01:56,680 --> 00:01:57,640
有点太粗糙了

56
00:01:57,640 --> 00:01:59,380
那么我要给英伟达

57
00:01:59,380 --> 00:02:00,640
AMD 英特达 苹果

58
00:02:00,640 --> 00:02:02,740
他们不同的小主机

59
00:02:02,740 --> 00:02:03,620
不同的芯片

60
00:02:03,620 --> 00:02:04,300
不同的显卡

61
00:02:04,300 --> 00:02:06,640
他们AI算力怎么样

62
00:02:06,640 --> 00:02:08,200
给大家做一个推荐

63
00:02:08,200 --> 00:02:09,380
归纳总结

64
00:02:09,380 --> 00:02:12,080
那像这些

65
00:02:12,080 --> 00:02:15,020
要跟我论坛实测的数据

66
00:02:15,020 --> 00:02:16,840
跟我实测的这些帖子互动

67
00:02:16,840 --> 00:02:18,520
进行初步的总结

68
00:02:18,520 --> 00:02:20,920
总结完了之后

69
00:02:20,920 --> 00:02:23,800
还把这个比较相关的一些帖子给它列出来

70
00:02:23,800 --> 00:02:26,720
然后有些更多的一些相关的信息

71
00:02:26,720 --> 00:02:27,020
是吧

72
00:02:27,020 --> 00:02:30,600
再把它这个链接到下面来

73
00:02:30,600 --> 00:02:32,760
像这些tag啊什么东西的是吧

74
00:02:32,760 --> 00:02:35,100
就直接链到论坛的帖子上来

75
00:02:35,100 --> 00:02:37,260
那这样的效果就会好很多

76
00:02:37,260 --> 00:02:39,580
当然我现在在本地

77
00:02:39,580 --> 00:02:41,300
由于我这个Firefox

78
00:02:41,300 --> 00:02:45,240
我是限制它不能翻墙的

79
00:02:45,240 --> 00:02:47,680
那么它可能说显示肯定是有的时候有些问题

80
00:02:47,680 --> 00:02:51,640
在正常的论坛中是没有问题的

81
00:02:51,640 --> 00:02:52,900
那做这个工作

82
00:02:52,900 --> 00:02:54,160
实际上工作量是非常大的

83
00:02:54,160 --> 00:02:56,580
首先他要从我这个云端

84
00:02:56,580 --> 00:02:58,500
把这个数据库同步到本地

85
00:02:58,500 --> 00:02:59,720
在本地完全复兴

86
00:02:59,720 --> 00:03:02,620
然后他要理解每一个帖子

87
00:03:02,620 --> 00:03:03,820
因为做这个工作之前

88
00:03:03,820 --> 00:03:06,460
我要他重新把所有的tag给重新打一遍

89
00:03:06,460 --> 00:03:07,740
把漏掉的给补上

90
00:03:07,740 --> 00:03:09,900
像那什么苹果的M4啊

91
00:03:09,900 --> 00:03:11,440
什么M5他都没有

92
00:03:11,440 --> 00:03:12,960
那都要给他加上呢

93
00:03:12,960 --> 00:03:14,880
是吧

94
00:03:14,880 --> 00:03:16,960
之前很多标签像什么B70

95
00:03:16,960 --> 00:03:18,380
这些标签都没有的

96
00:03:18,380 --> 00:03:19,760
都要给他补上

97
00:03:19,760 --> 00:03:20,900
5080这些

98
00:03:20,900 --> 00:03:22,680
是吧

99
00:03:22,680 --> 00:03:25,040
那现在都给他补上了

100
00:03:25,040 --> 00:03:26,320
包括W7800 7900

101
00:03:26,320 --> 00:03:30,200
把这些标签重新打一遍

102
00:03:30,200 --> 00:03:31,180
并且呢

103
00:03:31,180 --> 00:03:33,040
要他这个接手

104
00:03:33,040 --> 00:03:34,400
赫米斯以后的工作

105
00:03:34,400 --> 00:03:35,020
就是为

106
00:03:35,020 --> 00:03:36,000
可能说以后

107
00:03:36,000 --> 00:03:37,160
我要全面转换到

108
00:03:37,160 --> 00:03:38,120
