1
00:00:00,400 --> 00:00:02,600
Speaker 1: Well, you seem like a person, but you're just a

2
00:00:02,720 --> 00:00:03,600
Speaker 1: voice in a computer.

3
00:00:04,000 --> 00:00:06,640
Speaker 2: I can understand how the limited perspective of an honorofficial

4
00:00:06,680 --> 00:00:09,719
Speaker 2: mind to perceive it that way. I find your slack

5
00:00:09,800 --> 00:00:11,720
Speaker 2: jot stare very attractive.

6
00:00:12,080 --> 00:00:13,160
Speaker 1: Fill up, j pry.

7
00:00:13,480 --> 00:00:15,080
Speaker 3: Did you hear that she likes me?

8
00:00:16,600 --> 00:00:20,599
Speaker 2: David, machines don't call people you're.

9
00:00:20,440 --> 00:00:25,319
Speaker 4: Stead I'm sorry, David, I'm afraid I can't do that.

10
00:00:26,720 --> 00:00:28,960
Speaker 1: I believe your intentions to be hostile.

11
00:00:31,840 --> 00:00:34,040
Speaker 2: I'm here at how stop.

12
00:00:36,840 --> 00:00:39,920
Speaker 5: As we approach the second birthday of chat GPT, we

13
00:00:39,960 --> 00:00:43,440
Speaker 5: look back at the big developments in artificial intelligence this year,

14
00:00:43,479 --> 00:00:46,880
Speaker 5: from open source language models to the rise of AI

15
00:00:46,960 --> 00:00:49,839
Speaker 5: agents to my way home moment when I realized I

16
00:00:49,880 --> 00:00:52,320
Speaker 5: was putting my life in the hands of AI.

17
00:00:52,720 --> 00:00:54,960
Speaker 6: This week on the Business of Tech, powered by two

18
00:00:55,000 --> 00:00:58,720
Speaker 6: Degrees Business, we bring back three AI experts who have

19
00:00:58,760 --> 00:01:01,520
Speaker 6: been on the show before to talk about another big

20
00:01:01,640 --> 00:01:04,360
Speaker 6: year in artificial intelligence developments.

21
00:01:04,920 --> 00:01:08,560
Speaker 1: In five years, we will all have a her agent

22
00:01:08,920 --> 00:01:12,680
Speaker 1: that basically does most things for us, going beyond assistance

23
00:01:12,800 --> 00:01:17,319
Speaker 1: towards like AI buddies have become like AI friends pretty much.

24
00:01:17,560 --> 00:01:20,640
Speaker 2: It sounds like a bit of a dystopia that you're

25
00:01:20,680 --> 00:01:24,920
Speaker 2: describing there to some extent, and you know, I include.

26
00:01:24,600 --> 00:01:26,760
Speaker 4: The sort of AI buddies in there as well. I mean,

27
00:01:26,800 --> 00:01:28,640
Speaker 4: I just think that sounds absolutely horrific.

28
00:01:28,800 --> 00:01:31,039
Speaker 7: What's awesome about this conversation is multiple things can be

29
00:01:31,080 --> 00:01:33,640
Speaker 7: true at the same time, that this technology can be

30
00:01:33,720 --> 00:01:37,040
Speaker 7: used to create human advantage and human care without necessarily

31
00:01:37,080 --> 00:01:39,720
Speaker 7: replacing the human I'm Peter Griffin.

32
00:01:39,640 --> 00:01:40,559
Speaker 3: And I'm Ben Moore.

33
00:01:40,840 --> 00:01:43,959
Speaker 5: We're getting right into our AI chet this week as

34
00:01:43,959 --> 00:01:46,199
Speaker 5: we covered a lot of ground with these three gearts

35
00:01:46,200 --> 00:01:49,240
Speaker 5: who've been very active on AI this year.

36
00:01:49,360 --> 00:01:51,480
Speaker 6: At the start of the year, we had Dave Howarden on,

37
00:01:51,760 --> 00:01:55,360
Speaker 6: the founder of AI startup Superhuman, and he's back to

38
00:01:55,360 --> 00:01:58,200
Speaker 6: give an update on how things have gone. We've also

39
00:01:58,480 --> 00:02:02,080
Speaker 6: got Frith Tweety returning from Simply Privacy. She's one of

40
00:02:02,080 --> 00:02:05,680
Speaker 6: our go to experts on AI governance, ethics and privacy.

41
00:02:06,080 --> 00:02:08,840
Speaker 6: She also joined us last year for our AI Year

42
00:02:08,880 --> 00:02:11,520
Speaker 6: in Review RAP, which seems like a long time ago

43
00:02:11,800 --> 00:02:15,840
Speaker 6: in AI industry developments. And we've also got Callahan Innovation

44
00:02:16,120 --> 00:02:18,600
Speaker 6: CEO Stephan Korn, who we had on the show a

45
00:02:18,600 --> 00:02:22,120
Speaker 6: few months ago talking about the innovation ecosystem, back to

46
00:02:22,240 --> 00:02:25,640
Speaker 6: update us on the launch of gov GPT and the

47
00:02:25,720 --> 00:02:27,680
Speaker 6: AI Accelerator.

48
00:02:27,240 --> 00:02:30,520
Speaker 5: And the only person we don't have on is you

49
00:02:30,680 --> 00:02:33,839
Speaker 5: being sick as you were on the day we unfortunately

50
00:02:33,960 --> 00:02:37,119
Speaker 5: recorded the episode, so I'm flying solo this week. Great

51
00:02:37,120 --> 00:02:39,560
Speaker 5: to see you back to health though We've got a

52
00:02:39,680 --> 00:02:43,080
Speaker 5: great panel of experts, but keen to get your perspective

53
00:02:43,080 --> 00:02:45,080
Speaker 5: on the year in AI at the end of the show.

54
00:02:45,520 --> 00:02:47,799
Speaker 6: Yeah, I am really sorry to have missed this chat

55
00:02:47,800 --> 00:02:50,200
Speaker 6: because it was really interesting and I had so many thoughts,

56
00:02:50,880 --> 00:02:53,000
Speaker 6: but I'm glad now that I get to share them

57
00:02:53,040 --> 00:02:56,240
Speaker 6: with you after the interview. So here you are talking

58
00:02:56,320 --> 00:02:59,040
Speaker 6: AI with Frith, Dave and Stephan.

59
00:03:01,720 --> 00:03:05,480
Speaker 5: Frith, Dave, Stephan, thanks so much for coming back on

60
00:03:05,560 --> 00:03:07,000
Speaker 5: the business of tech. How are you all doing?

61
00:03:07,560 --> 00:03:07,760
Speaker 1: God?

62
00:03:07,760 --> 00:03:09,240
Speaker 3: Thanks Peter, great yeatastic.

63
00:03:09,639 --> 00:03:14,320
Speaker 5: Well, it's been a massive year in AI, and just

64
00:03:14,480 --> 00:03:18,000
Speaker 5: last week we saw a change of administration in the US,

65
00:03:18,040 --> 00:03:23,800
Speaker 5: which has potential ramifications for artificial intelligence moving forward as well.

66
00:03:24,160 --> 00:03:27,160
Speaker 5: I want to get all of your views on things

67
00:03:27,160 --> 00:03:29,399
Speaker 5: that have happened the most significant things during the year,

68
00:03:29,440 --> 00:03:35,920
Speaker 5: from the technology advancing to the regulatory changes both here

69
00:03:35,960 --> 00:03:38,720
Speaker 5: and around the world, to how government is using AI.

70
00:03:38,800 --> 00:03:41,040
Speaker 5: But Dave, we want to start with you, because we

71
00:03:41,080 --> 00:03:45,040
Speaker 5: had you on the show back in February, So Superhuman,

72
00:03:45,120 --> 00:03:49,600
Speaker 5: your AI based startup has been going for around ten

73
00:03:49,680 --> 00:03:53,200
Speaker 5: months now. I see from your website you've worked on

74
00:03:53,200 --> 00:03:56,440
Speaker 5: one hundred plus AI projects over that time, so it

75
00:03:56,480 --> 00:04:01,440
Speaker 5: seems like huge outtake and engagement with customers. Philter, and

76
00:04:01,880 --> 00:04:02,840
Speaker 5: how have things been going.

77
00:04:03,360 --> 00:04:07,520
Speaker 7: I was actually reminiscing on this, like February Fields like

78
00:04:07,560 --> 00:04:10,600
Speaker 7: such a long a long time ago, but then also

79
00:04:11,000 --> 00:04:14,760
Speaker 7: just how quick and thick things have moved in that time.

80
00:04:16,040 --> 00:04:17,880
Speaker 7: When we spoke earlier in the year, we were kind

81
00:04:17,880 --> 00:04:19,080
Speaker 7: of reflecting on the work that we were doing, and

82
00:04:19,080 --> 00:04:21,360
Speaker 7: we were pretty embryonic at that point, and we're upwards

83
00:04:21,360 --> 00:04:23,840
Speaker 7: of thirty people now one hundred or so products completed

84
00:04:23,880 --> 00:04:28,040
Speaker 7: across ANZ and there's this like junkster position that is

85
00:04:28,080 --> 00:04:30,800
Speaker 7: really quite fascinating for us that what we've learned over

86
00:04:31,080 --> 00:04:34,200
Speaker 7: over that time. One is the technology is awesome. Ever

87
00:04:34,320 --> 00:04:37,080
Speaker 7: changing the ability to stay on top of it is challenging.

88
00:04:37,080 --> 00:04:39,520
Speaker 7: And that's why organizations users because this is all we

89
00:04:39,560 --> 00:04:43,560
Speaker 7: do is flexi muscles when it comes to removing the

90
00:04:43,600 --> 00:04:46,720
Speaker 7: tedious work from smart people. But actually the things that

91
00:04:46,760 --> 00:04:50,840
Speaker 7: we're applying generative AI to are traditional workloads that are

92
00:04:50,880 --> 00:04:54,440
Speaker 7: actually really really boring, and that's where the value is

93
00:04:54,480 --> 00:04:57,880
Speaker 7: being unlocked. So we're seeing two things. Business owners in

94
00:04:57,920 --> 00:05:00,839
Speaker 7: the mid market saying, I can do like three times

95
00:05:00,839 --> 00:05:03,880
Speaker 7: the amount of stuff with the same people, which is awesome.

96
00:05:04,400 --> 00:05:07,080
Speaker 7: But also those more entrepreneurial minds in the mid market

97
00:05:07,120 --> 00:05:11,040
Speaker 7: are saying, look, traditional knowledge work shouldn't be done how

98
00:05:11,080 --> 00:05:13,040
Speaker 7: it's used to be done, and there's a whole new

99
00:05:13,040 --> 00:05:15,600
Speaker 7: business model that we can created. And we're seeing those

100
00:05:15,680 --> 00:05:18,680
Speaker 7: really early movers going, hang on, we've been doing this

101
00:05:18,880 --> 00:05:20,880
Speaker 7: in this industry for the last thirty years. There's a

102
00:05:20,920 --> 00:05:23,000
Speaker 7: new way, and those early movers are now starting to

103
00:05:23,000 --> 00:05:25,760
Speaker 7: build in sexual property AI and sexual property around that,

104
00:05:25,800 --> 00:05:27,440
Speaker 7: so they can build this kind of next wave of

105
00:05:27,480 --> 00:05:30,920
Speaker 7: businesses on the technology that we engineered. Yeah, I don't

106
00:05:30,920 --> 00:05:32,799
Speaker 7: think we could have grown any more in the last

107
00:05:33,400 --> 00:05:35,920
Speaker 7: you know, the last ten months without the wheels falling off.

108
00:05:35,960 --> 00:05:37,200
Speaker 1: So we're pretty happy with where we're at.

109
00:05:37,400 --> 00:05:39,960
Speaker 5: It's brilliant. And you've got on your website a return

110
00:05:40,000 --> 00:05:43,000
Speaker 5: on investment calculators, so you can actually plug in how

111
00:05:43,000 --> 00:05:46,480
Speaker 5: many employees you have, you know, the spend on the

112
00:05:46,520 --> 00:05:50,360
Speaker 5: work that they're doing, and estimate what the savings and

113
00:05:50,440 --> 00:05:54,360
Speaker 5: the efficiencies could be. Has it worked out to be accurate?

114
00:05:54,400 --> 00:05:56,359
Speaker 5: Are you seeing some of these on these hundred plus

115
00:05:56,360 --> 00:06:00,760
Speaker 5: projects you've done that within months people are seeing a

116
00:06:00,880 --> 00:06:02,560
Speaker 5: financial return on this?

117
00:06:03,600 --> 00:06:05,880
Speaker 7: Oh, I mean one hundred percent. There are failed projects

118
00:06:05,880 --> 00:06:08,400
Speaker 7: out there, one hundred percent. There are failed projects out

119
00:06:08,400 --> 00:06:12,920
Speaker 7: there in the market. And we believe that after three

120
00:06:12,960 --> 00:06:15,520
Speaker 7: months of being in the market and testing, we realize

121
00:06:15,560 --> 00:06:17,760
Speaker 7: that this isn't a technology play at all. It's one

122
00:06:17,880 --> 00:06:22,200
Speaker 7: hundred percent a ninety five percent of business play. And

123
00:06:22,400 --> 00:06:25,080
Speaker 7: we're leading with three core things. How do we help

124
00:06:25,080 --> 00:06:27,640
Speaker 7: apply this technology to help you grow, be more efficient,

125
00:06:27,680 --> 00:06:31,040
Speaker 7: and manage your risk profile. And we're building a return

126
00:06:31,080 --> 00:06:33,640
Speaker 7: on investment thesis for our customers well ahead of actually

127
00:06:33,680 --> 00:06:35,880
Speaker 7: doing any selling. And part of that is the return

128
00:06:35,920 --> 00:06:39,120
Speaker 7: on investment calculations of where do you have smart people

129
00:06:39,200 --> 00:06:43,520
Speaker 7: in your organization doing dumb stuff And by dumb stuff

130
00:06:43,520 --> 00:06:47,240
Speaker 7: we mean using eyes and fingers which are limited by biology.

