1
00:00:00,200 --> 00:00:03,040
Speaker 1: This is my great honor to welcome you to the

2
00:00:03,120 --> 00:00:08,760
Speaker 1: World Economic Firms Annual Meeting in twenty twenty five.

3
00:00:09,520 --> 00:00:14,800
Speaker 2: The cooperative world order we imagined twenty five years ago

4
00:00:15,600 --> 00:00:20,200
Speaker 2: has not turned into reality. Instead, we have entered a

5
00:00:20,239 --> 00:00:23,320
Speaker 2: new era of harsh geostrategic competition.

6
00:00:24,000 --> 00:00:28,600
Speaker 3: America is back at open for business under the Trump administration.

7
00:00:28,760 --> 00:00:31,520
Speaker 3: There will be no better place owner to create jobs,

8
00:00:31,560 --> 00:00:34,640
Speaker 3: bill factories, or grow a company than right here in

9
00:00:34,680 --> 00:00:35,800
Speaker 3: the good old USA.

10
00:00:36,360 --> 00:00:39,160
Speaker 2: If we have tit for tat retaliation, we are going

11
00:00:39,200 --> 00:00:42,680
Speaker 2: to see double digit global GDP losses.

12
00:00:43,479 --> 00:00:47,200
Speaker 4: This is the time of new technologies.

13
00:00:46,479 --> 00:00:51,600
Speaker 1: That Davos with a rich and powerful meat every year,

14
00:00:51,800 --> 00:00:54,320
Speaker 1: and this year they faced the dawn of a new

15
00:00:54,560 --> 00:00:58,440
Speaker 1: US presidency and an America First agenda that could reshape

16
00:00:58,600 --> 00:01:00,680
Speaker 1: the geopolitical landscape.

17
00:01:00,480 --> 00:01:03,560
Speaker 5: At Davos as a keynote speaker was Northland based Peter

18
00:01:03,720 --> 00:01:07,480
Speaker 5: Lucas Jones, the CEO of the Heku Media, a TV

19
00:01:07,600 --> 00:01:11,479
Speaker 5: and radio station network that's been undertaking some cutting edge

20
00:01:11,560 --> 00:01:14,240
Speaker 5: work on today AI language models.

21
00:01:14,480 --> 00:01:17,560
Speaker 6: There are many people that have wanted to teach computers

22
00:01:17,560 --> 00:01:20,360
Speaker 6: how to speak Maudi. There are many people, but we're

23
00:01:20,360 --> 00:01:22,399
Speaker 6: the people that did it. We're the people that did

24
00:01:22,440 --> 00:01:25,640
Speaker 6: it because we not only speak the language, we are

25
00:01:25,680 --> 00:01:29,480
Speaker 6: gathering corpus every day. We are tagging and labeling that

26
00:01:29,600 --> 00:01:33,480
Speaker 6: phonetical data every day, and the phonetical data that we

27
00:01:33,680 --> 00:01:37,399
Speaker 6: have is so closely related to other Pacific languages, the

28
00:01:37,520 --> 00:01:42,080
Speaker 6: model that we have developed can be reinterpreted into the

29
00:01:42,120 --> 00:01:44,000
Speaker 6: context of sister languages.

30
00:01:44,200 --> 00:01:46,720
Speaker 1: This week on the Business of Tech powered by two

31
00:01:46,720 --> 00:01:50,240
Speaker 1: Degrees Business, AI's place in the world and a fresh

32
00:01:50,320 --> 00:01:53,720
Speaker 1: perspective from one of Time Magazine's top one hundred leaders

33
00:01:53,760 --> 00:01:57,120
Speaker 1: in AI on our Taioha's edge when it comes to

34
00:01:57,280 --> 00:01:59,160
Speaker 1: harnessing AI for good.

35
00:01:59,080 --> 00:02:01,920
Speaker 5: And coloring the disc of AI at Davos, which is

36
00:02:01,960 --> 00:02:05,120
Speaker 5: a major theme, was the rising unease at least among

37
00:02:05,240 --> 00:02:09,320
Speaker 5: US tech leaders attending the forum about China's surprising advances

38
00:02:09,360 --> 00:02:10,880
Speaker 5: in AI model performance.

39
00:02:11,040 --> 00:02:14,920
Speaker 1: That unease sent tech stocks, including in video plunging this

40
00:02:14,960 --> 00:02:18,680
Speaker 1: week a total of a trillion dollars wiped off them. Initially,

41
00:02:18,680 --> 00:02:20,600
Speaker 1: that's come back, but it was on the news that

42
00:02:20,680 --> 00:02:23,639
Speaker 1: Deepseek has made a model to rival that of Open

43
00:02:23,680 --> 00:02:26,880
Speaker 1: Aiyes and metas despite it having limited access to high

44
00:02:26,880 --> 00:02:30,799
Speaker 1: performance computer chips due to US trade embargos.

45
00:02:30,360 --> 00:02:32,560
Speaker 5: YEP, It's thrown a huge spanner in the works in

46
00:02:32,600 --> 00:02:36,480
Speaker 5: the AI world globally. We's been so much discussion about it,

47
00:02:36,880 --> 00:02:38,520
Speaker 5: and we're going to have a little chat about it.

48
00:02:38,560 --> 00:02:41,480
Speaker 4: But just before we do, excuse my voice. I obviously

49
00:02:41,560 --> 00:02:42,880
Speaker 4: have a bit of a cold.

50
00:02:42,800 --> 00:02:44,360
Speaker 7: Not a high performance chip today.

51
00:02:44,720 --> 00:02:47,440
Speaker 5: No, no, so I might do a little less talking

52
00:02:47,480 --> 00:02:48,040
Speaker 5: than normal.

53
00:02:48,160 --> 00:02:51,280
Speaker 4: But well, let's actually save my voice a little bit more.

54
00:02:51,320 --> 00:02:55,480
Speaker 5: We'll throw it to an explanation from a very impressive YouTuber,

55
00:02:55,880 --> 00:02:59,800
Speaker 5: Dave's Garage about what exactly this deep seek model is.

56
00:03:00,600 --> 00:03:03,320
Speaker 8: Just as the launch of Sputank challenge assumptions about American

57
00:03:03,360 --> 00:03:07,200
Speaker 8: technological dominance in the twentieth century, deepsekr one is forcing

58
00:03:07,200 --> 00:03:09,799
Speaker 8: a reckoning in the twenty first. Not only does deepsekr

59
00:03:09,880 --> 00:03:12,280
Speaker 8: one meet or exceed the performance of the best American

60
00:03:12,320 --> 00:03:15,280
Speaker 8: AI models like open aish one, they did it on

61
00:03:15,320 --> 00:03:18,600
Speaker 8: the cheap, reportedly for under six million dollars. Because not

62
00:03:18,639 --> 00:03:20,440
Speaker 8: only is China claim to have done it cheaply, but

63
00:03:20,480 --> 00:03:22,920
Speaker 8: they reportedly did it without access to the latest of

64
00:03:23,000 --> 00:03:26,400
Speaker 8: nvidious chips. And just what is deep skr one. It's

65
00:03:26,440 --> 00:03:29,240
Speaker 8: a new language model designed to offer performance that punches

66
00:03:29,240 --> 00:03:32,160
Speaker 8: above its weight, trained on a smaller scale. But still

67
00:03:32,200 --> 00:03:36,280
Speaker 8: capable of answering questions, generating text, and understanding context. And

68
00:03:36,320 --> 00:03:38,640
Speaker 8: what sets it apart isn't just the capabilities, but the

69
00:03:38,680 --> 00:03:41,480
Speaker 8: way that it's been built. Deep Seek is designed to

70
00:03:41,520 --> 00:03:46,240
Speaker 8: be cheap, efficient and surprisingly resourceful, leveraging larger foundational ais

71
00:03:46,280 --> 00:03:49,680
Speaker 8: like open AI's GPT four or metaslama is scaffolding to

72
00:03:49,760 --> 00:03:53,640
Speaker 8: create something much larger. It's not perfect, it's not trying

73
00:03:53,680 --> 00:03:55,680
Speaker 8: to be, but it's a fascinating glimpse into what the

74
00:03:55,680 --> 00:03:58,160
Speaker 8: future of AI might look like. Lightweight, efficient and a

75
00:03:58,160 --> 00:04:00,000
Speaker 8: little rough around the edges, but full of potential.

76
00:04:00,040 --> 00:04:04,280
Speaker 1: A lot of researchers have been benchmarking this. This isn't

77
00:04:04,400 --> 00:04:08,760
Speaker 1: made up. What is unclear is the extent to which

78
00:04:09,080 --> 00:04:11,960
Speaker 1: the Chinese did or did not have access to high

79
00:04:12,000 --> 00:04:15,960
Speaker 1: performance chips. They claim that they were not using the

80
00:04:16,000 --> 00:04:19,560
Speaker 1: most powerful in video chips, which technically they're not allowed

81
00:04:19,560 --> 00:04:23,080
Speaker 1: to have. That it costs six million dollars. This is

82
00:04:23,160 --> 00:04:26,839
Speaker 1: the R one model that they've created, which is a tiny,

83
00:04:26,880 --> 00:04:31,000
Speaker 1: tiny fraction of what open ai, for instance, has spent

84
00:04:31,080 --> 00:04:35,719
Speaker 1: on its one model. So if that is true, that

85
00:04:35,880 --> 00:04:38,159
Speaker 1: is an absolute game changer. And look, my take on

86
00:04:38,200 --> 00:04:42,960
Speaker 1: this is this is exciting. This is incredibly positive for

87
00:04:43,000 --> 00:04:44,479
Speaker 1: the world. You know, we've been on this sort of

88
00:04:44,600 --> 00:04:49,640
Speaker 1: doom loop off more and more big spann. Only the

89
00:04:49,720 --> 00:04:53,560
Speaker 1: most powerful and well resourced in the world can join

90
00:04:53,600 --> 00:04:56,279
Speaker 1: this race. We talked last week about the five hundred

91
00:04:56,279 --> 00:05:01,480
Speaker 1: billion of investment that Trump announced with open and Oracle.

92
00:05:01,960 --> 00:05:04,559
Speaker 1: Trump this week, you know, surprisingly said this is great

93
00:05:04,839 --> 00:05:08,560
Speaker 1: because it's too expensive, and this allows us all to

94
00:05:08,680 --> 00:05:12,240
Speaker 1: do it cheaper. We can't forget that Open AI and

95
00:05:12,480 --> 00:05:16,760
Speaker 1: all those American companies still have superior technology. They're still

96
00:05:16,839 --> 00:05:20,240
Speaker 1: using superior chips. They will adopt this technology and make

97
00:05:20,279 --> 00:05:22,839
Speaker 1: their models better as well. But what it means is

98
00:05:22,839 --> 00:05:26,320
Speaker 1: that it's just lowers the threshold for what you can

99
00:05:26,360 --> 00:05:31,080
Speaker 1: do with really powerful large language models. That opens the

100
00:05:31,120 --> 00:05:35,120
Speaker 1: door to smaller company startups and countries like New Zealand.

101
00:05:35,520 --> 00:05:38,120
Speaker 5: Yeah, I mean, you've pretty much covered all of the

102
00:05:38,160 --> 00:05:40,440
Speaker 5: major shifts that we're going to see in terms of cost,

103
00:05:40,520 --> 00:05:43,839
Speaker 5: in terms of capability, in terms of expectation around where

104
00:05:43,880 --> 00:05:46,520
Speaker 5: the American companies go next. Just to give a little

105
00:05:46,520 --> 00:05:49,600
Speaker 5: sense of the difference that we're talking here in terms

106
00:05:49,640 --> 00:05:52,960
Speaker 5: of what is made available by this R one model,

107
00:05:53,000 --> 00:05:56,600
Speaker 5: I did a little experiment where I got chat gbts

108
00:05:57,600 --> 00:06:00,320
Speaker 5: free tier to create a logic puzzle form one of

109
00:06:00,320 --> 00:06:04,280
Speaker 5: those very simple logic puzzles, you know, list of names,

110
00:06:04,440 --> 00:06:06,960
Speaker 5: list of awards, who gets which award here, has a

111
00:06:06,960 --> 00:06:07,760
Speaker 5: list of clues.

