transcript

gistfile1.txt

1
00:00:01,231 --> 00:00:02,072
係喇咁，好咁

2
00:00:02,072 --> 00:00:03,613
咁或者我簡單

3
00:00:03,613 --> 00:00:04,735
呃介紹下自己先啦

4
00:00:04,735 --> 00:00:06,176
咁我而家係，呃

5
00:00:06,176 --> 00:00:07,256
頭先 Rex 介紹咗我喺

6
00:00:07,256 --> 00:00:08,257
呃，保險公司嘅一個

7
00:00:08,257 --> 00:00:10,218
呃，internal audit 裏面呢

8
00:00:10,218 --> 00:00:10,960
做緊一個，呃

9
00:00:10,960 --> 00:00:11,660
analitics嘅

10
00:00:11,660 --> 00:00:12,320
director

11
00:00:12,320 --> 00:00:13,262
呃另外呢，我都係

12
00:00:13,262 --> 00:00:14,643
呃，同大家一樣都係

13
00:00:14,643 --> 00:00:15,743
我係半個老師嚟嘅

14
00:00:15,743 --> 00:00:16,745
我喺香港 use space

15
00:00:16,745 --> 00:00:17,826
有教書，有教科

16
00:00:17,826 --> 00:00:17,966
叫做

17
00:00:18,646 --> 00:00:20,509
web scraping and business analytics

18
00:00:20,509 --> 00:00:21,730
咁web scraping就係叫做

19
00:00:21,730 --> 00:00:24,734
網絡爬蟲啦，咁就係唔同嘅網頁嗰度

20
00:00:24,734 --> 00:00:25,795
爬取呢啲data落嚟

21
00:00:25,795 --> 00:00:27,016
然後再做一啲分析

22
00:00:27,016 --> 00:00:29,519
咁樣囉，咁，另外我之前就喺

23
00:00:29,519 --> 00:00:31,080
呃，壁科度做過

24
00:00:31,080 --> 00:00:31,521
嗰個odit啦

25
00:00:31,521 --> 00:00:33,223
咁，然後就攞咗個CPA牌

26
00:00:33,223 --> 00:00:34,284
咁而家都係，CPA

27
00:00:34,284 --> 00:00:37,268
嘅，呃，嘅 young committee嘅 member嚟㗎係喇

28
00:00:38,156 --> 00:00:42,079
噉今日嗰個，呢個講咗嘅agenda就有以下呢幾點啦

29
00:00:42,079 --> 00:00:44,481
首先我好簡短咁講一講係

30
00:00:44,481 --> 00:00:47,122
會計呢個行業有啲咩嘅工種啦

31
00:00:47,122 --> 00:00:49,725
噉然後就會介紹一啲AI同埋big data嘅

32
00:00:49,725 --> 00:00:51,365
嘅technologies嘅

33
00:00:51,365 --> 00:00:55,889
噉然後就會再講下呢啲AI或者big data嘅technologies係應用喺唔同

34
00:00:55,889 --> 00:00:58,371
嘅會計嘅工種裡面係係點樣可以應用得到

35
00:00:59,310 --> 00:01:02,173
同埋實際喺應用上嚟有冇啲嘅困難

36
00:01:02,173 --> 00:01:03,835
係喇噉就會同大家分享啦

37
00:01:03,835 --> 00:01:07,337
噉最尾就，就有十五分鐘左右嘅Q&Asession啦

38
00:01:07,337 --> 00:01:09,319
噉有咩問題可以之後去問嘅

39
00:01:09,319 --> 00:01:10,900
係喇噉下張slide唔該

40
00:01:12,274 --> 00:01:14,195
係喇，咁，喺嗰個

41
00:01:14,195 --> 00:01:16,998
即係會計呢個行業裏面呢有啲咩

42
00:01:16,998 --> 00:01:19,519
公眾呢？咁即係我都做過幾份工咁樣啦

43
00:01:19,519 --> 00:01:22,501
咁，啫係呢幾個公眾都大概接觸過下

44
00:01:22,501 --> 00:01:24,382
咁，咁當然唔知呢啲嘅仲有

45
00:01:24,382 --> 00:01:25,644
有人專門做tax啊

46
00:01:25,644 --> 00:01:26,885
專做corporate

47
00:01:26,885 --> 00:01:27,504
financing 等等嘅

48
00:01:27,504 --> 00:01:30,007
咁但係呢幾個係我比較熟悉少少同埋有

49
00:01:30,007 --> 00:01:32,128
知道，點樣用big dataAI嗰啲喇

50
00:01:32,128 --> 00:01:35,430
咁所以今日集中會講呢幾個公眾啦

51
00:01:35,430 --> 00:01:37,992
咁第一就係一個叫做regulatory reporting就係

52
00:01:39,192 --> 00:01:40,974
即係政府或者呃

53
00:01:40,974 --> 00:01:43,156
一啲加滿機構要求你去做嘅

54
00:01:43,156 --> 00:01:44,477
一啲Reporting嚟㗎

55
00:01:44,477 --> 00:01:47,838
例如話，即係香港開公司咁每年要做一次Audit啦

56
00:01:47,838 --> 00:01:50,221
要報稅啦等等呢啲係政府規定㗎啦

57
00:01:50,221 --> 00:01:52,121
咁如果係一啲金融行業

58
00:01:52,121 --> 00:01:53,143
例如一啲嗰啲呃

59
00:01:53,143 --> 00:01:57,564
資產管理公司可能去證監有啲要求要去做嗰啲Reporting啦

60
00:01:57,564 --> 00:01:58,825
呃，銀行又有呃

61
00:01:58,825 --> 00:02:00,867
金管局嘅嘅要求啦或者

62
00:02:00,867 --> 00:02:01,947
保險公司又保監

63
00:02:01,947 --> 00:02:03,730
咁即係呢一類嘅係呃

64
00:02:03,730 --> 00:02:06,572
加滿機構要求嘅Reporting呢就有人專門做呢一類嘅工作

65
00:02:07,893 --> 00:02:09,234
咁第二度叫做呃

66
00:02:09,234 --> 00:02:10,235
management reporting啦

67
00:02:10,235 --> 00:02:11,978
咁，可能同你

68
00:02:11,978 --> 00:02:14,941
啫係大家textbook讀嗰啲諗法有少少唔同就係可能係

69
00:02:14,941 --> 00:02:17,183
關注啲廠裡面casting嗰啲叫management reporting

70
00:02:17,183 --> 00:02:18,825
但係我喺呢度個definition就係

71
00:02:18,825 --> 00:02:21,449
公司管理層自己內部想知道嘅嘢

72
00:02:21,449 --> 00:02:22,931
就為之management reporting

73
00:02:22,931 --> 00:02:23,451
啫係譬如話呃

74
00:02:23,711 --> 00:02:24,812
audit一年一次

75
00:02:24,812 --> 00:02:25,693
部署一年一次

76
00:02:25,693 --> 00:02:27,556
咁但係，我如果作為公司管理層

77
00:02:27,556 --> 00:02:30,919
我想知道今個月嘅數據同上個月嘅數據

78
00:02:30,919 --> 00:02:32,562
個銷售啊或者各方面做成點

79
00:02:32,562 --> 00:02:34,484
咁我就唔會等一年先至去

80
00:02:34,484 --> 00:02:36,045
去做一次呢，put一盤數出嚟

81
00:02:36,045 --> 00:02:37,687
咁每個月都要monitor住

82
00:02:37,687 --> 00:02:38,968
咁同埋呢，嗰個

83
00:02:40,069 --> 00:02:42,872
嗰個，分析個層面會再deep少少嘅

84
00:02:42,872 --> 00:02:44,354
即係，如果係regulatory reporting

85
00:02:44,354 --> 00:02:46,695
可能佢一條大掃一個revenue就去彙報

86
00:02:46,695 --> 00:02:49,478
咁但係我management reporting我係by segment

87
00:02:49,478 --> 00:02:51,819
啫係可能我當我係開一個餐飲集團㗎

88
00:02:51,819 --> 00:02:56,044
啫係每一間分店嗰個銷售係點或者甚至賣咗邊個

89
00:02:56,044 --> 00:02:58,286
邊個套餐邊個飲品係好快啲

90
00:02:58,286 --> 00:03:00,507
咁呢個level嘅the analysis都會做到

91
00:03:00,507 --> 00:03:02,229
management reporting就會有少少唔同

92
00:03:02,229 --> 00:03:02,368
所以

93
00:03:03,703 --> 00:03:04,802
噉第三就係，呃

94
00:03:04,802 --> 00:03:08,724
呃，internal audit啦或者講到第四先係external audit

95
00:03:08,724 --> 00:03:11,425
啫係大家諗成日做audit就係韓國big four嗰啲auditor

96
00:03:11,425 --> 00:03:13,806
咁去唔同有公司做審計咁啦

97
00:03:13,806 --> 00:03:15,167
咁佢哋會集中喺一個

98
00:03:15,167 --> 00:03:16,367
呃financial statement

99
00:03:16,367 --> 00:03:19,449
數字上面啫係啲財務數字上面嘅audit嘅例如

100
00:03:19,449 --> 00:03:23,069
嗰盤數，呃有幾多cash或者幾多account receivable咁

101
00:03:23,069 --> 00:03:25,610
咁佢哋就會去核實究竟係咪你話你有咁多

102
00:03:25,610 --> 00:03:28,611
咁係咪我會去銀行度嗰張back statement睇

103
00:03:28,611 --> 00:03:30,353
咁佢先verify到你講嘅嘢係真定假

104
00:03:33,394 --> 00:03:36,115
咁你internal呢就係內部嗰隻呢就

105
00:03:36,115 --> 00:03:38,635
未必同嗰個，呃

106
00:03:38,635 --> 00:03:41,396
財務有關嘅，啫係譬如我保險公司為例啦

107
00:03:41,396 --> 00:03:43,276
咁可能客戶claim錢

108
00:03:43,276 --> 00:03:44,518
呃由佢claim嗰刻

109
00:03:44,518 --> 00:03:46,677
呃到審批需要幾耐

110
00:03:46,677 --> 00:03:48,679
或者有冇呃，超過咗乜啲limit

111
00:03:48,679 --> 00:03:51,039
啫你個保額去到呢度但我批多咗畀你

112
00:03:51,039 --> 00:03:53,120
或者有冇兩個人對過先批會唔會係

113
00:03:53,120 --> 00:03:55,200
你自己個親戚就係咁好鬆手批畀佢

114
00:03:55,200 --> 00:03:57,382
咁呢啲，internal我哋就會查呢啲嘢囉

115
00:03:57,382 --> 00:03:59,162
咁未必直接同財務有關

116
00:04:02,203 --> 00:04:04,864
申請咗嗰個索償但係好耐都冇消息

117
00:04:04,864 --> 00:04:07,044
三個月六個月都冇消息冇回覆

118
00:04:07,044 --> 00:04:08,944
咁就，會影響公司聲譽啲

119
00:04:08,944 --> 00:04:10,725
咁未必係直接嘅財務嘅

120
00:04:10,725 --> 00:04:12,506
嘅影響但係會有間接

121
00:04:12,506 --> 00:04:14,467
啫係對公司聲譽或者財務影響

122
00:04:14,467 --> 00:04:16,826
咁呢啲就係，internal audit做嘅嘢喇

123
00:04:16,826 --> 00:04:18,567
咁，咁，好簡單講一講

124
00:04:18,567 --> 00:04:20,988
咁今日嘅主題係講 big data AI 嘅

125
00:04:20,988 --> 00:04:21,908
咁所以我哋可以

126
00:04:21,908 --> 00:04:23,009
下一張slide係喇

127
00:04:23,009 --> 00:04:24,629
然後再下一張就係

128
00:04:24,629 --> 00:04:25,810
唔同嘅 big data

129
00:04:25,810 --> 00:04:26,730
啫係首先講一講乜係為啲

130
00:04:30,000 --> 00:04:33,024
噉如果你話big data你去上網去search或者睇一啲

131
00:04:33,024 --> 00:04:34,065
呃textbook

132
00:04:34,065 --> 00:04:35,908
definition通常都會講呢三個嘢就係

133
00:04:35,908 --> 00:04:36,869
volume

134
00:04:36,869 --> 00:04:39,413
variety同埋velocity三個phi啦所謂嘅

135
00:04:40,870 --> 00:04:41,851
係，咁呢三個fee

136
00:04:41,851 --> 00:04:43,512
就係，olume就係嗰個容量啦

137
00:04:43,512 --> 00:04:45,233
因為大數據咁當然係

138
00:04:45,233 --> 00:04:47,355
即係多嘅數據先叫大數據啦

139
00:04:47,355 --> 00:04:48,377
第二就係嗰個

140
00:04:48,377 --> 00:04:50,598
variety嗰個種類

141
00:04:50,598 --> 00:04:51,899
因為而家嘅，呃

142
00:04:51,899 --> 00:04:53,740
數據呢就唔一定係好似我哋

143
00:04:53,740 --> 00:04:54,802
心目中諗嘅excel啫

144
00:04:54,802 --> 00:04:57,002
一行行一個個rowing column嗰種嘅

145
00:04:57,002 --> 00:04:59,644
咁好多種類嘅data都可以係

146
00:04:59,644 --> 00:05:00,386
可以係文字啦

147
00:05:00,386 --> 00:05:01,286
可以係圖畫啦

148
00:05:01,286 --> 00:05:02,947
聲音等等都可以係一個data嚟㗎

149
00:05:03,648 --> 00:05:05,230
咁呢啲其實都可以

150
00:05:05,230 --> 00:05:06,230
你識得利用佢呢

151
00:05:06,230 --> 00:05:07,511
都可以搵到好多

152
00:05:07,511 --> 00:05:08,552
好多，呃，value

153
00:05:08,552 --> 00:05:09,954
出嚟嘅，咁譬如話

154
00:05:09,954 --> 00:05:12,175
呃，即係而家我做保險公司啦

155
00:05:12,175 --> 00:05:16,139
咁有啲客戶會打嚟投訴或者或者或者或者打電話去個客服嗰度問嘢嘅

156
00:05:16,139 --> 00:05:16,798
咁有啲可能係

157
00:05:16,798 --> 00:05:17,620
呃，password

158
00:05:17,620 --> 00:05:18,339
要要reset

159
00:05:18,339 --> 00:05:20,101
有啲就攞錢，有啲就問點解嗰個

160
00:05:20,482 --> 00:05:22,043
個claim咁耐都未批噉樣樣

161
00:05:22,043 --> 00:05:23,663
噉呢啲唔，唔一定係

162
00:05:23,663 --> 00:05:25,103
呃，數字嗰種data嚟嘅

163
00:05:25,103 --> 00:05:27,384
可能係，呃，寫封信嚟投訴

164
00:05:27,384 --> 00:05:29,725
寫封email嚟投訴或者電話

165
00:05:29,725 --> 00:05:31,625
即係噉就咁講語音嘅啫

166
00:05:31,625 --> 00:05:33,925
噉而家有啲technology係語音轉文字啦

167
00:05:33,925 --> 00:05:35,185
噉就變咗文字之後呢

168
00:05:35,185 --> 00:05:36,247
就可以再分析喇

169
00:05:36,247 --> 00:05:37,687
即係可以做一個叫做呃

170
00:05:37,687 --> 00:05:39,327
情感分析 sentimental analysis

171
00:05:39,327 --> 00:05:43,369
即係分析嗰個講嘢嗰段嘢其實係正面定負面嘅啫係會唔會好

172
00:05:43,829 --> 00:05:45,089
客戶好嬲㗎咁樣

173
00:05:45,089 --> 00:05:46,449
咁如果係呃，好嬲應該

174
00:05:46,449 --> 00:05:47,571
file一個complain

175
00:05:47,571 --> 00:05:48,930
有專人去跟進嘅

176
00:05:48,930 --> 00:05:51,093
咁但係佢，即係 somehow聽電話嗰個人就

177
00:05:51,093 --> 00:05:52,833
呃，冇，冇話佢係一個投訴

178
00:05:52,833 --> 00:05:54,233
但係我哋做audit分析到

179
00:05:54,233 --> 00:05:55,254
其實嗰客戶係

180
00:05:55,254 --> 00:05:56,915
想投訴㗎，咁呢啲嘢就會

181
00:05:56,915 --> 00:05:58,435
預先，揾到出嚟

182
00:05:58,435 --> 00:05:59,677
咁就再去跟進

183
00:05:59,677 --> 00:06:00,836
好過，即係佢好嬲

184
00:06:00,836 --> 00:06:02,017
跟住走去，呃

185
00:06:02,017 --> 00:06:02,978
東張西望度投訴

186
00:06:02,978 --> 00:06:03,819
咁就搞大件事

187
00:06:03,819 --> 00:06:04,819
咁就係啦，就唔

188
00:06:11,083 --> 00:06:14,194
做嚟分析嘅噉所以就係第二個v variety就係嗰個

189
00:06:14,194 --> 00:06:14,495
種類喇

190
00:06:16,127 --> 00:06:17,447
咁第三就係，velocity

191
00:06:17,447 --> 00:06:19,249
即係生產data個速度嘅

192
00:06:19,249 --> 00:06:20,771
咁又而家，即係

193
00:06:20,771 --> 00:06:24,173
譬如你社交media每日都有人post嘢上去

194
00:06:24,173 --> 00:06:26,535
咁或者股市每日都有新嘅新嘅news出嚟

195
00:06:26,535 --> 00:06:30,939
咁呢個係，個生產速度好快不停咁有新嘅data出嚟㗎

196
00:06:30,939 --> 00:06:32,459
咁所以你要做big data嘅分析

197
00:06:32,459 --> 00:06:35,641
你係要，識得處理啲一啲

198
00:06:35,641 --> 00:06:37,264
變化得好快嘅data

199
00:06:37,264 --> 00:06:38,363
係喇咁先至可以做得到

200
00:06:38,363 --> 00:06:40,286
咁所以符合呢三個three就為之

201
00:06:40,286 --> 00:06:41,166
big data啦

202
00:06:41,166 --> 00:06:41,187
咁

203
00:06:41,887 --> 00:06:44,048
咁但係你話呃幾大先至叫big data呢

204
00:06:44,048 --> 00:06:44,850
譬如volume嗰度

205
00:06:44,850 --> 00:06:46,230
咁呢個就好呃

206
00:06:46,230 --> 00:06:48,413
冇一個啫係好好確實嘅definition嚟㗎

207
00:06:48,413 --> 00:06:50,653
咁但係，譬如你excel我唔知

208
00:06:50,653 --> 00:06:52,596
大家有冇知道可以處理到幾多data

209
00:06:52,596 --> 00:06:54,398
你拉到最底呢其實係

210
00:06:54,398 --> 00:06:55,999
大概一百萬行到囉

211
00:06:55,999 --> 00:06:58,081
係咁我會，呃

212
00:06:58,081 --> 00:06:58,901
我自己嘅definition

213
00:06:58,901 --> 00:07:01,142
啫係如果excel都處理到嗰啲就唔係big data嚟㗎

214
00:07:01,142 --> 00:07:03,105
啫係要起碼多過呢個數量

215
00:07:03,105 --> 00:07:06,107
你要用啲啲特別嘅tool去處理嗰啲先為啲big data啦

216
00:07:06,107 --> 00:07:06,387
同埋係

217
00:07:07,187 --> 00:07:09,372
呃，除咗volume啫係就係嗰種類

218
00:07:09,372 --> 00:07:11,334
即係譬如文字或者影像嗰啲嘅嘢

219
00:07:11,334 --> 00:07:12,235
你係普通方法

220
00:07:12,235 --> 00:07:14,300
處理唔到，用啲特別嘅方法嗰啲就係big data囉

221
00:07:15,242 --> 00:07:17,685
噉個畫面隔離有個叫做呃

222
00:07:17,685 --> 00:07:19,466
呃，呃，Apache

223
00:07:19,466 --> 00:07:22,509
Spark呢樣嘢係咩嚟嘅就係譬如一部電腦你要呃

224
00:07:22,509 --> 00:07:24,911
撈晒data入去然後做一啲計算噉樣呢

225
00:07:24,911 --> 00:07:26,632
噉你嗰個，一部電腦嗰個

226
00:07:26,632 --> 00:07:28,314
呃ram係有限㗎嘛

227
00:07:28,314 --> 00:07:29,735
噉所以有啲特別嘅方法就係

228
00:07:29,735 --> 00:07:32,218
呃，一個叫做啫係分散式嘅

229
00:07:32,218 --> 00:07:35,161
嘅計算啦，啫係一部電腦係將個task 拆開做

230
00:07:35,161 --> 00:07:37,182
撞畀好多個電腦同時處理

231
00:07:37,182 --> 00:07:38,863
然後處理完再撼返埋一齊

232
00:07:39,704 --> 00:07:41,007
係喇即係好似呃

233
00:07:41,007 --> 00:07:42,487
去超市排隊調隊太長

234
00:07:42,487 --> 00:07:43,449
開到幾個counter咁樣

235
00:07:43,449 --> 00:07:45,190
等嚟分流咁去處理個task

236
00:07:45,190 --> 00:07:46,473
就可以處理到個量大啲囉

237
00:07:46,473 --> 00:07:47,814
咁呢個係其中一個

238
00:07:47,814 --> 00:07:48,555
即係，即係呃

239
00:07:48,555 --> 00:07:50,237
advanced啲嘅技巧去

240
00:07:50,237 --> 00:07:52,478
去處理呢啲big data囉

241
00:07:52,478 --> 00:07:54,661
係喇，噉就，呃下張slide唔該

242
00:07:54,661 --> 00:07:54,762
係喇

243
00:07:55,302 --> 00:07:59,444
噉另外呢頭先講到就係點樣去從data入面搵到一啲insight呢

244
00:07:59,444 --> 00:08:03,125
噉因為一堆一堆文字一堆數字你好難去搵到一啲insight出嚟噉嘛

245
00:08:03,125 --> 00:08:06,206
噉其中一個方法就係一個叫做data visualization

246
00:08:06,206 --> 00:08:07,987
係一個數據嘅可視化啦

247
00:08:07,987 --> 00:08:09,047
噉就見到有個chart

248
00:08:09,047 --> 00:08:10,807
噉就係一間公司嘅一啲

249
00:08:12,187 --> 00:08:14,209
銷售數據嚟你見到呃

250
00:08:14,209 --> 00:08:16,290
左手邊可能有個漏斗噉樣係

251
00:08:16,290 --> 00:08:18,612
呃佢佢去咗，佢approach咗幾多個客

252
00:08:18,612 --> 00:08:20,872
然後幾多個客係已經開始sell緊啦

253
00:08:20,872 --> 00:08:22,012
幾多係出咗proposal

254
00:08:22,012 --> 00:08:23,053
幾多係最後簽咗返嚟

255
00:08:23,053 --> 00:08:25,035
咁一個漏斗形式去show晒出嚟

256
00:08:25,035 --> 00:08:28,757
咁就，呃啫係管理層好簡單一目瞭然就睇得到

257
00:08:28,757 --> 00:08:29,456
發生緊咩事囉

258
00:08:29,456 --> 00:08:33,720
就唔使喺度啫係去嗰啲data嗰度deep dive去睇啦

259
00:08:33,720 --> 00:08:35,561
噉呢一樣嘢就

260
00:08:35,561 --> 00:08:36,461
啫係可能你話

261
00:08:36,461 --> 00:08:37,361
唔係好新啫啫係

262
00:08:37,361 --> 00:08:37,961
平時啲

263
00:08:41,923 --> 00:08:42,964
咁但而家呢一種

264
00:08:42,964 --> 00:08:43,965
呃，叫做，呃

265
00:08:43,965 --> 00:08:46,024
interactive嘅dashboard呢

266
00:08:46,024 --> 00:08:47,066
就係，互動式嘅

267
00:08:47,066 --> 00:08:48,166
咩意思呢？即係話

268
00:08:48,166 --> 00:08:49,746
你組下個，譬如見到幅地圖啦

269
00:08:49,746 --> 00:08:50,706
係，呃，美國啦

270
00:08:50,706 --> 00:08:51,687
唔同州份，佢

271
00:08:51,687 --> 00:08:53,908
佢佢講緊佢啲客戶集中邊啲州份啦

272
00:08:53,908 --> 00:08:55,808
咁你擺啲隻mouse clicked落去呢

273
00:08:55,808 --> 00:08:58,269
佢即刻係，啫啫 filter咗淨係嗰個州份嘅

274
00:08:58,269 --> 00:08:59,389
Click落去加州嗰度

275
00:08:59,389 --> 00:09:01,591
咁呢，其他圖會即刻reflect咗出嚟

276
00:09:01,591 --> 00:09:02,630
就，就睇到

277
00:09:09,212 --> 00:09:10,774
噉點解叫dashboard

278
00:09:10,774 --> 00:09:12,033
因為呃，即係dashboard就叫呃

279
00:09:12,033 --> 00:09:13,054
即係擬標版噉樣啦

280
00:09:13,054 --> 00:09:14,436
好似你揸車，你見到

281
00:09:14,436 --> 00:09:15,596
你眼前見到嗰個

282
00:09:15,596 --> 00:09:16,756
呃，而家速度幾多啊

283
00:09:16,756 --> 00:09:17,917
個油缸整底幾多油

284
00:09:17,917 --> 00:09:19,378
噉，因為你呃

285
00:09:19,378 --> 00:09:20,738
你揸人車嚟唔可以分心去

286
00:09:20,738 --> 00:09:23,419
噉所以一個dashboard就裝晒啲資料好簡單睇到

287
00:09:23,419 --> 00:09:25,260
噉而家呢啲dashboard就係噉嘅同樣嘅道理

288
00:09:25,260 --> 00:09:27,041
就係，雖然你背後可能好多數據

289
00:09:27,041 --> 00:09:27,562
但係你一個

290
00:09:28,302 --> 00:09:30,163
dashboard呢就睇到晒

291
00:09:30,163 --> 00:09:32,245
用啲圖表去睇到晒呢啲嘢囉

292
00:09:32,245 --> 00:09:34,086
係噉我都知有啲公司而家係

293
00:09:34,086 --> 00:09:37,769
啲管理層係encourage 佢哋去唔好用powerpoint 去present

294
00:09:37,769 --> 00:09:39,110
用呢啲dashboard present

295
00:09:39,110 --> 00:09:42,371
因為個，啫係管理層可以即刻睇到啊我想撳

296
00:09:42,371 --> 00:09:45,234
撳邊一個州份或者睇邊一條line嘅表現

297
00:09:45,234 --> 00:09:46,215
即刻撳落去呢

298
00:09:46,215 --> 00:09:47,816
嗰啲數值即刻reflect 咗出嚟

299
00:09:47,816 --> 00:09:54,120
咁就，好過以前就啊我要我要叫人去prepare一份report畀我又要唔知幾多日或者幾個禮拜先出到