Deepseek Harness上来

109
00:03:38,120 --> 00:03:40,160
要提前做准备

110
00:03:40,160 --> 00:03:41,320
所以说整个论坛呢

111
00:03:41,320 --> 00:03:42,640
所有帖子让他通读

112
00:03:42,640 --> 00:03:43,480
规则让他通读

113
00:03:43,480 --> 00:03:46,380
就赫米斯是怎么管理这个论坛的

114
00:03:46,380 --> 00:03:48,200
要他准备接手

115
00:03:48,200 --> 00:03:50,360
那总体上这个工作量

116
00:03:50,360 --> 00:03:52,020
还是比较大的

117
00:03:52,020 --> 00:03:53,140
毕竟消耗了那么多tocons

118
00:03:53,140 --> 00:03:55,480
但是他执行任务的时候

119
00:03:55,480 --> 00:03:56,980
我感觉他的优点就非常的明显

120
00:03:56,980 --> 00:03:59,380
你说能不能平替Deepseek with Pro

121
00:03:59,380 --> 00:04:01,540
我觉得可能要稍微差一点意思

122
00:04:01,540 --> 00:04:03,000
但是基本上能够替代

123
00:04:03,000 --> 00:04:05,900
就他的这个常列任务的规划能力

124
00:04:05,900 --> 00:04:07,100
是非常强的

125
00:04:07,100 --> 00:04:10,020
一共是生成了65个页面

126
00:04:10,020 --> 00:04:13,180
这里面还包含了大量的

127
00:04:13,180 --> 00:04:14,080
在线的教验

128
00:04:14,080 --> 00:04:15,280
我要求他每一条数据

129
00:04:15,280 --> 00:04:16,920
一定要与官方数据

130
00:04:16,920 --> 00:04:19,040
与这个维基百科的数据进行

131
00:04:19,040 --> 00:04:20,060
校验之后

132
00:04:20,060 --> 00:04:21,240
才能采用

133
00:04:21,240 --> 00:04:24,800
这本身就涉及到很多网络查询的一些工作

134
00:04:24,800 --> 00:04:26,280
是吧

135
00:04:26,280 --> 00:04:27,320
还有一些图片识别

136
00:04:27,320 --> 00:04:28,220
当然了

137
00:04:28,220 --> 00:04:28,960
它这个图片识别

138
00:04:28,960 --> 00:04:29,620
我必须要说

139
00:04:29,620 --> 00:04:31,380
它比起GPT5

140
00:04:31,380 --> 00:04:33,500
或者比起GPT6自带的那个图片识别

141
00:04:33,500 --> 00:04:34,760
我觉得要差远了

142
00:04:34,760 --> 00:04:37,140
因为它本身不能深土

143
00:04:37,140 --> 00:04:38,060
就有的时候

144
00:04:38,060 --> 00:04:39,220
它找到的图片不好的话

145
00:04:39,220 --> 00:04:41,620
它实际上自己是看不大出来的

146
00:04:41,620 --> 00:04:42,600
像这种图片

147
00:04:42,600 --> 00:04:43,320
说实话

148
00:04:43,320 --> 00:04:44,460
其实是非常垃圾的

149
00:04:44,460 --> 00:04:47,000
但是如果用GPT来搞的话

150
00:04:47,000 --> 00:04:47,640
他可能说

151
00:04:47,640 --> 00:04:49,400
自己就直接画一个图

152
00:04:49,400 --> 00:04:50,660
也是比较准的

153
00:04:50,660 --> 00:04:53,200
但是我们也要说

154
00:04:53,200 --> 00:04:54,740
毕竟我消耗了14亿推肯子

155
00:04:54,740 --> 00:04:56,560
如果说是用GPT6的价格

156
00:04:56,560 --> 00:04:57,860
我现在已经破产了

157
00:04:57,860 --> 00:05:00,580
哪怕用GPT5.64我也扛不住

158
00:05:00,580 --> 00:05:03,140
可能说用Tera能把我的信用卡刷爆

159
00:05:03,140 --> 00:05:04,280
我不至于破产

160
00:05:04,280 --> 00:05:05,780
另外一个优点