112
00:06:07,839 --> 00:06:09,240
Speaker 4: You make it grit and you take it off.

113
00:06:09,480 --> 00:06:11,880
Speaker 5: And I did it, and of course, because it was

114
00:06:12,200 --> 00:06:14,920
Speaker 5: chat GPT's free model, it was unsolvable.

115
00:06:15,160 --> 00:06:16,239
Speaker 4: It didn't make any sense.

116
00:06:16,800 --> 00:06:20,000
Speaker 5: And so I then went around and I tossed that

117
00:06:20,480 --> 00:06:26,400
Speaker 5: puzzle back into chat GPT's for omni just to see

118
00:06:26,440 --> 00:06:28,680
Speaker 5: what it would come up with. And it's confidently said

119
00:06:28,720 --> 00:06:31,520
Speaker 5: these are the answers, and I went, obviously not.

120
00:06:31,560 --> 00:06:34,360
Speaker 4: Because it's actually there's a logical error in the puzzle.

121
00:06:34,480 --> 00:06:38,039
Speaker 5: So interesting confidence there, gem and I gave me the

122
00:06:38,080 --> 00:06:42,360
Speaker 5: same thing. Co pilot, Claude and Mistral all confidently came

123
00:06:42,360 --> 00:06:45,560
Speaker 5: out with their answers. I then threw it into chat

124
00:06:45,600 --> 00:06:49,560
Speaker 5: GPT's one model, the high Reasoning model. It took about

125
00:06:49,839 --> 00:06:52,760
Speaker 5: two and a half minutes of thinking and then finally

126
00:06:52,839 --> 00:06:56,600
Speaker 5: came back and said this is actually unsolvable. So that

127
00:06:56,720 --> 00:06:59,279
Speaker 5: was impressive, you know, it realized that it was unsolvable

128
00:07:00,000 --> 00:07:03,280
Speaker 5: through it into deep seek and it took maybe forty

129
00:07:03,320 --> 00:07:06,520
Speaker 5: seconds and I watched it reason through and reason through

130
00:07:06,880 --> 00:07:09,360
Speaker 5: and it came to the same conclusion. This is unsolvable.

131
00:07:09,400 --> 00:07:12,720
Speaker 5: There is an error in the actual logic puzzle differences.

132
00:07:13,040 --> 00:07:16,640
Speaker 5: That's the twenty dollars a month high level reasoning one

133
00:07:16,680 --> 00:07:19,160
Speaker 5: that we were talking about from Chatchiput, and that was

134
00:07:19,200 --> 00:07:24,320
Speaker 5: the completely free r one, just totally accessible deep Seek model.

135
00:07:24,640 --> 00:07:26,800
Speaker 5: So that's the level that we're getting from the deep

136
00:07:26,800 --> 00:07:30,800
Speaker 5: Seek model of reasoning capability for free. That's really going

137
00:07:30,840 --> 00:07:33,880
Speaker 5: to start to put pressure on those American companies in

138
00:07:34,000 --> 00:07:36,640
Speaker 5: terms of what these bots are expected to do.

139
00:07:37,920 --> 00:07:40,480
Speaker 1: Yeah, it's probably we're talking about. You know, if it's

140
00:07:40,520 --> 00:07:43,000
Speaker 1: so good, why would they open source it? You know?

141
00:07:43,080 --> 00:07:49,000
Speaker 1: And the reason really you open source something like Facebook

142
00:07:49,240 --> 00:07:54,000
Speaker 1: Meta has done with Lamma its large language model. They're

143
00:07:54,040 --> 00:07:57,160
Speaker 1: basically looking at the industry and saying, you know, open Ai, Microsoft,

144
00:07:57,200 --> 00:07:58,840
Speaker 1: do we really want to go head to head with

145
00:07:58,920 --> 00:08:01,640
Speaker 1: Anthropic and all these other companies, Let's open source it

146
00:08:01,720 --> 00:08:04,920
Speaker 1: because the real value will come elsewhere for us on

147
00:08:05,400 --> 00:08:08,840
Speaker 1: our social network platform and you'll see all these innovative

148
00:08:08,920 --> 00:08:12,560
Speaker 1: uses that will allow us to make money on our

149
00:08:12,600 --> 00:08:15,880
Speaker 1: platforms in different ways. We're not trying to sell subscriptions

150
00:08:15,880 --> 00:08:19,120
Speaker 1: to Lamma. So what is the motivation for deep Seek.

151
00:08:19,120 --> 00:08:23,800
Speaker 1: I haven't seen anything from the CEO really outlining his philosophy.

152
00:08:24,560 --> 00:08:27,880
Speaker 1: You know, he's a hedge fund trader, quantitative hedge fund

153
00:08:28,240 --> 00:08:31,000
Speaker 1: trading guy, so he is obviously interested in money. So

154
00:08:31,600 --> 00:08:34,120
Speaker 1: is the money going to be made elsewhere? Is deep

155
00:08:34,160 --> 00:08:38,760
Speaker 1: seek going to power Ali Baba and ten cents e

156
00:08:38,800 --> 00:08:42,520
Speaker 1: commerce engines and social media chatbots? And that is at

157
00:08:42,520 --> 00:08:46,880
Speaker 1: the angle, and he'll license the technology, but then you know,

158
00:08:46,920 --> 00:08:48,960
Speaker 1: if it's open source, everyone can pick it up anyway.

159
00:08:49,440 --> 00:08:53,160
Speaker 5: Yeah. Absolutely, And I think, you know, we'll start to

160
00:08:54,520 --> 00:08:57,079
Speaker 5: get a sense of that in the coming months as

161
00:08:57,080 --> 00:09:00,720
Speaker 5: we as we understand how the model play out. I

162
00:09:00,720 --> 00:09:05,720
Speaker 5: think they also already do have an API built into

163
00:09:05,720 --> 00:09:08,120
Speaker 5: deep Seek for people who just want something easy that

164
00:09:08,160 --> 00:09:11,080
Speaker 5: they can link into rather than having to take the

165
00:09:11,080 --> 00:09:13,640
Speaker 5: open source model and run it themselves. If you just

166
00:09:14,000 --> 00:09:17,160
Speaker 5: want to tap into a cheaper API that has similar

167
00:09:17,960 --> 00:09:21,760
Speaker 5: capabilities as open ai, but for much cheaper, it is

168
00:09:21,800 --> 00:09:24,000
Speaker 5: an option now, so there is a paid model there

169
00:09:24,000 --> 00:09:25,120
Speaker 5: in terms of API calls.

170
00:09:25,280 --> 00:09:28,000
Speaker 4: So it's very curious about how that's going to roll out.

171
00:09:28,080 --> 00:09:30,480
Speaker 4: Maybe he just wanted to prove that he could do it.

172
00:09:30,960 --> 00:09:31,439
Speaker 7: Or is that?

173
00:09:31,760 --> 00:09:35,360
Speaker 1: And I think you can't discount the political elements of

174
00:09:35,720 --> 00:09:38,760
Speaker 1: this as well. You know, it's interesting that they announce

175
00:09:38,840 --> 00:09:42,760
Speaker 1: Deep Sea Version three on Christmas Day and then on

176
00:09:42,840 --> 00:09:46,960
Speaker 1: Trump's inauguration are one the reasoning model. I think they're

177
00:09:46,960 --> 00:09:49,760
Speaker 1: making a point there that you know, the US doesn't

178
00:09:49,880 --> 00:09:55,319
Speaker 1: control AI in the world, that the Chinese have made

179
00:09:55,400 --> 00:09:58,400
Speaker 1: huge advances, and it could, I don't know speculating, it

180
00:09:58,440 --> 00:10:02,240
Speaker 1: could be literally that the politicians have said, wow, let's

181
00:10:02,360 --> 00:10:05,520
Speaker 1: use this as a tool and let's go go big

182
00:10:05,559 --> 00:10:10,040
Speaker 1: on this to make a political point. But regardless, I

183
00:10:10,040 --> 00:10:12,640
Speaker 1: think it's a real net positive for the world. And

184
00:10:12,679 --> 00:10:14,840
Speaker 1: it sort of goes into the themes of what we're

185
00:10:14,880 --> 00:10:18,000
Speaker 1: talking to Peter Lucas about, who's just come back from

186
00:10:18,200 --> 00:10:23,040
Speaker 1: from Davos, which is, you know, there's this massive AI

187
00:10:23,240 --> 00:10:25,120
Speaker 1: arms race going on, but there's actually a lot of

188
00:10:25,120 --> 00:10:28,880
Speaker 1: scope to do other things and other players to produce

189
00:10:29,120 --> 00:10:33,760
Speaker 1: really useful things like Tehiku Media is doing, and this

190
00:10:33,840 --> 00:10:35,560
Speaker 1: gives us the tools to do it. You know, we're

191
00:10:35,640 --> 00:10:40,480
Speaker 1: just about to start building a new public research organization

192
00:10:40,559 --> 00:10:43,520
Speaker 1: in New Zealand which is going to be focused on AI,

193
00:10:44,200 --> 00:10:48,840
Speaker 1: on quantum computing, on synthetic biology. AI is actually integral

194
00:10:48,880 --> 00:10:51,400
Speaker 1: to all of those three things. And I was thinking

195
00:10:51,480 --> 00:10:53,840
Speaker 1: last week when they announced it, I mean, what capability

196
00:10:53,920 --> 00:10:57,120
Speaker 1: do we have in AI? What infrastructure, what resource do

197
00:10:57,200 --> 00:11:01,559
Speaker 1: we have? And the answer is very little. But this

198
00:11:01,640 --> 00:11:05,120
Speaker 1: breakthrough suggests that you need very little to actually get

199
00:11:05,120 --> 00:11:08,920
Speaker 1: going on AI. And there will be other open source models.

200
00:11:08,920 --> 00:11:12,200
Speaker 1: So it's really changing the paradigm from having the big,

201
00:11:12,520 --> 00:11:16,199
Speaker 1: large language models that cost a lot to develop but

202
00:11:16,240 --> 00:11:18,080
Speaker 1: are very powerful, but you have to spend a lot

203
00:11:18,120 --> 00:11:21,600
Speaker 1: of money to access them, to potentially having dozens and

204
00:11:21,679 --> 00:11:26,559
Speaker 1: dozens off smaller, more efficient models that we can use

205
00:11:26,600 --> 00:11:30,839
Speaker 1: for specific domains and specific tasks. And that's potentially where

206
00:11:31,559 --> 00:11:34,199
Speaker 1: our edge lies in AI. We can't take on the giants,

207
00:11:34,240 --> 00:11:37,520
Speaker 1: but we can use these really efficient models to actually

208
00:11:37,520 --> 00:11:40,320
Speaker 1: come up with things that answer some of the problems

209
00:11:40,360 --> 00:11:41,559
Speaker 1: that we face as a country.

210
00:11:42,120 --> 00:11:45,959
Speaker 5: Yeah, and I would maybe encourage people at this point

211
00:11:46,360 --> 00:11:48,840
Speaker 5: to go back to the episode we did in March

212
00:11:49,120 --> 00:11:54,040
Speaker 5: last year with Kii and Dagua all about the value

213
00:11:54,040 --> 00:11:57,240
Speaker 5: of sovereign AI and how we could actually build our

214
00:11:57,240 --> 00:11:59,480
Speaker 5: own or host our own AA models and train our

215
00:11:59,520 --> 00:12:01,920
Speaker 5: own AM models within New Zealand, and he actually talks

216
00:12:01,920 --> 00:12:05,200
Speaker 5: about Teco Media and their work as well in that interview.