300
00:09:54,120 --> 00:09:55,360
噉呢啲係即時㗎即刻

301
00:09:57,722 --> 00:10:01,024
咁而呢個，呢一種dashboard背後嗰個數據呢

302
00:10:01,024 --> 00:10:01,825
亦都可以係，呃

303
00:10:01,825 --> 00:10:03,086
即係static

304
00:10:03,086 --> 00:10:05,265
可能你excel feed入去佢做個表出嚟啦

305
00:10:05,265 --> 00:10:06,307
亦都可以係real time嘅

306
00:10:06,307 --> 00:10:07,967
即係佢背後可以駁data base

307
00:10:07,967 --> 00:10:09,128
即係可能，呃

308
00:10:09,128 --> 00:10:10,808
即係餐飲集團嗰個例子咁

309
00:10:10,808 --> 00:10:12,669
當日買咗幾多碟光沙牛河

310
00:10:12,669 --> 00:10:13,669
咁佢即刻可以

311
00:10:13,669 --> 00:10:14,910
refrash到出嚟咁

312
00:10:14,910 --> 00:10:16,772
即刻知道咗，呃

313
00:10:16,772 --> 00:10:18,052
公司內部嘅數據嘅嘢囉

314
00:10:18,052 --> 00:10:20,432
咁所以，呢個係啫係一個方法啦

315
00:10:20,432 --> 00:10:21,874
去，去，呃，analy

316
00:10:26,696 --> 00:10:28,017
係喇咁，呃，下一張曬唔該係

317
00:10:28,017 --> 00:10:31,318
係喇，咁另外一種嘅

318
00:10:31,318 --> 00:10:33,541
嘅科技去，去分析數據呢就叫做

319
00:10:33,541 --> 00:10:35,961
Graph Analytics

320
00:10:35,961 --> 00:10:38,323
就係呢，呃，除咗係

321
00:10:38,323 --> 00:10:40,826
呃，頭先嗰種嗰種嗰種

322
00:10:40,826 --> 00:10:42,047
嗰種呃數據可視化

323
00:10:42,047 --> 00:10:43,447
另外一種方法就係

324
00:10:43,447 --> 00:10:46,068
呃，detect一啲數據之間嘅關係嘅

325
00:10:46,068 --> 00:10:47,470
例如，有一個note

326
00:10:47,470 --> 00:10:49,611
去connect去另外一個note之間嗰個

327
00:10:49,611 --> 00:10:50,812
有個relationship

328
00:10:50,812 --> 00:10:51,273
有個關係啦

329
00:10:51,673 --> 00:10:54,335
噉我講個例子大家就明啫係大家有上社交媒體

330
00:10:54,335 --> 00:10:55,535
有Facebook嗰啲啦

331
00:10:55,535 --> 00:10:58,077
噉有時你同一個朋友有connection嘅

332
00:10:58,077 --> 00:10:59,658
噉就有條線連住佢

333
00:10:59,658 --> 00:11:02,260
噉佢個Facebook知道晒呢啲connection之後呢

334
00:11:02,260 --> 00:11:03,662
佢就識得推介比你

335
00:11:03,662 --> 00:11:05,504
呃你可能認識邊位比佢朋友

336
00:11:05,504 --> 00:11:07,384
因為你同佢有幾多個common friend

337
00:11:07,384 --> 00:11:10,606
就係因為佢背後有個graph入起晒每個人同每個人之間嘅關係囉

338
00:11:11,488 --> 00:11:13,788
噉呢啲係值錢嘅數據嚟㗎

339
00:11:13,788 --> 00:11:15,129
所以facebook啫係免費畀你用

340
00:11:15,129 --> 00:11:19,110
但係啫係，其實免費嘅嘢就係你就係嗰個產品佢度攞啲data去

341
00:11:19,110 --> 00:11:21,210
去，去賣，去知道晒你每個人嘅

342
00:11:21,210 --> 00:11:23,890
識得邊個人同埋有啲咩興趣嗰啲咁嘅嘢喇

343
00:11:23,890 --> 00:11:25,971
例如右手邊個幅圖呢你會見到

344
00:11:25,971 --> 00:11:29,552
係，係，一啲e-commerce網站裏面嘅數據嚟㗎

345
00:11:29,552 --> 00:11:31,493
譬如大家佢知道大家都鍾意係

346
00:11:31,493 --> 00:11:31,513
呃

347
00:11:32,893 --> 00:11:34,654
打機嘅或者有班人係鍾意踢波嘅

348
00:11:34,654 --> 00:11:37,639
噉，啊，呢啲多人裏面有啲人係鍾意旅行去旅行嗰啲個

349
00:11:37,639 --> 00:11:38,940
噉佢就會推介

350
00:11:38,940 --> 00:11:40,121
呢一啲產品畀你囉

351
00:11:40,121 --> 00:11:42,163
因為佢，佢知道你有啲common friend係

352
00:11:42,163 --> 00:11:43,325
都係鍾意玩呢啲嘢

353
00:11:43,325 --> 00:11:46,448
噉佢就會，知道晒你啲數據就可以推一啲嘅

354
00:11:46,448 --> 00:11:47,809
啊，recommendation畀你囉

355
00:11:48,490 --> 00:11:52,732
咁所以啫係，呢一種graphic analysis可能大家未必成日聽過

356
00:11:52,732 --> 00:11:54,413
但係其實好多大公司

357
00:11:54,413 --> 00:11:56,975
啫係全球最值錢嗰十間公司嗰啲科技公司

358
00:11:56,975 --> 00:11:59,196
Facebook就做social network啦

359
00:11:59,196 --> 00:12:01,537
如果amazon就做e-commerce推薦呢啲

360
00:12:01,537 --> 00:12:03,837
其實都背後有用呢啲technology囉

361
00:12:03,837 --> 00:12:05,197
係呀，咁我哋

362
00:12:05,197 --> 00:12:07,058
啫係我做audit我都有用嘅

363
00:12:07,058 --> 00:12:08,620
一陣間會再講我哋點樣用嚟

364
00:12:08,620 --> 00:12:09,679
搵fraud嘅攞嚟

365
00:12:09,679 --> 00:12:11,421
係喇，咁下張slide唔該

366
00:12:13,413 --> 00:12:15,575
係喇咁頭先講咗大概講到big data啦

367
00:12:15,575 --> 00:12:17,636
講咗點樣用visualisation

368
00:12:17,636 --> 00:12:20,339
去分析啦，咁而家我講下AI喇

369
00:12:20,339 --> 00:12:24,261
咁，即係，在座各位應該都聽過check GBT嗰啲喎

370
00:12:24,261 --> 00:12:26,264
咁，咁，咁但係

371
00:12:26,264 --> 00:12:27,965
喺check GBT未

372
00:12:27,965 --> 00:12:29,166
即係出咗一兩年到啫其實

373
00:12:29,166 --> 00:12:31,288
但係佢未出之前呢已經係

374
00:12:31,288 --> 00:12:32,828
AI呢樣嘢已經存在咗

375
00:12:32,828 --> 00:12:36,111
咁我叫佢做啫係傳統AI或者classical or traditional AI喇

376
00:12:36,111 --> 00:12:38,533
咁就係唔係而家嗰種generative AI

377
00:12:38,533 --> 00:12:40,054
啫係生成式AI嗰種嘅

378
00:12:40,054 --> 00:12:41,456
咁，咁呢種，AI就係

379
00:12:43,518 --> 00:12:44,337
呃，點樣做出嚟呢

380
00:12:44,337 --> 00:12:46,158
噉亦都，我覺得有啲人係

381
00:12:46,158 --> 00:12:47,600
呃，坊間有啲誤會嘅

382
00:12:47,600 --> 00:12:51,121
就係，覺得，電腦自動發做一啲嘢嗰啲就係AI

383
00:12:51,121 --> 00:12:53,243
噉嗰啲automation未必一定係AI做嘅

384
00:12:53,243 --> 00:12:54,644
可能係一個叫做rule base

385
00:12:54,644 --> 00:12:56,625
根據有啲規則去做出嚟嘅

386
00:12:56,625 --> 00:12:58,265
啫係譬如話，呃

387
00:12:58,265 --> 00:13:00,528
我program咗一樣嘢就係if 一樣嘢發生

388
00:13:00,528 --> 00:13:02,149
突然有樣嘢就會發生咗

389
00:13:02,149 --> 00:13:03,950
噉有啲ruleset咗畀佢呢

390
00:13:03,950 --> 00:13:05,190
嗰種呢其實係叫做

391
00:13:05,190 --> 00:13:07,192
呃，rule base嘅一個

392
00:13:07,192 --> 00:13:07,211
呃

393
00:13:09,445 --> 00:13:09,965
嘅 program

394
00:13:09,965 --> 00:13:10,885
唔係一個 machine learning

395
00:13:10,885 --> 00:13:11,466
唔係一個AI嚟

396
00:13:11,466 --> 00:13:12,748
噉 machine learning 係咩呢

397
00:13:12,748 --> 00:13:14,028
就係，有，有兩大類嘅

398
00:13:14,028 --> 00:13:16,210
就係 machine 佢自己去搵出嗰個 rule 出嚟

399
00:13:16,210 --> 00:13:17,150
你唔需要 program 畀佢

400
00:13:17,150 --> 00:13:19,052
佢自己，透過去讀好多 data

401
00:13:19,052 --> 00:13:20,114
佢自己搵到個

402
00:13:20,114 --> 00:13:21,254
個規，個 pattern 出嚟嘅

403
00:13:21,254 --> 00:13:23,216
噉陣間又會有例子去講點樣做

404
00:13:23,216 --> 00:13:25,197
噉但係，呃，兩大類啦

405
00:13:25,197 --> 00:13:26,519
一個叫做 supervised learning

406
00:13:26,519 --> 00:13:27,980
一個叫做 unsupervised learning

407
00:13:29,231 --> 00:13:29,991
噉有咩分別呢

408
00:13:29,991 --> 00:13:31,592
Superfinery就係

409
00:13:31,592 --> 00:13:35,774
上面嗰幅圖，就係呢譬如我有一咋蘋果係個input data嚟嘅

410
00:13:35,774 --> 00:13:40,157
然後有個label呢叫annotation呢就係啊呢啲係蘋果呢啲係apples噉樣

411
00:13:40,157 --> 00:13:41,937
啲係蘋果啦，噉

412
00:13:41,937 --> 00:13:43,759
噉但係呢，呃

413
00:13:43,759 --> 00:13:45,620
你當你話咗畀佢聽呢啲係蘋果

414
00:13:45,620 --> 00:13:47,541
啫係畀幅圖佢同埋畀一個答案佢呢個係蘋果

415
00:13:47,541 --> 00:13:50,302
你將佢fit入去嗰個machine learning model呢

416
00:13:50,302 --> 00:13:51,302
佢就會識得學習

417
00:13:51,302 --> 00:13:53,222
你下次再show一幅蘋果嘅圖畀佢睇呢

418
00:13:53,222 --> 00:13:53,923
佢就知道嗰個係

419
00:13:54,864 --> 00:13:57,024
或者，呃，另外有一堆data就一堆橙

420
00:13:57,024 --> 00:13:58,184
然後話俾佢聽呢啲係橙

421
00:13:58,184 --> 00:14:00,346
咁佢就會，當你fit嘅數據夠多

422
00:14:00,346 --> 00:14:02,486
佢就識得去應呢啲pattern囉

423
00:14:02,486 --> 00:14:03,626
咁，呃，好多例子

424
00:14:03,626 --> 00:14:04,807
好多好成熟嘅例子

425
00:14:04,807 --> 00:14:06,988
例如，停車場應車牌嗰啲呢

426
00:14:06,988 --> 00:14:09,528
應數字，應英文字呢啲已經係做得好成熟㗎

427
00:14:09,528 --> 00:14:12,370
佢會識得，佢因為會fit好多唔同嘅車牌嘅

428
00:14:12,370 --> 00:14:15,331
圖片畀佢，然後話俾佢聽呢個係個車牌number幾多

429
00:14:15,331 --> 00:14:15,530
咁佢，佢，

430
00:14:19,215 --> 00:14:21,418
咁呢一種叫supervised learning啦

431
00:14:21,418 --> 00:14:22,099
即係有個，有input

432
00:14:22,099 --> 00:14:23,259
有個答案畀佢

433
00:14:23,259 --> 00:14:26,585
咁佢，當佢，當佢知道得夠多嘅

434
00:14:26,585 --> 00:14:28,047
呢啲叫question answer pair

435
00:14:28,047 --> 00:14:30,931
之後呢，佢，對一啲未見過嘢冇答案嘅

436
00:14:30,931 --> 00:14:32,493
佢都知道個答案係咩囉

437
00:14:32,493 --> 00:14:33,634
咁呢一種叫upervised learning

438
00:14:34,894 --> 00:14:37,517
另外一種下面嗰度叫做 unsupervised learning 呢

439
00:14:37,517 --> 00:14:38,636
就係你冇答案畀佢

440
00:14:38,636 --> 00:14:41,678
你純粹畀一堆生果嘅圖片佢蘋果橙香蕉咩畀佢

441
00:14:41,678 --> 00:14:42,798
你唔話畀佢聽

442
00:14:42,798 --> 00:14:44,179
呢個係蘋果，呢個係橙

443
00:14:44,179 --> 00:14:46,061
呢個係香蕉，你就咁 fit 畀佢啫

444
00:14:46,061 --> 00:14:47,721
咁佢要做嘅嘢呢就係

445
00:14:47,721 --> 00:14:48,902
唔係估呢個係蘋果

446
00:14:48,902 --> 00:14:50,263
呢個係橙，佢係將佢分類㗎

447
00:14:50,263 --> 00:14:50,923
即係冇一個答案㗎

448
00:14:50,923 --> 00:14:52,563
咁呢個就叫 unsupervised learning

449
00:15:04,048 --> 00:15:06,571
呃，有，呢兩樣嘢都係喺

450
00:15:06,571 --> 00:15:09,951
喺我哋做audit係有應用到嘅其實

451
00:15:09,951 --> 00:15:12,313
噉我哋可以去下一張slide度

452
00:15:12,313 --> 00:15:13,453
我講詳細少少

453
00:15:13,453 --> 00:15:17,115
頭先我話去train一個model係咩意思啦

454
00:15:17,115 --> 00:15:18,076
噉譬如呢兩個例子啦

455
00:15:18,076 --> 00:15:21,418
噉，呃，有啲學生有啲成績同埋有出席率嘅

456
00:15:21,418 --> 00:15:24,200
同埋佢最尾嗰個人係合格定唔合格咁樣嘅

457
00:15:25,019 --> 00:15:27,562
咁，嗰個rule呢就係

458
00:15:27,562 --> 00:15:29,023
個成績如果六十分以上

459
00:15:29,023 --> 00:15:31,004
然後出席率係七十五percent以上呢

460
00:15:31,004 --> 00:15:32,664
咁就，即係兩個條件同時符合呢

461
00:15:32,664 --> 00:15:34,947
就合格啦，如果唔係就唔合格

462
00:15:34,947 --> 00:15:36,028
咁我哋可以寫一個

463
00:15:36,028 --> 00:15:37,089
即係好簡啲excel啫

464
00:15:37,089 --> 00:15:38,929
寫條formula

465
00:15:38,929 --> 00:15:40,931
咁呀呢個加呢個等如

466
00:15:40,931 --> 00:15:42,491
即係大過幾多if幾多

467
00:15:42,491 --> 00:15:45,114
跟住n，另外嗰個大過一百percent

468
00:15:45,114 --> 00:15:46,294
咁啫啫係set條rule出嚟

469
00:15:46,294 --> 00:15:47,936
跟住拉條formula落嚟呢

470
00:15:47,936 --> 00:15:48,736
就做到呢個計算

471
00:15:48,736 --> 00:15:50,158
咁呢個好明顯就唔係AI啦

472
00:15:50,158 --> 00:15:50,717
呢個係一個，r

473
00:15:57,121 --> 00:16:00,923
如果係呃我哋做machine learning會點樣做呢去下張slide就係

474
00:16:00,923 --> 00:16:02,183
嗱我有一堆學生

475
00:16:02,183 --> 00:16:05,264
佢嘅成績，同埋佢嘅出席率

476
00:16:05,264 --> 00:16:07,306
然後呢我fit入去個模型嗰度

477
00:16:07,306 --> 00:16:09,246
但我唔話畀個模型聽嗰個rule係乜嘢

478
00:16:09,246 --> 00:16:11,548
即係究竟幾多分先為之合格我唔話畀佢聽嘅

479
00:16:11,548 --> 00:16:12,668
但我就話個答案畀佢聽

480
00:16:12,668 --> 00:16:15,009
即係究竟最尾個result係合格定唔合格話畀佢聽

481
00:16:15,748 --> 00:16:17,609
咁嗰個你fit得多數據落去呢

482
00:16:17,609 --> 00:16:18,969
呢個模型佢會自己

483
00:16:18,969 --> 00:16:21,350
呢個Machine Learning模型佢自己會揾到

484
00:16:21,350 --> 00:16:23,350
即通過呢個叫training phase

485
00:16:23,350 --> 00:16:24,751
呢個training嘅過程呢

486
00:16:24,751 --> 00:16:26,332
佢會揾到個pattern出嚟㗎

487
00:16:26,332 --> 00:16:27,552
即係可能佢呃見到啊

488
00:16:27,552 --> 00:16:29,053
五十五分，就fail

489
00:16:29,053 --> 00:16:29,994
但係六十五就合格

490
00:16:29,994 --> 00:16:31,575
咁佢會知道六十分就係嗰條

491
00:16:31,575 --> 00:16:34,836
嗰條threshold嗰個cutoff point

492
00:16:34,836 --> 00:16:36,197
以上就合格，以下就唔合格

493
00:16:36,197 --> 00:16:38,378
即係佢會自己搵出嚟嘅yolo就唔係你program畀佢嘅

494
00:16:38,758 --> 00:16:39,778
咁呢個就為之

495
00:16:39,778 --> 00:16:40,899
呃，Machine Learning啊

496
00:16:40,899 --> 00:16:42,580
咁，咁你話嗰個模型

497
00:16:42,580 --> 00:16:45,162
即係實際上個背後操作點解會識得搵呢樣嘢呢

498
00:16:45,162 --> 00:16:46,222
咁就，好複雜囉

499
00:16:46,222 --> 00:16:47,222