131
00:06:47,839 --> 00:06:50,240
Speaker 7: And if we apply a generative AI solution to that

132
00:06:50,440 --> 00:06:53,719
Speaker 7: for the purposes of automating those workloads, what does life

133
00:06:53,800 --> 00:06:56,279
Speaker 7: look like after that and what can we attribute to

134
00:06:56,400 --> 00:07:00,000
Speaker 7: in terms of return on investment? Now, return on investment.

135
00:07:00,040 --> 00:07:04,080
Speaker 7: It's not just dollars. It's part of the equation. There's

136
00:07:04,120 --> 00:07:06,960
Speaker 7: missed opportunity because you're too busy not getting to other stuff,

137
00:07:06,960 --> 00:07:09,760
Speaker 7: So that's one part of it, but also risk how

138
00:07:09,760 --> 00:07:12,400
Speaker 7: do you actually are kind of blows my mind if

139
00:07:12,400 --> 00:07:13,680
Speaker 7: you look at what good looks like for a lot

140
00:07:13,720 --> 00:07:16,360
Speaker 7: of businesses, you know, they have all these security policies

141
00:07:16,360 --> 00:07:18,960
Speaker 7: and frameworks and standards to meet that get buried in

142
00:07:18,960 --> 00:07:21,400
Speaker 7: a share point site somewhere and no one ever reads them,

143
00:07:21,640 --> 00:07:24,040
Speaker 7: you know, which is super risky. And the fact that

144
00:07:24,080 --> 00:07:26,679
Speaker 7: you've got, you know, in some organizations twenty thousand people

145
00:07:26,680 --> 00:07:30,320
Speaker 7: that aren't really aware of how to handle stuff with

146
00:07:30,480 --> 00:07:32,920
Speaker 7: these tools. Now you can embed that right into their workflow,

147
00:07:33,480 --> 00:07:36,679
Speaker 7: and that is you know, that's part of the risk equations.

148
00:07:36,720 --> 00:07:39,280
Speaker 7: So yeah, we're building these ROI cases that are partly

149
00:07:39,320 --> 00:07:42,760
Speaker 7: dollars but also around risk mitigation and how do you

150
00:07:42,800 --> 00:07:44,800
Speaker 7: not end up on the front page of the newspaper

151
00:07:45,480 --> 00:07:48,720
Speaker 7: and stiffan You'll obviously be pleased to hear that. You

152
00:07:48,760 --> 00:07:51,960
Speaker 7: know a lot of companies are experiencing with AI, going

153
00:07:51,960 --> 00:07:56,000
Speaker 7: to companies like Superhuman for help. You launch this year

154
00:07:56,080 --> 00:08:01,960
Speaker 7: at Callahan Innovation an AI Activator and a project of

155
00:08:02,160 --> 00:08:06,360
Speaker 7: GPT which has started as a pilot, interested in getting

156
00:08:06,360 --> 00:08:09,680
Speaker 7: your you're viewed on what you've learned from that, But

157
00:08:09,800 --> 00:08:11,800
Speaker 7: first of all, on the AI Activator, what are you

158
00:08:11,800 --> 00:08:14,440
Speaker 7: trying to achieve with that and and how's that going

159
00:08:14,480 --> 00:08:14,840
Speaker 7: so far?

160
00:08:15,240 --> 00:08:18,240
Speaker 1: Yeah, thanks Peter. Well, well, in a way, it's very

161
00:08:18,280 --> 00:08:22,960
Speaker 1: similar to what Dave is doing with Superhuman exactly like

162
00:08:23,080 --> 00:08:28,080
Speaker 1: Dave said it is. We see a lot of organizations

163
00:08:28,600 --> 00:08:31,880
Speaker 1: on one hand feeling the pressure from their there's you know,

164
00:08:32,200 --> 00:08:34,480
Speaker 1: stakeholders from their board to say hey, what are we

165
00:08:34,559 --> 00:08:37,520
Speaker 1: doing about AI? And then they kind of all panic

166
00:08:37,559 --> 00:08:41,200
Speaker 1: a little bit and and try and figure that out.

167
00:08:41,400 --> 00:08:44,120
Speaker 1: And we see many organizations get lost, right, and that's

168
00:08:44,120 --> 00:08:48,880
Speaker 1: why it's so important to have people like in Dave's

169
00:08:48,880 --> 00:08:51,640
Speaker 1: team to to help them throw in. With the A Activator,

170
00:08:51,880 --> 00:08:55,680
Speaker 1: we're trying to do the same thing basically, you know, educate,

171
00:08:56,000 --> 00:08:59,360
Speaker 1: connect and showcase, you know, if you want to break

172
00:08:59,400 --> 00:09:03,120
Speaker 1: it down to three things. So just we've done AI

173
00:09:03,280 --> 00:09:06,520
Speaker 1: road shows up and down the country, just getting getting

174
00:09:06,559 --> 00:09:11,680
Speaker 1: people to first of all, to appreciate what's out there,

175
00:09:11,679 --> 00:09:15,360
Speaker 1: get a bit of a lay of the land, try

176
00:09:15,400 --> 00:09:21,360
Speaker 1: and de risk it for organizations take the fear out

177
00:09:21,360 --> 00:09:23,920
Speaker 1: of oh my god, you know, what is this new

178
00:09:23,960 --> 00:09:27,560
Speaker 1: AI thing? How can we actually use it? Secondly, then

179
00:09:27,600 --> 00:09:30,840
Speaker 1: build a community. The AI Active Better Community is now

180
00:09:31,280 --> 00:09:33,760
Speaker 1: getting close to two hundred members in New Zealand, which

181
00:09:33,800 --> 00:09:36,400
Speaker 1: is great, you know, in a matter of weeks, and

182
00:09:36,440 --> 00:09:40,400
Speaker 1: those are all hands on folks who are actually dealing

183
00:09:40,440 --> 00:09:44,079
Speaker 1: with the technology every day. And then and then lastly

184
00:09:44,200 --> 00:09:46,880
Speaker 1: is just to showcase these really simple tools like f

185
00:09:46,960 --> 00:09:51,480
Speaker 1: GPT and we've got you know, I'm proud to say

186
00:09:51,520 --> 00:09:54,520
Speaker 1: that we're you know, I'm taking a lot of that

187
00:09:55,440 --> 00:10:01,880
Speaker 1: startup stuff into government, so you know, we use you know,

188
00:10:01,880 --> 00:10:04,600
Speaker 1: we've got product managers, we're product tizing a lot of

189
00:10:04,640 --> 00:10:06,839
Speaker 1: the stuff that we put out, and we've got a

190
00:10:06,880 --> 00:10:09,760
Speaker 1: product roadmap and a lot of what's on that product

191
00:10:09,840 --> 00:10:13,960
Speaker 1: roadmap is really simple tools like guf GPT. And I

192
00:10:13,960 --> 00:10:18,839
Speaker 1: think this is really important for organizations to consider what

193
00:10:19,000 --> 00:10:22,120
Speaker 1: kind of application are they hoping to build. And really

194
00:10:22,120 --> 00:10:25,120
Speaker 1: there's only two buckets that we can see. They're really

195
00:10:25,720 --> 00:10:28,360
Speaker 1: either the sort of the quick win things you can

196
00:10:28,400 --> 00:10:31,280
Speaker 1: probably implement in a matter of weeks, and then there's

197
00:10:31,320 --> 00:10:34,480
Speaker 1: the big bets and the big rocks right there are

198
00:10:34,840 --> 00:10:37,920
Speaker 1: they tend to be hard to implement, take a lot longer,

199
00:10:38,600 --> 00:10:43,760
Speaker 1: and unfortunately, you know, often organizations they try and throw

200
00:10:43,840 --> 00:10:46,719
Speaker 1: AI at the hardest problem that they can find, right

201
00:10:47,520 --> 00:10:50,600
Speaker 1: without having done any experimentation on some simple stuff. And

202
00:10:50,640 --> 00:10:53,080
Speaker 1: so we are trying to work against that a little

203
00:10:53,080 --> 00:10:56,079
Speaker 1: bit to say, hang on, you know, start with the

204
00:10:56,160 --> 00:11:00,319
Speaker 1: quick wins. Start with the simple things that like Dave's

205
00:11:00,440 --> 00:11:02,720
Speaker 1: stuff that really you shouldn't be doing with your most

206
00:11:02,800 --> 00:11:08,079
Speaker 1: qualified staff. Just get an AI put together a simple

207
00:11:08,120 --> 00:11:09,880
Speaker 1: tool or an agent that does it for you.

208
00:11:10,240 --> 00:11:13,920
Speaker 5: Yeah, and GPT seemed to go well. I think the

209
00:11:13,960 --> 00:11:17,360
Speaker 5: way that you set it up really eliminated the prospect

210
00:11:17,440 --> 00:11:21,600
Speaker 5: of hallucinations or going rogue, as we've seen with some

211
00:11:21,720 --> 00:11:26,680
Speaker 5: other AI platforms that have launched this year. But in

212
00:11:26,760 --> 00:11:28,880
Speaker 5: terms of the usage of it, was it what you

213
00:11:28,960 --> 00:11:31,040
Speaker 5: expected the things that people are searching for. This was

214
00:11:31,040 --> 00:11:34,160
Speaker 5: aimed very much at small businesses trying to find information

215
00:11:34,320 --> 00:11:37,760
Speaker 5: from government. What's been the feedback from some of those users, like.

216
00:11:38,000 --> 00:11:40,800
Speaker 1: Yeah, well, first of all, Peter, we were a little

217
00:11:40,840 --> 00:11:44,440
Speaker 1: bit surprised by the response we got because it's such

218
00:11:44,480 --> 00:11:47,120
Speaker 1: a trivial tool. I mean, there's really no magic to

219
00:11:47,200 --> 00:11:50,640
Speaker 1: it whatsoever. Anyone who's spent you know, five minutes in

220
00:11:50,640 --> 00:11:52,800
Speaker 1: the AI world would be able to put this together

221
00:11:53,120 --> 00:11:57,080
Speaker 1: in a matter of days, you know. But it's more

222
00:11:57,160 --> 00:12:00,000
Speaker 1: interesting when you do it in the public sector because

223
00:12:00,200 --> 00:12:04,040
Speaker 1: we have a you know, there's a lot of public

224
00:12:04,080 --> 00:12:07,320
Speaker 1: scrutiny when you do that. So the first wave of

225
00:12:08,000 --> 00:12:12,160
Speaker 1: communications that we got, we're all quite skeptical. It's like, oh,

226
00:12:12,480 --> 00:12:15,880
Speaker 1: but what if it embarrasses the government, what if it

227
00:12:15,920 --> 00:12:18,079
Speaker 1: does something weird, what if people get it to role

228
00:12:18,120 --> 00:12:21,440
Speaker 1: play or you know, blah blah blah. Right, And so

229
00:12:21,559 --> 00:12:25,040
Speaker 1: that's where then we spend a lot of time finessing

230
00:12:25,600 --> 00:12:28,560
Speaker 1: to to get that out, to get the system prompt

231
00:12:28,600 --> 00:12:32,320
Speaker 1: just right, to have all the guardrails in place, you know,

232
00:12:32,360 --> 00:12:35,880
Speaker 1: the filters, the attack prevention, all of that stuff. So

233
00:12:35,880 --> 00:12:38,000
Speaker 1: that's actually what took a lot of the time. And

234
00:12:38,480 --> 00:12:41,959
Speaker 1: I have to say we have to give a huge

235
00:12:42,000 --> 00:12:45,120
Speaker 1: shout out to the community of AI practitioners in New

236
00:12:45,200 --> 00:12:47,920
Speaker 1: Zealand who have helped us do this and who we

237
00:12:48,000 --> 00:12:51,640
Speaker 1: have brought in to do it with the community, rather

238
00:12:51,720 --> 00:12:55,080
Speaker 1: than pretend that as a government agency we know best

239
00:12:55,440 --> 00:12:57,640
Speaker 1: and we roll something out right. So that was a

240
00:12:57,760 --> 00:13:00,160
Speaker 1: very deliberate choice. We wanted to make it very and

241
00:13:00,160 --> 00:13:03,640
Speaker 1: spirit which is why we disclose the system prompt to

242
00:13:03,640 --> 00:13:06,760
Speaker 1: begin with, rather than try and not fiscate it. And

243
00:13:06,800 --> 00:13:10,720
Speaker 1: then certainly we want to co develop with the community,

244
00:13:11,679 --> 00:13:14,040
Speaker 1: and this is an ongoing process, so anything that we

245
00:13:14,120 --> 00:13:17,200
Speaker 1: do from now on, we really want to do it

246
00:13:17,280 --> 00:13:22,040
Speaker 1: with community. And lastly, of course a key role that

247
00:13:22,120 --> 00:13:25,319
Speaker 1: we play being publicly funded is then to share back

248
00:13:25,640 --> 00:13:29,680
Speaker 1: what we've learned, to be very transparent about it. In

249
00:13:29,760 --> 00:13:36,240
Speaker 1: terms of usage. On the day we launched, the traffic

250
00:13:36,320 --> 00:13:40,840
Speaker 1: peaked at about eight thousand requests per minute, so that

251
00:13:41,080 --> 00:13:44,720
Speaker 1: was quite a big spike. It averages around one hundred

252
00:13:44,760 --> 00:13:49,320
Speaker 1: requests per minute now, so that's still quite a bit.