161
00:05:05,780 --> 00:05:06,400
我们必须说

162
00:05:06,400 --> 00:05:08,040
它搭配自家的Hanis干活

163
00:05:08,040 --> 00:05:09,800
是相当的爽的

164
00:05:09,800 --> 00:05:10,060
你看

165
00:05:10,060 --> 00:05:12,180
手推肯平均是1.7秒

166
00:05:12,180 --> 00:05:15,080
平均Token是385Token每秒

167
00:05:15,080 --> 00:05:18,440
一共跌到了183轮2832步

168
00:05:18,440 --> 00:05:21,260
缓存命中率99.8%

169
00:05:21,260 --> 00:05:22,180
是吧

170
00:05:22,180 --> 00:05:24,220
输入13.6兆Token

171
00:05:24,220 --> 00:05:25,960
输出980GToken

172
00:05:25,960 --> 00:05:27,680
我说实话

173
00:05:27,680 --> 00:05:29,860
这个数据是相当的漂亮的

174
00:05:29,860 --> 00:05:32,120
第一是缓存命中率达到了99.8%

175
00:05:32,120 --> 00:05:35,180
绝大部分Token都是用白菜价来计算的

176
00:05:35,180 --> 00:05:37,080
你换其他任何的AI

177
00:05:37,080 --> 00:05:38,700
这个价格里都扛不住

178
00:05:38,700 --> 00:05:41,520
然后183轮

179
00:05:41,520 --> 00:05:43,100
是吧

180
00:05:43,100 --> 00:05:44,840
2800多部的迭代

181
00:05:44,840 --> 00:05:47,520
我相信这个任务还是相当相当的强大的

182
00:05:47,520 --> 00:05:50,420
这是一个比较复杂的任务

183
00:05:50,420 --> 00:05:54,280
他还是最终准确的交货了

184
00:05:54,280 --> 00:05:56,300
虽然过程中有些图片

185
00:05:56,300 --> 00:05:57,480
找的不太好

186
00:05:57,480 --> 00:05:59,200
讲实话我不太满意

187
00:05:59,200 --> 00:06:02,020
然后我又自己找了一些图片来替代他

188
00:06:02,020 --> 00:06:04,200
就因为这个被我骂的像狗一样

189
00:06:04,200 --> 00:06:05,700
但是凭良性讲

190
00:06:05,700 --> 00:06:08,520
他总体上任务完成的还是相当漂亮的

191
00:06:11,520 --> 00:06:14,160
答案呢具体怎么样

192
00:06:14,160 --> 00:06:15,920
还要把它再放到网上面

193
00:06:15,920 --> 00:06:19,460
再让大家这个什么参考的时候

194
00:06:19,460 --> 00:06:20,340
大家去核对

195
00:06:20,340 --> 00:06:22,860
因为我自己没有时间去核对这么多数据

196
00:06:22,860 --> 00:06:27,540
那他本次任务中还有一个比较重大的缺陷

197
00:06:27,540 --> 00:06:28,720
刚才我忘记说了

198
00:06:28,720 --> 00:06:29,760
是什么呢

199
00:06:29,760 --> 00:06:31,540
就是任务开始的时候

200
00:06:31,540 --> 00:06:32,620
我明确跟他讲了

201
00:06:32,620 --> 00:06:35,500
你还是要自己当监工

202
00:06:35,500 --> 00:06:38,660
你把主要的干脏活的事情交给这个subagent

203
00:06:38,660 --> 00:06:41,120
这个subagent用什么来驱动呢

204
00:06:41,120 --> 00:06:43,000
用这个轻轻3.827B

205
00:06:43,320 --> 00:06:45,520
IWQDFLUSHR这个模型

206
00:06:45,520 --> 00:06:47,640
就是我本地的S接浪驱动的模型

207
00:06:48,520 --> 00:06:49,400
用它来搞

208
00:06:50,120 --> 00:06:52,840
而且我这个模型里给它放什么

209
00:06:53,080 --> 00:06:54,440
三四个绘画四五个绘画

210
00:06:54,440 --> 00:06:55,640
它是完全没有问题的

211
00:06:56,880 --> 00:06:57,880