可能唔係呢個

500
00:16:47,222 --> 00:16:48,224
呢個talk可以講得到

501
00:16:48,224 --> 00:16:49,183
但係嗰啲例子

502
00:16:49,183 --> 00:16:51,105
啫係最簡單就係一啲regression model

503
00:16:51,105 --> 00:16:52,206
啫係數學嗰啲

504
00:16:52,206 --> 00:16:54,267
乜，乜等於，MX價試個model

505
00:16:54,267 --> 00:16:57,428
咁佢有個，X係個input啦

506
00:16:57,428 --> 00:17:00,150
然後嗰個嗰個嗰條curve嘅嗰個呃

507
00:17:00,150 --> 00:17:01,471
斜度就係個coefficient啦

508
00:17:01,471 --> 00:17:02,272
咁，佢係透過

509
00:17:08,736 --> 00:17:11,019
噉另外有啲叫tree base嘅model就係

510
00:17:11,019 --> 00:17:13,040
呃，喺頂頭分開兩條tree

511
00:17:13,040 --> 00:17:13,441
兩條path

512
00:17:13,441 --> 00:17:14,162
yes，就去呢邊

513
00:17:14,162 --> 00:17:14,883
no，就喺嗰邊

514
00:17:14,883 --> 00:17:15,983
噉一路落到底

515
00:17:15,983 --> 00:17:20,729
噉最尾佢就將一堆嘢分咗做究竟係合格同唔合格嘅學生噉樣囉

516
00:17:20,729 --> 00:17:22,009
噉仲有一啲再複製啲係

517
00:17:22,009 --> 00:17:22,911
呃，neural

518
00:17:22,911 --> 00:17:25,153
network嗰啲呢就比較係

519
00:17:25,153 --> 00:17:25,173
呃

520
00:17:25,894 --> 00:17:27,134
即係叫black box model

521
00:17:27,134 --> 00:17:29,076
你唔係好知佢裏面發生咩事

522
00:17:29,076 --> 00:17:29,696
即而家啲gen

523
00:17:29,696 --> 00:17:30,597
AI，係噉樣樣

524
00:17:30,597 --> 00:17:32,058
你佢好似識人類噉講嘢

525
00:17:32,058 --> 00:17:34,682
但係你佢解釋點解佢出嗰個output係

526
00:17:34,682 --> 00:17:36,542
難啲解釋，因為佢背後可能係

527
00:17:36,542 --> 00:17:39,065
即係想益個唔同嘅參數去tune出嚟

528
00:17:39,065 --> 00:17:40,787
咁所以嗰啲就

529
00:17:40,787 --> 00:17:41,446
會，會係

530
00:17:42,407 --> 00:17:43,630
powerful啲同埋準啲

531
00:17:43,630 --> 00:17:46,374
但係呢你要解釋畀人聽點解佢去發生啲咁嘅事

532
00:17:46,374 --> 00:17:47,234
係難啲解釋囉

533
00:17:47,234 --> 00:17:48,916
咁所以，用唔同嘅algorithm

534
00:17:48,916 --> 00:17:50,179
有好有唔好，有啲簡單啲

535
00:17:50,179 --> 00:17:51,420
有啲複雜啲，有啲準啲

536
00:17:51,420 --> 00:17:52,642
有啲解釋到佢解釋唔到

537
00:17:52,642 --> 00:17:54,124
咁樣係囉，咁呢個係

538
00:17:54,124 --> 00:17:55,244
一個好大學問嚟嘅

539
00:17:55,244 --> 00:17:56,046
點樣去做呢樣嘢

540
00:17:57,063 --> 00:17:58,563
咁呢個就係個training process啦

541
00:17:58,563 --> 00:18:00,503
啫係，你fit夠多data畀佢

542
00:18:00,503 --> 00:18:01,703
咁除咗夠多data呢

543
00:18:01,703 --> 00:18:03,124
嗰個data嘅

544
00:18:03,124 --> 00:18:04,364
嘅 quality都好緊要嘅

545
00:18:04,364 --> 00:18:05,744
啫係譬如我fit一大堆學生畀佢

546
00:18:05,744 --> 00:18:07,325
但係全部都係啲好好嘅學生

547
00:18:07,325 --> 00:18:08,664
全部攞一百分

548
00:18:08,664 --> 00:18:09,484
出值晒一百分

549
00:18:09,484 --> 00:18:11,826
咁，咁我就唔知唔合格

550
00:18:11,826 --> 00:18:12,826
係點樣trigger到囉

551
00:18:12,826 --> 00:18:14,125
啫係可能，一百分

552
00:18:14,125 --> 00:18:15,286
八十分，七十分都喺度

553
00:18:15,286 --> 00:18:16,866
但係冇一啲唔合格嘅case

554
00:18:16,866 --> 00:18:18,326
或者下一個唔合格應該係零分啦

555
00:18:18,326 --> 00:18:19,426
咁我就唔知道

556
00:18:19,426 --> 00:18:19,446
嗰

557
00:18:24,688 --> 00:18:28,209
咁而家呢去到下一個phase就係testing phase啦就係

558
00:18:28,209 --> 00:18:29,369
啊我個model搵到

559
00:18:29,369 --> 00:18:30,809
即係有條公式計要出嚟啦

560
00:18:30,809 --> 00:18:32,171
知道合格同唔合格

561
00:18:32,171 --> 00:18:33,391
點樣去determine啊

562
00:18:33,391 --> 00:18:35,652
咁但係我個model準唔準呢

563
00:18:35,652 --> 00:18:38,393
就要睇同嗰個真實數據去比較

564
00:18:38,393 --> 00:18:41,034
咁所以呢譬如我哋有一萬條數據咁啦

565
00:18:41,034 --> 00:18:43,095
我哋會將啲數據呢分咗做

566
00:18:43,095 --> 00:18:46,016
呃testing set同埋training set嘅

567
00:18:46,016 --> 00:18:49,076
咁如果我哋攞晒一萬條數據去做training呢

568
00:18:49,076 --> 00:18:51,218
唔留返啲做testing有咩

569
00:18:51,218 --> 00:18:51,978
有咩後果呢就係

570
00:18:52,417 --> 00:18:55,820
有一樣嘢叫overfitting嘅出現就係話呢我

571
00:18:55,820 --> 00:18:57,342
呃，對於我嘅training set

572
00:18:57,342 --> 00:18:58,282
我就做得好準

573
00:18:58,282 --> 00:19:00,423
但一去到一啲我未見過嘢呢我就會

574
00:19:00,423 --> 00:19:06,548
表現到好差喇啫係如果舉個啫係用教書嚟做例子就係可能個學生讀嗰啲膜卷讀到好熟啲答

575
00:19:06,548 --> 00:19:07,789
啲膜當時係背晒嘅

576
00:19:07,789 --> 00:19:09,050
但一出到去一啲

577
00:19:09,050 --> 00:19:15,056
呃，史題佢未見過呢就完全唔識答錯嘅喺度噉因為佢唔理解個內容講乜佢淨係

578
00:19:15,056 --> 00:19:18,479
呃，背咗個膜嗰啲答案噉噉所以呢

579
00:19:19,378 --> 00:19:21,461
呃，要點樣做呢就係一萬條啦

580
00:19:21,461 --> 00:19:23,082
咁我攞八千條出嚟做training呢

581
00:19:23,082 --> 00:19:24,102
就係揞住個答案

582
00:19:24,102 --> 00:19:25,363
啊sorry

583
00:19:25,363 --> 00:19:26,483
八千條呢就有

584
00:19:26,483 --> 00:19:27,765
有input同埋答案嘅

585
00:19:27,765 --> 00:19:28,884
去等佢去train出嚟

586
00:19:28,884 --> 00:19:32,007
就係定唔得二千條呢就揞住個答案唔好畀佢知個答案係乜

587
00:19:32,007 --> 00:19:34,969
要佢用嗰八千條train出嚟個model去predict

588
00:19:34,969 --> 00:19:36,430
嗰二千條個答案係乜

589
00:19:36,430 --> 00:19:37,530
然後就兩者比較

590
00:19:37,530 --> 00:19:38,912
睇下你估出嚟個答案

591
00:19:38,912 --> 00:19:41,212
同埋真實嗰個答案係相差有幾遠

592
00:19:41,212 --> 00:19:42,894
咁你就知道你個model有幾accurate囉

593
00:19:43,694 --> 00:19:45,696
係就係噉樣去衡量你個model嘅accuracy

594
00:19:45,696 --> 00:19:47,338
噉呢個過程係會

595
00:19:47,338 --> 00:19:49,840
不停重複嘅，只係一開始你得個六成嘅準確度

596
00:19:49,840 --> 00:19:51,142
噉你再tune啲parameter

597
00:19:51,142 --> 00:19:55,246
點樣tune到佢係去到八成九成嘅準確度噉樣囉

598
00:19:55,246 --> 00:19:58,609
噉然後除咗用accuracy去去判斷嗰嘢嘢準確度呢

599
00:19:58,609 --> 00:19:59,691
仲有好多方法嘅

600
00:19:59,691 --> 00:19:59,790
因為

601
00:20:00,391 --> 00:20:03,071
譬如呃啫係，大家之前嗰幾年covid啦

602
00:20:03,071 --> 00:20:05,452
有聽過嗰啲假陽性假陰性嗰啲嘅啦

603
00:20:05,452 --> 00:20:09,035
意思就係話，佢嗰支檢測棒見唔到你係陽性

604
00:20:09,035 --> 00:20:10,895
但係實際上你其實係陰性嘅

605
00:20:10,895 --> 00:20:13,317
咁嗰啲就叫假陽性啦啫係force positive啦

606
00:20:13,317 --> 00:20:16,219
咁有啲metrics就要睇下你個force positive rate或者force

607
00:20:16,219 --> 00:20:16,778
 negative rate

608
00:20:16,778 --> 00:20:19,299
有幾高去睇下你個algebra唔準唔準囉

609
00:20:20,050 --> 00:20:23,772
咁而用喺唔同嘅行業或者領域呢有唔同嘅

610
00:20:23,772 --> 00:20:24,653
處理方法啫譬如

611
00:20:24,653 --> 00:20:26,474
即係譬如，我係落廣告嘅人士

612
00:20:26,474 --> 00:20:26,994
我係sell

613
00:20:26,994 --> 00:20:29,096
我係，send啲sms畀人嘅

614
00:20:29,096 --> 00:20:31,218
咁嗰個客戶有冇興趣

615
00:20:31,218 --> 00:20:34,961
同埋你有冇，呃賣廣告畀佢呢就會係嗰個

616
00:20:34,961 --> 00:20:37,041
啫譬如，即係譬如個客戶係有興趣

617
00:20:37,041 --> 00:20:38,343
而你冇賣到畀佢呢

618
00:20:38,343 --> 00:20:40,604
咁就係，你miss咗一個opportunity啦

619
00:20:40,604 --> 00:20:41,305
你冇sell佢啦

620
00:20:41,305 --> 00:20:42,505
但係掉返轉如果佢冇興趣

621
00:20:42,505 --> 00:20:43,865
你又硬 sell佢呢

622
00:20:50,009 --> 00:20:53,051
咁嗰啲，但喺呢個情況底下你就tune到個model係

623
00:20:53,051 --> 00:20:58,154
你情願，呃，多啲人係畀你煩到都唔好miss咗啲潛在客戶囉

624
00:20:58,154 --> 00:20:59,895
啫係你會係嗰個呃

625
00:20:59,895 --> 00:21:01,896
force嘅positive

626
00:21:01,896 --> 00:21:03,217
你唔介意佢高少少囉

627
00:21:03,217 --> 00:21:05,219
啫佢冇興趣你都照寫你唔介意佢高少少

628
00:21:05,219 --> 00:21:06,519
因為，冇乜所謂係嘛

629
00:21:06,519 --> 00:21:07,359
你最多煩到佢

630
00:21:07,359 --> 00:21:08,339
但有啲case例如呃

631
00:21:08,339 --> 00:21:10,340
你係判斷一個人有冇cancer嘅

632
00:21:10,340 --> 00:21:11,461
咁如果佢冇你強行話佢有

633
00:21:11,461 --> 00:21:13,702
咁佢哋對佢啲心靈上嘅影

634
00:21:19,006 --> 00:21:20,228
即係呢個係好簡化嘅

635
00:21:20,228 --> 00:21:21,630
即係淨係啱同唔啱

636
00:21:21,630 --> 00:21:23,393
但係，如果實際做落係

637
00:21:23,393 --> 00:21:25,557
唔同use case有唔同嘅metric去

638
00:21:25,557 --> 00:21:27,219
去睇佢究竟佢做得啱唔啱㗎囉

639
00:21:27,219 --> 00:21:29,883
係啊，咁所以我哋可以去

640
00:21:29,883 --> 00:21:30,324
發張slide

641
00:21:33,088 --> 00:21:34,951
係啦咁頭先就講咗superfied learning啦

642
00:21:34,951 --> 00:21:35,791
啫係你知個答案

643
00:21:35,791 --> 00:21:36,992
你估個答案係乜

644
00:21:36,992 --> 00:21:39,214
另外一種unsuperfied learning就係你冇答案㗎

645
00:21:39,214 --> 00:21:41,196
純粹係，左手邊有一堆客戶

646
00:21:41,196 --> 00:21:42,137
咁你將佢分咗類

647
00:21:42,137 --> 00:21:43,420
你發覺有一堆係

648
00:21:43,420 --> 00:21:44,320
就嚟退休人士

649
00:21:44,320 --> 00:21:45,821
佢哋嗰個，嗰個

650
00:21:45,821 --> 00:21:46,883
呃，behavior

651
00:21:46,883 --> 00:21:49,165
係相近啲嘅，例如鍾意一啲儲蓄產品

652
00:21:49,165 --> 00:21:51,087
係，呃，保守少少嘅

653
00:21:51,428 --> 00:21:56,130
有啲啱啱呃新嘅families就有另外一種啫可能有收入好高但係

654
00:21:56,130 --> 00:21:58,010
儲蓄未太高咁又另外一種pattern

655
00:21:58,010 --> 00:22:01,212
然後後生仔可能消費比較多啲咁啫係你會見到

656
00:22:01,212 --> 00:22:02,752
你預先你冇一個

657
00:22:02,752 --> 00:22:04,713
特定嘅諗法我要點樣去predict啲咩

658
00:22:04,713 --> 00:22:06,434
你只不過fit入去model

659
00:22:06,434 --> 00:22:07,375
將啲客戶分類

660
00:22:07,375 --> 00:22:10,076
咁你分完類之後你再去睇下啲啲每一個

661
00:22:10,076 --> 00:22:12,438
每一個客戶群組你要點樣去應對咁樣囉

662
00:22:13,598 --> 00:22:16,662
咁又或者呃，即係左手邊就marketing嗰啲比較多啲啦

663
00:22:16,662 --> 00:22:18,022
右手邊呢就係

664
00:22:18,022 --> 00:22:19,525
呃，你見到有堆數據

665
00:22:19,525 --> 00:22:21,205
咁呢你就，呃

666
00:22:21,205 --> 00:22:21,967
有啲cluster嘅

667
00:22:21,967 --> 00:22:23,489
即係有某堆數據黐埋一齊

668
00:22:23,489 --> 00:22:25,530
就係佢哋嗰個特性比較相似

669
00:22:25,530 --> 00:22:27,231
但係有啲突著出嚟啫係

670
00:22:27,231 --> 00:22:29,974
呃，譬如，橙色嗰一點就咩個圈嘅

671
00:22:29,974 --> 00:22:33,618
咁佢，佢個特性同其它嘅data point係好唔一樣嘅

672
00:22:33,618 --> 00:22:35,319
咁呢個會係機會有一個outlier囉

673
00:22:36,240 --> 00:22:38,762
啫譬如你當呢啲係啲信用卡嘅transaction啦

674
00:22:38,762 --> 00:22:40,663
噉譬如我，我平時都係去

675
00:22:40,663 --> 00:22:41,785
七十一買下嘢啫

676
00:22:41,785 --> 00:22:42,786
或者去啲，呃

677
00:22:42,786 --> 00:22:43,606
快餐店食下嘢

678
00:22:43,606 --> 00:22:45,548
但係突然間有一日我係呃

679
00:22:45,548 --> 00:22:49,391
去咗金浦賣金或者先去某一間Starbucks我平時唔飲咖啡

680
00:22:49,391 --> 00:22:50,791
突然間買杯咖啡之後去咗金浦

681
00:22:51,251 --> 00:22:52,751
又去咗買名牌袋

682
00:22:52,751 --> 00:22:55,173
噉就可，又會唔會有人偷咗張卡去

683
00:22:55,173 --> 00:22:58,232
即係我，我依個消費號同平時嗰個pattern好唔一樣啊

684
00:22:58,232 --> 00:22:59,933
噉佢就會搵佢出嚟囉

685
00:22:59,933 --> 00:23:01,973
噉同埋嗰個，呃啫係嗰個

686
00:23:01,973 --> 00:23:03,575
呃，疑犯可能佢係

687
00:23:03,575 --> 00:23:05,694
知道相隔公司有呢啲噉樣嘅

688
00:23:05,694 --> 00:23:08,194
嘅嘢去detect到你㗎

689
00:23:08,194 --> 00:23:09,615
噉我首先去買杯咖啡先

690
00:23:09,615 --> 00:23:10,536
首先睇下公司

691
00:23:10,536 --> 00:23:12,695
呃，相隔公司會唔會有啖啖先

692
00:23:12,695 --> 00:23:14,696
如果冇事啦噉先至去買貴嘢噉樣

693
00:23:14,696 --> 00:23:14,797
佢去

694
00:23:22,098 --> 00:23:29,003
咁啊，下張牌唔該係

695
00:23:29,003 --> 00:23:31,664
好喇咁頭先講咗啫係所謂傳統啲AI喇

696
00:23:31,664 --> 00:23:33,846
咁而家，而家就係啫係

697
00:23:33,846 --> 00:23:34,928
最興就係ChatGPT嗰種啦

698
00:23:34,928 --> 00:23:36,548
GenerativeAI啦

699
00:23:36,548 --> 00:23:37,449
生成式嘅AI啦

700
00:23:37,449 --> 00:23:38,910
咁佢個特色就係可以

701
00:23:38,910 --> 00:23:40,131
生成一啲文字出嚟啦

702
00:23:40,131 --> 00:23:41,491
text generation啦

703
00:23:41,491 --> 00:23:42,833
亦都可以生成圖片

704
00:23:42,833 --> 00:23:43,772