217
00:12:05,320 --> 00:12:11,359
Speaker 1: Yeah, I guess you know the potential downsides around safety

218
00:12:11,640 --> 00:12:14,920
Speaker 1: and censorship and that you know some people, you know,

219
00:12:14,920 --> 00:12:18,200
Speaker 1: Paula Penfold had a play on the web version of

220
00:12:18,320 --> 00:12:20,520
Speaker 1: deep Seek, and I think there's some censorship going on,

221
00:12:20,600 --> 00:12:24,600
Speaker 1: which is not surprising given that it's a Chinese run

222
00:12:24,679 --> 00:12:27,800
Speaker 1: service on the On the web, pretty much everything is

223
00:12:27,840 --> 00:12:32,760
Speaker 1: censored there. That doesn't mean that you can't use deep

224
00:12:32,800 --> 00:12:37,360
Speaker 1: Seek setting it up yourself because it's open source and

225
00:12:37,440 --> 00:12:41,640
Speaker 1: avoid that censorship, but to what extent is censorship built

226
00:12:41,679 --> 00:12:45,400
Speaker 1: into the model itself. I think the delicious irony here

227
00:12:45,520 --> 00:12:49,120
Speaker 1: is that it seems as though they've basically used open

228
00:12:49,200 --> 00:12:52,280
Speaker 1: ai and all the other large language models really as

229
00:12:52,280 --> 00:12:55,800
Speaker 1: a scaffold to build deep Seek on top off. So

230
00:12:55,840 --> 00:12:58,520
Speaker 1: they've basically taken all the data and the answers from

231
00:12:59,160 --> 00:13:03,200
Speaker 1: those other model to inform their own model. And I

232
00:13:03,240 --> 00:13:06,560
Speaker 1: haven't seen Sam Altman directly attack them on that, because

233
00:13:06,600 --> 00:13:10,199
Speaker 1: he can't because he's done the exact same thing, harvesting

234
00:13:10,240 --> 00:13:14,080
Speaker 1: the entire Internet to train his own model. So you know,

235
00:13:14,480 --> 00:13:16,959
Speaker 1: I think he tweeted yesterday basically saying credit to them

236
00:13:16,960 --> 00:13:19,760
Speaker 1: this is legit, this is very good, and this is

237
00:13:19,800 --> 00:13:23,760
Speaker 1: going to encourage us to do better. But he must

238
00:13:23,800 --> 00:13:26,480
Speaker 1: be sort of thinking they've basically just stolen it, They've

239
00:13:26,480 --> 00:13:27,480
Speaker 1: copied what we've done.

240
00:13:28,080 --> 00:13:28,320
Speaker 2: Yeah.

241
00:13:28,440 --> 00:13:31,680
Speaker 5: I mean, look, if there is a pattern in consumer

242
00:13:31,720 --> 00:13:35,400
Speaker 5: products in history, it's that the Chinese make a cheaper

243
00:13:35,520 --> 00:13:39,839
Speaker 5: version of it and sell it en mass as you

244
00:13:39,880 --> 00:13:42,760
Speaker 5: can see, you know, with the explosion of Timuuinali expresses.

245
00:13:43,280 --> 00:13:47,400
Speaker 5: They take a great pattern that's working and they make

246
00:13:47,440 --> 00:13:50,440
Speaker 5: a copy of it that's legally distinct, or maybe not

247
00:13:50,520 --> 00:13:54,280
Speaker 5: so much in some cases, and then they sell it

248
00:13:54,320 --> 00:13:56,520
Speaker 5: to the world and maybe it's not as good, but

249
00:13:56,760 --> 00:14:00,319
Speaker 5: sometimes maybe it is. And yeah, I think I think

250
00:14:00,600 --> 00:14:03,640
Speaker 5: the pressure that deep seek will put on the American

251
00:14:03,640 --> 00:14:07,200
Speaker 5: companies will be good. And also having a bit of

252
00:14:07,400 --> 00:14:12,079
Speaker 5: really good competition for pricing, I think is important because you.

253
00:14:12,040 --> 00:14:14,480
Speaker 4: Know, look at the cost of Microsoft's co pilot.

254
00:14:14,520 --> 00:14:16,319
Speaker 7: It's just, yeah, it's ridiculous.

255
00:14:17,200 --> 00:14:18,719
Speaker 1: You know, there must be a lot of analysts and

256
00:14:18,760 --> 00:14:22,800
Speaker 1: people who value companies must just be going, oh my god,

257
00:14:23,200 --> 00:14:28,840
Speaker 1: how do we revalue you know, this industry. You've got billions,

258
00:14:29,000 --> 00:14:31,840
Speaker 1: hundreds of billions of dollars that a small number of

259
00:14:31,880 --> 00:14:37,560
Speaker 1: American companies are putting into this on the premise that

260
00:14:37,920 --> 00:14:42,840
Speaker 1: they need ever more capacity and larger language models to

261
00:14:42,880 --> 00:14:46,720
Speaker 1: stay ahead and to get to artificial general intelligence, which

262
00:14:46,760 --> 00:14:50,520
Speaker 1: is Sam Altman's stated goal. And then, you know, so

263
00:14:50,600 --> 00:14:53,360
Speaker 1: a lot of shrewd financial people will be going, well,

264
00:14:53,440 --> 00:14:57,080
Speaker 1: do we actually need all of that to create profitable

265
00:14:57,240 --> 00:15:00,880
Speaker 1: business models? Maybe we don't. Therefore, you know, in Nvidio

266
00:15:01,040 --> 00:15:03,280
Speaker 1: it's still down about ten percent as we record this.

267
00:15:03,800 --> 00:15:05,960
Speaker 1: Maybe it's never coming back to its highs. Maybe it's

268
00:15:06,000 --> 00:15:07,920
Speaker 1: actually got a lot more to lose, as well as

269
00:15:07,960 --> 00:15:11,760
Speaker 1: anthropic Microsoft and others. So I think that's going to

270
00:15:11,760 --> 00:15:15,080
Speaker 1: be the really interesting thing how it changes the value

271
00:15:15,080 --> 00:15:19,360
Speaker 1: proposition of AI the business model underpending things like co pilot,

272
00:15:19,480 --> 00:15:23,000
Speaker 1: where you know, we've talked to the CEO of New

273
00:15:23,080 --> 00:15:26,800
Speaker 1: Zealand who's like, this is really dying good value At

274
00:15:26,840 --> 00:15:28,520
Speaker 1: thirty dollars a month, you know it's going to make

275
00:15:28,560 --> 00:15:30,800
Speaker 1: you more efficient. Well, if it can make us more

276
00:15:30,800 --> 00:15:33,880
Speaker 1: efficient on deepseek for five dollars a month, you're going

277
00:15:33,920 --> 00:15:34,320
Speaker 1: to go with that.

278
00:15:34,960 --> 00:15:36,760
Speaker 4: Absolutely. Yeah.

279
00:15:36,800 --> 00:15:39,600
Speaker 5: And so that tension between the US and China isn't

280
00:15:39,680 --> 00:15:42,560
Speaker 5: likely to die down, but it doesn't have to define

281
00:15:42,680 --> 00:15:44,920
Speaker 5: how AI is developed or used.

282
00:15:45,280 --> 00:15:49,440
Speaker 1: Peter Lucas Jones has overseen globally recognized work at Kaiitia

283
00:15:49,520 --> 00:15:53,520
Speaker 1: based Tahiku Media. They've developed a Terreyo Marii speech to

284
00:15:53,600 --> 00:15:58,720
Speaker 1: text AI model that's ninety two percent accurate, way better

285
00:15:58,800 --> 00:16:00,760
Speaker 1: than anything that's been achieve before.

286
00:16:01,160 --> 00:16:03,960
Speaker 5: They are now using that capability they've developed to assist

287
00:16:04,000 --> 00:16:07,840
Speaker 5: other indigenous communities to preserve the integrity of their languages

288
00:16:07,880 --> 00:16:08,800
Speaker 5: in the AI era.

289
00:16:09,200 --> 00:16:12,720
Speaker 1: Peter Lucas was among indigenous leaders who spoke at Davos,

290
00:16:12,800 --> 00:16:14,680
Speaker 1: the only key we on the speaking bill as far

291
00:16:14,720 --> 00:16:17,960
Speaker 1: as we could tell. He gave a refreshing perspective on

292
00:16:18,000 --> 00:16:20,960
Speaker 1: the role technology like AI can play both for the

293
00:16:21,000 --> 00:16:22,680
Speaker 1: world and for New Zealand.

294
00:16:23,040 --> 00:16:26,520
Speaker 5: Here's Peter's interview with the Hiku Media's Peter Lucas Jones,

295
00:16:26,880 --> 00:16:29,760
Speaker 5: speaking from Geneva.

296
00:16:34,440 --> 00:16:38,120
Speaker 1: Peter Lucas Cuia, Welcome to the business of tech. You're

297
00:16:38,120 --> 00:16:41,640
Speaker 1: coming to us from Geneva, where you've spent the last

298
00:16:41,640 --> 00:16:45,760
Speaker 1: week or so at Davos, the World Economic Forums, big

299
00:16:46,240 --> 00:16:49,320
Speaker 1: gathering of the world's global elites.

300
00:16:49,680 --> 00:16:52,560
Speaker 7: Was that your first time at Davos.

301
00:16:52,200 --> 00:16:55,480
Speaker 6: Yeah, it was my first time at Davos, the World

302
00:16:55,520 --> 00:16:58,960
Speaker 6: Economic Forum. I mean, being invited to speak at Davos

303
00:16:59,000 --> 00:17:03,680
Speaker 6: in Switzerland has been a true honor. Meeting Vice President

304
00:17:03,720 --> 00:17:06,600
Speaker 6: Al Gore and of course the chairman of the Forum,

305
00:17:06,640 --> 00:17:11,360
Speaker 6: Clause Schwab, were certainly highlights. I was very different from

306
00:17:11,359 --> 00:17:14,040
Speaker 6: most of the people that are invited to speak at

307
00:17:14,040 --> 00:17:17,159
Speaker 6: the Forum because I live in Kaitaya and I spend

308
00:17:17,200 --> 00:17:22,280
Speaker 6: most of my time developing Maldi language speech technology and

309
00:17:22,320 --> 00:17:27,159
Speaker 6: creating ways to tell Mildi stories through content creation and distribution.

310
00:17:28,400 --> 00:17:32,399
Speaker 6: It's been an amazing time and so much as the

311
00:17:32,440 --> 00:17:40,439
Speaker 6: world has has some massive changes happening right now, and

312
00:17:40,480 --> 00:17:45,520
Speaker 6: the age of intelligence and our transition from the digital

313
00:17:45,600 --> 00:17:49,080
Speaker 6: age is one of the things that's being spoken about

314
00:17:49,119 --> 00:17:52,879
Speaker 6: in terms of economic growth here at the forum.

315
00:17:54,119 --> 00:17:57,000
Speaker 1: Yeah, and we're going to get to what Tahku Media

316
00:17:57,080 --> 00:17:59,240
Speaker 1: is doing some fantastic stuff, So we'll get a bit

317
00:17:59,240 --> 00:18:01,680
Speaker 1: of an update on some of the things that you're

318
00:18:01,720 --> 00:18:04,840
Speaker 1: working on. Give us a sense of what the vibe

319
00:18:04,840 --> 00:18:07,440
Speaker 1: really was like at Davos this year. I mean, it's

320
00:18:07,480 --> 00:18:13,080
Speaker 1: the intersection of geopolitics, trade technology. It's always fascinating the themes,

321
00:18:13,880 --> 00:18:17,440
Speaker 1: particularly how it sets the business agenda, what is trending,

322
00:18:17,480 --> 00:18:21,600
Speaker 1: what is topical in business? But this year Davos happened

323
00:18:21,680 --> 00:18:25,880
Speaker 1: right in the middle of President Trump's inauguration. I guess

324
00:18:25,920 --> 00:18:28,680
Speaker 1: a lot of people were with one eye keeping close

325
00:18:28,760 --> 00:18:31,840
Speaker 1: tabs on what Trump was saying and then analyzing at

326
00:18:31,920 --> 00:18:35,399
Speaker 1: Davos what this meant for the world literally as he

327
00:18:35,440 --> 00:18:39,639
Speaker 1: starts to implement his policies and rolls out his executive orders.