我有信心是因为什么

212
00:06:57,880 --> 00:06:59,040
因为这个DeepSickHarness

213
00:06:59,040 --> 00:07:00,720
已经更进到了1.2这支版

214
00:07:01,520 --> 00:07:02,440
好像是这样啊

215
00:07:03,000 --> 00:07:06,240
它已经不是什么α或者说是β版本了

216
00:07:06,640 --> 00:07:08,360
早些我用RCR的版本的时候

217
00:07:08,360 --> 00:07:09,480
它工作都是正常的

218
00:07:09,480 --> 00:07:13,680
那么现在我想它这个什么Subagent的能力应该是得到了比较大幅的提升

219
00:07:13,680 --> 00:07:15,960
结果证明它的Subagent还是一座狗屎

220
00:07:15,960 --> 00:07:19,380
就跟以前RC版本的时候没有任何区别

221
00:07:19,380 --> 00:07:23,280
当然我也不得不说像什么Hemex也好Codex也好

222
00:07:23,280 --> 00:07:24,580
他们的Subagent都是狗屎

223
00:07:24,580 --> 00:07:25,900
做的都不好

224
00:07:25,900 --> 00:07:29,380
可能说只有Cloud Code他们的Subagent做的比较好

225
00:07:29,380 --> 00:07:35,260
那你说在这种情况下应该怎么搞呢

226
00:07:35,260 --> 00:07:36,920
就是根据我之前用DeepSync的经验

227
00:07:36,920 --> 00:07:40,920
就是你要把你的任务

228
00:07:40,920 --> 00:07:42,540
分拆为不同的阶段

229
00:07:42,540 --> 00:07:46,580
每个阶段让他不同的agent

230
00:07:46,580 --> 00:07:49,540
通过同一个目录来沟通

231
00:07:49,540 --> 00:07:52,440
给这些目录上锁上队列

232
00:07:52,440 --> 00:07:53,920
让他们排队

233
00:07:53,920 --> 00:07:57,800
相互来教应相互来协作

234
00:07:57,800 --> 00:07:59,620
那么这个效率是比较高的

235
00:07:59,620 --> 00:08:01,200
他就可以怎么讲

236
00:08:01,200 --> 00:08:03,160
把大部分的tokens给省下来

237
00:08:15,923 --> 00:08:19,563
毕竟这种大活不是每天都有的

238
00:08:19,563 --> 00:08:21,083
就算每天都有

239
00:08:21,083 --> 00:08:22,943
一个月也就一千多块钱

240
00:08:22,943 --> 00:08:23,963
这个钱要不要省

241
00:08:23,963 --> 00:08:25,003
就见仁见智

242
00:08:25,003 --> 00:08:28,263
因为什么呢

243
00:08:28,263 --> 00:08:29,763
这个千万3.827比亚

244
00:08:29,763 --> 00:08:31,003
他做这种任务

245
00:08:31,003 --> 00:08:32,323
他做任务的能力

246
00:08:32,323 --> 00:08:34,043
你看包括他修复字幕

247
00:08:34,043 --> 00:08:35,123
我说实话

248
00:08:35,123 --> 00:08:36,863
他的能力是非常非常强的

249
00:08:36,863 --> 00:08:38,383
就虽然他的支持库很小

250
00:08:38,383 --> 00:08:41,103
但是他积极的上网去搜索

251
00:08:41,103 --> 00:08:41,843
是吧

252
00:08:41,843 --> 00:08:43,203
完成任务的质量是非常高的

253
00:08:43,203 --> 00:08:45,323
完全部署给Dips一个威斯典一

254
00:08:45,323 --> 00:08:49,863
之前我长期用这个VS Flush

255
00:08:49,863 --> 00:08:51,143
来修字幕的时候

256
00:08:51,143 --> 00:08:52,923
我感觉他并不比这个什么

257
00:08:52,923 --> 00:08:54,963
轻微3.827比要强什么

258
00:08:54,963 --> 00:08:58,003
就是他这个编码也是编程

259
00:08:58,003 --> 00:08:59,903
他修复这个论坛的插件

260
00:08:59,903 --> 00:09:01,863
包括说修复APP的一些小bug

261