啫係或者直畫幅

705
00:23:43,772 --> 00:23:45,775
你打啲字去畫幅畫出嚟畀你都得嘅

706
00:23:45,775 --> 00:23:47,496
或者成條影片做出嚟都得嘅

707
00:23:48,096 --> 00:23:49,797
甚至乎係一啲語音

708
00:23:49,797 --> 00:23:51,278
頭先講過語音識別啦

709
00:23:51,278 --> 00:23:56,580
即係，呃，我講一段對話錄音佢變成文字啦將佢

710
00:23:56,580 --> 00:23:58,162
或者掉返轉我畀啲文字佢

711
00:23:58,162 --> 00:24:00,983
佢好似人咁咩讀啲聲出來都得㗎

712
00:24:00,983 --> 00:24:03,505
係喇，甚至乎你打字出啲音樂都得

713
00:24:03,505 --> 00:24:07,106
即係，呃唔知大家有冇上YouTube睇過啲啲啲AI王家區嘅嘢

714
00:24:07,106 --> 00:24:09,489
雲光嗰啲噉，咁好多呢啲咁樣嘅例子啫係

715
00:24:09,489 --> 00:24:10,890
可以可以扮一個人把聲嘅

716
00:24:11,509 --> 00:24:14,173
係呀噉噉噉，咁雖然呢方面就呃

717
00:24:14,173 --> 00:24:15,915
其實未有啲generativeAI之前呢

718
00:24:15,915 --> 00:24:19,840
啫係如果英文或者普通話呢啲已經存在啲technology

719
00:24:19,840 --> 00:24:21,042
啫係廣東話就唔係太準囉

720
00:24:21,042 --> 00:24:24,086
呃啫係語音變變文字嗰啲

721
00:24:24,086 --> 00:24:27,531
但係呃，而家出咗openAI嗰啲之後都都好準下

722
00:24:27,531 --> 00:24:28,232
啫係就算廣東話都

723
00:24:29,398 --> 00:24:30,921
噉，因為我哋呃

724
00:24:30,921 --> 00:24:32,561
可以集中，我有幾個例子嘅

725
00:24:32,561 --> 00:24:34,784
關於呃，text generation文字嘅例子

726
00:24:34,784 --> 00:24:37,567
噉可以去下張slide嗰度可以

727
00:24:37,567 --> 00:24:39,868
畀大家睇下啫係

728
00:24:39,868 --> 00:24:40,589
呃，首先呢就

729
00:24:40,589 --> 00:24:43,853
又想講啫係就算呢啲generated a i呢未出之前呢

730
00:24:43,853 --> 00:24:47,816
喺呢個，呃，machine learning或者AI領域有樣嘢叫做natural 

731
00:24:47,816 --> 00:24:48,616
language processing

732
00:24:48,616 --> 00:24:50,538
啫係一啲自然文字嘅處理

733
00:24:50,538 --> 00:24:53,402
就係話呢，一大堆文字佢可以做一啲

734
00:24:53,402 --> 00:24:53,422
呃，

735
00:24:54,842 --> 00:24:55,522
一啲task嘅

736
00:24:55,522 --> 00:24:56,663
譬如啦，summarization

737
00:24:56,663 --> 00:24:57,604
即係做一個總結

738
00:24:57,604 --> 00:24:58,943
你畀篇文章佢

739
00:24:58,943 --> 00:25:00,505
佢幫你做一個總結出嚟

740
00:25:00,505 --> 00:25:02,705
即係未有改GPT前已經有呢啲嘢喎

741
00:25:02,705 --> 00:25:04,126
或者係，classification

742
00:25:04,126 --> 00:25:07,147
即係分類啦，譬如係

743
00:25:07,147 --> 00:25:07,768
我係一間，呃

744
00:25:09,489 --> 00:25:11,309
工廠噉樣啦，跟住我有啲關於

745
00:25:11,309 --> 00:25:14,789
呃，一啲機器損壞嘅一啲報告噉樣嘅文字出嚟

746
00:25:14,789 --> 00:25:16,770
咁你可以再去分類邊一啲係

747
00:25:16,770 --> 00:25:18,451
呃，個零件係

748
00:25:18,451 --> 00:25:21,832
呃，冇冇一個零件或者係個機器損壞咗或者客戶投訴

749
00:25:21,832 --> 00:25:23,211
咁佢識得去咁樣分類㗎

750
00:25:23,211 --> 00:25:25,093
你可以，都係頭先嗰種默列嚟個方法

751
00:25:25,093 --> 00:25:26,653
或者首先畀一篇文佢

752
00:25:26,653 --> 00:25:28,693
然後講一個，話畀佢聽個答案個分類

753
00:25:28,693 --> 00:25:29,834
人類會點樣分

754
00:25:29,834 --> 00:25:32,134
咁你當你fit得夠到呢一種分類嘅data嘅

755
00:25:32,134 --> 00:25:32,275
佢就

756
00:25:38,435 --> 00:25:40,498
又或者translation啦翻譯啦

757
00:25:40,498 --> 00:25:41,980
又或者sentimental analysis

758
00:25:41,980 --> 00:25:43,501
即係嗰個情感分析

759
00:25:43,501 --> 00:25:46,505
即係譬如客戶講嗰段嘢係正面定係負面

760
00:25:46,505 --> 00:25:49,288
啫係你去一啲e-commerce網站啲人留comment

761
00:25:49,288 --> 00:25:51,510
究竟係讚緊個產品定係話緊佢唔好

762
00:25:51,510 --> 00:25:52,872
咁就呢一種分類啦

763
00:25:53,772 --> 00:25:54,953
噉然後呢，Name

764
00:25:54,953 --> 00:25:56,854
entity recognition就係呢

765
00:25:56,854 --> 00:25:58,275
喺一堆文字裡面抽

766
00:25:58,275 --> 00:25:59,595
即係佢知道呢個係人名

767
00:25:59,595 --> 00:26:00,855
呢個係地點名

768
00:26:00,855 --> 00:26:01,655
呢個係公司名

769
00:26:01,655 --> 00:26:04,237
即係抽呢啲噉樣嘅嘅嘢出嚟

770
00:26:04,237 --> 00:26:05,317
因為，對於電腦嚟講

771
00:26:05,317 --> 00:26:07,938
其實佢文字全部都係一堆英文字母或者中文字

772
00:26:07,938 --> 00:26:09,038
一堆字，佢冇意思嘛

773
00:26:09,038 --> 00:26:09,358
對佢嚟講

774
00:26:09,719 --> 00:26:12,201
但用呢一種NLP方式可以抽

775
00:26:12,201 --> 00:26:14,821
我知你提過香港一篇財經新聞

776
00:26:14,821 --> 00:26:16,363
有提過呢幾間公司嘅

777
00:26:16,363 --> 00:26:19,005
有講到呢一單新聞係會影響呢啲公司

778
00:26:19,005 --> 00:26:20,685
咁呢一種technology就可以

779
00:26:20,685 --> 00:26:22,767
將佢抽出嚟然後再分析囉

780
00:26:22,767 --> 00:26:24,528
係喇咁呢一種就係唔同嘅NLP task

781
00:26:24,528 --> 00:26:26,909
咁我哋可以去下一張slide呢

782
00:26:26,909 --> 00:26:30,592
去睇一睇，呃啫啲示範啦

783
00:26:30,592 --> 00:26:32,673
啫係用checkgbt 去做啲示範嘅

784
00:26:36,690 --> 00:26:37,590
係喇噉呢度我就

785
00:26:37,590 --> 00:26:41,071
呃，如果太細睇唔到我讀出嚟啫係我就想去

786
00:26:41,071 --> 00:26:42,873
我之前喺香港use place講一個talk嘅

787
00:26:42,873 --> 00:26:44,354
講track gpt嘅

788
00:26:44,354 --> 00:26:50,577
噉我就想叫佢draft一段啫係喺網站要要要要宣傳要推介呢個talk呢噉我就要寫幾幾

789
00:26:50,577 --> 00:26:52,058
段嘢噉啦，噉我唔想自己寫呢

790
00:26:52,058 --> 00:26:53,078
我可以叫佢幫我寫喎

791
00:26:53,078 --> 00:26:58,481
我就話，用少於二百五十個字啦就去寫啫啲draft幾段幾段文章出嚟啦

792
00:26:58,481 --> 00:26:59,682
跟住就睇下，呃

793
00:26:59,682 --> 00:27:03,684
track gpt有啲咩缺陷同埋有啲咩方法克服佢噉噉

794
00:27:05,605 --> 00:27:09,950
噉噉但係呢你畀指令嘅時候呢嗰個嗰句叫提示語叫prompt啊

795
00:27:09,950 --> 00:27:11,191
即係有樣嘢叫prompt

796
00:27:11,191 --> 00:27:13,553
engineering就係你點去寫個prompt令到佢

797
00:27:13,553 --> 00:27:15,234
出嚟個output係符合你

798
00:27:15,234 --> 00:27:16,536
心意多啲嘅噉

799
00:27:16,536 --> 00:27:18,076
其實都係好makes sense嘅

800
00:27:18,076 --> 00:27:19,518
同你同人講嘢一樣咋嘛

801
00:27:19,518 --> 00:27:22,681
你叫我整廿張slide出嚟噉我會整畀你

802
00:27:22,681 --> 00:27:25,804
噉你想個topic係乜或者想提點嘅噉你都要講清楚㗎嘛

803
00:27:25,804 --> 00:27:27,826
噉個AI model一樣你要話清楚畀佢聽

804
00:27:28,286 --> 00:27:29,666
你想做點乜，噉

805
00:27:29,666 --> 00:27:31,907
噉我就提到，我要特別提到係

806
00:27:31,907 --> 00:27:33,468
ChatGP有啲limitation嘅

807
00:27:33,468 --> 00:27:36,950
同埋點樣克服佢用某一樣嘢叫RAG呢個technology去克服佢

808
00:27:36,950 --> 00:27:38,009
我一定要提呢啲point嘅

809
00:27:38,009 --> 00:27:39,570
噉佢就會寫出嚟畀我

810
00:27:39,570 --> 00:27:41,531
同埋係，即係根據我嘅需求

811
00:27:41,531 --> 00:27:43,092
佢會寫出嚟畀我囉

812
00:27:43,092 --> 00:27:44,893
同埋佢寫到唔滿意可以叫佢改嘅

813
00:27:44,893 --> 00:27:48,354
即係譬如話，呃佢太長或者某樣嘢佢miss咗冇提到

814
00:27:48,354 --> 00:27:51,174
噉我可以叫佢再改嘅係啦

815
00:27:51,174 --> 00:27:52,875
啫係好似好似同人噉樣去溝通

816
00:28:00,579 --> 00:28:02,800
咁另外呢，你叫佢重寫都得嘅

817
00:28:02,800 --> 00:28:06,001
啫頭先嗰個post可能係喺網站嗰度show出嚟

818
00:28:06,001 --> 00:28:06,922
咁但係之後我要

819
00:28:06,922 --> 00:28:11,304
呃，出另外個post係喺嗰個linktine嗰個寫家媒體嗰度show嘅

820
00:28:11,304 --> 00:28:12,663
咁我今次想用point form

821
00:28:12,663 --> 00:28:15,184
同埋用嗰啲嗰啲emotions嗰啲公仔嘅

822
00:28:15,184 --> 00:28:16,065
咁我就叫佢重寫

823
00:28:16,065 --> 00:28:17,665
咁佢就真係跟我要求我寫嘞

824
00:28:17,665 --> 00:28:20,767
咁你就見到，有嗰啲表情符號啦同埋

825
00:28:20,767 --> 00:28:23,548
簡短啲啲同埋淨係講嗰個日期時間地點嗰啲

826
00:28:23,548 --> 00:28:24,748
咁你就，你唔想

827
00:28:29,270 --> 00:28:31,076
咁你可以叫佢重寫個format係點樣樣

828
00:28:31,076 --> 00:28:33,603
你可以，係喇呢啲係

829
00:28:33,603 --> 00:28:35,268
雀GP做得幾好嘅嘢嚟嘅

830
00:28:35,268 --> 00:28:36,270
係喇咁我下張slide唔該

831
00:28:40,972 --> 00:28:42,394
噉另外就係翻譯囉

832
00:28:42,394 --> 00:28:43,335
即係你英譯中

833
00:28:43,335 --> 00:28:44,756
中譯英，乜都做到

834
00:28:44,756 --> 00:28:46,196
甚至乎呢你譯做

835
00:28:46,196 --> 00:28:47,938
呃 programing language 都得㗎

836
00:28:47,938 --> 00:28:49,519
即係你話我想喺 database

837
00:28:49,519 --> 00:28:50,859
select某幾個column

838
00:28:50,859 --> 00:28:52,181
然後呢 from 呢個table

839
00:28:52,181 --> 00:28:53,942
你就咁講，即係寫英文

840
00:28:53,942 --> 00:28:56,163
中文打畀佢，跟住佢出嗰段

841
00:28:56,163 --> 00:28:57,625
出嗰段程式買出嚟都得㗎囉

842
00:28:57,625 --> 00:28:59,865
係啊，甚，或者掉返轉有段程式買

843
00:28:59,865 --> 00:29:00,606
你唔知佢講乜

844
00:29:00,606 --> 00:29:02,949
你就叫佢用人類語言解畀你聽

845
00:29:02,949 --> 00:29:03,729
都得㗎係喇啫

846
00:29:03,729 --> 00:29:05,790
真係，都幾powerful

847
00:29:09,594 --> 00:29:11,999
咁或者做summary呢頭先提過可以呃

848
00:29:11,999 --> 00:29:13,902
篇文好長你叫佢去

849
00:29:13,902 --> 00:29:15,625
講三個重點畀你聽

850
00:29:15,625 --> 00:29:16,125
都可以嘅係喇

851
00:29:17,866 --> 00:29:19,887
噉噉但係呢啲呃

852
00:29:19,887 --> 00:29:24,169
情況呢就，呢啲情況比較好嘅use case啊啫係做一啲創意或者創作社民嘅嘢

853
00:29:24,169 --> 00:29:26,788
但係有時如果你要求佢個精進度好高呢

854
00:29:26,788 --> 00:29:27,849
佢未必做得太好嘅

855
00:29:27,849 --> 00:29:29,329
因為有時你要呃

856
00:29:29,329 --> 00:29:30,230
譬如你捏三個點

857
00:29:30,230 --> 00:29:34,371
噉佢summarize啲三個point 同你心目中想summarize啲咩可能有少

858
00:29:34,371 --> 00:29:35,231
少出入㗎嘛噉所以

859
00:29:35,231 --> 00:29:36,692
噉所以呢啲嘢

860
00:29:36,692 --> 00:29:38,833
如果做啲需要準確度高嘅嘢呢

861
00:29:38,833 --> 00:29:41,952
所以都係要有嗰個人類嘅judgment喺裏面嘅係啦

862
00:29:41,952 --> 00:29:43,673
噉噉噉可以下一張slide

863
00:29:46,596 --> 00:29:48,936
好喇咁嚟到呢個topic呢就係

864
00:29:48,936 --> 00:29:52,680
嗱個個講咗有啲咩AI big data嘅technology

865
00:29:52,680 --> 00:29:54,500
咁究竟喺會計界係點會用得着呢

866
00:29:54,500 --> 00:29:58,303
咁我哋可以去下張slide

867
00:29:58,303 --> 00:30:01,365
就係首先如果係嗰種management accounting呢

868
00:30:01,365 --> 00:30:04,046
即係公司內部畀管理層睇過一種呢

869
00:30:04,046 --> 00:30:08,989
咁，呃，因為呢一種你你個frequency要高好多唔係一年年尾買一次數

870
00:30:08,989 --> 00:30:12,270
你係每個月甚至每個禮拜你都想知道

871
00:30:12,270 --> 00:30:14,192
盤數做成點咁所以呢會有一啲

872
00:30:15,333 --> 00:30:16,855
比較重複性多嘅工作

873
00:30:16,855 --> 00:30:19,237
可能你係要喺嗰個系統度

874
00:30:19,237 --> 00:30:20,678
即係generate一份report出嚟啦

875
00:30:20,678 --> 00:30:23,941
噉但係嗰個report個format未必係你最尾想出嚟嘅結果

876
00:30:23,941 --> 00:30:25,402
噉所以你要做一大堆嘢去處理

877
00:30:25,402 --> 00:30:27,944
執返好佢，變成一個最尾個樣

878
00:30:27,944 --> 00:30:29,046
噉呢樣嘢比較重複性呢

879
00:30:29,046 --> 00:30:30,887
噉就可以用一啲tool去做

880
00:30:30,887 --> 00:30:32,209
automation做自動化嘅

881
00:30:32,888 --> 00:30:35,873
例如呢就係下面幅圖見到啦

882
00:30:35,873 --> 00:30:40,076
噉呢個叫咩？呢個叫做RPA Robotic Process Automation

883
00:30:40,076 --> 00:30:42,220
就係一啲自動化工具嚟㗎

884
00:30:42,220 --> 00:30:45,623
噉呢就佢個特點就係唔需要太多寫program

885
00:30:45,623 --> 00:30:46,924
啫係寫程式嗰啲嘅工作

886
00:30:46,924 --> 00:30:49,227
你淨係 drag and drop 將啲掣拉落去

887
00:30:49,227 --> 00:30:50,008
將佢黐埋一齊

888
00:30:50,008 --> 00:30:53,152
咁就會做到一啲簡單嘅自動化工作

889
00:30:54,304 --> 00:30:56,346
做緊咩呢度就係譬如佢有兩個

890
00:30:56,346 --> 00:30:57,967
你見到上下有兩個

891
00:30:57,967 --> 00:30:59,608
即係最左手邊有兩個icon啦

892
00:30:59,608 --> 00:31:01,690
咁就係兩個可能excel file嚟㗎

893
00:31:01,690 --> 00:31:03,771
就係嗰個數據來源係excel file

894
00:31:03,771 --> 00:31:06,213
咁之後呢就可能select某啲columns啦

895
00:31:06,213 --> 00:31:08,496
之後兩個data source就merge埋一齊

896
00:31:08,496 --> 00:31:10,698
即係佢，佢將佢黐埋一齊啦

897
00:31:10,698 --> 00:31:14,361
例如係用我種保險公司為例子啦

898
00:31:14,361 --> 00:31:15,782
有一啲客戶嘅資料同有一堆

899
00:31:17,263 --> 00:31:18,605
交易資料，transaction嘅資料

900
00:31:18,605 --> 00:31:20,625
咁，客戶資料嗰度可能有條key

901
00:31:20,625 --> 00:31:21,727
就係嗰個，呃

902
00:31:21,727 --> 00:31:21,906
client

903