328
00:18:39,920 --> 00:18:43,359
Speaker 6: Yeah, most certainly. But I think that the highlight in

329
00:18:43,480 --> 00:18:47,400
Speaker 6: terms of that dialogue was when he addressed the forum directly,

330
00:18:48,119 --> 00:18:51,760
Speaker 6: and so he was really quite clear around what his

331
00:18:52,040 --> 00:18:56,560
Speaker 6: focus and priorities are over this year and of course

332
00:18:57,040 --> 00:19:01,760
Speaker 6: the next few days, weeks, months, and theng of his term.

333
00:19:01,960 --> 00:19:05,720
Speaker 6: One of the things being spoken about was his target

334
00:19:05,840 --> 00:19:10,480
Speaker 6: on oil and other resources being a priority in terms

335
00:19:10,480 --> 00:19:16,000
Speaker 6: of mining, and that, of course, balanced with the conversation

336
00:19:16,119 --> 00:19:21,359
Speaker 6: around climatic change and environmental issues and the intersection of

337
00:19:22,840 --> 00:19:28,360
Speaker 6: economic growth and the need to urgently address environmental issues

338
00:19:28,480 --> 00:19:33,040
Speaker 6: was certainly a conversation point. Highlighted within all of that

339
00:19:33,200 --> 00:19:39,600
Speaker 6: was certainly the development of artificial intelligence, data governance, and

340
00:19:39,720 --> 00:19:43,680
Speaker 6: of course things like the digital divide, which, of course,

341
00:19:43,720 --> 00:19:46,040
Speaker 6: in the age of intelligence, will be a divide of

342
00:19:46,119 --> 00:19:49,600
Speaker 6: dignity when we think about the one billion people that

343
00:19:49,800 --> 00:19:55,080
Speaker 6: have access to compute and the seven that don't. When

344
00:19:55,119 --> 00:19:58,359
Speaker 6: we think about that, that's one to advance and seven

345
00:19:58,400 --> 00:20:02,080
Speaker 6: to be left behind. People thought that that just wasn't

346
00:20:02,119 --> 00:20:07,119
Speaker 6: good enough, and so it's raised to people's attention the

347
00:20:07,200 --> 00:20:13,280
Speaker 6: need for greater collaboration, greater collaboration across countries, across industries.

348
00:20:13,880 --> 00:20:22,679
Speaker 6: And the President Trump certainly invited people to participate in

349
00:20:22,880 --> 00:20:27,159
Speaker 6: growing their business in the United States and that they

350
00:20:27,160 --> 00:20:32,240
Speaker 6: would experience the very reduced tax and they would not

351
00:20:32,480 --> 00:20:36,399
Speaker 6: be paying tariffs. But if they chose not to do

352
00:20:36,520 --> 00:20:41,040
Speaker 6: their business or advance their industry in the United States,

353
00:20:41,080 --> 00:20:46,520
Speaker 6: then they would certainly be participating and contributing to the

354
00:20:46,560 --> 00:20:50,720
Speaker 6: coffers of the United States, and of course paying down

355
00:20:50,760 --> 00:20:51,280
Speaker 6: their debt.

356
00:20:51,600 --> 00:20:54,240
Speaker 1: Yeah, it was very clear from his speech and his

357
00:20:54,320 --> 00:20:58,040
Speaker 1: subsequent sort of speeches that it's going to be America first.

358
00:20:58,160 --> 00:21:01,680
Speaker 1: So if it's good for America, sure we'll partner with

359
00:21:01,720 --> 00:21:04,919
Speaker 1: you and we'll trade with you, but very much a

360
00:21:05,040 --> 00:21:08,240
Speaker 1: shift towards it's got to be seen through the lens

361
00:21:08,240 --> 00:21:11,480
Speaker 1: of how does this advance America's interest, whether it be

362
00:21:11,600 --> 00:21:16,720
Speaker 1: trade or national security. So that's interesting. You talked about

363
00:21:17,119 --> 00:21:21,000
Speaker 1: artificial intelligence there reading up the World Economic Forums basically

364
00:21:21,000 --> 00:21:24,000
Speaker 1: saying that the big focus on AI this year is

365
00:21:24,359 --> 00:21:29,280
Speaker 1: moving really from the investing in the fundamental capability, particularly

366
00:21:29,320 --> 00:21:33,520
Speaker 1: around generative AI, to the application of AI. What was

367
00:21:33,800 --> 00:21:38,040
Speaker 1: the sense you got about where big businesses that were

368
00:21:38,040 --> 00:21:42,600
Speaker 1: present at Davos really are placed when it comes to AI.

369
00:21:42,800 --> 00:21:45,760
Speaker 1: So much money has gone into this area in the

370
00:21:45,840 --> 00:21:48,560
Speaker 1: last couple of years, and even the week you were there,

371
00:21:49,240 --> 00:21:54,119
Speaker 1: five hundred billion dollar investment between soft Bank Oracle Open AI.

372
00:21:54,200 --> 00:21:57,600
Speaker 1: Trump very proudly in the White House announced that do

373
00:21:57,680 --> 00:22:01,400
Speaker 1: you see this huge amount of investment just continuing as

374
00:22:01,480 --> 00:22:06,000
Speaker 1: this need for a centric data centers just really becomes

375
00:22:06,080 --> 00:22:06,800
Speaker 1: quite intense.

376
00:22:07,080 --> 00:22:11,040
Speaker 6: I think it's really important to understand the difference between

377
00:22:11,200 --> 00:22:15,919
Speaker 6: use of AI and development of AI. And use of AI,

378
00:22:16,119 --> 00:22:19,399
Speaker 6: of course might enhance the way that you do your work,

379
00:22:19,720 --> 00:22:23,400
Speaker 6: or create new jobs or reduce the need for some jobs,

380
00:22:23,720 --> 00:22:26,239
Speaker 6: but the reality is that the real money is going

381
00:22:26,320 --> 00:22:29,760
Speaker 6: to be made and developing AI. And in New Zealand

382
00:22:29,880 --> 00:22:33,159
Speaker 6: we've yet to really take hold of that opportunity and

383
00:22:33,280 --> 00:22:37,520
Speaker 6: understand its true meaning. And I think that means we

384
00:22:37,640 --> 00:22:42,439
Speaker 6: need to understand the opportunity to bring services in house

385
00:22:43,080 --> 00:22:46,280
Speaker 6: and not just be the users of technology, but the

386
00:22:46,359 --> 00:22:50,159
Speaker 6: creators of technology. We need to carve out a place

387
00:22:50,200 --> 00:22:54,480
Speaker 6: in the world that isn't just based on primary industries.

388
00:22:55,040 --> 00:22:59,240
Speaker 6: And a big part of Donald Trump's speech was about

389
00:22:59,320 --> 00:23:04,359
Speaker 6: the development of data centers. Now, the reality is data

390
00:23:04,400 --> 00:23:09,440
Speaker 6: centers consume a lot of energy, and when we think

391
00:23:09,480 --> 00:23:13,480
Speaker 6: about where we are in New Zealand, I think it's

392
00:23:13,520 --> 00:23:18,000
Speaker 6: important that we understand that AI and data science advances

393
00:23:18,920 --> 00:23:23,800
Speaker 6: these technologies, they're going to demand more compute power, demand

394
00:23:23,880 --> 00:23:29,159
Speaker 6: more energy, demand more water for cooling, and balancing economic

395
00:23:29,200 --> 00:23:32,000
Speaker 6: progress and the fight against climate change. Of course, it's

396
00:23:32,040 --> 00:23:37,520
Speaker 6: not just necessary insurgent, but we need to also understand

397
00:23:37,520 --> 00:23:41,720
Speaker 6: the opportunity there to grow our economy not just through

398
00:23:41,960 --> 00:23:45,480
Speaker 6: using the technology, but developing it. And I think that's

399
00:23:45,520 --> 00:23:49,800
Speaker 6: where the true partnerships can evolve for us as New Zealanders.

400
00:23:51,040 --> 00:23:54,439
Speaker 6: And it's a special opportunity for us to think about

401
00:23:54,440 --> 00:23:58,240
Speaker 6: how do we be part of the group of people

402
00:23:58,280 --> 00:24:01,080
Speaker 6: in the world that are going to be advancing economic

403
00:24:01,160 --> 00:24:06,680
Speaker 6: growth through the age of intelligence. Because as we move

404
00:24:06,720 --> 00:24:10,520
Speaker 6: into the age of intelligence and new divides start to emerge,

405
00:24:10,920 --> 00:24:14,960
Speaker 6: without action, many people will be left behind. And that's

406
00:24:15,000 --> 00:24:18,240
Speaker 6: why we need a governance model that takes into account

407
00:24:18,240 --> 00:24:22,720
Speaker 6: the need to be inclusive, innovative, and impactful. And so

408
00:24:22,760 --> 00:24:25,639
Speaker 6: when we think about the need to address things like

409
00:24:25,760 --> 00:24:30,359
Speaker 6: future jobs and the skills our people as New Zealanders

410
00:24:30,400 --> 00:24:35,399
Speaker 6: need to get them and require to get them, it

411
00:24:35,440 --> 00:24:38,760
Speaker 6: requires us not only to be optimistic but also realistic

412
00:24:38,800 --> 00:24:42,760
Speaker 6: as well, because innovation is the key driver for long

413
00:24:42,840 --> 00:24:47,040
Speaker 6: term economic growth and the writings on the Wall, artificial

414
00:24:47,119 --> 00:24:51,359
Speaker 6: intelligence is an area of economic growth that we cannot

415
00:24:51,400 --> 00:24:56,000
Speaker 6: only bring something unique to. It's an opportunity for us

416
00:24:56,040 --> 00:25:00,440
Speaker 6: to grow our capabilities. So when we think about language models,

417
00:25:00,440 --> 00:25:04,320
Speaker 6: now you referred to large language models or you know,

418
00:25:04,400 --> 00:25:08,240
Speaker 6: but there are also small, large, small models, and whether

419
00:25:08,280 --> 00:25:11,080
Speaker 6: they're large or small, they can still provide benefit to

420
00:25:11,280 --> 00:25:15,280
Speaker 6: decision makers that are discovering new and meaningful ways to

421
00:25:15,560 --> 00:25:20,760
Speaker 6: solve problems. And so Mildi language is more than a

422
00:25:20,800 --> 00:25:27,080
Speaker 6: tool of communication. It's what was being described in conversations

423
00:25:27,560 --> 00:25:32,640
Speaker 6: as an intangible asset that holds traditional knowledge, wisdom, and identity.