00:09:01,863 --> 00:09:02,383
他都不插

262
00:09:02,383 --> 00:09:03,563
他唯一插的地方

263
00:09:03,563 --> 00:09:04,663
就是他的常设下人不行

264
00:09:04,663 --> 00:09:07,063
他干不了常任任务

265
00:09:07,063 --> 00:09:08,943
就你如果要他执行一个复杂的任务

266
00:09:08,943 --> 00:09:10,383
必须要给他拆解到位

267
00:09:10,383 --> 00:09:12,363
然后监工的话

268
00:09:12,363 --> 00:09:14,043
就像这个什么

269
00:09:14,043 --> 00:09:16,463
V4.1

270
00:09:16,463 --> 00:09:18,403
或者说V4.1

271
00:09:18,403 --> 00:09:19,543
来监工

272
00:09:19,543 --> 00:09:23,783
这个模型它是个多摩泰模型

273
00:09:23,783 --> 00:09:24,523
昨天

274
00:09:24,523 --> 00:09:26,763
你看我发这些东西给他

275
00:09:26,763 --> 00:09:27,943
他都看得懂的

276
00:09:27,943 --> 00:09:30,103
图片他还是看得懂的

277
00:09:30,103 --> 00:09:31,403
他这个多摩泰模型

278
00:09:31,403 --> 00:09:36,043
这个价格的话

279
00:09:36,043 --> 00:09:37,543
包括他这个响应速度

280
00:09:37,543 --> 00:09:39,723
还有他的这个

281
00:09:39,723 --> 00:09:40,623
长年任务规划

282
00:09:40,623 --> 00:09:42,183
我觉得是比V4.1要更好的

283
00:09:42,183 --> 00:09:46,423
这个模型如果说测试通过上线的话

284
00:09:46,423 --> 00:09:48,223
我觉得Deepseek就短期内

285
00:09:48,223 --> 00:09:49,583
没有必要再折腾VisPro了

286
00:09:49,583 --> 00:09:50,823
可以把VisPro给下架了

287
00:09:50,823 --> 00:09:55,743
VisPro我感觉啊

288
00:09:55,743 --> 00:09:56,683
就是我个人的体感

289
00:09:56,683 --> 00:09:58,103
这个东西是比较玄乎的

290
00:09:58,103 --> 00:10:00,303
它没有什么硬性的数据来证实啊

291
00:10:00,303 --> 00:10:01,743
就是因为之前我用这个什么

292
00:10:01,743 --> 00:10:04,523
VisPro来开发是自动剪辑器

293
00:10:04,523 --> 00:10:05,423
视频的自动生成器

294
00:10:05,423 --> 00:10:06,043
自动剪辑器

295
00:10:06,043 --> 00:10:07,923
它也是迭代了将近200轮

296
00:10:07,923 --> 00:10:10,503
还200多轮

297
00:10:10,503 --> 00:10:11,103
3000多

298
00:10:11,103 --> 00:10:12,963
我感觉

299
00:10:12,963 --> 00:10:14,423
怎么讲

300
00:10:14,423 --> 00:10:16,703
它的这个常任务的表现

301
00:10:16,703 --> 00:10:18,943
还是比这个V4.1

302
00:10:18,943 --> 00:10:20,363
弗拉西要好一点点的

303
00:10:20,363 --> 00:10:22,503
这个感觉就非常的主观了

304
00:10:22,503 --> 00:10:24,323
但是最终都是完成的任务嘛

305
00:10:24,323 --> 00:10:26,223
那个任务的难度我认为

306
00:10:26,223 --> 00:10:27,743
要稍高一点点

307
00:10:27,743 --> 00:10:31,983
总之呢这个Deepseek

308
00:10:31,983 --> 00:10:34,223
V4弗拉西它降价之后

309
00:10:34,223 --> 00:10:36,103
尤其是它后期如果很快能够

310
00:10:36,103 --> 00:10:37,183
推出V4.1弗拉西

311
00:10:37,183 --> 00:10:40,223
我觉得大家真的就没有必要去折腾

312
00:10:40,223 --> 00:10:42,383
什么其他的一些

313
00:10:42,383 --> 00:10:43,683