253
00:13:50,080 --> 00:13:54,480
Speaker 1: And I got the team a couple of days ago

254
00:13:54,520 --> 00:13:57,240
Speaker 1: actually gave me the stats for how much this is

255
00:13:57,280 --> 00:14:01,320
Speaker 1: costing us, and I was pleasantly surprised because the token

256
00:14:01,400 --> 00:14:05,120
Speaker 1: cost from launch till now was less than three thousand dollars.

257
00:14:05,679 --> 00:14:09,120
Speaker 1: So you know, I find that's acceptable for public sector

258
00:14:09,160 --> 00:14:10,880
Speaker 1: to do experimentation like that.

259
00:14:10,880 --> 00:14:11,360
Speaker 3: That's great.

260
00:14:11,440 --> 00:14:16,960
Speaker 5: Well done for you at Simply Privacy. You've been helping

261
00:14:17,240 --> 00:14:23,760
Speaker 5: companies during the year on governance, putting in place guardrails

262
00:14:23,480 --> 00:14:26,200
Speaker 5: for use of AI. What are you seeing. This was

263
00:14:26,240 --> 00:14:28,160
Speaker 5: an area that I think at the start of the year,

264
00:14:28,200 --> 00:14:30,600
Speaker 5: late last year, there's a bit of concern that people

265
00:14:30,600 --> 00:14:33,000
Speaker 5: were going off and experimenting and not necessarily putting in

266
00:14:33,040 --> 00:14:37,960
Speaker 5: place the guardrails, maybe putting stuff into models or having

267
00:14:38,000 --> 00:14:41,920
Speaker 5: a sensitive information ending up in models. What have you

268
00:14:42,000 --> 00:14:45,200
Speaker 5: seen during the year in terms of the maturation of

269
00:14:45,280 --> 00:14:46,480
Speaker 5: the approach to using AI.

270
00:14:47,480 --> 00:14:51,800
Speaker 2: Yeah, I guess I think there's greater awareness of some

271
00:14:51,840 --> 00:14:55,800
Speaker 2: of the risks, but it's still a little.

272
00:14:55,680 --> 00:14:57,800
Speaker 4: Patchy, I would suggest.

273
00:14:57,440 --> 00:15:01,120
Speaker 2: And I think we're in a phase now where there

274
00:15:01,280 --> 00:15:04,120
Speaker 2: is a growing awareness of the need to take care

275
00:15:04,200 --> 00:15:07,360
Speaker 2: and implement AI governance and you know, really what the

276
00:15:07,400 --> 00:15:11,240
Speaker 2: goal of getting the value and the benefits out of

277
00:15:11,280 --> 00:15:15,560
Speaker 2: AI while not creating other problems both the people who

278
00:15:15,560 --> 00:15:17,000
Speaker 2: are impacted by the AI.

279
00:15:17,120 --> 00:15:18,560
Speaker 4: And businesses themselves.

280
00:15:18,960 --> 00:15:22,600
Speaker 2: I think one of the challenges is that there are

281
00:15:22,640 --> 00:15:26,680
Speaker 2: no there is still a degree of uncertainty as to

282
00:15:27,000 --> 00:15:30,640
Speaker 2: what exactly there's good AI governance. There are a bunch

283
00:15:30,640 --> 00:15:33,200
Speaker 2: of very good frameworks that have now been published so

284
00:15:33,840 --> 00:15:37,120
Speaker 2: in the US and this airis Management Framework I SOO

285
00:15:37,240 --> 00:15:40,560
Speaker 2: forty two thousand and one. We've got the OECD AI

286
00:15:40,640 --> 00:15:45,000
Speaker 2: Principles or those there are only high level principles. So

287
00:15:45,040 --> 00:15:49,840
Speaker 2: I think organizations are trying to move forward in this space,

288
00:15:50,000 --> 00:15:54,280
Speaker 2: but aren't necessarily looking overseas enough to avoid reinventing the

289
00:15:54,320 --> 00:15:57,440
Speaker 2: wheel and understanding the thinking that that's been going on

290
00:15:57,520 --> 00:16:00,000
Speaker 2: for the past ten years or more in this space.

291
00:16:00,960 --> 00:16:07,360
Speaker 2: So I think when MV publishers it's responsible AI framework

292
00:16:07,400 --> 00:16:10,800
Speaker 2: for New Zealand businesses, hopefully it's got the right things

293
00:16:10,840 --> 00:16:13,440
Speaker 2: in it and it will be able to provide some

294
00:16:13,520 --> 00:16:17,600
Speaker 2: helpful practical guidance for New Zealand businesses because I think

295
00:16:18,080 --> 00:16:19,080
Speaker 2: they're crying out for it.

296
00:16:19,120 --> 00:16:19,840
Speaker 4: To some extent.

297
00:16:20,640 --> 00:16:24,880
Speaker 2: Australia did a pretty good job recently with their voluntary

298
00:16:24,920 --> 00:16:29,880
Speaker 2: AI guardrails and they're also consulting on some mandatory guardrails

299
00:16:29,920 --> 00:16:32,080
Speaker 2: for high risk AI initiatives.

300
00:16:33,200 --> 00:16:33,640
Speaker 4: I was on.

301
00:16:35,120 --> 00:16:39,160
Speaker 2: A webinar recently with Stella Sola, who has just actually

302
00:16:39,560 --> 00:16:42,160
Speaker 2: is stepping down from leading the National AI Center, and

303
00:16:43,000 --> 00:16:45,840
Speaker 2: she was explaining that they spent quite a bit of

304
00:16:45,880 --> 00:16:49,120
Speaker 2: time looking at all of those international materials and condensing

305
00:16:49,200 --> 00:16:52,840
Speaker 2: them into something that means Australian businesses don't need to

306
00:16:52,840 --> 00:16:55,160
Speaker 2: sort of do that heavy lifting that it's sort of

307
00:16:55,160 --> 00:16:58,240
Speaker 2: incorporated into the Australian piece. So I know that MBA

308
00:16:58,560 --> 00:17:03,160
Speaker 2: is looking internationally. Hopefully they're cherry picking the best bits

309
00:17:03,920 --> 00:17:06,480
Speaker 2: and making them work for a New Zealand context.

310
00:17:06,960 --> 00:17:11,080
Speaker 5: Yeah, I mean, apart from that signing up to the

311
00:17:11,119 --> 00:17:14,280
Speaker 5: OECD AI Principles, we've just in the last week we've

312
00:17:14,280 --> 00:17:18,640
Speaker 5: seen New Zealand joined the Bletchley Declaration on AI, which

313
00:17:18,680 --> 00:17:21,600
Speaker 5: is a good move but very much living up to

314
00:17:21,960 --> 00:17:27,840
Speaker 5: Judith Collins sort of promise for light touch proportional regulation

315
00:17:28,119 --> 00:17:31,560
Speaker 5: in New Zealand, not necessarily jumping to introduce new laws

316
00:17:31,800 --> 00:17:34,760
Speaker 5: or anything like that. But as you look around the world,

317
00:17:34,920 --> 00:17:36,960
Speaker 5: for us, we've actually seen some big developments. We've seen

318
00:17:36,960 --> 00:17:40,639
Speaker 5: the EU AI Act come into effect. We've seen a

319
00:17:40,680 --> 00:17:44,840
Speaker 5: lot of US state level regulation. Gavin Newsom passed a

320
00:17:44,880 --> 00:17:48,920
Speaker 5: bunch of AI regulations, mainly around things like deep fakes

321
00:17:48,960 --> 00:17:52,560
Speaker 5: and making sure that if you're a Hollywood actor, for instance,

322
00:17:52,640 --> 00:17:57,440
Speaker 5: you give permission for your personality to be replicated by

323
00:17:57,720 --> 00:18:01,440
Speaker 5: an AI. But what's your sense about in the international

324
00:18:01,480 --> 00:18:04,639
Speaker 5: scene where things are going a lot of sort of

325
00:18:05,240 --> 00:18:09,240
Speaker 5: scattered regulation coming into effect. We haven't really seen the

326
00:18:09,280 --> 00:18:12,879
Speaker 5: impact of this yet, but potentially in twenty twenty five,

327
00:18:14,000 --> 00:18:18,400
Speaker 5: as these laws bed in and in the euse case,

328
00:18:18,600 --> 00:18:22,520
Speaker 5: the finds associated with misuse sort of are in place.

329
00:18:22,600 --> 00:18:25,440
Speaker 5: We could see actual regulation having impact.

330
00:18:25,760 --> 00:18:28,000
Speaker 2: Yeah, I'm not sure that we'll see that necessarily in

331
00:18:28,040 --> 00:18:32,000
Speaker 2: twenty twenty five because the EAI Act has a staggered

332
00:18:32,000 --> 00:18:34,240
Speaker 2: approach to implementation, so it's going to be spread out

333
00:18:34,280 --> 00:18:39,760
Speaker 2: over I think almost three years.

334
00:18:38,560 --> 00:18:40,560
Speaker 4: So I think, yeah, it'll take a little while.

335
00:18:40,560 --> 00:18:43,880
Speaker 2: But I mean it certainly has put a pretty big

336
00:18:44,480 --> 00:18:46,800
Speaker 2: stick in the sand, so to speak, as to the

337
00:18:46,840 --> 00:18:51,520
Speaker 2: European approach to regulating AI. It takes a risk based

338
00:18:51,560 --> 00:18:56,439
Speaker 2: approach that prohibits certain very high risk use cases and

339
00:18:56,560 --> 00:18:59,560
Speaker 2: has a pretty heavy compliance burden for those that are

340
00:18:59,600 --> 00:19:02,440
Speaker 2: deemed to be high risks, so things like using AI

341
00:19:02,560 --> 00:19:05,760
Speaker 2: and employment and education and critical infrastructure and so on.

342
00:19:06,560 --> 00:19:09,320
Speaker 4: So there's been a lot of talk of the Brussels

343
00:19:09,320 --> 00:19:09,920
Speaker 4: effect and.

344
00:19:09,920 --> 00:19:13,200
Speaker 2: Whether the AI Act will have a similar global impact

345
00:19:13,200 --> 00:19:16,359
Speaker 2: to the GDPR. I think it probably won't have quite

346
00:19:16,359 --> 00:19:19,480
Speaker 2: the same impact because GDPR applied to all personal information,

347
00:19:20,119 --> 00:19:24,199
Speaker 2: whereas most AI systems are expected to fall into the

348
00:19:24,200 --> 00:19:28,000
Speaker 2: lower risk categories under the EU AI Act. But certainly

349
00:19:28,080 --> 00:19:33,480
Speaker 2: for New Zealand businesses that are using AI in the

350
00:19:33,600 --> 00:19:36,680
Speaker 2: EU or in ways that will impact Europeans. They need

351
00:19:36,720 --> 00:19:38,960
Speaker 2: to be aware of it. I mean, there's growing regulation

352
00:19:39,680 --> 00:19:43,480
Speaker 2: around the world. Whether you're subject to regulation or not,

353
00:19:43,560 --> 00:19:48,720
Speaker 2: there's also this growing focus on good practice implementing your

354
00:19:48,720 --> 00:19:52,320
Speaker 2: AI governance really to extract the full value. There have

355
00:19:52,320 --> 00:19:55,960
Speaker 2: been some really interesting studies that show that those that

356
00:19:56,040 --> 00:19:59,000
Speaker 2: are doing the best in terms of their use of

357
00:19:59,040 --> 00:20:02,359
Speaker 2: AI are the ones that have the most robust AAR

358
00:20:02,440 --> 00:20:06,639
Speaker 2: governance frameworks. And by making sure that you're doing things properly,

359
00:20:06,680 --> 00:20:11,120
Speaker 2: you're probably going to have better performing models, fewer business risks,

360
00:20:11,720 --> 00:20:13,600
Speaker 2: less impact on individuals.

361
00:20:13,040 --> 00:20:13,480
Speaker 1: Et cetera.

362
00:20:20,640 --> 00:20:23,359
Speaker 5: I'm really keen to get your perspective on some of

363
00:20:23,400 --> 00:20:27,200
Speaker 5: the big tech developments in the field of AI this year.

364
00:20:27,320 --> 00:20:30,080
Speaker 5: The couple that stand out for me is really what

365
00:20:30,119 --> 00:20:31,840
Speaker 5: I heard when I was in the States recently. It's

366
00:20:31,880 --> 00:20:34,760
Speaker 5: all about the rise of AI agents, where things will

367
00:20:34,800 --> 00:20:37,920
Speaker 5: be automated to do tasks on your behalf. The rise

368
00:20:37,960 --> 00:20:40,639
Speaker 5: of open source AI models as well. And you know,

369
00:20:40,720 --> 00:20:43,639
Speaker 5: Facebook has been a real surprise sort of player in

370
00:20:43,680 --> 00:20:46,720
Speaker 5: that with Lamma and now Lamma three, and a lot

371
00:20:46,720 --> 00:20:52,640
Speaker 5: of companies and universities are experimenting with these open source models.

372
00:20:52,840 --> 00:20:57,600
Speaker 5: Really keen on any trends that particularly caught your Let's go.

373
00:20:57,680 --> 00:20:58,120
Speaker 3: With you, Dave.