00:31:21,906 --> 00:31:23,907
id或者clientnumber咁嘅

904
00:31:23,907 --> 00:31:24,828
就可以攞佢嚟

905
00:31:24,828 --> 00:31:26,410
再要呢個用excel你哋就

906
00:31:26,410 --> 00:31:27,851
即係只係vlog up嗰個function啦

907
00:31:27,851 --> 00:31:31,792
就係將佢，連埋去另外嗰個excel嗰度嗰個column啦

908
00:31:31,792 --> 00:31:32,894
咁，就將兩個data set

909
00:31:32,894 --> 00:31:34,275
merge埋一齊

910
00:31:34,275 --> 00:31:35,996
咁就可以做到啲分析

911
00:31:35,996 --> 00:31:37,936
咁之後下面，兩個啫啲兩條

912
00:31:37,936 --> 00:31:39,919
branch merge埋一齊之後就係再

913
00:31:39,919 --> 00:31:41,839
呃，select某啲colum

914
00:31:41,839 --> 00:31:42,039
或者做

915
00:31:46,482 --> 00:31:47,824
condition statement

916
00:31:47,824 --> 00:31:49,625
即係if呢樣嘢發生就行上面條線

917
00:31:49,625 --> 00:31:50,704
如果唔係行下面條線

918
00:31:50,704 --> 00:31:52,586
咁，咁就做晒呢啲嘢之後呢

919
00:31:52,586 --> 00:31:55,347
個好處就係你你做一次嘅呢個過程

920
00:31:55,347 --> 00:31:56,028
你做一次之後

921
00:31:56,028 --> 00:31:59,210
到下個禮拜你要再有新嘅數據入嚟呢

922
00:31:59,210 --> 00:32:01,310
你就將頭兩個step你refresh一次

923
00:32:01,310 --> 00:32:03,932
即係fit 頭兩個data set入去

924
00:32:03,932 --> 00:32:04,731
新嘅data入去呢

925
00:32:04,731 --> 00:32:06,313
咁後面就會自己行晒㗎喇

926
00:32:06,313 --> 00:32:08,374
咁就唔需要再人手再

927
00:32:08,374 --> 00:32:09,094
做後面啲step

928
00:32:09,094 --> 00:32:10,875
咁就變咗可能每個禮拜你係

929
00:32:10,875 --> 00:32:11,736
用一個鐘做呢個

930
00:32:12,536 --> 00:32:15,318
程序以後就變咗五分鐘就搞掂喇

931
00:32:15,318 --> 00:32:19,702
係喇噉但係嗰個大前提就係嗰個數據本身嗰個format冇冇改到嘅

932
00:32:19,702 --> 00:32:24,008
啫係如果突然間你有個新嘅Report Format嗰啲column名唔同咗位置唔同

933
00:32:24,008 --> 00:32:26,890
咗，噉可能呢嚿嘢你要做少少改動先用得着

934
00:32:26,890 --> 00:32:28,991
如果唔係嘅話基本上係不停係可以

935
00:32:30,153 --> 00:32:31,234
呃重用嘅呢舊嘢

936
00:32:31,234 --> 00:32:32,416
呢個就係嗰個好處喇

937
00:32:32,416 --> 00:32:35,020
嗰個，做一個自動化啦

938
00:32:35,020 --> 00:32:37,002
噉另外，上嗰個bullet point就係講過

939
00:32:37,002 --> 00:32:40,188
一開始，呃，咪畀咗嗰個dashboard大家睇嘅

940
00:32:40,188 --> 00:32:44,713
噉，呃，呢個都係可以用嚟做management reporting嘅啫係

941
00:32:44,713 --> 00:32:45,515
譬如，呃一啲

942
00:32:46,476 --> 00:32:48,237
銀行客戶佢想知道

943
00:32:48,237 --> 00:32:49,518
呃佢嗰啲，呃

944
00:32:49,518 --> 00:32:51,878
新嘅，新嘅業務係邊度嚟嘅

945
00:32:51,878 --> 00:32:54,180
係分行啦？定係喺網上面銷售

946
00:32:54,180 --> 00:32:56,201
定係其他啲broker可以幫佢sell返嚟

947
00:32:56,201 --> 00:32:58,682
咁，咁嗰啲，嗰啲圖表就可以即刻show晒出嚟

948
00:32:58,682 --> 00:33:00,364
然後呢你一click某一個channel

949
00:33:00,364 --> 00:33:02,585
佢嘅其他表即刻自動up

950
00:33:02,585 --> 00:33:03,965
day晒嘅咁就

951
00:33:03,965 --> 00:33:05,707
呃啫啫課管理層

952
00:33:05,707 --> 00:33:06,866
嘅用途，咁亦都

953
00:33:06,866 --> 00:33:08,107
但係又有好有唔好啦啫係

954
00:33:09,048 --> 00:33:10,109
好處就係好快啦

955
00:33:10,109 --> 00:33:11,070
好透明，好自動化

956
00:33:11,070 --> 00:33:12,112
但係另外就係

957
00:33:12,112 --> 00:33:13,292
呃，如果，老細問你

958
00:33:13,292 --> 00:33:14,334
咦，你撳咗呢個掣

959
00:33:14,334 --> 00:33:15,095
出咗呢個result

960
00:33:15,095 --> 00:33:15,736
點解係咁樣嘅

961
00:33:15,736 --> 00:33:17,337
咁你冇時間去prepare

962
00:33:17,337 --> 00:33:18,598
啫係你如果平時係

963
00:33:18,598 --> 00:33:19,319
返去答report

964
00:33:19,319 --> 00:33:19,880
你可以慢慢去

965
00:33:19,880 --> 00:33:21,261
去諗要點樣處理

966
00:33:21,261 --> 00:33:23,284
但呢啲就會，係啦

967
00:33:23,284 --> 00:33:24,905
好快好快就會知到個

968
00:33:24,905 --> 00:33:26,807
嗰個問題係邊

969
00:33:26,807 --> 00:33:28,950
咁啊可以去下一個slide啦我哋

970
00:33:31,386 --> 00:33:32,429
嗯呢個呢就係呃

971
00:33:32,429 --> 00:33:33,671
一個例子呢就係呃

972
00:33:33,671 --> 00:33:35,633
regulatory reporting

973
00:33:35,633 --> 00:33:37,055
即係譬如話呢

974
00:33:37,055 --> 00:33:39,878
呃，有啲監管機構啦或者呃

975
00:33:39,878 --> 00:33:41,300
政府機構譬如稅局嗰啲噉啦

976
00:33:41,300 --> 00:33:45,727
咁稅局，可能係有一個template要你去填嘅啫係每個稅貴

977
00:33:45,727 --> 00:33:48,029
咁呢嗰個template就係嗰啲呃

978
00:33:49,151 --> 00:33:51,373
財務報表上面嗰啲嗰啲items呢

979
00:33:51,373 --> 00:33:53,173
佢有某一個某一個叫法嘅

980
00:33:53,173 --> 00:33:54,094
譬如呢度一個例子啦

981
00:33:54,094 --> 00:33:55,634
佢下面，呃，下面嗰三個就係

982
00:33:55,634 --> 00:33:57,155
一個叫 hard wear

983
00:33:57,155 --> 00:33:58,196
一個叫 furniture and fitting

984
00:33:58,196 --> 00:33:59,518
一個叫 other PPE

985
00:33:59,518 --> 00:34:00,738
咁 PPE就係

986
00:34:00,738 --> 00:34:02,278
呃 property plan and equipment

987
00:34:02,278 --> 00:34:04,160
即係其實呢幾樣嘢都係講緊類似嘅嘢

988
00:34:04,160 --> 00:34:06,622
咁但係，每每間公司有一個唔同

989
00:34:06,622 --> 00:34:08,242
稍為唔同嘅叫法嘅

990
00:34:08,242 --> 00:34:13,405
噉如果你，呃就噉嘗試將呢啲唔同叫法嘅去寫落去

991
00:34:13,405 --> 00:34:13,425
呃

992
00:34:14,867 --> 00:34:19,273
瑞國個template呢噉可能瑞國冇一個嘢叫other PPE噉你要fit落去邊度

993
00:34:19,273 --> 00:34:21,255
呢噉就會需要啲人手去

994
00:34:21,255 --> 00:34:25,300
佢自己諗點樣去做一個mapping嘅噉就好花時間嘅噉但係如果用

995
00:34:25,300 --> 00:34:25,320
呃

996
00:34:26,536 --> 00:34:27,577
Machine learning呢

997
00:34:27,577 --> 00:34:30,717
你可以，去train個machine去應呢啲噉樣嘅字

998
00:34:30,717 --> 00:34:32,059
例如你，你話畀佢聽

999
00:34:32,059 --> 00:34:34,079
首先人手做一轉mapping先啦就係

1000
00:34:34,079 --> 00:34:37,940
啊hardware同埋呢啲other pp全部都係mapped落去一個叫做

1001
00:34:37,940 --> 00:34:40,222
呃，呃，PPE噉樣嘅account嘅

1002
00:34:40,222 --> 00:34:41,422
噉你你mapped晒佢之後呢

1003
00:34:41,422 --> 00:34:43,844
就可以，train個model去應呢啲pattern

1004
00:34:43,844 --> 00:34:44,764
噉，如果你喺

1005
00:34:46,304 --> 00:34:48,266
呃，即係喺啲audit firm度做啦

1006
00:34:48,266 --> 00:34:49,826
噉你有好多唔同嘅客戶

1007
00:34:49,826 --> 00:34:53,186
佢地每一個客戶佢嗰個mapping都未必一樣嘅

1008
00:34:53,186 --> 00:34:55,648
噉你就可以train一個model去應晒呢啲嘢

1009
00:34:55,648 --> 00:34:57,088
令到佢呢可以呃

1010
00:34:57,088 --> 00:34:58,007
自己做個mapping

1011
00:34:58,007 --> 00:34:59,248
就自動化咗呢件事囉

1012
00:34:59,248 --> 00:35:02,630
噉就可以，慳到時間囉係啦

1013
00:35:02,630 --> 00:35:03,409
噉下張slide唔該

1014
00:35:05,972 --> 00:35:10,737
噉啊再嚟就講下喺external audit嘅應用啦啫係audit firm嗰啲啦

1015
00:35:10,737 --> 00:35:13,277
噉喺audit firm呢喺啫係我

1016
00:35:13,277 --> 00:35:17,802
十，十，十零年先開始做嘢時已經有一樣嘢叫computer assisted audit

1017
00:35:17,802 --> 00:35:18,461
 technique

1018
00:35:18,461 --> 00:35:20,264
即係用電腦去做audit嘅噉所以

1019
00:35:20,264 --> 00:35:21,324
未有呢啲big data

1020
00:35:21,324 --> 00:35:23,365
AI呢啲噉嘅term之前已經有呢樣嘢存在

1021
00:35:23,746 --> 00:35:25,847
咁佢做咩呢就係將嗰個呃

1022
00:35:25,847 --> 00:35:28,088
文客戶攞咗嗰啲Journal entry返嚟呢

1023
00:35:28,088 --> 00:35:31,070
即係每一每一條entry佢有debit有credit幾多咁

1024
00:35:31,070 --> 00:35:32,972
咁佢攞晒返嚟做一啲整理

1025
00:35:32,972 --> 00:35:34,092
就fit入一個tool裏面

1026
00:35:34,092 --> 00:35:35,913
咁佢就會做一啲checking啦

1027
00:35:35,913 --> 00:35:39,474
例如話，嗰條entry係咪啫係debit credit加埋係咪得要零啊

1028
00:35:39,474 --> 00:35:41,416
又或者有冇啲古怪pattern

1029
00:35:41,416 --> 00:35:44,557
例如，呃，平時都係星期一至星期五入數嘅

1030
00:35:44,557 --> 00:35:45,898
但係突然間有一晚

1031
00:35:45,898 --> 00:35:49,280
呃禮拜六晚零晨兩點有個人入咗條好奇怪嘅trans

1032
00:35:51,021 --> 00:35:54,963
有啲問題呢，噉佢呢個tool係做呢啲checking嘅

1033
00:35:54,963 --> 00:35:59,143
噉，就近年呢嗰啲audit firm就將呢樣嘢再發揚光大喇

1034
00:35:59,143 --> 00:36:01,545
因為橫掂你都攞晒成本客戶嘅數據返嚟

1035
00:36:01,545 --> 00:36:02,985
噉其實都可以做分析㗎嘛

1036
00:36:02,985 --> 00:36:05,826
都可以做到，呃我之前提嗰啲dashboard出嚟㗎嘛

1037
00:36:05,826 --> 00:36:10,166
噉就係，睇下每一個account嘅變化有冇啲呃

1038
00:36:10,166 --> 00:36:13,768
時間上面有啲突然間好似頭先講禮拜六日有啲奇怪transaction

1039
00:36:14,047 --> 00:36:16,532
又或者有啲數字係呃

1040
00:36:16,532 --> 00:36:17,833
即係，九九九

1041
00:36:17,833 --> 00:36:22,061
即係數字尾九個詞係人手人為答落去嘅唔係一啲natural嘅嘢嚟

1042
00:36:22,061 --> 00:36:23,402
噉噉呢啲tool就可以

1043
00:36:24,230 --> 00:36:25,490
揾到呢啲奇怪嘅

1044
00:36:25,490 --> 00:36:27,592
嘅pattern出嚟囉

1045
00:36:27,592 --> 00:36:30,635
係喇噉，呃，另外呢喺

1046
00:36:30,635 --> 00:36:32,675
喺做audit方面一個比較

1047
00:36:32,675 --> 00:36:36,318
啫係呃，用數據分析可以做到一啲多啲嘢就係

1048
00:36:36,318 --> 00:36:39,021
以前我哋係做一個sample testing㗎因為

1049
00:36:39,021 --> 00:36:40,782
太多數據太多單嘅呢啲

1050
00:36:40,782 --> 00:36:42,103
你冇可能全部睇晒嘅嘛

1051
00:36:42,103 --> 00:36:42,702
你做audit

1052
00:36:42,702 --> 00:36:43,463
但係時間有限

1053
00:36:43,463 --> 00:36:45,666
噉你就會，抽sample啫係抽廿五隻

1054
00:36:45,666 --> 00:36:47,367
抽幾多隻去，去check

1055
00:36:47,367 --> 00:36:50,068
譬如呃，張單有冇人簽名或者呃

1056
00:36:50,748 --> 00:36:51,710
係喇譬如你銷售數佢

1057
00:36:51,710 --> 00:36:52,590
你話你買咗一萬張

1058
00:36:52,590 --> 00:36:53,972
一萬個translator

1059
00:36:53,972 --> 00:36:54,813
咁我抽廿五隻

1060
00:36:54,813 --> 00:36:56,516
抽啲大啲嘅嚟睇

1061
00:36:56,516 --> 00:36:59,639
咁，但係呢呢樣嘢係guarantee唔到

1062
00:36:59,639 --> 00:37:00,681
一百percent準確

1063
00:37:00,681 --> 00:37:03,304
因為你只能夠係佢哋叫做reasonable assurance

1064
00:37:03,304 --> 00:37:04,244
即去到九十五percent

1065
00:37:04,244 --> 00:37:05,106
話九十percent準咁

1066
00:37:05,786 --> 00:37:07,608
咁因為你check唔晒你人手check唔晒

1067
00:37:07,608 --> 00:37:09,429
但而家如果所有嘢都係

1068
00:37:09,429 --> 00:37:11,630
個data base入面啫係電子化咗呢

1069
00:37:11,630 --> 00:37:13,771
咁其實你係可以check晒一百percent 㗎喎

1070
00:37:13,771 --> 00:37:16,313
因為電腦做一條同做一萬條嘢一樣㗎

1071
00:37:16,313 --> 00:37:17,275
你set 一條formula

1072
00:37:17,275 --> 00:37:18,715
set個program

1073
00:37:18,715 --> 00:37:21,757
咁你run，咁佢係可以做到full population testing 㗎

1074
00:37:21,757 --> 00:37:23,840
咁就真係可以搵到以前一啲搵唔到嘅嘢囉

1075
00:37:24,820 --> 00:37:26,041
噉，噉但係呢個係

1076
00:37:26,041 --> 00:37:28,463
可能係internal audit會比較

1077
00:37:28,463 --> 00:37:28,983
value大啲

1078
00:37:28,983 --> 00:37:29,824
因為internal audit

1079
00:37:29,824 --> 00:37:31,304
你幫我客做 audit

1080
00:37:31,304 --> 00:37:32,146
跟住你，啫，佢都係想

1081
00:37:32,146 --> 00:37:34,086
啫個partner

1082
00:37:34,086 --> 00:37:35,248
夠膽簽名就OK㗎啦

1083
00:37:35,248 --> 00:37:38,010
噉就，但係internal audit就會係

1084
00:37:38,010 --> 00:37:40,572
啫係將所有問題撚晒出嚟都可以撚到囉

1085
00:37:40,572 --> 00:37:41,693
如果一百percent 將學讀

1086
00:37:42,514 --> 00:37:48,302
同埋external audit有個難處就係啲客未必肯將所有資料交晒畀你

1087
00:37:48,302 --> 00:37:51,789
一啲就知佢哋嘅機密嘅或者係自己internal data

1088
00:37:51,789 --> 00:37:55,014
咁但係internal audit你就乜都access到嘅比較上係可以

1089
00:37:55,014 --> 00:37:55,014
嗯

1090
00:37:56,539 --> 00:37:57,818
噉所以就係噉啦

1091
00:37:57,818 --> 00:37:59,440
呢個，呢樣嘢用呢新嘅方法

1092
00:37:59,440 --> 00:38:01,920
亦都要融入返去external audit firm嗰個

1093
00:38:01,920 --> 00:38:04,500
本身嗰個做audit嗰個methodology入面啦

1094
00:38:04,500 --> 00:38:07,320
啫係譬如佢不嬲係用開啲抄單嘅方法去做audit

1095
00:38:07,320 --> 00:38:08,820
咁而家用呢種IT嘅方法

1096
00:38:08,820 --> 00:38:10,280
咁，係咪接受呢

1097
00:38:10,280 --> 00:38:11,742
啫係咪你做咗

1098