424
00:25:33,359 --> 00:25:36,560
Speaker 6: But when it comes to data sovereignty, the conversation often

425
00:25:36,600 --> 00:25:40,320
Speaker 6: stops at things like intellectual property rights. And as we

426
00:25:40,480 --> 00:25:45,840
Speaker 6: broaden the dialogue to include the emerging language economy of

427
00:25:45,960 --> 00:25:51,159
Speaker 6: indigenous languages, and I mentioned this because people are looking. Certainly,

428
00:25:51,240 --> 00:25:55,840
Speaker 6: the conversation at Davos was focused on looking to indigenous

429
00:25:55,880 --> 00:26:02,439
Speaker 6: and traditional knowledge to bring something unique to the climate conversation,

430
00:26:04,200 --> 00:26:08,480
Speaker 6: the conversation around innovation in medicines, and of course the

431
00:26:08,520 --> 00:26:14,200
Speaker 6: preservation of biodiversity and our unique knowledge about the ocean

432
00:26:14,280 --> 00:26:16,280
Speaker 6: when we think about the place of New Zealand and

433
00:26:16,320 --> 00:26:20,320
Speaker 6: the Pacific. So those were some of the things that

434
00:26:21,119 --> 00:26:23,640
Speaker 6: I was speaking about. And I had the unique privilege

435
00:26:23,680 --> 00:26:28,040
Speaker 6: of sharing the story of Tehiku media ground, our groundbreaking

436
00:26:28,160 --> 00:26:32,359
Speaker 6: work and developing speech technology for Tedi or Miudi, and

437
00:26:32,520 --> 00:26:38,000
Speaker 6: advocating for technology that can scale and support indigenous language

438
00:26:38,000 --> 00:26:42,440
Speaker 6: systems across the Pacific and of course benefit indigenous languages

439
00:26:42,480 --> 00:26:47,880
Speaker 6: throughout the world. So I think that's why we need

440
00:26:47,920 --> 00:26:50,919
Speaker 6: to be part of these types of conversations, particularly when

441
00:26:50,960 --> 00:26:55,080
Speaker 6: we're talking about economic growth, because there is an opportunity

442
00:26:55,160 --> 00:26:58,480
Speaker 6: for us to even have smaller scale data centers that

443
00:26:58,560 --> 00:27:03,560
Speaker 6: are focused on our uniqueness in New Zealand and our

444
00:27:03,600 --> 00:27:07,800
Speaker 6: special offering, our point of difference, our value proposition that

445
00:27:07,880 --> 00:27:09,760
Speaker 6: we can provide to the world.

446
00:27:10,040 --> 00:27:12,960
Speaker 1: It really resonates with me what you're saying that we

447
00:27:12,960 --> 00:27:15,920
Speaker 1: don't necessarily have to rely on that infrastructure. Having our

448
00:27:15,960 --> 00:27:20,200
Speaker 1: own infrastructure is really important. When you were featured in

449
00:27:20,480 --> 00:27:24,440
Speaker 1: the Time one hundred Top Leaders in AI last year,

450
00:27:24,640 --> 00:27:28,280
Speaker 1: I was actually at the Salesforce conference flicking through Time

451
00:27:28,320 --> 00:27:31,840
Speaker 1: magazine and I saw the blurb about you in there.

452
00:27:31,840 --> 00:27:34,399
Speaker 1: I couldn't believe it. I thought, how great, there's a

453
00:27:34,440 --> 00:27:36,679
Speaker 1: key we featured in the top one hundred here. But

454
00:27:36,800 --> 00:27:41,000
Speaker 1: what you said to Time Magazine again really resonated with me.

455
00:27:41,040 --> 00:27:43,639
Speaker 1: He said, if we do not have controlled governance and

456
00:27:43,760 --> 00:27:48,560
Speaker 1: ongoing guardianship of our data as indigenous people, we will

457
00:27:48,560 --> 00:27:52,639
Speaker 1: be landless in the digital world too. Really powerful, and

458
00:27:52,680 --> 00:27:55,640
Speaker 1: I think this goes even for New New Zealand. If

459
00:27:55,640 --> 00:27:58,440
Speaker 1: we just give all of our data way to train

460
00:27:58,560 --> 00:28:03,399
Speaker 1: large language models and sit on platforms that are owned

461
00:28:03,400 --> 00:28:07,280
Speaker 1: by offshore companies that have developed these models and are

462
00:28:07,320 --> 00:28:10,400
Speaker 1: extracting all the value, you know, what is the future

463
00:28:10,600 --> 00:28:13,160
Speaker 1: for us. We will give it all away. And that's

464
00:28:13,640 --> 00:28:16,920
Speaker 1: something that you know. Hopefully the likes of you, Tahiku

465
00:28:17,000 --> 00:28:19,720
Speaker 1: Media and others. We're not seeing so much of it yet,

466
00:28:19,800 --> 00:28:22,880
Speaker 1: but we really need to get that going, that innovation

467
00:28:22,960 --> 00:28:25,960
Speaker 1: that allows us to keep control of the data that

468
00:28:26,040 --> 00:28:29,159
Speaker 1: are so important to the AI world.

469
00:28:29,240 --> 00:28:32,040
Speaker 6: Yeah, well, there is no data science without the data.

470
00:28:32,720 --> 00:28:37,080
Speaker 6: And the reality is understanding and knowing your data really

471
00:28:37,160 --> 00:28:46,560
Speaker 6: well provides exceptional support to develop models that are not

472
00:28:46,600 --> 00:28:53,280
Speaker 6: only high quality, they're focused on precision and accuracy. And

473
00:28:53,320 --> 00:28:56,480
Speaker 6: those are the sorts of tools that people want to

474
00:28:56,520 --> 00:29:01,760
Speaker 6: rely on. And that reason is because throughout any process

475
00:29:01,840 --> 00:29:05,920
Speaker 6: of use, you're always building trust. So when we think

476
00:29:05,960 --> 00:29:11,640
Speaker 6: about how much of our GDP is actually focused on consumers,

477
00:29:12,320 --> 00:29:16,280
Speaker 6: consumers buying things, people want to buy things that they

478
00:29:16,360 --> 00:29:19,520
Speaker 6: can trust, and that's the point of difference that we're

479
00:29:19,560 --> 00:29:25,080
Speaker 6: focused on. And when we reimagine growth, we're constantly thinking

480
00:29:25,120 --> 00:29:29,320
Speaker 6: about investing in our people because part of the ability

481
00:29:29,440 --> 00:29:34,560
Speaker 6: to grow is focused on your ability to retain talent. Now,

482
00:29:34,560 --> 00:29:38,200
Speaker 6: when we think about artificial intelligence, when we think about

483
00:29:38,240 --> 00:29:43,120
Speaker 6: the need to grow talent, we must also retain talent.

484
00:29:43,560 --> 00:29:49,640
Speaker 6: Talent is one of our biggest resources. And pardon me,

485
00:29:50,440 --> 00:29:58,320
Speaker 6: I think about the proverb of meeting he tangua he

486
00:29:58,480 --> 00:30:01,520
Speaker 6: tangata tangata. What is the greatest thing in this world?

487
00:30:01,640 --> 00:30:04,440
Speaker 6: It is people. It is people. It is people. And

488
00:30:05,320 --> 00:30:09,920
Speaker 6: our team at Tecumedia is a team that's not only

489
00:30:10,160 --> 00:30:15,720
Speaker 6: focused on the age of intelligence, but it's also focused

490
00:30:15,760 --> 00:30:21,040
Speaker 6: on safeguarding our planet and how do we use artificial

491
00:30:21,160 --> 00:30:28,880
Speaker 6: intelligence also to reduce the strain on our natural resources.

492
00:30:29,440 --> 00:30:32,959
Speaker 6: We know that artificial intelligence has already been used by

493
00:30:33,080 --> 00:30:39,080
Speaker 6: mining companies to navigate and make decisions around where deposits

494
00:30:39,520 --> 00:30:44,640
Speaker 6: of minerals and oil and other resources might be. We

495
00:30:44,800 --> 00:30:48,640
Speaker 6: know that it's being used to fast track the innovation

496
00:30:48,840 --> 00:30:56,960
Speaker 6: around medicine. Likewise, technology that is developed based on traditional

497
00:30:57,040 --> 00:31:02,520
Speaker 6: knowledge can provide a way to answer questions around biodiversity,

498
00:31:02,960 --> 00:31:10,120
Speaker 6: to answer questions around climate response strategies, to answer questions

499
00:31:10,160 --> 00:31:16,800
Speaker 6: around how to provide a new philosophical point of view,

500
00:31:18,080 --> 00:31:23,600
Speaker 6: to inspire a change in the way that we develop

501
00:31:23,760 --> 00:31:28,280
Speaker 6: technologies and the way that we address things like divides

502
00:31:29,240 --> 00:31:32,840
Speaker 6: and those that will benefit and those that will not benefit.

503
00:31:33,400 --> 00:31:35,520
Speaker 1: When you were away the same week, all this was

504
00:31:35,600 --> 00:31:38,160
Speaker 1: going on a big announcement in New Zealand about the

505
00:31:38,240 --> 00:31:41,560
Speaker 1: future of the science sector here, the merging of some

506
00:31:41,720 --> 00:31:46,760
Speaker 1: of the Crown Research institutes, the disestablishment of Callahan Innovation,

507
00:31:47,320 --> 00:31:49,360
Speaker 1: so a bit of consternation around all of that, But

508
00:31:50,240 --> 00:31:52,200
Speaker 1: one of the bright spots and all of that was

509
00:31:52,280 --> 00:31:55,920
Speaker 1: the creation of this new public research organization yet to

510
00:31:56,040 --> 00:32:00,479
Speaker 1: be named, but it will focus on fundamental AI research,

511
00:32:00,640 --> 00:32:05,479
Speaker 1: quantum computing, synthetic biology, and that really caught my attention.

512
00:32:05,640 --> 00:32:07,720
Speaker 1: This has been a gap in our arsenal for a

513
00:32:07,800 --> 00:32:11,920
Speaker 1: long time to have a dedicated research focus on some

514
00:32:12,040 --> 00:32:14,480
Speaker 1: of the emerging technologies that are going to sort of

515
00:32:14,520 --> 00:32:17,560
Speaker 1: define the twenty first century. Interested in your take on that,

516
00:32:17,680 --> 00:32:21,200
Speaker 1: and you said, we do have a unique angle on AI,

517
00:32:21,280 --> 00:32:24,920
Speaker 1: which you've just articulated. It's focusing on those things like equity,

518
00:32:25,560 --> 00:32:29,160
Speaker 1: retaining data, sovereignty, and I think that that's going to

519
00:32:29,200 --> 00:32:33,920
Speaker 1: be increasingly important as people realize this is not an

520
00:32:34,000 --> 00:32:37,080
Speaker 1: equitable proposition when we give away all our data, sometimes

521
00:32:37,160 --> 00:32:40,680
Speaker 1: without our permission, and get sold services back to us

522
00:32:40,840 --> 00:32:45,600
Speaker 1: using our intellectual property. So I guess that as a

523
00:32:45,600 --> 00:32:48,600
Speaker 1: philosophy around how we do AI can infuse it. But

524
00:32:48,960 --> 00:32:51,520
Speaker 1: what are your thoughts about when it comes to our

525
00:32:51,640 --> 00:32:55,800
Speaker 1: unique selling point around developing AI that is cutting edge

526
00:32:55,800 --> 00:32:56,280
Speaker 1: and different.

527
00:32:56,400 --> 00:32:58,560
Speaker 7: What are some of the areas you think we have

528
00:32:58,640 --> 00:32:59,400
Speaker 7: a real edge.

529
00:33:00,080 --> 00:33:07,600
Speaker 6: Of our edges is our understanding of our intangible assets. Now,

530
00:33:07,760 --> 00:33:11,240
Speaker 6: data is often described as an intangible asset, but the

531
00:33:11,360 --> 00:33:16,080
Speaker 6: reality is is it has commercial value. Our data that

532
00:33:16,320 --> 00:33:20,200
Speaker 6: we hold regarding the ocean in the Pacific, the ocean

533
00:33:20,280 --> 00:33:26,080
Speaker 6: around New Zealand, our resources, our minerals, our whole biodiversity,

534
00:33:26,280 --> 00:33:31,440
Speaker 6: which is of course connected to our conversations like Y

535
00:33:31,560 --> 00:33:34,120
Speaker 6: two six to two, which is of course is a

536
00:33:34,320 --> 00:33:40,080
Speaker 6: white tribunal claim. But that sort of information, that intangible

537
00:33:40,800 --> 00:33:46,520
Speaker 6: asset conversation, is something we could probably learn to describe differently,

538
00:33:47,280 --> 00:33:50,959
Speaker 6: learn to measure differently, learn to value differently. And our

539
00:33:51,040 --> 00:33:54,040
Speaker 6: way to articulate that to the world is not just

540
00:33:54,160 --> 00:33:56,800
Speaker 6: important now, it's going to be important in the future.