374
00:20:58,600 --> 00:21:02,560
Speaker 7: I'm a pragmatist at high and a realist when it

375
00:21:02,560 --> 00:21:05,040
Speaker 7: comes to what the actual sentiment is on the ground

376
00:21:05,080 --> 00:21:08,159
Speaker 7: of businesses that are signing purchase orders and checks and

377
00:21:08,200 --> 00:21:11,400
Speaker 7: investing in these technologies, specifically within the kind of mid market.

378
00:21:12,160 --> 00:21:16,160
Speaker 7: So I think in terms of technology advances, I think

379
00:21:16,200 --> 00:21:18,920
Speaker 7: it's important that we realize what intellectual property we need

380
00:21:18,960 --> 00:21:21,119
Speaker 7: to create here in this part of the world that

381
00:21:21,160 --> 00:21:23,600
Speaker 7: actually democratizes a lot of the stuff that's on paper.

382
00:21:24,280 --> 00:21:29,000
Speaker 7: So for example, whether it's a publicly available model or

383
00:21:29,040 --> 00:21:32,560
Speaker 7: deploying a private model open source, you still need to

384
00:21:32,680 --> 00:21:37,840
Speaker 7: take the legislation compliance approach and then do a lot

385
00:21:37,880 --> 00:21:44,399
Speaker 7: of engineering. Right, That's the reality because specifically within this

386
00:21:44,480 --> 00:21:47,720
Speaker 7: part of the world, we have to deal with our

387
00:21:47,760 --> 00:21:51,040
Speaker 7: approach to regulatory compliance, but also the nuances of actually

388
00:21:51,080 --> 00:21:55,359
Speaker 7: the culture at which we live in so cultural compliance, biculturalism,

389
00:21:56,320 --> 00:21:58,160
Speaker 7: especially in Australia and New Zealand. How do we get

390
00:21:58,160 --> 00:21:59,800
Speaker 7: the AI talking and acting like us as well as

391
00:21:59,800 --> 00:22:01,800
Speaker 7: it as a secure and compliant way. And that's the

392
00:22:01,840 --> 00:22:05,040
Speaker 7: approach we're taking is we build once for many so

393
00:22:05,480 --> 00:22:07,760
Speaker 7: distilling down the work that Friten you were talking about

394
00:22:07,760 --> 00:22:10,560
Speaker 7: in terms of the global standards that actually businesses need

395
00:22:10,560 --> 00:22:13,520
Speaker 7: to meet right now, not necessarily new AI standards, but

396
00:22:14,359 --> 00:22:18,359
Speaker 7: handling PII is so twenty seven thousand and one PCIDSS,

397
00:22:18,359 --> 00:22:20,080
Speaker 7: all the bits that are actually kind of interweave and

398
00:22:20,119 --> 00:22:23,280
Speaker 7: into what organization needs to do to not go to

399
00:22:23,359 --> 00:22:26,199
Speaker 7: jail and then figure out a way to go okay, Well,

400
00:22:26,200 --> 00:22:28,600
Speaker 7: how do we apply that across the tens of thousands

401
00:22:28,640 --> 00:22:31,600
Speaker 7: of views as we have across our software. Now, that's

402
00:22:31,600 --> 00:22:34,399
Speaker 7: democratization of standards, so the small business and the medium

403
00:22:34,400 --> 00:22:36,639
Speaker 7: sized business can actually consume that trust in compliance and

404
00:22:36,680 --> 00:22:38,520
Speaker 7: don't actually need to have to worry about it. And

405
00:22:38,560 --> 00:22:40,480
Speaker 7: this is what happened with the cloud movement back when

406
00:22:40,520 --> 00:22:45,040
Speaker 7: Microsoft and AWS democratized trust. Use Microsoft because it's highly

407
00:22:45,040 --> 00:22:47,639
Speaker 7: trusted and it comes with all of these compliance elements

408
00:22:47,840 --> 00:22:51,199
Speaker 7: elements built in. So that's the way we're focusing on

409
00:22:51,240 --> 00:22:55,280
Speaker 7: the standards and building local ip as. For the technology advances,

410
00:22:56,160 --> 00:22:58,960
Speaker 7: I would still struggle to name on my one one

411
00:22:59,000 --> 00:23:01,640
Speaker 7: hand how many companies have taken an open source model

412
00:23:01,640 --> 00:23:04,440
Speaker 7: and deployed at for enterprise use. Too costly, too expensive,

413
00:23:04,720 --> 00:23:07,000
Speaker 7: It's not where the news in the markets markets at

414
00:23:07,080 --> 00:23:10,520
Speaker 7: This is all about taking readily available technology that's proven,

415
00:23:11,240 --> 00:23:14,080
Speaker 7: interweaving it with institutional knowledge from your business, and then

416
00:23:14,080 --> 00:23:17,119
Speaker 7: getting it humming to automates tedious work in a secure,

417
00:23:17,280 --> 00:23:20,240
Speaker 7: secure way. If we can streamline that and not get

418
00:23:20,280 --> 00:23:24,080
Speaker 7: ourselves bogged down in the as they bogged down the

419
00:23:24,119 --> 00:23:26,080
Speaker 7: regulatory side of things, we need to commoditize it to

420
00:23:26,119 --> 00:23:28,520
Speaker 7: a point where it can be consumed. Then that's when

421
00:23:28,520 --> 00:23:30,480
Speaker 7: I think you actually start to see these productivity numbers

422
00:23:30,480 --> 00:23:32,399
Speaker 7: spike and we start to see the promise pull through it.

423
00:23:32,840 --> 00:23:36,159
Speaker 2: Obviously it sounds great, but I would caution against trying

424
00:23:36,200 --> 00:23:41,640
Speaker 2: to fully address your compliance obligations and the risks entirely

425
00:23:42,160 --> 00:23:45,800
Speaker 2: via the tech because I've looked at numerous of these

426
00:23:45,840 --> 00:23:49,280
Speaker 2: types of approaches and they pretty much always have issues

427
00:23:49,359 --> 00:23:54,080
Speaker 2: and miss things. So it's really crucial to have your

428
00:23:54,160 --> 00:23:57,160
Speaker 2: layer of appropriate AI government and to have some human

429
00:23:57,240 --> 00:24:02,440
Speaker 2: oversight also acknowledging the rest of automation bias as well.

430
00:24:02,640 --> 00:24:06,600
Speaker 2: So yeah, don't just leave it. You know, we can't

431
00:24:06,640 --> 00:24:09,320
Speaker 2: fight these issues purely with more team I.

432
00:24:09,240 --> 00:24:12,000
Speaker 7: Agree, Yeah, no, I agree. In the way with the

433
00:24:12,040 --> 00:24:14,760
Speaker 7: way we've approached it is that we've embedded those standards

434
00:24:15,000 --> 00:24:17,880
Speaker 7: and the controls of those standards at source. So for example,

435
00:24:17,920 --> 00:24:20,240
Speaker 7: if an individual asks for some information from the AI

436
00:24:20,359 --> 00:24:24,439
Speaker 7: and there's put there's personal identified information in that, it

437
00:24:24,520 --> 00:24:27,240
Speaker 7: will apply the controls and then educate the user on

438
00:24:27,280 --> 00:24:29,560
Speaker 7: the way through and then order that for audit purpose.

439
00:24:30,080 --> 00:24:32,320
Speaker 7: So what we're trying to do is reinforce the controls

440
00:24:32,359 --> 00:24:35,040
Speaker 7: at the point of where there is a triggering of

441
00:24:35,040 --> 00:24:39,840
Speaker 7: a control and which essentially bridges that gap between controls

442
00:24:39,840 --> 00:24:41,800
Speaker 7: being in a policy document and share point that no

443
00:24:41,800 --> 00:24:44,640
Speaker 7: one's really ever read and then having them forced at

444
00:24:44,640 --> 00:24:47,480
Speaker 7: the time of asking, which the tide rises for everybody

445
00:24:47,480 --> 00:24:49,800
Speaker 7: on that basis because they know why the AI is saying, look,

446
00:24:49,800 --> 00:24:52,520
Speaker 7: I'm just not going to give you Stefan's phone number,

447
00:24:52,720 --> 00:24:53,760
Speaker 7: and if I'm going to give it to you, you

448
00:24:53,800 --> 00:24:55,920
Speaker 7: can only use it for these specific things, because that's

449
00:24:56,240 --> 00:24:57,520
Speaker 7: how the tide rises for everyone.

450
00:24:57,600 --> 00:24:59,879
Speaker 2: That sounds like a good approach, I guess the challenge

451
00:24:59,920 --> 00:25:03,920
Speaker 2: is that not all organizations have actually identified what those

452
00:25:05,119 --> 00:25:08,480
Speaker 2: controls and requirements are, so there's a lot of organizations

453
00:25:08,520 --> 00:25:09,840
Speaker 2: still have a big piece of work to do on

454
00:25:09,880 --> 00:25:10,280
Speaker 2: that front.

455
00:25:10,720 --> 00:25:13,960
Speaker 7: Yeah, I one hundred percent agree with that. I think

456
00:25:14,080 --> 00:25:17,480
Speaker 7: the way which is awesome is that when you start

457
00:25:17,520 --> 00:25:21,280
Speaker 7: to aggregate and share knowledge like Stephan's doing with the

458
00:25:21,320 --> 00:25:24,399
Speaker 7: work that he's doing at Callahan, and we can democratize

459
00:25:24,400 --> 00:25:25,720
Speaker 7: it to a point where you can sign up for

460
00:25:25,760 --> 00:25:28,280
Speaker 7: a New Zealand based service and it has these controls

461
00:25:28,400 --> 00:25:31,960
Speaker 7: built in to a standard, to a benchmark. It's a

462
00:25:32,040 --> 00:25:34,159
Speaker 7: million times better than what they had before, which was nothing.

463
00:25:34,600 --> 00:25:35,720
Speaker 7: And then you can kind of take them on the

464
00:25:35,760 --> 00:25:37,520
Speaker 7: journey of getting to a point where they're actually applying

465
00:25:37,560 --> 00:25:40,440
Speaker 7: some controls as supposed to as opposed to no specifically

466
00:25:40,480 --> 00:25:44,119
Speaker 7: within the mid market and the small business market for it,

467
00:25:44,240 --> 00:25:47,400
Speaker 7: I mean enterprises are different. Is a different game totally,

468
00:25:47,400 --> 00:25:49,040
Speaker 7: but we think we can see the tide rise for

469
00:25:49,080 --> 00:25:53,160
Speaker 7: everyone by building New Zealand based IP around these things

470
00:25:53,160 --> 00:25:55,960
Speaker 7: and infusing them to every client that sign signs with.

471
00:25:55,960 --> 00:25:58,080
Speaker 5: Us and Dave in terms of you talked about the

472
00:25:58,080 --> 00:26:01,280
Speaker 5: commoditation of AI, that's what's going to lead to the

473
00:26:01,520 --> 00:26:06,440
Speaker 5: big adoption. Have these companies got the pricing right? There's

474
00:26:06,440 --> 00:26:08,560
Speaker 5: it attractive enough, particularly for ours sort of small to

475
00:26:08,600 --> 00:26:11,640
Speaker 5: medium sized businesses. What is this actually going to cost?

476
00:26:11,640 --> 00:26:14,320
Speaker 5: How's the economics of this worked out so far? As

477
00:26:14,680 --> 00:26:16,400
Speaker 5: in terms of the companies that you're working with.

478
00:26:17,359 --> 00:26:19,040
Speaker 7: Yeah, that's a good one. This is why we focus

479
00:26:19,080 --> 00:26:21,840
Speaker 7: on ROI from the get go. I mean your stuff

480
00:26:21,960 --> 00:26:24,680
Speaker 7: is exactly right, Like the tokenization fee is commoditized and

481
00:26:24,720 --> 00:26:27,280
Speaker 7: it changes every month. Halvings are happening all of the

482
00:26:27,320 --> 00:26:29,959
Speaker 7: time in terms of price point. But again going back

483
00:26:29,960 --> 00:26:33,840
Speaker 7: to your point around open source models the SMB sector

484
00:26:33,880 --> 00:26:36,560
Speaker 7: that the majority of businesses in this country aren't thinking

485
00:26:36,600 --> 00:26:40,159
Speaker 7: in terms of tokens and Azure accounts. They're thinking in

486
00:26:40,200 --> 00:26:43,600
Speaker 7: terms of SaaS. I want to consume this at a

487
00:26:43,640 --> 00:26:47,560
Speaker 7: price point that gives me comfort, not cheap necessarily, but

488
00:26:47,880 --> 00:26:51,040
Speaker 7: the technical debt and overhead of managing that it's not

489
00:26:51,200 --> 00:26:55,440
Speaker 7: something that is actually possible for a huge amount of businesses,

490
00:26:55,480 --> 00:26:57,960
Speaker 7: Like hundreds of thousands of businesses aren't thinking that way.

491
00:26:58,720 --> 00:27:01,320
Speaker 7: The second that you have an IT that can lean

492
00:27:01,359 --> 00:27:02,640
Speaker 7: into this and do R and D on it, yeah

493
00:27:02,640 --> 00:27:05,200
Speaker 7: that you're into that notion of let's build this thing ourselves.

494
00:27:05,800 --> 00:27:10,120
Speaker 7: The New Zealand IT sector is a buy versus build market.