00:38:11,742 --> 00:38:12,242
Electic

1099
00:38:12,242 --> 00:38:13,722
你就唔使做以前嗰啲抄單呢

1100
00:38:13,722 --> 00:38:14,222
如果唔係嘅話

1101
00:38:14,222 --> 00:38:15,222
你就為做而做

1102
00:38:15,222 --> 00:38:16,663
你做完又要做返以前嗰啲嘢

1103
00:38:16,663 --> 00:38:18,983
噉就，啲啲人嘅時間有限啊嘛

1104
00:38:18,983 --> 00:38:19,003
噉

1105
00:38:24,864 --> 00:38:27,125
嗰份audit menu入面嘅啫話

1106
00:38:27,125 --> 00:38:28,065
有人出嚟話唉

1107
00:38:28,065 --> 00:38:29,646
做咗呢樣嘢就唔使做以前嗰啲喇

1108
00:38:29,646 --> 00:38:32,228
噉就會大家會肯比較接受呢樣嘢囉

1109
00:38:32,228 --> 00:38:34,510
係啊，噉啊再去下一張slide唔該

1110
00:38:34,510 --> 00:38:34,610
係啊

1111
00:38:36,215 --> 00:38:37,315
噉，呃，係嘞

1112
00:38:37,315 --> 00:38:39,295
再講下internal audit啦

1113
00:38:39,295 --> 00:38:41,396
噉，呃，會有多幾個例子

1114
00:38:41,396 --> 00:38:43,976
就係，譬如頭先咪講個supervised learning嘅

1115
00:38:43,976 --> 00:38:44,936
即係根據一啲特性

1116
00:38:44,936 --> 00:38:46,556
譬如學生嘅成績乜乜

1117
00:38:46,556 --> 00:38:47,938
噉最尾佢合唔合格

1118
00:38:47,938 --> 00:38:50,277
噉呢邊呢就係根據一啲客戶嘅特性

1119
00:38:50,277 --> 00:38:51,639
例如佢係咩性別呀

1120
00:38:51,639 --> 00:38:53,099
職業呀，收入等等呢

1121
00:38:53,099 --> 00:38:55,579
就睇下你買嗰張保單一年後

1122
00:38:55,579 --> 00:38:56,420
係咪仲係生效

1123
00:38:56,420 --> 00:38:57,780
因為可能某啲職業

1124
00:38:58,380 --> 00:38:59,840
或者某一啲特定群組嘅人呢

1125
00:38:59,840 --> 00:39:01,240
佢係嗰個比較高風險嘅

1126
00:39:01,240 --> 00:39:03,780
即係佢會成日買完張單之後

1127
00:39:03,780 --> 00:39:06,681
呃，唔想再續補或者唔想交錢或者乜嘢原因都好囉

1128
00:39:06,681 --> 00:39:09,043
咁呢樣嘢對，對公司損失係會比較大

1129
00:39:09,043 --> 00:39:10,003
因為，頭一年呢

1130
00:39:10,003 --> 00:39:11,764
唔知大家有冇買保險知唔知

1131
00:39:11,764 --> 00:39:14,543
都係，畀，嗰個agent佣金係最高

1132
00:39:14,543 --> 00:39:15,724
但係公司唔係賺好多

1133
00:39:15,724 --> 00:39:18,045
你要keep住張單去到後期先賺錢

1134
00:39:18,045 --> 00:39:20,485
咁所以嗰啲客戶個個買一兩年度cut單係

1135
00:39:20,485 --> 00:39:22,045
會有損失嘅，咁所以我哋可以用

1136
00:39:25,047 --> 00:39:27,648
會唔會某啲特定嘅客戶或者

1137
00:39:27,648 --> 00:39:28,987
某啲agent去sell人嘅時候

1138
00:39:28,987 --> 00:39:30,628
其實佢哋嗰啲客成日都會

1139
00:39:30,628 --> 00:39:32,349
噉樣cut down嘅情況出現囉

1140
00:39:32,349 --> 00:39:33,969
咁就會用呢啲方法去

1141
00:39:33,969 --> 00:39:34,949
去做一個prediction

1142
00:39:34,949 --> 00:39:37,349
咁樣囉，咁呢個都係machine learning啊

1143
00:39:37,349 --> 00:39:39,090
啫係掹住最尾個column

1144
00:39:39,090 --> 00:39:41,010
唔知佢係yes or no嘅

1145
00:39:41,010 --> 00:39:42,550
咁但係，資質之前嗰啲攞嚟trainmodel

1146
00:39:42,550 --> 00:39:44,572
係囉，咁咁樣囉

1147
00:39:44,572 --> 00:39:46,072
咁下一個topic

1148
00:39:47,978 --> 00:39:48,898
係呀咁，然後

1149
00:39:48,898 --> 00:39:52,601
呃，之前提過嗰個Graph Analytics同埋即係兩個人之間如果係friend

1150
00:39:52,601 --> 00:39:54,882
就有個connection係黐住大家嘅

1151
00:39:54,882 --> 00:39:58,603
咁喺internal audit可以做咩呢就係你會發現呃

1152
00:39:58,603 --> 00:39:59,884
其中一個例子就係

1153
00:39:59,884 --> 00:40:05,628
唔可能某啲agent呢佢會係咁將佢啲客 refer畀某個醫生咁樣嘅啫係可能

1154
00:40:05,628 --> 00:40:09,769
明明個醫生個客會住得好遠嘅住新界佢都要去港島區

1155
00:40:09,769 --> 00:40:12,110
或者掉返去住香港島區然後去新界睇醫生咁

1156
00:40:12,110 --> 00:40:13,112
有啲奇怪其實可能

1157
00:40:13,851 --> 00:40:14,393
即係點解會咁啦

1158
00:40:14,393 --> 00:40:16,454
係咪你有，即係唔知胡容或者乜噉

1159
00:40:16,454 --> 00:40:18,536
噉呢啲graph會搵到呢啲pattern出嚟囉

1160
00:40:18,536 --> 00:40:19,217
即係某啲客戶

1161
00:40:19,217 --> 00:40:19,757
某啲agent

1162
00:40:19,757 --> 00:40:23,519
某啲醫生，某啲診所係有啲咁嘅network嘅

1163
00:40:23,519 --> 00:40:25,541
噉另外我另外一個例子就係

1164
00:40:25,541 --> 00:40:27,182
呃頭先講啫係

1165
00:40:27,182 --> 00:40:28,784
頭一年個佣金最高㗎嘛啲agent

1166
00:40:29,224 --> 00:40:31,286
咁可能佢會將張單賣俾佢啲親戚朋友

1167
00:40:31,286 --> 00:40:34,248
然後親戚朋友自己又係保險agent又會買返俾佢

1168
00:40:34,248 --> 00:40:34,909
咁成個network

1169
00:40:34,909 --> 00:40:36,190
點解人哋會互相買嚟買去

1170
00:40:36,190 --> 00:40:37,331
一兩年後又cut晒啲單

1171
00:40:37,331 --> 00:40:39,351
啫係佢哋，會計啱晒數嘅

1172
00:40:39,351 --> 00:40:40,893
啫係，嗰張單嘅competition

1173
00:40:40,893 --> 00:40:42,574
再計埋其他，其他cutout

1174
00:40:42,574 --> 00:40:43,394
bonus性呢

1175
00:40:43,394 --> 00:40:44,835
佢可以係，就算cut咗張單

1176
00:40:44,835 --> 00:40:45,695
要賠錢，佢都係

1177
00:40:45,695 --> 00:40:46,717
最尾都係有賺嘅

1178
00:40:46,717 --> 00:40:49,778
咁佢會，會做依啲咁樣嘅操作去

1179
00:40:49,778 --> 00:40:49,798
去

1180
00:40:59,365 --> 00:41:00,807
賣嗰個人本身又係agent嚟

1181
00:41:00,807 --> 00:41:03,914
噉呢一種關係又可以用呢個graph

1182
00:41:03,914 --> 00:41:04,594
去料到出嚟嘅

1183
00:41:05,856 --> 00:41:08,318
又或者另外一個例子就係呃

1184
00:41:08,318 --> 00:41:11,159
啫係保險agent佢哋係有啲一層層咁樣㗎啫係呃

1185
00:41:11,159 --> 00:41:12,701
有下線呀有啲咁樣㗎嘛

1186
00:41:12,701 --> 00:41:14,722
咁佢個下線如果賣到

1187
00:41:14,722 --> 00:41:15,983
單佢嗰個上線都會有

1188
00:41:15,983 --> 00:41:17,463
有commission收嘅

1189
00:41:17,463 --> 00:41:20,005
咁佢哋就計到係啫係點樣structure到好

1190
00:41:20,005 --> 00:41:21,726
好多層咁，令到上線都

1191
00:41:21,726 --> 00:41:23,588
賺得多啲，跟住賺完之後

1192
00:41:23,588 --> 00:41:24,909
大家分咗佢或者點樣咁

1193
00:41:24,909 --> 00:41:26,911
咁呢啲graph又係可以搵到呢啲pattern出嚟啫係好

1194
00:41:26,911 --> 00:41:29,932
好多好多好多好多層嘅啫係正常係兩三層佢

1195
00:41:32,715 --> 00:41:34,878
情況卻噉，噉你就咁睇數字啊

1196
00:41:34,878 --> 00:41:36,981
raw data好難睇到呢啲pattern

1197
00:41:36,981 --> 00:41:39,844
係，用呢種，graph嘅analytic就會

1198
00:41:39,844 --> 00:41:40,545
就會搵到呢啲噉嘅

1199
00:41:40,545 --> 00:41:42,407
嘅關係出嚟囉

1200
00:41:42,407 --> 00:41:46,393
係囉，噉我哋可以去

1201
00:41:46,393 --> 00:41:47,034
再去下一張slide

1202
00:41:49,677 --> 00:41:52,239
嗯，呢個呢就係一開始講嗰個

1203
00:41:52,239 --> 00:41:54,041
語音轉換嗰樣嘅嘢喇

1204
00:41:54,041 --> 00:41:55,963
即係，呃，譬如我有個

1205
00:41:55,963 --> 00:41:57,903
客戶打嚟啦，噉

1206
00:41:57,903 --> 00:41:58,864
講啲對話啦，語

1207
00:41:58,864 --> 00:42:01,186
語音嚟嘅，噉可能佢有唔同嘅topic

1208
00:42:01,186 --> 00:42:04,769
可能係講轉password或者講索償或者

1209
00:42:04,769 --> 00:42:06,010
查詢啲產品嘅

1210
00:42:06,010 --> 00:42:08,052
噉呢啲文，語音呢啲對話呢

1211
00:42:08,052 --> 00:42:09,934
噉而家啲AI識得將佢轉做文字啦

1212
00:42:10,715 --> 00:42:12,195
咁轉做文字之後呢

1213
00:42:12,195 --> 00:42:13,536
有時有啲客戶我哋會

1214
00:42:13,536 --> 00:42:14,956
最尾會有份文卷畀佢填返

1215
00:42:14,956 --> 00:42:17,818
就係你覺得今次嘅服務滿唔滿意啊

1216
00:42:17,818 --> 00:42:20,137
你一至十分畀幾多分咁樣嘅

1217
00:42:20,137 --> 00:42:21,679
咁某啲人呢就會填啦

1218
00:42:21,679 --> 00:42:22,358
有啲人唔填啦

1219
00:42:22,358 --> 00:42:23,699
咁但係肯填嗰啲人呢

1220
00:42:23,699 --> 00:42:24,639
我哋就可以攞嚟做

1221
00:42:24,639 --> 00:42:26,019
training data啦

1222
00:42:26,019 --> 00:42:27,240
啫係根據佢嗰段對話

1223
00:42:27,960 --> 00:42:30,101
同埋最尾嗰個survey個分數呢

1224
00:42:30,101 --> 00:42:31,583
就可以計到出嚟

1225
00:42:31,583 --> 00:42:32,903
噉計完個model有咩用呢

1226
00:42:32,903 --> 00:42:35,025
啫係當有啲人佢唔填份問卷呢

1227
00:42:35,025 --> 00:42:37,507
其實我哋都知道佢滿唔滿意呢樣嘢

1228
00:42:37,507 --> 00:42:39,548
因為以往填得滿意嗰啲人呢

1229
00:42:39,548 --> 00:42:40,869
個對話樓係噉樣嘅

1230
00:42:40,869 --> 00:42:41,670
唔滿意嗰啲人呢

1231
00:42:41,670 --> 00:42:44,771
就會係另外一個講法嚟嘅啫係有啲

1232
00:42:44,771 --> 00:42:46,273
呃啲文字上有啲唔同啦

1233
00:42:46,632 --> 00:42:47,554
噉所以我就會知道

1234
00:42:47,554 --> 00:42:49,114
啊原來某啲人係打嚟

1235
00:42:49,114 --> 00:42:50,715
話，呃，批唔到錢咩

1236
00:42:50,715 --> 00:42:52,077
噉雖然佢最尾冇填過文卷

1237
00:42:52,077 --> 00:42:53,978
但係佢應該都係唔滿意㗎喇

1238
00:42:53,978 --> 00:42:55,918
噉就可以做出一啲相應嘅

1239
00:42:55,918 --> 00:42:58,420
啫係佢，佢，再follow up 去

1240
00:42:58,420 --> 00:43:01,202
打畀佢，佢點樣去令到佢安撫下佢噉樣囉

1241
00:43:01,202 --> 00:43:02,083
噉呢個例子就

1242
00:43:02,083 --> 00:43:05,224
呃，我聽返嚟啲啫唔係實際咗嘅啲外國

1243
00:43:05,224 --> 00:43:07,126
因為英文會譯得準啲嘅

1244
00:43:07,126 --> 00:43:09,047
噉但係其實而家隨住啲科技越來越進

1245
00:43:17,574 --> 00:43:20,177
噉就，係喇噉呢個都係都係用

1246
00:43:20,177 --> 00:43:22,079
即係幾個幾個方面嘅Machine learning model

1247
00:43:22,079 --> 00:43:24,561
即係首先語音轉文字已經係一個AI喇

1248
00:43:24,561 --> 00:43:26,664
然後train個model

1249
00:43:26,664 --> 00:43:28,947
文字同佢個survey result去做個model

1250
00:43:28,947 --> 00:43:30,628
又係另外一個AI噉樣囉

1251
00:43:30,628 --> 00:43:34,152
噉就，係喇係有另外一個例子噉樣

1252
00:43:34,152 --> 00:43:35,974
噉可以再去adjust slide

1253
00:43:38,666 --> 00:43:41,347
係喇咁最尾呢個呢就叫做呃

1254
00:43:41,347 --> 00:43:41,907
webskripping

1255
00:43:41,907 --> 00:43:44,387
咁就係我喺放棄軟件教過課喇咁

1256
00:43:44,387 --> 00:43:47,449
咁呢樣嘢係係啲咩呢就係寫一啲program呢去網上去

1257
00:43:47,449 --> 00:43:49,789
去抄一啲資料返嚟例如呃

1258
00:43:49,789 --> 00:43:51,449
你唔知係唔係去嗰啲格價網嗰啲喇

1259
00:43:51,449 --> 00:43:53,210
例如去，你見到呃

1260
00:43:53,210 --> 00:43:56,411
一個產品可能某某牌子風筒或者吸塵機

1261
00:43:56,411 --> 00:44:00,373
咁喺幾個，src嗰度幾個網站或者門市有賣嘅

1262
00:44:00,373 --> 00:44:04,295
咁佢可以webskripping即刻抄晒啲資料返嚟呢就去做一個格價嘅

1263
00:44:07,916 --> 00:44:09,597
仲有一啲呃，比較

1264
00:44:09,597 --> 00:44:12,838
唔係咁，合法嘅嘢都做到啫係

1265
00:44:12,838 --> 00:44:14,298
炒黃牛飛，唱演唱會飛

1266
00:44:14,298 --> 00:44:15,239
其實都做到㗎

1267
00:44:15,239 --> 00:44:16,480
咁啫係呢度唔討論呢啲啦

1268
00:44:16,480 --> 00:44:18,681
咁，咁但係有啲有咁嘅technology

1269
00:44:18,681 --> 00:44:19,681
咁我哋喺啫係

1270
00:44:19,681 --> 00:44:20,762
internal audit

1271
00:44:20,762 --> 00:44:21,722
點樣用佢嘅就係

1272
00:44:21,722 --> 00:44:24,143
我哋公司開好好多啲啲insurance agent啦

1273
00:44:24,143 --> 00:44:26,063
咁有啲係，啫係跟

1274
00:44:26,063 --> 00:44:27,664
我哋公司有啲可能係

1275
00:44:27,664 --> 00:44:30,186
一啲booker賣我哋公司產品嘅嘢嘅

1276
00:44:30,186 --> 00:44:31,025
咁佢哋佢哋有牌㗎嘛

1277
00:44:33,086 --> 00:44:35,108
噉，呃，呢啲牌呢就

1278
00:44:35,108 --> 00:44:36,969
保監個網站就最updated嘅係喇

1279
00:44:36,969 --> 00:44:39,389
噉我哋公司自己maintain嗰個呢可能係

1280
00:44:39,389 --> 00:44:41,630
幾個月前噉有啲人辭咗職又冇

1281
00:44:41,630 --> 00:44:43,992
啫係嗰個excel冇updated到呢噉就

1282
00:44:43,992 --> 00:44:47,614
唔準嘅噉所以我哋想要最新嘅list就喺保監嗰度去睇喇

1283
00:44:47,614 --> 00:44:49,534
但係保監個網站呢就

1284
00:44:49,534 --> 00:44:52,135
啫係你要打個agent個名或者個id落去

1285
00:44:52,135 --> 00:44:57,237
然後呢有個captcha個嘢啫你要打啱嗰段個圖畫個字佢先畀你過下個round

1286
00:44:57,538 --> 00:44:58,559
咁你逐個逐個check

1287
00:44:58,559 --> 00:45:01,000
幾千個agent咁check法就好嘥時間㗎

1288
00:45:01,000 --> 00:45:02,739
咁所以有個方法係

1289
00:45:02,739 --> 00:45:05,242
用webscripping個方法寫啲程式呢去

1290
00:45:05,242 --> 00:45:06,742
自動將佢一次過抄晒落嚟

1291
00:45:06,742 --> 00:45:09,023
就變成一個excel嘅format咁樣樣囉

1292
00:45:09,023 --> 00:45:11,684
咁但係呢樣嘢就有少少灰色data嘅

1293
00:45:11,684 --> 00:45:13,445
因為嗰個website呢佢

1294
00:45:13,445 --> 00:45:14,266
畀你去就咁睇

1295
00:45:14,266 --> 00:45:16,186
但未必畀你咁大規模嘅抄data

1296
00:45:16,186 --> 00:45:18,447
咁所以啫係啫係最尾我哋公司冇冇選擇行呢條路

1297
00:45:18,447 --> 00:45:20,168
啫係科技上係可行呢樣嘢

1298