541
00:33:57,240 --> 00:34:00,600
Speaker 6: And I raise that to our attention because as AI

542
00:34:00,760 --> 00:34:05,120
Speaker 6: and data science evolves, we can't run away from the

543
00:34:05,200 --> 00:34:09,719
Speaker 6: fact that they're driving innovation across all sectors. But this

544
00:34:09,960 --> 00:34:14,120
Speaker 6: progress demands, like you said, more compute power, more energy

545
00:34:14,360 --> 00:34:18,759
Speaker 6: and water for cooling, especially in data centers. But we

546
00:34:18,880 --> 00:34:24,000
Speaker 6: can't forget about the local economic benefits. And AI and

547
00:34:24,239 --> 00:34:27,440
Speaker 6: data centers are going to create jobs both in the

548
00:34:27,560 --> 00:34:32,360
Speaker 6: area of technology, whether that's engineering or data analysis or

549
00:34:33,480 --> 00:34:36,200
Speaker 6: other new jobs that we might not even know about.

550
00:34:36,680 --> 00:34:40,080
Speaker 6: Reskilling and upskilling of our people is going to be

551
00:34:40,320 --> 00:34:44,760
Speaker 6: a major priority. And if I was to predict something,

552
00:34:45,160 --> 00:34:47,920
Speaker 6: it would be the best skill that we could all

553
00:34:48,040 --> 00:34:52,920
Speaker 6: acquire now is the ability to learn, learn, relearn and

554
00:34:53,120 --> 00:34:59,520
Speaker 6: learn again, and that's going to contribute really meaningful impact.

555
00:34:59,640 --> 00:35:02,600
Speaker 6: I think when we start to talk about investment and

556
00:35:02,760 --> 00:35:08,680
Speaker 6: data infrastructure and how that can attract global companies, boosting

557
00:35:08,880 --> 00:35:14,560
Speaker 6: local economies by increasing not only tax revenue but employment

558
00:35:15,000 --> 00:35:20,120
Speaker 6: and regional development. And while we do that, we can't

559
00:35:20,160 --> 00:35:22,719
Speaker 6: run away from the fact that there's going to be

560
00:35:22,920 --> 00:35:27,440
Speaker 6: more demand for energy, which means higher carbon footprints, large

561
00:35:27,480 --> 00:35:31,840
Speaker 6: scale cooling uses and substantial water resources are going to

562
00:35:31,920 --> 00:35:37,880
Speaker 6: be required. These pose challenges, especially in areas around the

563
00:35:37,960 --> 00:35:45,960
Speaker 6: globe that of course have experienced water restrictions water rain

564
00:35:46,400 --> 00:35:50,520
Speaker 6: diminishing in areas. But as a nation, New Zealand is

565
00:35:50,560 --> 00:35:54,800
Speaker 6: surrounded by the ocean, so we have a special value

566
00:35:54,840 --> 00:35:58,359
Speaker 6: proposition and I think it's important that we also look

567
00:35:58,400 --> 00:36:01,840
Speaker 6: at how we can reduce arb and footprints. We are

568
00:36:01,960 --> 00:36:04,480
Speaker 6: in the Sunbolt in the far north of New Zealand.

569
00:36:04,960 --> 00:36:10,400
Speaker 6: We know that solar is a way to harness energy

570
00:36:10,640 --> 00:36:13,320
Speaker 6: and then of course that could be fed onto the

571
00:36:13,400 --> 00:36:18,239
Speaker 6: grid or even established next to data centers. So there's

572
00:36:18,320 --> 00:36:22,600
Speaker 6: ways for us to innovate. And let's not forget that

573
00:36:22,760 --> 00:36:26,759
Speaker 6: innovation is the backbone of long term economic growth.

574
00:36:26,920 --> 00:36:31,120
Speaker 1: The attention you got, you know, the Time magazine feature,

575
00:36:31,400 --> 00:36:33,800
Speaker 1: and prior to that, you've had a lot of attention

576
00:36:34,280 --> 00:36:40,160
Speaker 1: at Tahiku Media for your Mari speech AI models, speech

577
00:36:40,200 --> 00:36:43,919
Speaker 1: to text models ninety two percent accuracy, so incredibly high

578
00:36:44,000 --> 00:36:49,200
Speaker 1: quality models. Your automatic speech recognition models are obviously very

579
00:36:49,280 --> 00:36:52,759
Speaker 1: high quality. You've been working on that for a few

580
00:36:52,840 --> 00:36:55,879
Speaker 1: years now, and Video wasn't involved in that as well.

581
00:36:55,920 --> 00:36:58,960
Speaker 1: You're using their GPUs to work on that. Where are

582
00:36:59,040 --> 00:37:02,320
Speaker 1: things at at the moment in terms of the models

583
00:37:02,360 --> 00:37:05,359
Speaker 1: that you're working on and the applications that run off

584
00:37:05,440 --> 00:37:06,040
Speaker 1: those models.

585
00:37:06,320 --> 00:37:10,600
Speaker 6: So we're focused on speech to text and speech to

586
00:37:10,719 --> 00:37:16,280
Speaker 6: text is an important feature of any sort of language modeling.

587
00:37:17,320 --> 00:37:21,719
Speaker 6: Our ability to transcribe Teddy or Maudi is at a

588
00:37:21,840 --> 00:37:26,600
Speaker 6: ninety two percent accuracy rate, which is exceptional. It outperforms

589
00:37:26,760 --> 00:37:32,080
Speaker 6: any attempts by big tech internationally. We also have it

590
00:37:32,200 --> 00:37:36,399
Speaker 6: operating bilingually, so focused on New Zealand English because as

591
00:37:36,480 --> 00:37:39,960
Speaker 6: New Zealanders we actually speak English differently too. We have

592
00:37:40,120 --> 00:37:43,640
Speaker 6: our own accent. Often when we speak to a device

593
00:37:44,239 --> 00:37:48,239
Speaker 6: and use other models, it may not recognize our New

594
00:37:48,320 --> 00:37:53,040
Speaker 6: Zealand accent. So we're focused on ensuring that our platform

595
00:37:53,120 --> 00:37:57,520
Speaker 6: and our model cannot only provide Teddy or Maudi, but

596
00:37:57,719 --> 00:38:05,440
Speaker 6: also provides optional English New Zealand English transcription. We've developed

597
00:38:05,560 --> 00:38:14,640
Speaker 6: speech synthesis, so that's converting text to spoken word. We've

598
00:38:14,719 --> 00:38:17,760
Speaker 6: got several voices that we've developed in Tede, l Mildi.

599
00:38:18,120 --> 00:38:22,800
Speaker 6: We are now reaching out to sister languages in the

600
00:38:22,880 --> 00:38:27,480
Speaker 6: Pacific because the ability to develop a multi language model.

601
00:38:27,520 --> 00:38:30,160
Speaker 6: You and me just spoke about small language models in

602
00:38:30,320 --> 00:38:35,719
Speaker 6: large language models. Our ability at Teikumedia to now grow

603
00:38:35,840 --> 00:38:40,239
Speaker 6: into the Pacific and scale provides not only benefit for

604
00:38:40,360 --> 00:38:43,200
Speaker 6: New Zealand, but benefit for our sister languages who do

605
00:38:43,320 --> 00:38:48,080
Speaker 6: not have that technology. Collectively, we know that we're connected linguistically,

606
00:38:48,320 --> 00:38:51,680
Speaker 6: and that's a well known fact. We also know that

607
00:38:51,800 --> 00:38:57,320
Speaker 6: our traditional knowledge of the Pacific Ocean is very much related,

608
00:38:57,880 --> 00:39:01,000
Speaker 6: and so when we have multi language model from the Pacific,

609
00:39:01,440 --> 00:39:05,440
Speaker 6: it provides new and meaningful ways to not only access information,

610
00:39:05,920 --> 00:39:10,279
Speaker 6: but apply that information as we address things like the

611
00:39:10,320 --> 00:39:15,440
Speaker 6: acidification of the ocean. So our ability to grow economically

612
00:39:15,600 --> 00:39:19,400
Speaker 6: here in Alau and throughout the globe, I think is

613
00:39:19,560 --> 00:39:25,720
Speaker 6: about presenting the special point of difference, the special value

614
00:39:25,800 --> 00:39:30,160
Speaker 6: proposition that we bring to the conversation through language, and

615
00:39:30,880 --> 00:39:36,000
Speaker 6: we are about to start working with all Hawaii native

616
00:39:36,000 --> 00:39:43,040
Speaker 6: Hawaiian language based in Hawaiian. We have created some very

617
00:39:43,200 --> 00:39:50,440
Speaker 6: meaningful relationships with indigenous language groups in the Americas, and

618
00:39:50,760 --> 00:39:54,680
Speaker 6: we are also looking at how we can partner with

619
00:39:55,040 --> 00:39:58,520
Speaker 6: other indigenous groups throughout the world, including the Sami based

620
00:39:58,560 --> 00:40:02,520
Speaker 6: in Sweden who have have a totally unrelated language. But

621
00:40:02,719 --> 00:40:05,759
Speaker 6: of course a're interested in our methodologies and the way

622
00:40:05,840 --> 00:40:10,400
Speaker 6: that we've created created these models from the ground up.

623
00:40:10,920 --> 00:40:15,239
Speaker 6: But again, data science is about knowing your data. We

624
00:40:15,360 --> 00:40:18,160
Speaker 6: don't have the benefit of big data. What we have

625
00:40:18,560 --> 00:40:22,600
Speaker 6: is less data, but quality data. We've tagged and labeled

626
00:40:22,640 --> 00:40:25,160
Speaker 6: our data so no one knows the data like we do,

627
00:40:26,200 --> 00:40:30,960
Speaker 6: and what that has given us is a really really

628
00:40:31,000 --> 00:40:36,560
Speaker 6: good platform to create tools that people can trust. And

629
00:40:36,760 --> 00:40:40,040
Speaker 6: like I said before, we've focused on accuracy and precision

630
00:40:40,400 --> 00:40:44,360
Speaker 6: qualities everything when you're providing a service, and the service

631
00:40:44,440 --> 00:40:50,759
Speaker 6: that we provide through z is used by content creators

632
00:40:50,800 --> 00:40:54,800
Speaker 6: in New Zealand. But like I mentioned before, content creators

633
00:40:54,880 --> 00:40:58,719
Speaker 6: and other indigenous languages may well find it meaningful and

634
00:40:59,000 --> 00:40:59,880
Speaker 6: of benefit to them.

635
00:41:00,280 --> 00:41:02,279
Speaker 1: Obviously, in the far North there you you have a

636
00:41:02,440 --> 00:41:08,200
Speaker 1: very substantial operation there around radio and TV. Really interested

637
00:41:08,280 --> 00:41:11,920
Speaker 1: in where you see that business going. You know, just

638
00:41:12,800 --> 00:41:16,480
Speaker 1: last week entered me which which owns business desk and

639
00:41:16,680 --> 00:41:19,480
Speaker 1: the business off tech you know, and painfully had to

640
00:41:19,880 --> 00:41:23,919
Speaker 1: announce up to forty jobs are going there. Twenty twenty

641
00:41:23,960 --> 00:41:28,279
Speaker 1: four was a real disastrous year for the media. You

642
00:41:28,360 --> 00:41:31,120
Speaker 1: seem to have a thriving sort of operation up there,

643
00:41:31,520 --> 00:41:33,000
Speaker 1: but what's the what's the.

644
00:41:33,480 --> 00:41:36,120
Speaker 7: Environment like for you? Where do you see things going?

645
00:41:36,239 --> 00:41:40,120
Speaker 7: And AI? How does it come into your business.