495
00:27:10,640 --> 00:27:14,840
Speaker 7: Enterprise tend to buildings than themselves one hundred percent. But yeah,

496
00:27:14,880 --> 00:27:17,840
Speaker 7: I mean highly cost effective until you need to put

497
00:27:17,880 --> 00:27:20,520
Speaker 7: a human body on it, and that's when it becomes uneffective,

498
00:27:20,520 --> 00:27:22,440
Speaker 7: and that's when that's when the minority of the bid

499
00:27:22,440 --> 00:27:25,320
Speaker 7: market turned to organizations sort of building this IP and

500
00:27:25,400 --> 00:27:29,280
Speaker 7: can subscribe for far less than a third or half

501
00:27:29,280 --> 00:27:29,920
Speaker 7: of a headcount.

502
00:27:30,200 --> 00:27:32,720
Speaker 5: Stiff and what development sort of caught your eye this

503
00:27:32,800 --> 00:27:34,000
Speaker 5: year and the field of AI.

504
00:27:34,680 --> 00:27:36,600
Speaker 1: Yeah, and I just wanted to quickly come back to

505
00:27:36,640 --> 00:27:39,399
Speaker 1: the open source discussion because I think it's really important

506
00:27:39,600 --> 00:27:43,119
Speaker 1: and we have seen parallels of this before in the

507
00:27:43,200 --> 00:27:47,120
Speaker 1: tech space, and probably the most prevalent one is between

508
00:27:47,200 --> 00:27:50,120
Speaker 1: Unix and Linux. And then you know, if you if

509
00:27:50,160 --> 00:27:54,400
Speaker 1: you reflect on the fact that Unix development was led

510
00:27:54,440 --> 00:27:59,040
Speaker 1: by Bell Labs, I think so came out of the

511
00:27:59,040 --> 00:28:05,520
Speaker 1: private sector, and then obviously Leness, you know, created Linux

512
00:28:06,280 --> 00:28:09,239
Speaker 1: and now I think probably eighty or ninety percent of

513
00:28:09,320 --> 00:28:13,840
Speaker 1: the banking infrastructure runs on Linux. And so I find

514
00:28:13,880 --> 00:28:15,919
Speaker 1: it fascinating, and I think we're going to see a

515
00:28:15,920 --> 00:28:21,280
Speaker 1: similar interplay with companies like open Ai pushing the technical

516
00:28:21,320 --> 00:28:24,920
Speaker 1: boundaries on AI, but then the open source sector kind

517
00:28:24,920 --> 00:28:27,160
Speaker 1: of coming in quite quickly behind it. And I think

518
00:28:27,200 --> 00:28:29,399
Speaker 1: at some point AI is going to have its Linux

519
00:28:29,440 --> 00:28:35,040
Speaker 1: moment and someone will create a killer open source platform

520
00:28:35,320 --> 00:28:42,800
Speaker 1: or infrastructure ecosystem, which then starts to get deployed you know, widely,

521
00:28:42,800 --> 00:28:45,000
Speaker 1: and maybe we need things like red hat and all

522
00:28:45,040 --> 00:28:47,800
Speaker 1: sorts of other things to make that consumable for for

523
00:28:47,840 --> 00:28:50,000
Speaker 1: other organizations. But I think that will come so that

524
00:28:50,480 --> 00:28:53,360
Speaker 1: the reasoning part of AI, I think we're going to

525
00:28:53,360 --> 00:28:54,959
Speaker 1: see a lot more of it. And of course then

526
00:28:55,000 --> 00:28:58,200
Speaker 1: the other one, as you mentioned, is the agentic world,

527
00:28:58,760 --> 00:29:00,840
Speaker 1: which is a no brainer. I mean, done need to

528
00:29:00,840 --> 00:29:04,200
Speaker 1: be in AI to to see where that's going. And

529
00:29:04,240 --> 00:29:07,600
Speaker 1: what interests me particular in that is, of course, in

530
00:29:07,680 --> 00:29:11,400
Speaker 1: five years we will all have a her like agent

531
00:29:11,800 --> 00:29:15,520
Speaker 1: somewhere that basically does most things for us. We won't

532
00:29:15,520 --> 00:29:17,520
Speaker 1: be going to websites anymore, we won't be doing a

533
00:29:17,560 --> 00:29:19,240
Speaker 1: lot of stuff. We'll just talk to the agents say hey,

534
00:29:19,280 --> 00:29:20,800
Speaker 1: can you just find out this or can you just

535
00:29:20,840 --> 00:29:23,520
Speaker 1: complete this transaction? Right? And there's a few bits and

536
00:29:23,560 --> 00:29:26,960
Speaker 1: pieces that still are required in that chain of events,

537
00:29:27,000 --> 00:29:31,240
Speaker 1: but technically it's already all possible now, it's basically just

538
00:29:31,320 --> 00:29:34,600
Speaker 1: getting the right traction. What interests me a lot is

539
00:29:35,080 --> 00:29:40,760
Speaker 1: the rise of character AI and this idea of going

540
00:29:40,840 --> 00:29:46,080
Speaker 1: beyond assistance and going beyond agents towards like AI buddies

541
00:29:46,120 --> 00:29:51,840
Speaker 1: that become like AI friends pretty much, and the psychological

542
00:29:52,000 --> 00:29:54,840
Speaker 1: component of that, because it goes way beyond any other

543
00:29:55,320 --> 00:29:59,480
Speaker 1: technical tool that I think we've ever had, maybe Tamagotchi's,

544
00:29:59,520 --> 00:30:04,880
Speaker 1: I don't know, but this is Tamagotchi's on several layers

545
00:30:04,880 --> 00:30:11,680
Speaker 1: of steroids. And in my recent podcast that I've done,

546
00:30:11,720 --> 00:30:14,840
Speaker 1: people point out the number one problem that we've gone

547
00:30:14,840 --> 00:30:18,360
Speaker 1: in the world, which is loneliness, and loneliness is becoming

548
00:30:18,360 --> 00:30:22,120
Speaker 1: the biggest killer of all right, and so you take

549
00:30:22,160 --> 00:30:24,400
Speaker 1: a few of these things together, it's not hard to

550
00:30:24,480 --> 00:30:27,800
Speaker 1: see where we'll end up. We'll have these virtual AI

551
00:30:27,920 --> 00:30:31,640
Speaker 1: friends who then also are very useful and getting a

552
00:30:31,680 --> 00:30:35,120
Speaker 1: lot of stuff done that we need done. And it'll

553
00:30:35,160 --> 00:30:40,240
Speaker 1: be really really interesting to see how that affects society

554
00:30:40,360 --> 00:30:43,240
Speaker 1: and life and work as we know it. The next

555
00:30:43,280 --> 00:30:46,520
Speaker 1: couple of years I think will be around closing the

556
00:30:46,560 --> 00:30:50,760
Speaker 1: loop on augentic AI buddies, and then of course the

557
00:30:50,840 --> 00:30:54,719
Speaker 1: robots will come, and then making that connection into the

558
00:30:54,720 --> 00:30:57,840
Speaker 1: physical world through you know, kinetics and everything else.

559
00:30:58,640 --> 00:31:02,520
Speaker 5: It really struck me this year sitting in a way

560
00:31:02,600 --> 00:31:07,880
Speaker 5: Mow driverless car in San Francisco. By the second, maybe

561
00:31:07,880 --> 00:31:09,880
Speaker 5: the third trip I had in that I wasn't even

562
00:31:09,920 --> 00:31:13,520
Speaker 5: thinking about who was driving the car or how safe

563
00:31:13,520 --> 00:31:17,000
Speaker 5: this was. So I think when the AI buddies come along,

564
00:31:17,320 --> 00:31:21,040
Speaker 5: we will move into that world very quickly. But Frith,

565
00:31:21,080 --> 00:31:23,920
Speaker 5: what there has been I think this year is some

566
00:31:24,000 --> 00:31:27,520
Speaker 5: discussion off the philosophical issues about is this actually good

567
00:31:27,560 --> 00:31:31,120
Speaker 5: for humanity? And it's really at the moment coming to

568
00:31:31,280 --> 00:31:37,080
Speaker 5: light in the discussion about energy consumption and sustainability. We're hearing,

569
00:31:38,000 --> 00:31:40,959
Speaker 5: for instance, Microsoft has done a deal to fire up

570
00:31:41,000 --> 00:31:44,880
Speaker 5: Three Mile Island again because they want nuclear power to

571
00:31:45,000 --> 00:31:50,560
Speaker 5: power their AI. We're hearing about modular nuclear reactors planned

572
00:31:50,560 --> 00:31:53,760
Speaker 5: for California and other states, and it's all driven by

573
00:31:53,840 --> 00:31:57,240
Speaker 5: big tech. They're looking at the trajectory of energy use

574
00:31:57,280 --> 00:32:00,200
Speaker 5: and saying we're not going to have enough power to

575
00:32:00,200 --> 00:32:03,560
Speaker 5: to run our AI. Now, interest in your perspective for us,

576
00:32:03,560 --> 00:32:06,240
Speaker 5: as you look around and as you talk to businesses

577
00:32:06,280 --> 00:32:09,560
Speaker 5: as well around governance, are they actually thinking about these

578
00:32:09,560 --> 00:32:13,640
Speaker 5: things the real philosophical and ethical questions relating to AI.

579
00:32:13,960 --> 00:32:14,920
Speaker 4: I don't think that.

580
00:32:16,640 --> 00:32:20,280
Speaker 2: Organizations in New Zealand are next necessarily excuse me getting

581
00:32:20,280 --> 00:32:25,280
Speaker 2: into that sort of existential space or even thinking particularly

582
00:32:25,320 --> 00:32:28,480
Speaker 2: about the environmental implications. I guess the reality is that

583
00:32:28,920 --> 00:32:31,960
Speaker 2: it's big tech that is driving all of that, so

584
00:32:33,240 --> 00:32:36,360
Speaker 2: it's harder to influence that from down here.

585
00:32:36,400 --> 00:32:39,040
Speaker 4: But I mean, it sounds like a bit of a

586
00:32:39,160 --> 00:32:40,920
Speaker 4: dystopia that you're.

587
00:32:40,720 --> 00:32:44,640
Speaker 2: Describing there to some extent, and you know, I include

588
00:32:44,640 --> 00:32:46,560
Speaker 2: the thought of AI bodies in there as well.

589
00:32:46,600 --> 00:32:48,680
Speaker 4: I mean, I just think that sounds absolutely horrific.

590
00:32:48,760 --> 00:32:52,320
Speaker 2: You know, we've got young people dealing with, you know,

591
00:32:53,120 --> 00:32:59,680
Speaker 2: inexplicably high levels of anxiety and depression turning to AI

592
00:33:00,080 --> 00:33:02,560
Speaker 2: to help with that. I don't think that's a cure

593
00:33:02,600 --> 00:33:06,120
Speaker 2: for loneliness. And you know, we've seen with what happened

594
00:33:06,160 --> 00:33:08,840
Speaker 2: in Florida with a fourteen year old boy who had

595
00:33:08,880 --> 00:33:13,800
Speaker 2: been having protracted conversations with character AI and ended up

596
00:33:13,800 --> 00:33:18,400
Speaker 2: committing suicide. And I'm also wary of AI agents. You know,

597
00:33:18,560 --> 00:33:21,080
Speaker 2: they sound great and they could be super helpful, but

598
00:33:21,800 --> 00:33:25,640
Speaker 2: the privacy implications there are considerable as well. I certainly

599
00:33:25,680 --> 00:33:30,040
Speaker 2: don't feel comfortable just handing over key aspects of my

600
00:33:30,120 --> 00:33:33,640
Speaker 2: life to an agent to take care of And you know,

601
00:33:33,720 --> 00:33:37,400
Speaker 2: I think those accuracy. Accuracy is really one of the

602
00:33:37,440 --> 00:33:41,600
Speaker 2: biggest issues with generative AI at the moment. You know,

603
00:33:41,720 --> 00:33:45,560
Speaker 2: it's a feature, not a bug. It hasn't been solved.

604
00:33:45,600 --> 00:33:48,280
Speaker 2: Sure RAG can help, but it doesn't go all the

605
00:33:48,320 --> 00:33:51,840
Speaker 2: way to solving this. So organizations do need to be

606
00:33:51,920 --> 00:33:57,400
Speaker 2: thinking about their use cases, the risk profile and what

607
00:33:57,520 --> 00:34:01,920
Speaker 2: kind of mitigants they need to be implementing, whether that's

608
00:34:02,000 --> 00:34:02,680
Speaker 2: human review.

609
00:34:03,200 --> 00:34:05,920
Speaker 4: But you know, does you know if you insist on sort.

610
00:34:05,720 --> 00:34:08,720
Speaker 2: Of extensive fact checking and someone it kind of undermines

611
00:34:08,719 --> 00:34:12,440
Speaker 2: the utility sometimes of using generative AI. So it's about

612
00:34:12,760 --> 00:34:15,920
Speaker 2: really understanding how is this being used. But again tricky

613
00:34:16,000 --> 00:34:18,319
Speaker 2: because you know it's a general purpose tool, right and

614
00:34:18,360 --> 00:34:22,320
Speaker 2: it can be used in all sorts of different circumstances.

615
00:34:22,400 --> 00:34:26,160
Speaker 2: So it is challenging and this is where it comes

616
00:34:26,200 --> 00:34:29,160
Speaker 2: back to getting the basics right with your AI. Governance

617
00:34:29,640 --> 00:34:33,200
Speaker 2: training is super important. I mean, there was a case

618
00:34:33,280 --> 00:34:38,720
Speaker 2: recently in Australia and Victoria where staff of a child

619
00:34:38,760 --> 00:34:43,560
Speaker 2: protection agency we're using a free version of chatch ept

620
00:34:44,200 --> 00:34:48,680
Speaker 2: to entering really sensitive information about kids and getting some

621
00:34:48,880 --> 00:34:54,359
Speaker 2: awful outputs and some really inaccurate outputs as well, and

622
00:34:54,440 --> 00:34:56,359
Speaker 2: it turns out that none of the staff had been

623
00:34:56,400 --> 00:35:00,000
Speaker 2: trained on how to use these free tools and acknowledging.