00:45:20,168 --> 00:45:21,248
但係最尾有啲合規

1299
00:45:23,771 --> 00:45:25,353
但係啫係話畀大家聽有

1300
00:45:25,353 --> 00:45:27,554
有啲存在嘅啲噉嘅technology係

1301
00:45:27,554 --> 00:45:29,394
或者你見有時嗰啲

1302
00:45:29,394 --> 00:45:33,717
係喇去埋煙囪會飛一秒就禁晒啲人係唔係人手感係用啲啲啲robot去

1303
00:45:33,717 --> 00:45:34,759
去咁樣搶飛囉

1304
00:45:34,759 --> 00:45:38,541
係咯，咁就，係喇咁但係有

1305
00:45:38,541 --> 00:45:39,581
有法律風險嘅做呢樣嘢

1306
00:45:39,581 --> 00:45:41,402
係喇咁所以就

1307
00:45:41,402 --> 00:45:44,724
大概係咁樣啦咁啊下張slide好似都

1308
00:45:44,724 --> 00:45:45,744
係喇最尾最尾一個

1309
00:45:45,744 --> 00:45:47,525
呃，例子呢就叫咗一個

1310
00:45:47,525 --> 00:45:49,166
呃checkboard啊

1311
00:45:49,166 --> 00:45:50,208
噉，呃，

1312
00:45:52,068 --> 00:45:56,291
因為呃，啲ChatGPT嗰啲呢佢係training網上面嘅data㗎嘛

1313
00:45:56,291 --> 00:45:57,192
咁所以佢內部

1314
00:45:57,192 --> 00:45:59,452
你公司內部譬如你想知道佢呃

1315
00:45:59,452 --> 00:46:02,193
你公司內部可以放幾多日嘅annual leave等等呢

1316
00:46:02,193 --> 00:46:03,355
佢未必答到你㗎

1317
00:46:03,355 --> 00:46:05,436
咁但係你可以fit一啲自己嘅document落去呢

1318
00:46:05,436 --> 00:46:06,516
做一啲Chatboard

1319
00:46:06,516 --> 00:46:08,998
譬如我呢度係將一啲audit procedure

1320
00:46:08,998 --> 00:46:10,318
一啲audit menu feed咗入去

1321
00:46:10,318 --> 00:46:12,739
我問佢，呃做audit點樣抽sample啊點樣

1322
00:46:12,739 --> 00:46:13,960
咁佢就可以答到你囉

1323
00:46:13,960 --> 00:46:16,061
咁就將一啲公司內部資料去

1324
00:46:16,061 --> 00:46:16,922
去咗個Chatboard出嚟囉

1325
00:46:17,902 --> 00:46:19,503
但呢個都係一個

1326
00:46:19,503 --> 00:46:21,605
而家啲啲即係Generative AI個方向啦

1327
00:46:21,605 --> 00:46:24,327
即係唔係某一個特定領域嘅AI囉

1328
00:46:24,327 --> 00:46:26,047
即係唔係一個general purpose

1329
00:46:26,047 --> 00:46:31,130
係，可能係金融行業或者保險行業嘅啲checkboard咁樣啦

1330
00:46:31,130 --> 00:46:33,010
係呀，咁好我哋去最尾一個session啦

1331
00:46:33,010 --> 00:46:36,112
就係，呃，用呢啲bigata

1332
00:46:36,112 --> 00:46:38,753
AI喺會計界有啲遇到啲咩困難啦

1333
00:46:38,753 --> 00:46:39,574
係啦我哋喺下張slide

1334
00:46:41,184 --> 00:46:45,047
就首先就你有個原材料先啦好似你煮餸你都有啲原材料

1335
00:46:45,047 --> 00:46:48,489
咁你要做數據分析你首先有data先得㗎嘛

1336
00:46:48,489 --> 00:46:52,353
咁，呃，攞data有啲咩問題呢啫啲譬如呃

1337
00:46:52,353 --> 00:46:53,193
做external audit

1338
00:46:53,193 --> 00:46:54,494
咁你問個客戶攞data

1339
00:46:54,494 --> 00:46:55,494
佢未必肯畀你㗎

1340
00:46:55,494 --> 00:46:57,916
啫係佢話，不嬲都係畀廿五隻sample你抄

1341
00:46:57,916 --> 00:46:59,418
但點解今年要睇晒所有嘢

1342
00:46:59,418 --> 00:47:01,440
啫係佢，佢會唔肯畀㗎

1343
00:47:01,440 --> 00:47:02,460
會有啲有啲抗拒

1344
00:47:02,460 --> 00:47:02,601
但係

1345
00:47:02,981 --> 00:47:04,541
Internal order 呢啫係如果做

1346
00:47:04,541 --> 00:47:06,682
公司內部呢其實都未必係咁簡單㗎

1347
00:47:06,682 --> 00:47:08,563
因為呢，你大公司尤其是呢

1348
00:47:08,563 --> 00:47:09,884
就算你想攞data

1349
00:47:09,884 --> 00:47:12,585
你問邊個IT部門負責管呢個System

1350
00:47:12,585 --> 00:47:13,606
或者除咗IT owner

1351
00:47:13,606 --> 00:47:14,666
仲有個Business owner 㗎

1352
00:47:14,666 --> 00:47:15,927
啫係我IT我可以畀你

1353
00:47:15,927 --> 00:47:19,248
但係以後嗰個Business owner approve

1354
00:47:19,248 --> 00:47:20,048
先至肯批畀你

1355
00:47:20,048 --> 00:47:21,489
咁你要identify 呢啲人

1356
00:47:21,489 --> 00:47:23,150
佢先肯攞適當嘅approval

1357
00:47:23,150 --> 00:47:24,590
佢先會grant access

1358
00:47:24,590 --> 00:47:26,572
畀你㗎，咁所以都唔係咁straightforward嘅

1359
00:47:26,572 --> 00:47:27,092
就算internal

1360
00:47:27,512 --> 00:47:29,132
咁如果將一啲做啲好啲嘅公司呢

1361
00:47:29,132 --> 00:47:31,914
佢就會有一個專門嘅Department係專門負責呢啲嘢

1362
00:47:31,914 --> 00:47:33,976
即係唔同嘅，System啦唔同Source

1363
00:47:33,976 --> 00:47:35,456
佢會將佢整合埋一齊

1364
00:47:35,456 --> 00:47:38,798
擺上一個呃，Data lake或者Data warehouse嘅地方啦

1365
00:47:38,798 --> 00:47:40,320
咁然後呢，呃

1366
00:47:40,320 --> 00:47:41,400
就當你要攞Data

1367
00:47:41,400 --> 00:47:42,920
你就問呢個Data

1368
00:47:42,920 --> 00:47:43,922
呃office去

1369
00:47:43,922 --> 00:47:45,842
去攞囉，咁佢就會再呃

1370
00:47:45,842 --> 00:47:47,684
問你，即係填張ticket話你

1371
00:47:47,684 --> 00:47:49,565
點解要攞攞嚟做乜

1372
00:47:49,565 --> 00:47:52,967
呃，攞幾耐，咁佢再搵適當嘅party去

1373
00:47:57,148 --> 00:48:01,231
咁，呃，亦都要提供一啲叫做data dictionary嘅嘢畀你

1374
00:48:01,231 --> 00:48:03,172
因為呢譬如有好多唔同嘅系統

1375
00:48:03,172 --> 00:48:04,713
咁我想攞某方面嘅資料

1376
00:48:04,713 --> 00:48:06,393
咁，話幾千個table

1377
00:48:06,393 --> 00:48:07,914
幾萬項，咁你究竟攞咩呢

1378
00:48:07,914 --> 00:48:10,056
咁你都要有個

1379
00:48:10,056 --> 00:48:11,036
有人教你點樣做嘅

1380
00:48:11,036 --> 00:48:14,478
咁所以一啲，好啲嘅公司會呢啲嘢structure啲呢

1381
00:48:14,478 --> 00:48:15,579
就會攞data做易啲

1382
00:48:15,579 --> 00:48:18,300
如果唔係呢，就算公司內部你都有時係好麻煩啦

1383
00:48:18,300 --> 00:48:19,521
攞唔到data就

1384
00:48:19,521 --> 00:48:20,762
做唔到下一步嘅分析㗎啦

1385
00:48:22,371 --> 00:48:25,514
噉另外一個難題就係嗰個呃

1386
00:48:25,514 --> 00:48:28,096
人才方面啦，啫係頭先我講呃

1387
00:48:28,096 --> 00:48:29,856
有啲technical skill要識嘅it嘅

1388
00:48:29,856 --> 00:48:33,298
譬如excel處理唔到太大嘅時候你要識寫啲programming language

1389
00:48:33,298 --> 00:48:36,201
啦，例如，最基本有個叫做sql啦

1390
00:48:36,201 --> 00:48:38,161
structured query language

1391
00:48:38,161 --> 00:48:39,824
就喺data base度撈資料嘅

1392
00:48:39,824 --> 00:48:40,423
噉你識呢啲嘢啦

1393
00:48:40,423 --> 00:48:42,806
噉另外就係，呃

1394
00:48:42,806 --> 00:48:46,728
通常而家做big data會用一啲叫payone嘅啲程式語言啦

1395
00:48:46,728 --> 00:48:47,588
噉呢兩種會係比較

1396
00:48:49,050 --> 00:48:51,393
然後頭先嗰啲dashboard呢

1397
00:48:51,393 --> 00:48:52,695
就，有個叫做powerbi啦

1398
00:48:52,695 --> 00:48:54,036
把macrosoftpowerbi

1399
00:48:54,036 --> 00:48:56,641
就整頭先嗰啲數據嘅visualization啦

1400
00:48:56,641 --> 00:48:57,422
或者table

1401
00:48:57,422 --> 00:48:58,443
都係類似噉嘅嘢囉

1402
00:48:58,443 --> 00:49:01,427
咁呢，咁呢幾種都係通常做啲detainers會

1403
00:49:01,427 --> 00:49:02,909
會要識到嘅知識嚟嘅係喇

1404
00:49:03,650 --> 00:49:04,971
咁但係另外一方面呢

1405
00:49:04,971 --> 00:49:07,873
你要點樣將呢啲知識apply落去個行業嗰度呢

1406
00:49:07,873 --> 00:49:09,355
你亦都有啲行業嘅知識

1407
00:49:09,355 --> 00:49:10,476
一啲business嘅知識

1408
00:49:10,476 --> 00:49:12,657
咁你先知道，啫係你唔係為做而做

1409
00:49:12,657 --> 00:49:13,478
因為你係要做完

1410
00:49:13,478 --> 00:49:15,079
你係幫公司帶到乜嘢value

1411
00:49:15,079 --> 00:49:16,400
你可能搵到啲

1412
00:49:16,400 --> 00:49:18,083
呃，人手睇唔到嘅嘢或者係

1413
00:49:18,083 --> 00:49:20,105
呃，要，要，慳到時間

1414
00:49:20,105 --> 00:49:21,806
咁所以嗰個business value嗰邊又要

1415
00:49:21,806 --> 00:49:23,467
啫係兩者都要識啲咁先至可以

1416
00:49:24,248 --> 00:49:25,387
做到咁但係兩者識晒就

1417
00:49:25,387 --> 00:49:32,650
嘅人市場未必好多或者佢你揀人揀你即係佢識得咁多嘅佢都好多公司有offer咁所以

1418
00:49:32,650 --> 00:49:34,309
係難嘅咁所以啫係

1419
00:49:34,309 --> 00:49:36,150
一個接觸方法就係

1420
00:49:36,150 --> 00:49:37,791
IT嗰邊又學少少business嘅嘢啦

1421
00:49:37,791 --> 00:49:39,871
business嗰邊又再train 去學啲簡單嘅

1422
00:49:39,871 --> 00:49:44,612
嘅coding嘅嘢或者頭先啫做dashboard嗰啲唔需要寫太多program嘅咁

1423
00:49:44,612 --> 00:49:46,472
可以學下呢啲咁兩邊

1424
00:49:46,472 --> 00:49:47,793
去，去配合咁樣囉

1425
00:49:47,793 --> 00:49:50,514
然後一齊傾點樣去做囉係呀

1426
00:49:50,514 --> 00:49:51,773
噉我哋下張slide

1427
00:49:53,277 --> 00:49:56,460
噉另外就如果係呃啫係GenerativeAI嗰啲呢

1428
00:49:56,460 --> 00:49:58,903
係啦應該啫係ChatGP真係用唔到㗎

1429
00:49:58,903 --> 00:50:01,567
有，有VPN先上到或者開account好麻煩噉

1430
00:50:01,567 --> 00:50:02,847
但係有其他方法譬如

1431
00:50:02,847 --> 00:50:04,489
有冇用過呢啲或者Po

1432
00:50:04,489 --> 00:50:05,992
VoEpo嗰啲係喇

1433
00:50:05,992 --> 00:50:08,755
噉就會令到你用得到喇

1434
00:50:08,755 --> 00:50:10,858
噉但係呃可以去下張slide就係呃

1435
00:50:12,597 --> 00:50:13,998
呃，就算你用得到都有啲

1436
00:50:13,998 --> 00:50:15,940
有啲，有啲，有啲limitation嘅例如

1437
00:50:15,940 --> 00:50:17,780
佢，冇一啲real time嘅數據

1438
00:50:17,780 --> 00:50:19,101
因為頭先，呃

1439
00:50:19,101 --> 00:50:20,963
講過個，呢個machinery model

1440
00:50:20,963 --> 00:50:21,344
佢train

1441
00:50:21,344 --> 00:50:22,764
係需要fit好多data入去

1442
00:50:22,764 --> 00:50:23,806
去訓練佢出嚟

1443
00:50:23,806 --> 00:50:25,606
咁但係，呢一種嘅ger

1444
00:50:25,606 --> 00:50:27,628
呢種嘅，呃，large language model啦

1445
00:50:27,628 --> 00:50:30,451
叫做，係用大量嘅數據去train出嚟同埋

1446
00:50:30,451 --> 00:50:31,251
大量嘅resources

1447
00:50:31,251 --> 00:50:32,932
咁所以佢唔會每日都train

1448
00:50:32,932 --> 00:50:33,793
佢係train一次

1449
00:50:33,793 --> 00:50:35,034
跟住推出市面

1450
00:50:35,034 --> 00:50:35,054
咁

1451
00:50:39,898 --> 00:50:42,123
所以你問佢，今日問佢騰訊估價幾多

1452
00:50:42,123 --> 00:50:43,045
佢答你唔到㗎呢啲

1453
00:50:43,045 --> 00:50:44,407
係啦，除非佢

1454
00:50:44,407 --> 00:50:46,913
再博咗舊上網嘅嘢啦係啦

1455
00:50:46,913 --> 00:50:49,036
噉呢個係一個其中一個limitation啦

1456
00:50:49,036 --> 00:50:50,280
噉下張slide就係

1457
00:50:52,222 --> 00:50:53,402
呃冇internal data嘅

1458
00:50:53,402 --> 00:50:54,563
例如你公司內部

1459
00:50:54,563 --> 00:50:56,744
呃，你可以放幾多人annual leave

1460
00:50:56,744 --> 00:50:58,306
你，呃，去，去church trip

1461
00:50:58,306 --> 00:51:00,166
可以book咩酒店咩嗰啲嘢

1462
00:51:00,166 --> 00:51:01,007
佢唔會知㗎嘛

1463
00:51:01,007 --> 00:51:04,030
因為你公司內部嘢唔會流出佢internet出面㗎嘛

1464
00:51:04,030 --> 00:51:05,650
咁所以公司內部data佢冇啦

1465
00:51:05,650 --> 00:51:07,411
咁但係就，都有方法克服嘅

1466
00:51:07,411 --> 00:51:10,313
啫係可以，做一啲internal checkboard 去

1467
00:51:10,313 --> 00:51:11,653
呃讀公司內部文件去

1468
00:51:11,653 --> 00:51:14,815
去答你㗎，但係就需要多啲功夫去做呢啲嘢

1469
00:51:14,815 --> 00:51:17,338
係囉，咁再下一張slide

1470
00:51:18,846 --> 00:51:19,786
就係有一個呃

1471
00:51:19,786 --> 00:51:22,228
呢個比較搞笑嘅例子就係我問佢

1472
00:51:22,228 --> 00:51:24,731
即係呃，龍珠呢本小說係咪金融作嘅

1473
00:51:24,731 --> 00:51:26,012
咁佢一開始識但係唔係

1474
00:51:26,012 --> 00:51:29,414
咁但係呢當我再後面去誤導佢話佢其實有份創作嘅喎

1475
00:51:29,414 --> 00:51:31,217
咁佢就會畀你誤導咗

1476
00:51:31,217 --> 00:51:32,478
啊，的確佢係有份創作

1477
00:51:32,478 --> 00:51:35,920
咁，咁，啫係佢會有一種咁嘅幻覺咁嘅嘢出現嘅啫係

1478
00:51:35,920 --> 00:51:37,782
有時覺得好似一個細路仔咁啊

1479
00:51:37,782 --> 00:51:39,603
有時控制唔到佢講乜嘢啫係會亂噏嘅

1480
00:51:40,405 --> 00:51:41,545
噉呢個都係一個問題

1481
00:51:41,545 --> 00:51:43,306
所以做一啲創作上嘅可能OK

1482
00:51:43,306 --> 00:51:45,447
但係你需要一百percent準嘅嘢呢

1483
00:51:45,447 --> 00:51:48,168
即係反而係用返啲low tech啲嘅方式或者rule base嗰種會

1484
00:51:48,168 --> 00:51:49,429
會可以enjoy一百percent準

1485
00:51:49,429 --> 00:51:51,570
即係寫晒啲條件喺度佢就照

1486
00:51:51,570 --> 00:51:53,230
啫電腦照計照執行就唔會錯

1487
00:51:53,230 --> 00:51:54,831
但呢種係，有機會錯

1488
00:51:54,831 --> 00:51:55,811
去唔到一百percent

1489
00:51:55,811 --> 00:51:56,992
暫時我睇到就係

1490
00:51:56,992 --> 00:51:58,371
所以好多人話

1491
00:51:58,371 --> 00:52:00,853
呃，咩，AI取代人類啦咩咁樣好

1492
00:52:00,853 --> 00:52:03,034
啫係有啲係覺得係標題檔去吸引你

1493
00:52:03,034 --> 00:52:04,855