646
00:41:40,239 --> 00:41:44,160
Speaker 1: We've seen things like notebook mL from from Google where

647
00:41:44,200 --> 00:41:50,200
Speaker 1: you can auto create podcasts with real convincing sounding hosts

648
00:41:50,640 --> 00:41:54,879
Speaker 1: that can summarize documents and then have a conversation about

649
00:41:54,920 --> 00:41:59,120
Speaker 1: them literally to AI hosts talking about a complex piece

650
00:41:59,160 --> 00:42:02,359
Speaker 1: of information, which is pretty incredible. Interest in your views

651
00:42:02,400 --> 00:42:05,200
Speaker 1: on the economic model around this business that we're in

652
00:42:05,719 --> 00:42:09,160
Speaker 1: and how do we use AI to improve that or

653
00:42:09,239 --> 00:42:12,000
Speaker 1: to create new opportunities rather than see them the whole

654
00:42:12,040 --> 00:42:12,719
Speaker 1: thing destroyed.

655
00:42:13,040 --> 00:42:17,359
Speaker 6: Our birth was as one of the twenty one EWE

656
00:42:17,520 --> 00:42:21,200
Speaker 6: radio stations in New Zealand, and as one of the

657
00:42:21,239 --> 00:42:26,160
Speaker 6: twenty one EWEI radio stations we have been broadcasting Interedel

658
00:42:26,320 --> 00:42:32,080
Speaker 6: Maudi for more than thirty years. We started off looking

659
00:42:32,160 --> 00:42:37,120
Speaker 6: at how our archival data could be transcribed to provide

660
00:42:37,400 --> 00:42:42,400
Speaker 6: learning material, educational material not only in Tedel Maudi, but

661
00:42:42,680 --> 00:42:49,800
Speaker 6: focused on local histories, local biodiversity, local solutions for local problems.

662
00:42:50,400 --> 00:42:54,200
Speaker 6: And as we embarked on that journey, we quickly realized

663
00:42:54,239 --> 00:42:59,640
Speaker 6: that there were simply not another of us to transcribe

664
00:42:59,680 --> 00:43:04,359
Speaker 6: thirty years of archival material. So we decided to teach

665
00:43:04,440 --> 00:43:07,640
Speaker 6: computers how to speak Mildi. That's how our journey began.

666
00:43:08,200 --> 00:43:12,520
Speaker 6: But we quickly realized that through bringing services in house,

667
00:43:12,600 --> 00:43:17,120
Speaker 6: we could also reduce overheads, We could also reduced cost

668
00:43:17,760 --> 00:43:20,160
Speaker 6: and we could play a special role and not just

669
00:43:20,400 --> 00:43:23,560
Speaker 6: using technology. Like I said before, we wanted to be

670
00:43:23,680 --> 00:43:26,680
Speaker 6: the developers of it, and now we're internationally known for that.

671
00:43:27,320 --> 00:43:29,840
Speaker 6: But there is no data science, like I said before,

672
00:43:30,239 --> 00:43:33,400
Speaker 6: without the data. And we have a data pipeline. Our

673
00:43:33,560 --> 00:43:37,200
Speaker 6: data pipeline is our EWE radio station. If we didn't

674
00:43:37,280 --> 00:43:40,200
Speaker 6: have an EWE radio station, we would not have a

675
00:43:40,280 --> 00:43:44,080
Speaker 6: data science project. There are many people that have wanted

676
00:43:44,160 --> 00:43:48,040
Speaker 6: to teach computers how to speak Milding. There are many people,

677
00:43:48,680 --> 00:43:51,319
Speaker 6: but we're the people that did it. We're the people

678
00:43:51,400 --> 00:43:54,400
Speaker 6: that did it. Because we not only speak the language,

679
00:43:54,920 --> 00:43:58,360
Speaker 6: we are gathering corpus every day. We are tagging and

680
00:43:58,520 --> 00:44:03,360
Speaker 6: labeling that phonetical every day, and the phonetical data that

681
00:44:03,560 --> 00:44:07,360
Speaker 6: we have is so closely related to other specific languages.

682
00:44:07,880 --> 00:44:12,439
Speaker 6: The model that we have developed can be reinterpreted into

683
00:44:12,520 --> 00:44:16,840
Speaker 6: the context of sister languages. Language is the key to culture,

684
00:44:17,280 --> 00:44:21,880
Speaker 6: and culture is the home of the philosophical worldview and

685
00:44:22,000 --> 00:44:27,319
Speaker 6: the traditional knowledge in which that language lives. And whether

686
00:44:27,440 --> 00:44:31,600
Speaker 6: that's the landscape, the environment, or the biodiversity that you

687
00:44:31,680 --> 00:44:36,440
Speaker 6: and men talked about earlier, it's all important information. And

688
00:44:36,680 --> 00:44:40,680
Speaker 6: the tools that we've created could well become part of

689
00:44:41,000 --> 00:44:46,840
Speaker 6: you know, agents, could well become part of people's new ideas.

690
00:44:47,840 --> 00:44:52,080
Speaker 6: We want to partner with people that have like minds.

691
00:44:52,680 --> 00:44:59,040
Speaker 6: We want to work with international corporations that can provide

692
00:44:59,719 --> 00:45:03,200
Speaker 6: way for us to contribute to saving the planet. We

693
00:45:03,400 --> 00:45:07,759
Speaker 6: believe that there are ways to reinvent the work that

694
00:45:07,880 --> 00:45:11,719
Speaker 6: we've done in different contexts, and we believe that from

695
00:45:11,800 --> 00:45:15,440
Speaker 6: our special context and Kaitaia, building something from the ground

696
00:45:15,560 --> 00:45:19,680
Speaker 6: up wasn't just groundbreaking, it was breaking a ceiling. And

697
00:45:19,800 --> 00:45:21,759
Speaker 6: we don't want to be the only people that break

698
00:45:21,880 --> 00:45:24,359
Speaker 6: that ceiling, and we hope that the work that we've

699
00:45:24,440 --> 00:45:29,200
Speaker 6: achieved can benefit others as well. And that was what

700
00:45:29,360 --> 00:45:32,720
Speaker 6: I was talking about at Davos, which is largely about

701
00:45:32,800 --> 00:45:40,239
Speaker 6: safety and accountability, responsibility, obligations, but also growing trust, maintaining

702
00:45:40,360 --> 00:45:45,240
Speaker 6: trust through delivering a quality service, a quality products.

703
00:45:45,280 --> 00:45:49,880
Speaker 1: It's incredibly important Mahi that you've done up there huge

704
00:45:50,560 --> 00:45:54,759
Speaker 1: value for New Zealand. As you left Davos, where you optimistic.

705
00:45:54,880 --> 00:45:57,160
Speaker 1: There's a lot going on in the world. There are

706
00:45:57,280 --> 00:46:01,600
Speaker 1: wars happening. Trump is going to create a lot of

707
00:46:01,680 --> 00:46:07,399
Speaker 1: uncertainty geopolitically, just talk of taris. How did you feel

708
00:46:07,480 --> 00:46:11,840
Speaker 1: sort of leaving Davos, particularly meeting with all those indigenous

709
00:46:11,920 --> 00:46:15,520
Speaker 1: leaders who are really worried about some of the same

710
00:46:15,600 --> 00:46:19,320
Speaker 1: issues that you've been talking about, the environment, the preservation

711
00:46:19,520 --> 00:46:23,759
Speaker 1: of indigenous cultures. Do you feel that we do have

712
00:46:24,360 --> 00:46:28,719
Speaker 1: the power and the opportunity to make a difference, a

713
00:46:28,760 --> 00:46:32,280
Speaker 1: positive difference in a world that is becoming more complex

714
00:46:32,360 --> 00:46:33,040
Speaker 1: and more chaotic.

715
00:46:33,239 --> 00:46:36,600
Speaker 6: I think that everybody there was concerned about the environment.

716
00:46:37,200 --> 00:46:42,080
Speaker 6: Everyone there was reflective on the fact that climate's changing,

717
00:46:42,719 --> 00:46:48,400
Speaker 6: that events that happened annually used to happen less often.

718
00:46:49,640 --> 00:46:56,360
Speaker 6: The scale it, which you know, catastrophic weather events happen

719
00:46:56,440 --> 00:46:59,600
Speaker 6: in different parts of the world, was certainly on the

720
00:46:59,640 --> 00:47:03,960
Speaker 6: tip of most people's tongue. But when I had the

721
00:47:04,040 --> 00:47:10,320
Speaker 6: opportunity to discuss things with presidents, vice presidents, CEOs of

722
00:47:10,480 --> 00:47:14,560
Speaker 6: some of the biggest companies in the world, I was

723
00:47:14,640 --> 00:47:21,120
Speaker 6: having those conversations as somebody that had achieved something in

724
00:47:21,400 --> 00:47:25,800
Speaker 6: my own right. And whilst I was taking with me,

725
00:47:25,960 --> 00:47:28,759
Speaker 6: of course, the card or the thinking and the philosophical

726
00:47:28,880 --> 00:47:34,600
Speaker 6: worldview of my own upbringing, I was very open to

727
00:47:34,680 --> 00:47:37,880
Speaker 6: the fact that others too had their own perspective on

728
00:47:37,960 --> 00:47:42,400
Speaker 6: the world. And that was important because largely the conversation

729
00:47:42,680 --> 00:47:46,479
Speaker 6: was about innovation and economic growth. It didn't really matter

730
00:47:46,640 --> 00:47:51,879
Speaker 6: which part of the forum you were involved with. Much

731
00:47:51,960 --> 00:47:57,560
Speaker 6: of the conversation was about sustainable AI practices and how

732
00:47:57,800 --> 00:48:02,480
Speaker 6: can long term economic resilience and positioning local economies as well,

733
00:48:02,600 --> 00:48:07,440
Speaker 6: not just global economies but local economies, and those leaders

734
00:48:07,600 --> 00:48:12,480
Speaker 6: in both the conversation around innovation and the conversation around sustainability,

735
00:48:12,880 --> 00:48:18,640
Speaker 6: because they're interconnected, and the reality is is AI is

736
00:48:19,000 --> 00:48:25,400
Speaker 6: very intensive in its consumption of resources. So the conversation

737
00:48:25,640 --> 00:48:31,200
Speaker 6: that I was particularly interested was how do we reduce that.

738
00:48:32,200 --> 00:48:36,880
Speaker 6: How do we continue to encourage innovation and enhance the

739
00:48:36,960 --> 00:48:39,440
Speaker 6: work that we're doing in terms of the advancement of

740
00:48:39,560 --> 00:48:44,880
Speaker 6: AI while also looking at the fact that energy consumption

741
00:48:45,560 --> 00:48:48,040
Speaker 6: at some stage needs to be solved.

742
00:48:48,400 --> 00:48:51,080
Speaker 1: Well, Peter Lucas, thanks so much for coming on the

743
00:48:51,120 --> 00:48:55,759
Speaker 1: business of tech. Fascinating insights there, safe travels home, and

744
00:48:56,320 --> 00:48:59,200
Speaker 1: all the best for what's ahead with ta Hiku media.

745
00:48:59,360 --> 00:49:01,960
Speaker 6: Well, thank you. So I'd just like to finish by saying,

746
00:49:02,000 --> 00:49:04,480
Speaker 6: you know, like the key is going to be smart

747
00:49:04,640 --> 00:49:09,799
Speaker 6: investment and local economies. We can all leverage AIS growth

748
00:49:09,920 --> 00:49:13,759
Speaker 6: by embracing sustainability. It's an opportunity for us to create

749
00:49:13,880 --> 00:49:20,200
Speaker 6: new jobs, attract investment into our regions and build a

750
00:49:20,360 --> 00:49:24,520
Speaker 6: future proof, eco conscious economy. And I think as part

751
00:49:24,760 --> 00:49:28,040
Speaker 6: of the Pacific New Zealand, we've always been leaders in

752
00:49:28,120 --> 00:49:30,680
Speaker 6: that and let's continue to lead in that. Kilder.