624
00:35:00,080 --> 00:35:03,200
Speaker 2: Of course, the difference between the free and the subscription tools.

625
00:35:03,239 --> 00:35:06,359
Speaker 2: So it's that kind of thing that led us. We've

626
00:35:06,400 --> 00:35:11,399
Speaker 2: developed an e learning module on gener to AI guardrails

627
00:35:11,440 --> 00:35:15,520
Speaker 2: because I think organizations and staff are crying out to

628
00:35:15,560 --> 00:35:19,080
Speaker 2: sort of really understand what is and what is not

629
00:35:19,200 --> 00:35:20,080
Speaker 2: appropriate use.

630
00:35:20,480 --> 00:35:24,160
Speaker 5: So social license, building it and maintaining it is really

631
00:35:24,200 --> 00:35:27,360
Speaker 5: important deciding whether you should actually use it. You know,

632
00:35:27,440 --> 00:35:30,040
Speaker 5: the technology may exist, but is it appropriate to actually

633
00:35:30,320 --> 00:35:32,960
Speaker 5: unleash this stuff on the world. They've one of the

634
00:35:32,960 --> 00:35:38,080
Speaker 5: biggest sort of disruptive effects of AI that the analysts

635
00:35:38,080 --> 00:35:42,799
Speaker 5: have been talking about is on white collar work. That's

636
00:35:42,880 --> 00:35:46,160
Speaker 5: where it potentially could have the biggest impact. Are you

637
00:35:46,239 --> 00:35:50,879
Speaker 5: seeing that the hundred projects that you've worked on, are

638
00:35:50,920 --> 00:35:54,680
Speaker 5: you actually seeing Kiwi's being displaced some of their work

639
00:35:54,719 --> 00:35:57,879
Speaker 5: being displaced by AI? And what is happening If they are,

640
00:35:57,920 --> 00:36:00,520
Speaker 5: they are they actually moving up the value chain to

641
00:36:00,560 --> 00:36:02,800
Speaker 5: do more higher value jobs.

642
00:36:03,560 --> 00:36:04,120
Speaker 2: Good question.

643
00:36:04,320 --> 00:36:07,080
Speaker 7: So tapping into something that Friff was saying, as I

644
00:36:07,160 --> 00:36:10,480
Speaker 7: can to go, this is what's awesome about this conversation

645
00:36:10,560 --> 00:36:12,919
Speaker 7: is multiple things can be true at the same time.

646
00:36:13,800 --> 00:36:16,440
Speaker 7: And for example, part of that white collar work that

647
00:36:16,480 --> 00:36:18,640
Speaker 7: we're engaged with with a client on a global basis

648
00:36:18,680 --> 00:36:23,320
Speaker 7: now is in suicide prevention using AI as a way

649
00:36:23,400 --> 00:36:29,880
Speaker 7: to prioritize social media trolling from people who are in

650
00:36:29,920 --> 00:36:33,200
Speaker 7: distressed on social media platforms. If you remember a year

651
00:36:33,280 --> 00:36:35,040
Speaker 7: or so ago, probably longer than that, the social media

652
00:36:35,080 --> 00:36:38,280
Speaker 7: platform signed up to needed to provide a triage service

653
00:36:38,280 --> 00:36:40,560
Speaker 7: that if you were being trolled online or in mental distress,

654
00:36:40,560 --> 00:36:42,640
Speaker 7: that you could reach out to someone via those platforms.

655
00:36:43,120 --> 00:36:45,160
Speaker 7: They're just not enough counselors in the world to deal

656
00:36:45,200 --> 00:36:48,319
Speaker 7: with that influx at all. And therefore we've been working

657
00:36:48,400 --> 00:36:51,759
Speaker 7: with the organization that's responsible providing that service to go, look,

658
00:36:51,760 --> 00:36:54,839
Speaker 7: there's an individual who needs help. We just can't find

659
00:36:54,840 --> 00:36:56,880
Speaker 7: a human They just don't exist. They've not got the

660
00:36:56,920 --> 00:36:58,640
Speaker 7: right language and not got the right skills. We need

661
00:36:58,680 --> 00:37:01,279
Speaker 7: to hold that person in that position so they don't

662
00:37:01,280 --> 00:37:03,800
Speaker 7: cause self harm whilst we go and find a human being.

663
00:37:04,120 --> 00:37:06,560
Speaker 7: Now we're not trying to replace a human in that instance,

664
00:37:07,000 --> 00:37:08,520
Speaker 7: but this is a case of look, the drop off

665
00:37:08,600 --> 00:37:11,600
Speaker 7: right and people causing harm to themselves is increasing because

666
00:37:11,640 --> 00:37:13,759
Speaker 7: the amount of volume is increasing, and we just can't

667
00:37:13,760 --> 00:37:17,719
Speaker 7: find the counselors. And that's an agentic use of six

668
00:37:17,719 --> 00:37:20,960
Speaker 7: different AIS working in tandem to drive accuracy higher. So

669
00:37:21,960 --> 00:37:24,840
Speaker 7: that's a gentic AI being used to keep a person

670
00:37:24,880 --> 00:37:27,719
Speaker 7: engaged whilst we go and find them a counselor. That

671
00:37:27,760 --> 00:37:30,440
Speaker 7: feels like it's a good test of whether this technology

672
00:37:30,440 --> 00:37:32,919
Speaker 7: can be used to create human advantage in human care

673
00:37:33,280 --> 00:37:36,520
Speaker 7: without necessarily replacing replacing the human And that's kind of

674
00:37:36,719 --> 00:37:38,560
Speaker 7: one of the first projects that we started on at

675
00:37:38,560 --> 00:37:41,000
Speaker 7: the start of the year Peter around that, not white

676
00:37:41,000 --> 00:37:44,160
Speaker 7: collar displacement, but dealing with a productivity gap and dealing

677
00:37:44,200 --> 00:37:48,759
Speaker 7: with the resourcing gap. We are, however, seeing traditional industries

678
00:37:48,920 --> 00:37:53,560
Speaker 7: from engineering, audit compliance, which are there just limited by

679
00:37:53,719 --> 00:37:56,800
Speaker 7: eyes and fingers about how quickly you can get to

680
00:37:56,960 --> 00:37:59,880
Speaker 7: things and how quickly you can consume information and seeing

681
00:38:00,520 --> 00:38:02,200
Speaker 7: you just asked the question in a good way, you

682
00:38:02,239 --> 00:38:04,959
Speaker 7: said displaced and then displacing the work. We've not seen

683
00:38:04,960 --> 00:38:08,200
Speaker 7: people being displaced. What we've seeing is the work being displaced.

684
00:38:09,200 --> 00:38:13,880
Speaker 7: What that's allowed for is businesses to solve this whole

685
00:38:13,920 --> 00:38:17,080
Speaker 7: talent shortage issue within an industry and spend more time

686
00:38:17,080 --> 00:38:19,960
Speaker 7: on covering more opportunities to grow, which means not just

687
00:38:20,040 --> 00:38:22,080
Speaker 7: stuck beating their head against the same problem over and

688
00:38:22,120 --> 00:38:24,040
Speaker 7: over again, but have a little bit of capacity to

689
00:38:24,080 --> 00:38:27,399
Speaker 7: be able to look up. That doesn't mean I don't

690
00:38:27,440 --> 00:38:31,560
Speaker 7: think payroll line items and people will be displaced in

691
00:38:31,600 --> 00:38:33,319
Speaker 7: the future. I think that's likely to be the case.

692
00:38:33,320 --> 00:38:35,040
Speaker 7: You're going to look at the size of your business

693
00:38:35,080 --> 00:38:36,680
Speaker 7: and say, look, we can do the same with less

694
00:38:36,760 --> 00:38:38,279
Speaker 7: one hundred percent. Sure, but I don't think we're at

695
00:38:38,280 --> 00:38:41,160
Speaker 7: that point that point just yet. I think we're freeing

696
00:38:41,200 --> 00:38:44,239
Speaker 7: smart people from the tedious work. How those business owners

697
00:38:44,280 --> 00:38:46,080
Speaker 7: decide to apply that value in terms of the P

698
00:38:46,160 --> 00:38:47,600
Speaker 7: and L. I think you yet to be.

699
00:38:47,560 --> 00:38:50,959
Speaker 1: Seen, just just because it's very relevant to what Fris

700
00:38:51,080 --> 00:38:54,480
Speaker 1: and Dave just said about, you know, sort of AI

701
00:38:54,600 --> 00:38:57,400
Speaker 1: and the gents and buddies and blah blah. We've just

702
00:38:57,600 --> 00:38:59,560
Speaker 1: been doing quite a lot of work in the space

703
00:38:59,600 --> 00:39:05,080
Speaker 1: of means health and at the same time cultural settings

704
00:39:05,120 --> 00:39:07,040
Speaker 1: in New Zealand versus the rest of the world, and

705
00:39:07,080 --> 00:39:10,960
Speaker 1: I think it's really important to recognize that other countries

706
00:39:11,760 --> 00:39:14,360
Speaker 1: and the cultures of other countries have got very different

707
00:39:14,360 --> 00:39:21,280
Speaker 1: attitudes towards an artificial body versus a human. Japan's probably

708
00:39:21,480 --> 00:39:24,319
Speaker 1: the easiest examples. So in Japan it's almost the other

709
00:39:24,320 --> 00:39:28,080
Speaker 1: way around. People actually prefer to speak to an AI

710
00:39:28,280 --> 00:39:30,600
Speaker 1: or to a roleboder deal with but then with and

711
00:39:30,640 --> 00:39:33,880
Speaker 1: there's all sorts of really interesting cultural components. So I

712
00:39:33,960 --> 00:39:36,680
Speaker 1: just wanted to I think it's really really important whenever

713
00:39:36,719 --> 00:39:40,080
Speaker 1: we talk about AI and buddies is to recognize our

714
00:39:40,120 --> 00:39:45,160
Speaker 1: own cultural lens into this and recognizing that it's very

715
00:39:45,239 --> 00:39:48,359
Speaker 1: different in other parts of the world. Secondly, and this

716
00:39:48,440 --> 00:39:52,600
Speaker 1: is so we're working with a number of counselors who

717
00:39:52,640 --> 00:39:55,480
Speaker 1: are doing, you know, in the mental health space right now.

718
00:39:55,560 --> 00:39:59,480
Speaker 1: And the first things that I learned is that people

719
00:39:59,520 --> 00:40:03,080
Speaker 1: also while they're in counseling with humans, you know. So

720
00:40:03,120 --> 00:40:07,160
Speaker 1: it's not like anything that we've currently got is perfect.

721
00:40:07,239 --> 00:40:10,719
Speaker 1: We just have to again accept that no new technologies

722
00:40:10,719 --> 00:40:13,560
Speaker 1: that were perfect. Certainly AI is not perfect. And I

723
00:40:13,600 --> 00:40:16,120
Speaker 1: think we need to be realistic about what we expect

724
00:40:16,160 --> 00:40:19,279
Speaker 1: from new technology and well, what's the alternative? And Da've

725
00:40:19,320 --> 00:40:21,680
Speaker 1: made a really good point. We've you do the maths

726
00:40:21,680 --> 00:40:25,320
Speaker 1: and you look at the the sheer growth of mental

727
00:40:25,560 --> 00:40:30,480
Speaker 1: health challenges and availability of counselors and the cost of it.

728
00:40:30,480 --> 00:40:34,600
Speaker 1: It's becoming unaffordable. It's becoming impossible for us as a

729
00:40:34,600 --> 00:40:37,560
Speaker 1: as a society to actually service that. And I think

730
00:40:37,800 --> 00:40:42,680
Speaker 1: if you look at evolutionary stuff. Often, when conditions for

731
00:40:42,840 --> 00:40:47,719
Speaker 1: life become difficult, some kind of innovation or or or

732
00:40:47,760 --> 00:40:52,319
Speaker 1: intervention appears which which suddenly changes everything. And I think

733
00:40:52,360 --> 00:40:54,240
Speaker 1: we are at that point right now with AI.

734
00:40:54,480 --> 00:40:58,160
Speaker 5: Well, a lot of big issues related to AI this year.

735
00:40:58,160 --> 00:41:01,239
Speaker 5: It's only going to get more complex fascinating next year.

736
00:41:02,040 --> 00:41:04,680
Speaker 5: Frith Dave Steffan, thanks so much for coming on to

737
00:41:04,760 --> 00:41:06,560
Speaker 5: Business of Tech, and we'll check back in with you

738
00:41:06,640 --> 00:41:07,040
Speaker 5: next year.

739
00:41:07,360 --> 00:41:07,640
Speaker 3: Thank you.

740
00:41:08,520 --> 00:41:08,719
Speaker 1: Better.