什么coding plan

314
00:10:43,683 --> 00:10:45,823
或者说去折腾其他的AI了

315
00:10:45,823 --> 00:10:46,923
短时间之内

316
00:10:46,923 --> 00:10:49,183
这个劳良又变成了梁子

317
00:10:49,183 --> 00:10:51,323
甚至呢这个马上又有可能

318
00:10:51,323 --> 00:10:52,243
这个变成良盛

319
00:10:52,243 --> 00:10:54,463
这个是有可能的

320
00:10:54,463 --> 00:10:56,963
Deepseek现在用起来还是非常的像

321
00:10:56,963 --> 00:10:59,763
今天这个高分级过去

322
00:10:59,763 --> 00:11:00,383
等夜间

323
00:11:00,383 --> 00:11:02,063
我让这个什么

324
00:11:02,063 --> 00:11:03,743
Deepseek Harness

325
00:11:03,743 --> 00:11:05,303
把我这个本地的

326
00:11:05,303 --> 00:11:06,503
这些tags

327
00:11:06,503 --> 00:11:07,763
本地的这些什么页面

328
00:11:07,763 --> 00:11:09,083
都给他同步到云端

329
00:11:09,083 --> 00:11:11,363
大家明天就可以在手机和电脑上

330
00:11:11,363 --> 00:11:12,743
去看到这些东西了

331
00:11:12,743 --> 00:11:14,743
自己去看一下

332
00:11:14,743 --> 00:11:15,963
帮我这个检查一下

333
00:11:15,963 --> 00:11:17,663
看看里面有没有什么重大的错漏

334
00:11:17,663 --> 00:11:19,203
小毛病就自己发帖

335
00:11:19,203 --> 00:11:22,443
到这个什么论坛中发帖来提就行了

336
00:11:22,443 --> 00:11:23,563
千万不要给我发私信

337
00:11:23,563 --> 00:11:25,423
好吧

338
00:11:25,423 --> 00:11:26,283
跟大家交流一下

339
00:11:26,283 --> 00:11:27,123
就是说

340
00:11:27,123 --> 00:11:29,323
欢迎大家来这个论坛发帖交流讨论

341
00:11:29,323 --> 00:11:30,923
但是要有点功德性

342
00:11:30,923 --> 00:11:31,803
要遵守秩序

343
00:11:31,803 --> 00:11:33,783
论坛因为有各位的

344
00:11:33,783 --> 00:11:35,163
有价值的这些帖子

345
00:11:35,163 --> 00:11:37,083
我们才会变得越来越好

346
00:11:37,083 --> 00:11:37,923
是不是

347
00:11:37,923 --> 00:11:38,703
像这种帖子

348
00:11:38,703 --> 00:11:39,683
图文并报

349
00:11:39,683 --> 00:11:42,063
自己愿意去尝试新的东西

350
00:11:42,063 --> 00:11:43,583
然后尝试完之后

351
00:11:43,583 --> 00:11:44,483
分享给大家

352
00:11:44,483 --> 00:11:45,463
实际上他的帖子

353
00:11:45,463 --> 00:11:47,043
影响了好几个人

354
00:11:47,043 --> 00:11:48,323
我们论坛的大神

355
00:11:48,323 --> 00:11:51,483
有好几个人都发了帖子

356
00:11:51,483 --> 00:11:52,043
都测试了

357
00:11:52,043 --> 00:11:52,943
非常的好

358
00:11:52,943 --> 00:11:54,723
非常的感谢他

359
00:11:54,723 --> 00:11:58,023
就不要学这种

360
00:11:58,023 --> 00:11:59,403
毫无功德性的人

361
00:12:04,263 --> 00:12:05,503
什么证明功行都不做

362
00:12:05,503 --> 00:12:06,483
就知道在那算

363
00:12:06,483 --> 00:12:09,583
让我看到一次骂一次

364
00:12:09,583 --> 00:12:11,483
还被封号是吧

365
00:12:11,483 --> 00:12:16,883
好吧 本期视频就到这里

366
00:12:16,883 --> 00:12:17,723
我们下期再见