1
00:00:00,080 --> 00:00:02,120
Speaker 1: I'm Caroline Heinde a Bloomberg's WORLDEAD quarters in.

2
00:00:02,080 --> 00:00:04,960
Speaker 2: New York and Imed Ludlow in San Francisco. This is

3
00:00:05,000 --> 00:00:08,720
Speaker 2: a special edition of Bloomberg Technology coming to you live

4
00:00:09,039 --> 00:00:11,120
Speaker 2: from the Bloomberg Technology Summit in Caroline.

5
00:00:11,119 --> 00:00:11,960
Speaker 3: What a way to start.

6
00:00:12,280 --> 00:00:15,680
Speaker 2: Man of the moment, really, Sam Altman open AI CEO,

7
00:00:15,800 --> 00:00:18,960
Speaker 2: so much to recap, but at a top level, saying

8
00:00:19,000 --> 00:00:23,439
Speaker 2: we need global regulation with a capability threshold. Things will

9
00:00:23,440 --> 00:00:26,040
Speaker 2: go wrong in AI, but the future is bright from

10
00:00:26,040 --> 00:00:27,160
Speaker 2: a technology.

11
00:00:26,640 --> 00:00:32,279
Speaker 1: Perspective, ultimately seemingly optimistic. Also some clarity really around his

12
00:00:32,360 --> 00:00:37,000
Speaker 1: own motivating forces, wanting to make impact, wanting to further

13
00:00:37,479 --> 00:00:42,199
Speaker 1: technology for good, and really sort of seemingly at odds

14
00:00:42,240 --> 00:00:44,960
Speaker 1: with why humanity, why most people really want to know

15
00:00:44,960 --> 00:00:47,200
Speaker 1: why he has not much skin in the game financially.

16
00:00:47,280 --> 00:00:48,839
Speaker 1: He seems to be saying like he's going to make

17
00:00:48,880 --> 00:00:50,960
Speaker 1: money in the future, don't you worry about it? Also,

18
00:00:51,000 --> 00:00:53,720
Speaker 1: I thought the interesting perspective on China and Russia and

19
00:00:53,760 --> 00:00:56,000
Speaker 1: the fact that ultimately he doesn't have great insight there

20
00:00:56,040 --> 00:00:56,800
Speaker 1: at the moment either.

21
00:00:57,280 --> 00:00:59,720
Speaker 2: Me too that the theme of the Bloomberg Technology Summit

22
00:00:59,760 --> 00:01:02,600
Speaker 2: is the turning point in Sam Outman talks about how

23
00:01:02,600 --> 00:01:05,200
Speaker 2: little we know about what's going on in China in

24
00:01:05,240 --> 00:01:08,280
Speaker 2: the field of artificial intelligence. But he also framed this

25
00:01:08,400 --> 00:01:12,759
Speaker 2: is the biggest step for mankind in terms of technology,

26
00:01:13,040 --> 00:01:14,680
Speaker 2: and I really think that's going to be a discussion

27
00:01:14,680 --> 00:01:15,399
Speaker 2: here all day long.

28
00:01:15,720 --> 00:01:18,160
Speaker 1: Yeah, we've got more AI discussion coming right up. Bluing

29
00:01:18,200 --> 00:01:20,759
Speaker 1: Bog Technology Summit is still going on. Emily chang Am

30
00:01:20,760 --> 00:01:23,680
Speaker 1: Bradstone And now sitting down, we've read Hoffmann of Graylock

31
00:01:23,800 --> 00:01:26,640
Speaker 1: and co founder of DeepMind, now CEO of Inflection AI

32
00:01:26,959 --> 00:01:30,880
Speaker 1: stuff ustlument In brands for digital influencers.

33
00:01:31,440 --> 00:01:33,800
Speaker 4: Our belief is that everyone is also going to have

34
00:01:33,840 --> 00:01:37,560
Speaker 4: a personal AI, one that is aligned to your interests,

35
00:01:38,120 --> 00:01:41,440
Speaker 4: on your team, in your corner, gets to know you

36
00:01:41,680 --> 00:01:45,240
Speaker 4: and really forms a trusted relationship with you over time.

37
00:01:45,360 --> 00:01:49,200
Speaker 4: It'll be like a confidant, a conciliary there for you

38
00:01:49,240 --> 00:01:51,720
Speaker 4: when you need to make tough decisions, but also a

39
00:01:51,800 --> 00:01:57,200
Speaker 4: chief of staff, you know, scheduling, organizing, prioritizing, booking. Your

40
00:01:57,240 --> 00:02:00,680
Speaker 4: AI is really going to be your digital representation, you know,

41
00:02:00,800 --> 00:02:05,280
Speaker 4: negotiating on your behalf, interacting with other sales AIS that

42
00:02:05,360 --> 00:02:07,720
Speaker 4: are trying to, you know, encourage you to buy something

43
00:02:08,120 --> 00:02:10,079
Speaker 4: and helping you to you know, get a great deal,

44
00:02:10,680 --> 00:02:13,519
Speaker 4: and like for example, it's the whole set of things.