753
00:49:37,800 --> 00:49:39,680
Speaker 1: So, there you have it been Peter Lucas. Jones has

754
00:49:39,719 --> 00:49:42,560
Speaker 1: really been loaded for great work on this speech. Totext

755
00:49:43,560 --> 00:49:48,320
Speaker 1: AI model to hicky media has developed, and you know,

756
00:49:48,560 --> 00:49:50,480
Speaker 1: I think you probably put a bit of an upbeat

757
00:49:50,560 --> 00:49:53,319
Speaker 1: spin on what was actually happening at Davos. There may

758
00:49:53,360 --> 00:49:56,920
Speaker 1: have been a lot of optimism there, but really the

759
00:49:57,080 --> 00:49:59,919
Speaker 1: forces that are converging in the world at the moment

760
00:50:00,080 --> 00:50:05,879
Speaker 1: with Trump's inauguration, Davos is all about free trade, collaboration,

761
00:50:06,800 --> 00:50:10,640
Speaker 1: international cooperation, all of that is really at threat now.

762
00:50:11,520 --> 00:50:14,680
Speaker 1: Looming over all of that, you have technological change and

763
00:50:14,760 --> 00:50:20,359
Speaker 1: particularly AI, and we didn't really talk about deep SEEK specifically.

764
00:50:20,440 --> 00:50:23,120
Speaker 1: That was sort of just blew up really in the

765
00:50:23,239 --> 00:50:26,919
Speaker 1: last couple of days, but clearly Peter Lucas is really

766
00:50:26,960 --> 00:50:30,839
Speaker 1: interested in sort of narrow AI rather than general AI,

767
00:50:31,000 --> 00:50:34,239
Speaker 1: trying to compete because we can't compete on building large

768
00:50:34,320 --> 00:50:39,279
Speaker 1: language models smaller uses of smaller AI models, and that's

769
00:50:39,320 --> 00:50:44,200
Speaker 1: been incredibly useful. He's pursuing that for really worthy purposes

770
00:50:44,680 --> 00:50:50,719
Speaker 1: around retaining cultural identity and language across the Pacific and elsewhere.

771
00:50:51,040 --> 00:50:52,800
Speaker 1: And that's just a really good example. I think of

772
00:50:53,120 --> 00:50:57,600
Speaker 1: what our value at is doing safe, reliable, responsible AI

773
00:50:58,719 --> 00:51:01,640
Speaker 1: and doing it in secretive niches or niches that are

774
00:51:01,719 --> 00:51:03,759
Speaker 1: underserved at the moment, and that you know with our

775
00:51:03,840 --> 00:51:07,840
Speaker 1: public research organization that's coming I startup's use of AI.

776
00:51:07,960 --> 00:51:10,640
Speaker 1: I think that's quite instructive about what our approach should

777
00:51:10,640 --> 00:51:10,920
Speaker 1: be like.

778
00:51:11,640 --> 00:51:14,680
Speaker 5: Yeah, and what is unique about New Zealand as well,

779
00:51:15,400 --> 00:51:21,480
Speaker 5: having the strong Indigenous population that are becoming increasingly technically

780
00:51:21,600 --> 00:51:25,680
Speaker 5: capable thanks to a long history of you know, committed

781
00:51:25,719 --> 00:51:29,759
Speaker 5: efforts across everybody in New Zealand, across all areas of

782
00:51:29,840 --> 00:51:33,400
Speaker 5: New Zealand to make sure that we are empowering everybody

783
00:51:33,480 --> 00:51:35,320
Speaker 5: within New Zealand. So to see that paying off on

784
00:51:35,400 --> 00:51:37,320
Speaker 5: a global stage as well is really exciting.

785
00:51:37,640 --> 00:51:40,200
Speaker 1: Yeah, so we'll keep an eye on to HIKU media

786
00:51:40,320 --> 00:51:44,000
Speaker 1: and on what AI developments are are going to come

787
00:51:44,160 --> 00:51:46,360
Speaker 1: in the research sector. You know, I've been sort of

788
00:51:46,400 --> 00:51:50,200
Speaker 1: asking around my AI contacts this week, you know who's

789
00:51:50,200 --> 00:51:52,399
Speaker 1: around who could lead this, you know, who's the crack

790
00:51:52,480 --> 00:51:56,600
Speaker 1: team that they could potentially gather together, And got a

791
00:51:56,680 --> 00:52:00,359
Speaker 1: lot of long pauses over the phone when I when

792
00:52:00,400 --> 00:52:01,600
Speaker 1: I asked them that it's not that we don't have

793
00:52:01,719 --> 00:52:04,160
Speaker 1: good people here. It's just like, if you're trying to

794
00:52:04,200 --> 00:52:07,520
Speaker 1: build a capability that the csi R in Australia has

795
00:52:07,560 --> 00:52:10,960
Speaker 1: they have over a thousand AI researchers, you know, it's

796
00:52:11,000 --> 00:52:13,840
Speaker 1: going to take quite a while. But the suggestion to

797
00:52:13,920 --> 00:52:16,560
Speaker 1: me is that we if we try to focus on

798
00:52:17,120 --> 00:52:20,719
Speaker 1: you know, real research intensive we're going to fail in

799
00:52:20,800 --> 00:52:23,360
Speaker 1: that organization. It has to be engineering based, so it

800
00:52:23,440 --> 00:52:28,920
Speaker 1: has to be taking problems from society or economic or

801
00:52:29,000 --> 00:52:33,480
Speaker 1: social or environmental problems, having our companies go to this

802
00:52:33,719 --> 00:52:38,400
Speaker 1: new public research organization saying I've got this problem, can

803
00:52:38,520 --> 00:52:42,040
Speaker 1: AI help me solve it, and then using that intellectual

804
00:52:42,120 --> 00:52:47,120
Speaker 1: property to try and better the entire country. So more

805
00:52:47,200 --> 00:52:51,440
Speaker 1: practical than real deep research focus, because we're just never

806
00:52:51,480 --> 00:52:54,200
Speaker 1: going to be able to build that capability quickly, that's

807
00:52:54,280 --> 00:52:58,719
Speaker 1: the suggestion, and focus very much on the narrow AI stuff.

808
00:52:58,719 --> 00:53:00,840
Speaker 1: If we have an environmental problem and we need to solve,

809
00:53:01,440 --> 00:53:05,319
Speaker 1: throw some of the best AI researchers in this organization

810
00:53:05,480 --> 00:53:07,600
Speaker 1: at it, but in conjunction with the people who are

811
00:53:07,600 --> 00:53:08,080
Speaker 1: going to use it.

812
00:53:08,760 --> 00:53:10,960
Speaker 5: Yeah, I think you're right, And there needs to be

813
00:53:11,200 --> 00:53:17,399
Speaker 5: a close relationship with those more broad AI researchers as well,

814
00:53:17,560 --> 00:53:20,719
Speaker 5: who might be able to offer surprising insights that you

815
00:53:20,800 --> 00:53:23,360
Speaker 5: don't get when you're only focused on the specific So,

816
00:53:23,480 --> 00:53:26,600
Speaker 5: whether that is some of these CSIRO people or other

817
00:53:26,680 --> 00:53:32,000
Speaker 5: academics within international universities who have really broad knowledges about

818
00:53:32,080 --> 00:53:35,759
Speaker 5: AI related issues, that's going to be super useful as well.

819
00:53:35,840 --> 00:53:38,719
Speaker 5: So hopefully that's part of the agenda, And.

820
00:53:38,760 --> 00:53:41,319
Speaker 1: That's a great thing about open sources. You know, when

821
00:53:41,320 --> 00:53:44,280
Speaker 1: you get a community behind it, everyone shares the library,

822
00:53:44,400 --> 00:53:48,920
Speaker 1: shares the techniques, and that will all be accessible to

823
00:53:49,360 --> 00:53:52,760
Speaker 1: people down here as well. So I think positive development

824
00:53:53,280 --> 00:53:57,000
Speaker 1: this week. There's still a few caveats and things to

825
00:53:57,160 --> 00:53:59,520
Speaker 1: be seen. And the one thing we didn't discuss really

826
00:53:59,719 --> 00:54:04,359
Speaker 1: is Trumps welcomed this news this week of deep Seek

827
00:54:04,480 --> 00:54:08,120
Speaker 1: being more efficient and cheaper to run. Ultimately, he still wants,

828
00:54:08,920 --> 00:54:10,840
Speaker 1: you know, the US to win in this race. So

829
00:54:11,080 --> 00:54:13,800
Speaker 1: is he just going to pull that lever off banning

830
00:54:14,239 --> 00:54:18,000
Speaker 1: deep Seek from being used in the US as the

831
00:54:18,320 --> 00:54:21,719
Speaker 1: you know, the previous administration did with TikTok, So are

832
00:54:21,760 --> 00:54:24,520
Speaker 1: we going to see that come into play basically, the

833
00:54:25,840 --> 00:54:31,560
Speaker 1: the technology being restricted from US companies and people, And

834
00:54:32,440 --> 00:54:34,680
Speaker 1: you know, I think what deep Seek has shown is

835
00:54:34,760 --> 00:54:40,160
Speaker 1: that the trade and bargo escalation banning of the export

836
00:54:40,239 --> 00:54:43,000
Speaker 1: of technology hasn't really worked in this case because obviously

837
00:54:43,560 --> 00:54:45,600
Speaker 1: Deep Seek were able to get around that and and

838
00:54:46,600 --> 00:54:50,920
Speaker 1: do it with more lower power chips. So I think

839
00:54:51,320 --> 00:54:54,719
Speaker 1: he's realizing now that banning his way to supremacy is

840
00:54:54,800 --> 00:54:56,719
Speaker 1: not the most effective way to do it.

841
00:54:56,960 --> 00:54:59,120
Speaker 7: You've got to win you've got to be better innovation.

842
00:55:00,000 --> 00:55:03,640
Speaker 4: Absolutely. Necessity is the mother of innovation, isn't it.

843
00:55:03,719 --> 00:55:06,480
Speaker 5: So if you are in a restrained environment, you make

844
00:55:06,560 --> 00:55:08,400
Speaker 5: it work with what you have if you can, and

845
00:55:08,480 --> 00:55:10,640
Speaker 5: that's what China has done exactly.

846
00:55:11,440 --> 00:55:14,560
Speaker 1: So thanks so much to Peter Lucas Jones for coming

847
00:55:14,640 --> 00:55:17,000
Speaker 1: on the Business of Tech. Details in the show notes

848
00:55:17,320 --> 00:55:20,360
Speaker 1: about the AI models and speech to text application to

849
00:55:20,440 --> 00:55:22,240
Speaker 1: Hicku Media has developed.

850
00:55:22,200 --> 00:55:24,160
Speaker 5: Head to businessdesk, dot co, dot and Z had to

851
00:55:24,239 --> 00:55:26,560
Speaker 5: check those out and our reading list of the big

852
00:55:26,680 --> 00:55:27,879
Speaker 5: stories in tech this week.

853
00:55:28,239 --> 00:55:31,280
Speaker 1: Subscribe to the Business of Tech on your podcast platform

854
00:55:31,320 --> 00:55:34,480
Speaker 1: of choice. We're also streaming on iHeartRadio.

855
00:55:34,200 --> 00:55:36,520
Speaker 5: And get in touch with your feedback and topic suggestions.

856
00:55:36,560 --> 00:55:38,600
Speaker 5: We're on LinkedIn and Blue Sky.

857
00:55:38,840 --> 00:55:42,040
Speaker 1: And catch us again next week for another episode of

858
00:55:42,200 --> 00:55:43,040
Speaker 1: the Business of Tech.

859
00:55:43,280 --> 00:55:44,680
Speaker 4: Till then, have a great week.