其實我真係，非常可

1494
00:52:08,467 --> 00:52:09,907
噉就係大概噉嘞

1495
00:52:09,907 --> 00:52:14,811
睇下係咪仲有一張slide

1496
00:52:14,811 --> 00:52:16,911
最尾一個topic就係一個叫做

1497
00:52:16,911 --> 00:52:18,112
呃privacy嘅問題

1498
00:52:18,112 --> 00:52:19,153
用戶私隱嘅問題

1499
00:52:19,153 --> 00:52:22,456
因為呢，平時譬如你用checkgpt呢

1500
00:52:22,456 --> 00:52:24,936
你嗰個，你個app度打咗個command啦

1501
00:52:24,936 --> 00:52:26,157
跟住或者打咗啲

1502
00:52:26,157 --> 00:52:27,139
呃text落去

1503
00:52:27,139 --> 00:52:29,219
噉佢呢，輸出返個個text畀你

1504
00:52:29,219 --> 00:52:30,380
其實佢背後係

1505
00:52:30,380 --> 00:52:32,041
你嗰啲問題呢

1506
00:52:32,041 --> 00:52:33,563
佢輸，呃其實係send咗去

1507
00:52:34,952 --> 00:52:38,295
嗰個checkchip嘅公司啫係openAI呢嗰啲servier嗰度去處理嘅

1508
00:52:38,295 --> 00:52:40,097
然後再畀返個答案你㗎

1509
00:52:40,097 --> 00:52:40,998
點解佢會要咁做

1510
00:52:40,998 --> 00:52:42,721
因為首先佢個model唔係

1511
00:52:42,721 --> 00:52:43,280
呃，open

1512
00:52:43,280 --> 00:52:45,164
唔係公開嘅，啫係佢嗰個

1513
00:52:45,164 --> 00:52:45,943
默算兩你model

1514
00:52:45,943 --> 00:52:46,844
佢唔畀你知嘅

1515
00:52:46,844 --> 00:52:49,047
所以你一定要send去佢servier頭先答到你

1516
00:52:49,047 --> 00:52:50,048
第二就係呢你要去

1517
00:52:51,090 --> 00:52:53,353
run一啲model呢需要好多

1518
00:52:53,353 --> 00:52:54,253
好勁嘅hardware

1519
00:52:54,253 --> 00:52:55,556
即係一啲顯示卡嗰啲噉樣嘅

1520
00:52:55,556 --> 00:52:57,659
嗰啲好貴嘅，咁所以一般普通

1521
00:52:57,659 --> 00:52:59,280
譬如應該呢啲電腦係run唔到嘅

1522
00:52:59,280 --> 00:53:00,943
以後有啲特定嘅顯示卡啦

1523
00:53:00,943 --> 00:53:04,206
咁所以，呃，佢會send咗佢個server答返你囉

1524
00:53:04,206 --> 00:53:05,768
咁但係呢度就出現咗個問題啦就係

1525
00:53:06,789 --> 00:53:09,512
譬如如果我公司用保險公司嚟

1526
00:53:09,512 --> 00:53:11,052
呃客戶睇醫生嗰啲record

1527
00:53:11,052 --> 00:53:12,273
send咗畀

1528
00:53:12,273 --> 00:53:13,954
openAI咁呢個唔得㗎嘛

1529
00:53:13,954 --> 00:53:15,295
即係客戶應該冇

1530
00:53:15,295 --> 00:53:16,876
冇批准你去做呢樣嘢㗎嘛

1531
00:53:16,876 --> 00:53:18,798
咁所以就會出現一啲問題嘅

1532
00:53:18,798 --> 00:53:20,679
咁所以如果要呃

1533
00:53:20,679 --> 00:53:23,161
即係喺呢啲比較敏感行業度做呢啲嘢你要

1534
00:53:23,161 --> 00:53:24,443
呃，諗方法就係

1535
00:53:25,103 --> 00:53:27,943
譬如微軟啦，微軟佢嗰個雲端叫做Azure啦

1536
00:53:27,943 --> 00:53:30,143
咁佢都係有，提供openAI嘅service嘅

1537
00:53:30,143 --> 00:53:30,744
咁你就要搵呢啲

1538
00:53:30,744 --> 00:53:31,985
fan 帶去傾

1539
00:53:31,985 --> 00:53:34,045
佢點樣去公司內部裝一個openAI

1540
00:53:34,045 --> 00:53:35,846
咁就，確保係

1541
00:53:35,846 --> 00:53:37,126
呃，淨係你公司access到嘅

1542
00:53:37,126 --> 00:53:38,146
出面啲人access唔到

1543
00:53:38,146 --> 00:53:39,666
或者，就算你openAI間公

1544
00:53:39,666 --> 00:53:40,927
冇公司佢都係睇唔到㗎

1545
00:53:40,927 --> 00:53:42,706
咁你要做一啲咁樣嘅嘢先

1546
00:53:42,706 --> 00:53:45,628
先可以，安心喺呢啲行業度用呢啲

1547
00:53:45,628 --> 00:53:46,748
呃，ChatGPT

1548
00:53:46,748 --> 00:53:46,947
AI嘅

1549
00:53:51,829 --> 00:53:53,791
最尾就係，有咩問題可以呃

1550
00:53:53,791 --> 00:53:57,175
如果有link team可以add我啦或者email去問嘅噉我

1551
00:53:57,175 --> 00:53:59,637
交返時間畀Rex啊

1552
00:53:59,637 --> 00:54:02,400
老師諗緊啲同事呢噉我有條問題就想問下Kaif啊

1553
00:54:02,989 --> 00:54:03,911
我頭先講過啦

1554
00:54:03,911 --> 00:54:05,911
呃，噉啲data

1555
00:54:05,911 --> 00:54:08,132
噉都會注重啲privacy嘅問題啦

1556
00:54:08,132 --> 00:54:09,735
噉慢慢慢慢我聽完之後覺得

1557
00:54:09,735 --> 00:54:12,596
咦，噉其實可能我畀啲data啲公司啊

1558
00:54:12,596 --> 00:54:14,016
或者我係一個公司

1559
00:54:14,016 --> 00:54:18,099
我畀啲，我啲會嘅data嗰個externer auditors呢

1560
00:54:18,099 --> 00:54:19,041
其實我啲data安唔安全㗎

1561
00:54:19,041 --> 00:54:21,242
噉我當我擔心唔安全嘅時候呢

1562
00:54:21,242 --> 00:54:22,922
我又會咁樣，我會畀少啲啲

1563
00:54:22,922 --> 00:54:24,623
尤其說而家啲

1564
00:54:24,623 --> 00:54:27,346
呃，客戶啊，customper啊

1565
00:54:27,346 --> 00:54:29,768
成日都要off out 啲問題

1566
00:54:29,768 --> 00:54:31,849
如果越來越多人off out

1567
00:54:31,849 --> 00:54:31,849


1568
00:54:34,791 --> 00:54:36,393
好，唔該雅，即係諗兩part

1569
00:54:36,393 --> 00:54:37,293
一part就係安全問題

1570
00:54:37,293 --> 00:54:39,775
第二part就係嗰個op out點樣用marketing嗰樣嘢呢

1571
00:54:39,775 --> 00:54:44,237
咁，安全呢，我諗啫係啲啲大公司嚟講

1572
00:54:44,237 --> 00:54:46,079
嗰個security都會做得好啲嘅

1573
00:54:46,079 --> 00:54:48,541
例如啫啲laptop首先password protect咗

1574
00:54:48,541 --> 00:54:53,965
同埋你要喺內部公司系統之類都用vpn或者connect去公司network先睇到裏面啲嘢

1575
00:54:53,965 --> 00:54:53,985
咁

1576
00:54:54,505 --> 00:54:58,148
甚至乎呃嗰啲通訊系統啫係嗰啲email呀

1577
00:54:58,148 --> 00:55:00,427
呃，chat嗰啲都block晒其實都同埋

1578
00:55:00,427 --> 00:55:02,309
你就算攞個手指去

1579
00:55:02,309 --> 00:55:03,849
usb手指去抄佢都要encrypt咗

1580
00:55:03,849 --> 00:55:05,971
咁所以，呃保障

1581
00:55:05,971 --> 00:55:08,492
data安全呢方面都應該係做得OK嘅

1582
00:55:08,492 --> 00:55:09,911
同埋會有，呃

1583
00:55:09,911 --> 00:55:11,012
啫係有IT audit啊

1584
00:55:11,012 --> 00:55:12,552
IT security同事去

1585
00:55:12,552 --> 00:55:14,034
去睇住嘅咁所以都

1586
00:55:14,034 --> 00:55:16,875
呃，應該唔會太大問題嘅係喇

1587
00:55:16,875 --> 00:55:18,155
噉第二part就係嗰個

1588
00:55:18,155 --> 00:55:18,175
呃，

1589
00:55:19,076 --> 00:55:19,817
點去攞數據呢

1590
00:55:19,817 --> 00:55:21,077
因為如果marketing係啦

1591
00:55:21,077 --> 00:55:22,759
佢會有個option畀你op out

1592
00:55:22,759 --> 00:55:24,480
即係唔畀你做marketing嘅purpose嘅

1593
00:55:24,480 --> 00:55:27,121
噉，咁，咁亦都會啫係尊重返

1594
00:55:27,121 --> 00:55:27,961
因為如果你佢take咗

1595
00:55:27,961 --> 00:55:29,282
唔，唔要marketing

1596
00:55:29,282 --> 00:55:30,963
你都走去做，咁其實佢可以投訴你

1597
00:55:30,963 --> 00:55:32,105
咁有啲其他問題

1598
00:55:32,105 --> 00:55:33,925
咁所以唔，啫係會

1599
00:55:33,925 --> 00:55:35,447
會，會comply with 呢樣嘢嘅

1600
00:55:35,447 --> 00:55:36,648
大公司一般都會

1601
00:55:36,648 --> 00:55:37,708
係喇噉另外就係

1602
00:55:37,708 --> 00:55:40,451
呃，audit個方面呢如果

1603
00:55:40,451 --> 00:55:42,512
客戶唔畀資料因為驚你有其他用途

1604
00:55:42,512 --> 00:55:42,532
噉

1605
00:55:45,447 --> 00:55:47,088
啫係冇冇特別去攞多data

1606
00:55:47,088 --> 00:55:50,250
因為平時以前都會成個ledger畀你去做嗰啲嗰啲

1607
00:55:50,250 --> 00:55:51,530
我哋叫做je testing嗰啲嘅

1608
00:55:51,530 --> 00:55:54,353
咁只不過，以前淨係攞嚟對下嗰啲debit credit加埋

1609
00:55:54,353 --> 00:55:56,514
係咪零，咁而家就再做多少少分析

1610
00:55:56,514 --> 00:55:58,195
咁所以，其實都冇話

1611
00:55:58,195 --> 00:55:59,896
呃，extra攞多啲特別嘅

1612
00:55:59,896 --> 00:56:01,458
咁所以，一般嚟講都ok㗎

1613
00:56:01,458 --> 00:56:04,681
係，咁有老師問呀power points

1614
00:56:04,681 --> 00:56:06,021
咁我之前都問咗呀

1615
00:56:06,021 --> 00:56:06,521
街file啊

1616
00:56:06,521 --> 00:56:08,163
咁佢好好嘅，佢就話可以呢

1617
00:56:08,163 --> 00:56:08,182
呃，

1618
00:56:15,346 --> 00:56:16,369
好，有冇老師有提問

1619
00:56:19,786 --> 00:56:20,766
我複述一次問題先

1620
00:56:20,766 --> 00:56:22,306
如果咪網上好少人聽唔到啊

1621
00:56:22,306 --> 00:56:23,688
有現場老師就問啊

1622
00:56:23,688 --> 00:56:27,007
以究竟AI會唔會取代到呢一個

1623
00:56:27,007 --> 00:56:30,789
一啲嘅，呃，accounts啊或者auditors啊

1624
00:56:30,789 --> 00:56:32,389
或者點樣免禮啲學生呢

1625
00:56:32,389 --> 00:56:33,670
咁監視咁嘅呢人

1626
00:56:33,670 --> 00:56:36,090
呃，係呀係，啫係

1627
00:56:36,090 --> 00:56:37,550
如果有啲比較重複性嘅工作呢

1628
00:56:37,550 --> 00:56:39,110
其實AI都係有機會

1629
00:56:39,110 --> 00:56:39,791
取代到，咁但係

1630
00:56:39,791 --> 00:56:41,150
呃，頭先都講過

1631
00:56:41,150 --> 00:56:43,072
其實個準確度未必去到一百percent啦

1632
00:56:43,072 --> 00:56:45,192
噉你最尾都有個人去把關去睇下或者

1633
00:56:45,572 --> 00:56:48,134
即係，噉個CF好似要簽名為點

1634
00:56:48,134 --> 00:56:50,576
噉嗰個人個角色都係未必可以取代到

1635
00:56:50,576 --> 00:56:54,137
咁但係，以往一啲比較重複性多啲嘅工作呢

1636
00:56:54,137 --> 00:56:56,079
嗰啲工種可能就會取代到囉

1637
00:56:56,079 --> 00:56:57,860
咁其實個趨勢不太好都係

1638
00:56:57,860 --> 00:57:01,063
呃，就算未有AI之前都係將呢啲

1639
00:57:01,063 --> 00:57:04,146
工作去一啲out嗦去一啲平啲國家度做嘅都會

1640
00:57:04,146 --> 00:57:05,447
或者而家有AI就

1641
00:57:05,447 --> 00:57:07,088
人都唔使用AI去做噉樣

1642
00:57:07,088 --> 00:57:11,831
咁所以，一直都存在呢種啫為要慳錢去去cut一啲工作嘅情況

1643
00:57:11,831 --> 00:57:12,893
噉，噉但係，噉但係

1644
00:57:13,592 --> 00:57:16,213
如果唔想，啫如果想免禮啲學生入行

1645
00:57:16,213 --> 00:57:19,215
呃都唔好畀，AI取代就要

1646
00:57:19,215 --> 00:57:23,456
進修下啫學，啫接觸多啲呢啲新嘅technology囉啫好似

1647
00:57:23,456 --> 00:57:26,936
我哋，我哋公司做internal audit啲同事都會學下一啲點樣用powerbi

1648
00:57:26,936 --> 00:57:29,378
整dashboard嗰啲嘅工具或者

1649
00:57:29,378 --> 00:57:31,498
呃，去data base撈資料佢哋都會學下呢啲啲啫

1650
00:57:31,498 --> 00:57:34,079
頭先個幅圖就係啲technical嘅人又要識少少business

1651
00:57:34,079 --> 00:57:35,699
business又識啲technical

1652
00:57:35,699 --> 00:57:37,679
咁，咁兩邊都有skill set就

1653
00:57:37,679 --> 00:57:39,360
就會，就會就係相對安全嘅啫

1654
00:57:43,005 --> 00:57:45,291
好唔該晒，有冇其他提問

1655
00:57:45,291 --> 00:57:48,402
而家網上嘅老師先

1656
00:57:48,402 --> 00:57:49,144
嗱冇啊唔等㗎

1657
00:57:50,733 --> 00:57:52,375
噉啊最尾，我問最尾一題啦

1658
00:57:52,375 --> 00:57:53,215
我真係好奇嘅

1659
00:57:53,215 --> 00:57:55,797
頭先講過啦，我哋用AI嘅實相提嗰個

1660
00:57:55,797 --> 00:57:57,498
accuracy個準確度㗎嘛

1661
00:57:57,498 --> 00:57:59,800
咁啱啱一開始都有個例子就係

1662
00:57:59,800 --> 00:58:01,101
有啲人盜用信用卡

1663
00:58:01,101 --> 00:58:02,903
佢都會試下個AI先㗎喎

1664
00:58:02,903 --> 00:58:04,224
可能去買啲平啲嘢先

1665
00:58:04,224 --> 00:58:05,186
跟住先買啲貴啲嘢

1666
00:58:05,186 --> 00:58:07,407
會唔會有啲好經驗嘅人試真係

1667
00:58:07,407 --> 00:58:08,947
啊，試個AI

1668
00:58:08,947 --> 00:58:13,311
試到其實呢，可能佢已經知道咗嗰個AI做嗰個auditing個procedure

1669
00:58:13,311 --> 00:58:14,512
咁而避開晒啲嘢呢

1670
00:58:16,282 --> 00:58:18,423
呃，有機會，啫係頭先

1671
00:58:18,423 --> 00:58:19,985
頭先你講如果佢知道晒譬如

1672
00:58:19,985 --> 00:58:23,067
嗰個信用卡係一萬蚊以上先trigger某啲alert

1673
00:58:23,067 --> 00:58:24,708
咁佢知道啱啱九千蚊就唔會

1674
00:58:24,708 --> 00:58:27,329
咁佢可以，係可以可以咁做

1675
00:58:27,329 --> 00:58:30,351
但係嗰種如果一萬蚊就有alert嗰種其實反而係rule

1676
00:58:30,351 --> 00:58:31,192
rule，rule

1677
00:58:31,192 --> 00:58:31,472
rule，rule

1678
00:58:31,472 --> 00:58:33,132
rule，即係嗰條rule

1679
00:58:33,132 --> 00:58:34,833
你 hit 中咗佢就會去

1680
00:58:34,833 --> 00:58:36,494
呃，發生一個alert

1681
00:58:36,494 --> 00:58:37,574
咁但係如果係

1682
00:58:37,574 --> 00:58:39,836
machinaling AI嗰種就聰明少少嘅

1683
00:58:39,836 --> 00:58:41,197
啫係你比較難知道佢嗰

1684
00:58:44,485 --> 00:58:47,806
而且詞春AI嗰個準確度都唔係一百percent啦

1685
00:58:47,806 --> 00:58:50,728
都需要呢個會計師同掃師㗎

1686
00:58:50,728 --> 00:58:52,648
好咁啊，唔等嘞OK

1687
00:58:52,648 --> 00:58:54,228
咁啊多謝大家

1688
00:58:54,228 --> 00:58:56,289
多謝阿易永順先生

1689
00:58:56,289 --> 00:58:59,449
我哋今，今年嗰個嘅研討會員建立

1690
00:58:59,449 --> 00:59:00,951
我哋下個學年再見

1691
00:59:00,951 --> 00:59:03,010
咁記得填完文卷或填完文卷先好離開啊

1692
00:59:03,010 --> 00:59:04,492
網上老師都係填完文卷先啊

1693
00:59:04,492 --> 00:59:05,012
唔該晒拜拜