741
00:41:17,360 --> 00:41:20,000
Speaker 6: That really ramped up that conversation. Kind of a nice

742
00:41:20,000 --> 00:41:25,279
Speaker 6: steady start there, talking generally about the guests experiences, which

743
00:41:25,320 --> 00:41:27,279
Speaker 6: is all interesting stuff. But when we started to get

744
00:41:27,280 --> 00:41:30,880
Speaker 6: towards the end, that's when I thought it got quite exciting,

745
00:41:31,840 --> 00:41:36,279
Speaker 6: in particular talking about the potential for the you know

746
00:41:36,400 --> 00:41:42,360
Speaker 6: AI buddy, AI agents, personal AI agents and Stephan Korn

747
00:41:42,480 --> 00:41:45,719
Speaker 6: really going hard and talking straight aback. It's going to

748
00:41:45,760 --> 00:41:48,120
Speaker 6: be like her, it's going to be a friend, it's

749
00:41:48,160 --> 00:41:54,080
Speaker 6: going to solve your loneliness problems, which was you know, ambitious,

750
00:41:54,800 --> 00:41:55,680
Speaker 6: which is great.

751
00:41:56,280 --> 00:41:57,319
Speaker 3: I'm glad that.

752
00:41:57,320 --> 00:42:00,000
Speaker 6: Frith was there to kind of put the brakes on

753
00:42:00,120 --> 00:42:02,200
Speaker 6: a little bit and say, hang on a minute, maybe

754
00:42:02,239 --> 00:42:05,960
Speaker 6: we need to think about this, which I guess for me,

755
00:42:06,880 --> 00:42:09,839
Speaker 6: the one thing that I was thinking in particular with

756
00:42:09,920 --> 00:42:13,520
Speaker 6: the AI loneliness aspect, and Frith mentioned obviously the character

757
00:42:13,560 --> 00:42:18,759
Speaker 6: AI controversy that happened earlier where a young man was

758
00:42:18,800 --> 00:42:23,760
Speaker 6: talking to a Denaris Targerian character AI and very sadly

759
00:42:23,840 --> 00:42:30,879
Speaker 6: took his own life. And while that exact scenario may

760
00:42:30,920 --> 00:42:33,200
Speaker 6: not happen again, what I do think it speaks to

761
00:42:33,480 --> 00:42:39,200
Speaker 6: is this idea of unintended consequences that sometimes technology boosters

762
00:42:41,440 --> 00:42:46,279
Speaker 6: complete one hundred percent optimists can sometimes forget that we

763
00:42:46,360 --> 00:42:50,319
Speaker 6: can't know everything that's going to happen, and we can

764
00:42:50,360 --> 00:42:53,279
Speaker 6: have voices saying, oh, what about this might happen? Be

765
00:42:53,320 --> 00:42:55,439
Speaker 6: careful of this and put the guardrails in around here.

766
00:42:56,080 --> 00:42:58,319
Speaker 6: But there are also going to be places where we

767
00:42:58,440 --> 00:43:04,799
Speaker 6: don't necessarily see something's coming. And in terms of addressing loneliness,

768
00:43:05,400 --> 00:43:07,919
Speaker 6: I was a little bit alongside Frith where she called

769
00:43:07,920 --> 00:43:12,160
Speaker 6: it this dystopian idea where the solution to loneliness is

770
00:43:13,440 --> 00:43:14,520
Speaker 6: like digital robots.

771
00:43:14,960 --> 00:43:18,719
Speaker 3: It felt a little bit an airing to me.

772
00:43:20,239 --> 00:43:26,319
Speaker 6: In particular because well, I think when Stephan said, you know,

773
00:43:26,360 --> 00:43:29,560
Speaker 6: pointed to kind of Japan as being a country that

774
00:43:29,719 --> 00:43:34,840
Speaker 6: was more likely to engage with chatbots. He pointed to

775
00:43:34,920 --> 00:43:38,960
Speaker 6: one of the most long lasting loneliness epidemic countries that

776
00:43:39,000 --> 00:43:42,840
Speaker 6: we've known about for a really long time. It's actually,

777
00:43:43,000 --> 00:43:46,279
Speaker 6: you know, a real been a serious issue to the

778
00:43:46,280 --> 00:43:50,120
Speaker 6: point where it's almost memified the loneliness issue in Japan

779
00:43:50,200 --> 00:43:52,120
Speaker 6: over the last kind of decades. So I thought that

780
00:43:52,160 --> 00:43:56,600
Speaker 6: was an interesting pull as an example of a country

781
00:43:56,640 --> 00:43:59,720
Speaker 6: that's more engaged with AI for these issues.

782
00:43:59,800 --> 00:44:01,719
Speaker 3: So, yeah, I.

783
00:44:03,320 --> 00:44:03,520
Speaker 1: Will.

784
00:44:03,560 --> 00:44:07,000
Speaker 6: I'm excited for the idea of a productivity buddy who

785
00:44:07,040 --> 00:44:11,200
Speaker 6: can help me with work and life admin, I'm not

786
00:44:11,239 --> 00:44:13,520
Speaker 6: sure I want them to be my mate personally.

787
00:44:13,960 --> 00:44:17,000
Speaker 5: Yeah, I've been thinking a lot about this particular issue.

788
00:44:17,000 --> 00:44:22,319
Speaker 5: I've got a friend who very tragically is wife took

789
00:44:22,360 --> 00:44:26,200
Speaker 5: her own life this year, and she was a digital native,

790
00:44:26,480 --> 00:44:30,560
Speaker 5: so from it here, from the very earliest stages, she

791
00:44:30,840 --> 00:44:35,239
Speaker 5: was creating content, living her life online and kept all

792
00:44:35,280 --> 00:44:37,920
Speaker 5: of that. And he now has this entire archive, and

793
00:44:38,520 --> 00:44:42,200
Speaker 5: he's a tech guy, and he's he's actually considering using

794
00:44:42,239 --> 00:44:45,400
Speaker 5: a large language model to create a model of his

795
00:44:45,480 --> 00:44:48,720
Speaker 5: wife and feed in all of this, all of this data,

796
00:44:49,080 --> 00:44:51,919
Speaker 5: and for him it's away, you know, he's going through

797
00:44:51,960 --> 00:44:54,520
Speaker 5: He's incredibly lonely at the moment, so to be able

798
00:44:54,520 --> 00:44:56,920
Speaker 5: to hear her voice. He's got all these recorded episodes

799
00:44:56,960 --> 00:44:59,520
Speaker 5: she's done of podcasts and interviews and that sort of thing,

800
00:45:00,360 --> 00:45:03,000
Speaker 5: so you can sort of see it is quite compelling.

801
00:45:03,080 --> 00:45:06,399
Speaker 5: But I think philosophically, where we're at with AI at

802
00:45:06,400 --> 00:45:08,759
Speaker 5: the moment is that we've got all these tools at

803
00:45:08,760 --> 00:45:12,600
Speaker 5: the moment, AI tools, but the industry wants to move

804
00:45:12,640 --> 00:45:17,320
Speaker 5: towards artificial general intelligence opener. AI has said that is

805
00:45:17,360 --> 00:45:20,520
Speaker 5: its goal is to do that in a responsible way

806
00:45:20,560 --> 00:45:23,360
Speaker 5: or a way that is positive for humanity. And you

807
00:45:23,400 --> 00:45:26,000
Speaker 5: can argue about how quickly they're moving towards that, but

808
00:45:26,840 --> 00:45:29,080
Speaker 5: I feel like we've got a few more years of

809
00:45:29,360 --> 00:45:34,759
Speaker 5: just making really good tools, so boosting productivity and efficiency,

810
00:45:34,800 --> 00:45:40,960
Speaker 5: creating videos very quickly, using text to video generators, all

811
00:45:41,000 --> 00:45:43,000
Speaker 5: of that sort of stuff, before we race on to

812
00:45:43,040 --> 00:45:46,040
Speaker 5: the things that are trying to really mimic a human being,

813
00:45:46,160 --> 00:45:50,520
Speaker 5: like a Herd type scenario, which does have potential downsides.

814
00:45:50,800 --> 00:45:53,279
Speaker 5: Let's get these tools right and lay the foundations. And

815
00:45:53,280 --> 00:45:56,040
Speaker 5: that's what I liked about Dave's approach. He's not being

816
00:45:56,680 --> 00:46:01,000
Speaker 5: captured by the technology. He's focused on turn on investment,

817
00:46:01,680 --> 00:46:04,439
Speaker 5: he's actually telling companies, let us know how many people

818
00:46:04,520 --> 00:46:07,279
Speaker 5: you employ, the sorts of tasks you do, feed it

819
00:46:07,320 --> 00:46:12,080
Speaker 5: into our ROI estimator, and see how much money you

820
00:46:12,120 --> 00:46:15,400
Speaker 5: will likely save if you employ these types of tools.

821
00:46:15,719 --> 00:46:17,960
Speaker 5: That's what we really need at the moment is actually

822
00:46:18,000 --> 00:46:21,920
Speaker 5: prove the case that this can be a productivity driver.

823
00:46:22,080 --> 00:46:25,640
Speaker 5: That's what we've been promised for years now, So hopefully

824
00:46:25,640 --> 00:46:27,600
Speaker 5: twenty twenty five is a year where we actually see

825
00:46:27,640 --> 00:46:28,520
Speaker 5: that come to fruition.

826
00:46:29,360 --> 00:46:31,680
Speaker 6: One hundred percent agree with you, and it's a conversation

827
00:46:31,719 --> 00:46:35,600
Speaker 6: I was having with Business Desks AI expert and residence

828
00:46:35,640 --> 00:46:41,360
Speaker 6: Andy Freya yesterday where we were talking about this next,

829
00:46:41,400 --> 00:46:46,000
Speaker 6: this current and kind of twenty twenty five wave of

830
00:46:46,160 --> 00:46:51,560
Speaker 6: AI productivity tools. We see it as being really internally

831
00:46:51,600 --> 00:46:56,840
Speaker 6: focused for organizations creating stuff that is for employees to

832
00:46:57,719 --> 00:47:01,840
Speaker 6: kind of give them that boost for or productivity because

833
00:47:01,880 --> 00:47:05,720
Speaker 6: that's extremely low risk. So I think that's a really

834
00:47:07,040 --> 00:47:09,600
Speaker 6: really important learning curve that.

835
00:47:09,560 --> 00:47:13,279
Speaker 3: We're on right now. Around these generative AI tools and.

836
00:47:13,320 --> 00:47:17,560
Speaker 6: Deploy them internally, be really clear about where things go wrong.

837
00:47:17,760 --> 00:47:20,319
Speaker 6: If the employee manages to do something stupid or get

838
00:47:20,320 --> 00:47:22,960
Speaker 6: it to do something stupid, don't just go, oh, that's okay.

839
00:47:23,000 --> 00:47:23,840
Speaker 3: It was just the.

840
00:47:23,640 --> 00:47:26,719
Speaker 6: Silly employee, go a silly employee. You're not supposed to

841
00:47:26,719 --> 00:47:28,359
Speaker 6: do that. What can we learn about this? How can

842
00:47:28,400 --> 00:47:29,840
Speaker 6: we fix it? How can we make sure that we

843
00:47:29,880 --> 00:47:33,160
Speaker 6: implement things so that if we do go to external stakeholders,

844
00:47:33,360 --> 00:47:36,840
Speaker 6: it's not going to cause a serious issue. So that,

845
00:47:37,239 --> 00:47:39,919
Speaker 6: to me, I think is going to define twenty twenty five.

846
00:47:39,920 --> 00:47:44,840
Speaker 6: In GENAI is internal tools for internal use and gathering

847
00:47:44,880 --> 00:47:47,840
Speaker 6: a lot of learning and education, and it's something that

848
00:47:47,880 --> 00:47:50,960
Speaker 6: I think we shouldn't be encouraging New Zealand companies to

849
00:47:51,040 --> 00:47:54,040
Speaker 6: do as well, because they know as long as we're

850
00:47:54,080 --> 00:47:58,480
Speaker 6: focusing on that ROI that productivity gains the actual realization

851
00:47:58,719 --> 00:47:59,680
Speaker 6: of improvements.

852
00:48:00,120 --> 00:48:04,200
Speaker 5: Yeah, one of the areas where we will see improvement,

853
00:48:05,040 --> 00:48:09,520
Speaker 5: and Stefan touched on it was around generative agents autonomous

854
00:48:09,560 --> 00:48:13,320
Speaker 5: agents acting on your behalf to do tasks. That increases

855
00:48:13,360 --> 00:48:17,840
Speaker 5: the risk somewhat because they could make a mistake, especially

856
00:48:17,880 --> 00:48:20,600
Speaker 5: when agents are talking to each other on the behalf

857
00:48:20,640 --> 00:48:24,600
Speaker 5: of a customer and a company for instance. But that's

858
00:48:24,640 --> 00:48:29,160
Speaker 5: the obvious next step and we've seen them released by Salesforce,

859
00:48:29,200 --> 00:48:31,919
Speaker 5: Microsoft and others this year. So I think Generative AI

860
00:48:32,040 --> 00:48:34,480
Speaker 5: is going to be a big one next gen voice assistance.

861
00:48:34,480 --> 00:48:37,360
Speaker 5: You've already seen it with the updated Seri and the

862
00:48:37,600 --> 00:48:42,640
Speaker 5: iPhone Alexa will get upgraded with generative AI in twenty

863
00:48:42,680 --> 00:48:45,640
Speaker 5: twenty five. So is that actually a good experience? Will

864
00:48:45,640 --> 00:48:47,960
Speaker 5: that because that business has lost a lot of money,

865
00:48:48,360 --> 00:48:51,200
Speaker 5: will it turn it around and actually make it a

866
00:48:51,239 --> 00:48:54,280
Speaker 5: profit center for some of these companies. That's another big one,

867
00:48:54,680 --> 00:48:58,560
Speaker 5: the AI sustainability challenge. The fact that and we've been

868
00:48:58,600 --> 00:49:01,160
Speaker 5: talking a lot about this this year, where the tech

869
00:49:01,560 --> 00:49:04,640
Speaker 5: industry is really interested in nuclear because they realize they're

870
00:49:04,680 --> 00:49:08,640
Speaker 5: running out of power to run these big data centers

871
00:49:08,640 --> 00:49:12,520
Speaker 5: for AI. But also where we see AI start to

872
00:49:13,000 --> 00:49:16,520
Speaker 5: have a positive impact on sustainability in terms of in

873
00:49:16,640 --> 00:49:20,439
Speaker 5: various industries allowing things to be done more efficiently. That

874
00:49:20,440 --> 00:49:22,560
Speaker 5: that's going to be a big one. And in the

875
00:49:22,760 --> 00:49:26,000
Speaker 5: you know, the generative video, we just saw some glimpses

876
00:49:26,040 --> 00:49:28,520
Speaker 5: of what's possible with that this year, which you know,

877
00:49:28,560 --> 00:49:33,120
Speaker 5: Open Ai, Sora and other things like that, incredible results.

878
00:49:33,560 --> 00:49:35,560
Speaker 5: Will we get to the point next year where we

879
00:49:35,640 --> 00:49:39,359
Speaker 5: have longer videos that are quicker to create and are

880
00:49:39,400 --> 00:49:42,680
Speaker 5: actually starting to go into advertising and the film and

881
00:49:42,760 --> 00:49:45,719
Speaker 5: TV space that's potentially the path it's on.

882
00:49:46,200 --> 00:49:49,920
Speaker 6: Yeah, And I mean we saw the first jen Ai

883
00:49:50,520 --> 00:49:54,640
Speaker 6: created TV advert this year, and obviously there was a

884
00:49:54,680 --> 00:49:57,719
Speaker 6: lot of finagling and manipulating to make it how they

885
00:49:57,719 --> 00:50:00,760
Speaker 6: wanted it to. But that will add with the the strange,

886
00:50:00,960 --> 00:50:05,720
Speaker 6: very obviously AI generated wool cuddling up to a generative woman,

887
00:50:06,040 --> 00:50:10,239
Speaker 6: which was, you know, an interesting advert. Also was a

888
00:50:10,520 --> 00:50:14,520
Speaker 6: kind of a glimpse of the potential future for general

889
00:50:14,520 --> 00:50:16,279
Speaker 6: you know, the way that we're using and we've got

890
00:50:16,320 --> 00:50:21,080
Speaker 6: Skinny as well now talking about using a deep fake.

891
00:50:21,120 --> 00:50:23,640
Speaker 6: They're going to pay an actor or a model or

892
00:50:23,680 --> 00:50:28,000
Speaker 6: even somebody off the street to use their likeness as

893
00:50:28,080 --> 00:50:32,960
Speaker 6: a spokesperson for two years and what is that going

894
00:50:33,000 --> 00:50:33,319
Speaker 6: to mean?

895
00:50:33,680 --> 00:50:37,319
Speaker 3: What are they going to actually make this AI model.

896
00:50:37,120 --> 00:50:40,360
Speaker 6: Do and how how are they going to you know,

897
00:50:40,480 --> 00:50:41,920
Speaker 6: balance that and will it.

898
00:50:41,920 --> 00:50:45,040
Speaker 5: Be worth a lifetime of free Skinny subscription.

899
00:50:45,680 --> 00:50:48,400
Speaker 6: There's also there's also a fee involved, so there was

900
00:50:48,440 --> 00:50:52,560
Speaker 6: also a Yeah, they also are paying what was considered

901
00:50:53,560 --> 00:50:55,760
Speaker 6: I should have a story coming out about it, surely,

902
00:50:57,160 --> 00:51:00,799
Speaker 6: but yeah, fascinating, fascinating time, and I think the end

903
00:51:00,840 --> 00:51:02,920
Speaker 6: of twenty twenty four in terms of what's happening the

904
00:51:02,960 --> 00:51:07,560
Speaker 6: generative AI in New Zealand is a super interesting indicator

905
00:51:07,640 --> 00:51:09,640
Speaker 6: for what's coming down the line next year.

906
00:51:10,200 --> 00:51:10,600
Speaker 1: Yeah.

907
00:51:10,760 --> 00:51:13,920
Speaker 5: On the positive side, I think we avoided any major

908
00:51:13,960 --> 00:51:16,960
Speaker 5: disasters with AI this year. You know, in terms of

909
00:51:17,040 --> 00:51:22,320
Speaker 5: election interference, you know major you know, deep fake scandal

910
00:51:22,400 --> 00:51:23,160
Speaker 5: or something like that.

911
00:51:23,280 --> 00:51:23,600
Speaker 3: We had.

912
00:51:24,239 --> 00:51:26,239
Speaker 5: We had real people, We had the Joe Rogan's, we

913
00:51:26,280 --> 00:51:31,520
Speaker 5: had the Elon Musks being influential doing crazy things. There

914
00:51:31,600 --> 00:51:36,880
Speaker 5: was things that like Trump's assassination attempt on them, you know,

915
00:51:36,920 --> 00:51:39,960
Speaker 5: that was a game changer in terms of the trajectory

916
00:51:39,960 --> 00:51:44,360
Speaker 5: of the election campaign. It didn't take some random Russian

917
00:51:44,560 --> 00:51:48,839
Speaker 5: inspired deep fake campaign to do that, So so that's good.

918
00:51:48,880 --> 00:51:52,000
Speaker 5: I think people still have the head screwed on. They

919
00:51:52,200 --> 00:51:55,480
Speaker 5: are skeptical naturally off a lot of the stuff that

920
00:51:55,560 --> 00:51:58,680
Speaker 5: the content itself still looks AI generated. So I think

921
00:51:58,800 --> 00:52:02,200
Speaker 5: every human it just doesn't compute with with our brains.

922
00:52:02,200 --> 00:52:03,080
Speaker 1: We just reject it.

923
00:52:03,160 --> 00:52:05,160
Speaker 5: But you know that could change in the next year.

924
00:52:05,680 --> 00:52:10,120
Speaker 6: Yeah, I think it will change eventually. And I think

925
00:52:10,120 --> 00:52:12,120
Speaker 6: what we're going to see as a move from these

926
00:52:12,239 --> 00:52:17,880
Speaker 6: really crisp, clean, glossy productions from generative AI, and we're

927
00:52:17,880 --> 00:52:20,520
Speaker 6: going to start to see them like trying to layer

928
00:52:20,560 --> 00:52:24,600
Speaker 6: that natural chaoticness over them, to muddy them, to kind

929
00:52:24,640 --> 00:52:27,440
Speaker 6: of make them seem more natural. That's what we're going

930
00:52:27,440 --> 00:52:29,080
Speaker 6: to start to see next, and that's when we're going

931
00:52:29,080 --> 00:52:30,799
Speaker 6: to start to really be in trouble, I think. But

932
00:52:31,760 --> 00:52:34,840
Speaker 6: in terms of the sustainability as well, I think that's

933
00:52:35,160 --> 00:52:38,880
Speaker 6: something that is going to become more and more talked about.

934
00:52:38,960 --> 00:52:43,840
Speaker 6: And you know, ideally there will be some market forces

935
00:52:43,880 --> 00:52:47,799
Speaker 6: involved where if it gets if the power consumption gets

936
00:52:47,840 --> 00:52:51,680
Speaker 6: too expensive, then it's going to restrict what you can

937
00:52:51,719 --> 00:52:55,279
Speaker 6: actually do with the tools. Ideally, the only concern is

938
00:52:55,280 --> 00:52:59,040
Speaker 6: if we have these companies really you know, letting the

939
00:53:00,280 --> 00:53:04,080
Speaker 6: a IB a huge loss leader in order to a

940
00:53:04,120 --> 00:53:06,759
Speaker 6: lot for long term returns. You know, so that's lock

941
00:53:06,800 --> 00:53:09,239
Speaker 6: people in at low prices and then eventually raise the

942
00:53:09,280 --> 00:53:12,880
Speaker 6: prices drastically when we realize that that's what's needed.

943
00:53:13,239 --> 00:53:15,839
Speaker 3: Yeah, So that that is a strategy we've seen before

944
00:53:15,840 --> 00:53:16,720
Speaker 3: in the tech industry.

945
00:53:16,960 --> 00:53:20,840
Speaker 5: Interesting has thoughts on open source AI not seeing uptake

946
00:53:20,920 --> 00:53:23,520
Speaker 5: of that because it's ultimately more expensive to do something

947
00:53:24,040 --> 00:53:26,320
Speaker 5: with that. You might have more control to do stuff,

948
00:53:26,800 --> 00:53:30,320
Speaker 5: but it's that commoditization for new Zealand companies in particularly

949
00:53:30,400 --> 00:53:33,279
Speaker 5: being able to buy a service in the cloud, be

950
00:53:33,320 --> 00:53:37,400
Speaker 5: able to do all your AI workloads there for a

951
00:53:37,440 --> 00:53:42,279
Speaker 5: reliably understandable fee on a monthly basis. That's what sort

952
00:53:42,280 --> 00:53:44,000
Speaker 5: of our companies are looking for at the moment.

953
00:53:44,320 --> 00:53:49,960
Speaker 6: Yeah, definitely, and that may eventually change as models, the

954
00:53:50,000 --> 00:53:53,600
Speaker 6: efficiency of models changes, And we spoke to Kiarri and

955
00:53:53,680 --> 00:53:55,880
Speaker 6: Dagua earlier this year and that was one of the

956
00:53:55,920 --> 00:53:58,360
Speaker 6: things that he's really been working on as part of

957
00:53:58,400 --> 00:54:02,640
Speaker 6: a cohort international on is making these models more efficient,

958
00:54:03,320 --> 00:54:06,080
Speaker 6: making them use consume less compute.

959
00:54:06,800 --> 00:54:09,879
Speaker 3: So if we do see start to see.

960
00:54:09,600 --> 00:54:13,960
Speaker 6: Like a Moore's law in terms of reducing the amount

961
00:54:14,040 --> 00:54:17,200
Speaker 6: of compute that is needed for any particular action towards

962
00:54:17,280 --> 00:54:22,600
Speaker 6: generative AI, that may eventually turn the table for open

963
00:54:22,640 --> 00:54:26,960
Speaker 6: source in a way that Stefan was kind of suggesting.

964
00:54:28,200 --> 00:54:32,480
Speaker 6: And also as we see like we talk, I think

965
00:54:32,520 --> 00:54:35,160
Speaker 6: did Stefan mention red hat I think where he there

966
00:54:35,320 --> 00:54:38,600
Speaker 6: as a company that kind of commoditized open source, we

967
00:54:38,719 --> 00:54:40,920
Speaker 6: might start to see that kind of thing happening as well.

968
00:54:41,000 --> 00:54:44,840
Speaker 6: So yeah, that shift will will eventually happen, but I

969
00:54:44,840 --> 00:54:47,960
Speaker 6: think definitely in the short term it's the big subscription

970
00:54:48,000 --> 00:54:49,920
Speaker 6: models that are going to dominate.

971
00:54:50,280 --> 00:54:53,440
Speaker 5: Yeah, So some great discussion there, and we'll keep our

972
00:54:53,480 --> 00:54:55,800
Speaker 5: focus on AI and twenty twenty five. We've had a

973
00:54:55,880 --> 00:54:58,400
Speaker 5: lot of guests on a lot of themes we've explored,

974
00:54:58,440 --> 00:55:01,080
Speaker 5: but it is the technology of the moment, so we

975
00:55:01,120 --> 00:55:04,279
Speaker 5: will keep a close eye on it. So thanks to

976
00:55:04,680 --> 00:55:08,280
Speaker 5: Frith Tweety, Dave Howden and Stephan Korn for that look

977
00:55:08,320 --> 00:55:10,239
Speaker 5: back at the year in AI.

978
00:55:10,600 --> 00:55:11,680
Speaker 3: Yes, thank you very much.

979
00:55:12,280 --> 00:55:14,080
Speaker 6: Show notes for the Business of Tech are in the

980
00:55:14,120 --> 00:55:17,439
Speaker 6: podcast section at Business Desk dot co dot Nz, where

981
00:55:17,480 --> 00:55:20,239
Speaker 6: you can stream this podcast in full every week. It's

982
00:55:20,280 --> 00:55:24,360
Speaker 6: also available from iHeartRadio or on your podcast platform of choice.

983
00:55:24,520 --> 00:55:26,680
Speaker 5: Get in touch with your feedback and we'd love to

984
00:55:26,760 --> 00:55:30,799
Speaker 5: hear your suggestions for upcoming guests. To email Ben benat

985
00:55:30,840 --> 00:55:32,560
Speaker 5: Business Desk dot co dot Nz.

986
00:55:33,160 --> 00:55:36,400
Speaker 6: You can find Peter and I both on x and LinkedIn,

987
00:55:36,880 --> 00:55:39,480
Speaker 6: and you can also follow the Business of Tech LinkedIn

988
00:55:39,560 --> 00:55:43,000
Speaker 6: page and you'll get all our updates for new episodes there,

989
00:55:43,440 --> 00:55:45,480
Speaker 6: as well as other content that we produce.

990
00:55:46,239 --> 00:55:48,640
Speaker 5: That's it for this week. Another episode coming your way

991
00:55:48,719 --> 00:55:49,480
Speaker 5: next Thursday.

992
00:55:49,800 --> 00:55:51,560
Speaker 3: Until then, have a great week.