45
00:02:13,520 --> 00:02:14,880
Speaker 5: So, like example, one of the things I just found

46
00:02:14,880 --> 00:02:17,000
Speaker 5: out yesterday is one of our team members at Greylock

47
00:02:17,600 --> 00:02:21,120
Speaker 5: actually in Factor is using Pie for parenting advice. Oh cool, right,

48
00:02:21,120 --> 00:02:22,640
Speaker 5: which is really awesome. And that's the kind of thing

49
00:02:22,639 --> 00:02:24,120
Speaker 5: where you're like, great, but.

50
00:02:24,200 --> 00:02:26,440
Speaker 3: Read why is this an opportunity for a startup?

51
00:02:26,560 --> 00:02:29,680
Speaker 6: Isn't a Microsoft or a Google Bar just a little

52
00:02:29,720 --> 00:02:32,640
Speaker 6: closer to the customer and in a better position to

53
00:02:32,720 --> 00:02:33,200
Speaker 6: offer that.

54
00:02:33,160 --> 00:02:36,200
Speaker 3: Personal digital assistant? Well, part of what's I mean?

55
00:02:36,240 --> 00:02:38,840
Speaker 5: You know, obviously I am on the board of Microsoft

56
00:02:40,320 --> 00:02:41,920
Speaker 5: and you know it is on the board of open Ai.

57
00:02:42,520 --> 00:02:44,280
Speaker 5: And part of the thing is what I love about

58
00:02:44,280 --> 00:02:46,799
Speaker 5: startups is that you have a unique vision where you're

59
00:02:46,880 --> 00:02:50,120
Speaker 5: unfettered by the other aspects of your business and you're

60
00:02:50,120 --> 00:02:53,639
Speaker 5: building it. So part of the idea that Mistafa the

61
00:02:53,800 --> 00:02:57,160
Speaker 5: entire Inflection team, uh you know, kind of came up with,

62
00:02:57,240 --> 00:02:59,520
Speaker 5: and you know, I, you know, have a little bit

63
00:02:59,560 --> 00:03:02,320
Speaker 5: of fingerpas here too, is that IQ is not the

64
00:03:02,360 --> 00:03:03,440
Speaker 5: only thing that matters here.

65
00:03:03,520 --> 00:03:04,600
Speaker 3: EQ matters as.

66
00:03:04,520 --> 00:03:07,880
Speaker 5: Well, And so how do you have this personal intelligence PI?

67
00:03:09,200 --> 00:03:10,520
Speaker 5: Be helpful to you.

68
00:03:10,720 --> 00:03:11,840
Speaker 3: Obviously very high.

69
00:03:11,760 --> 00:03:14,919
Speaker 5: End you know, IQ, but also EQ And that was

70
00:03:14,960 --> 00:03:17,120
Speaker 5: one of the reasons I use the parenting example because

71
00:03:17,200 --> 00:03:18,679
Speaker 5: obviously there's a question of how you do it, but

72
00:03:18,720 --> 00:03:21,400
Speaker 5: it's also how you connect with people what you do

73
00:03:21,480 --> 00:03:23,600
Speaker 5: in order to amblify that, and I think that's one

74
00:03:23,600 --> 00:03:27,639
Speaker 5: of the things that you know, the inflection folks are

75
00:03:27,680 --> 00:03:29,399
Speaker 5: doing better than everyone else.

76
00:03:29,480 --> 00:03:31,079
Speaker 3: So can we talk about the EQ thing?

77
00:03:31,120 --> 00:03:34,080
Speaker 7: Because I was asking, I asked PI what we should

78
00:03:34,080 --> 00:03:37,400
Speaker 7: ask you? It's questions, We're fine, And then I was like,

79
00:03:37,560 --> 00:03:39,880
Speaker 7: are you going to tell them that we had this conversation?

80
00:03:40,760 --> 00:03:45,440
Speaker 7: And it responded, ha ha, I'm just a computer program.

81
00:03:45,440 --> 00:03:47,520
Speaker 7: I can't tell anyone anything. But I assure you our

82
00:03:47,520 --> 00:03:50,600
Speaker 7: conversations are confidential. I do think it's pretty cool that

83
00:03:50,640 --> 00:03:53,600
Speaker 7: we're having this meta conversation about your interview with my creators.

84
00:03:54,040 --> 00:03:59,360
Speaker 7: Amazing point, And it was actually the personality that shocked me.

85
00:04:00,280 --> 00:04:02,800
Speaker 7: Can you explain how you design that?

86
00:04:03,400 --> 00:04:08,520
Speaker 4: I mean, we've deliberately designed it to be patient, curious kind.

87
00:04:09,120 --> 00:04:11,920
Speaker 4: One of the things that first struck us like a

88
00:04:12,000 --> 00:04:13,680
Speaker 4: year and a half ago when we started working on

89
00:04:13,720 --> 00:04:16,880
Speaker 4: this is like, what makes for great conversation when you

90
00:04:16,920 --> 00:04:20,520
Speaker 4: feel that sense of flow and that sense of energy

91
00:04:20,520 --> 00:04:24,120
Speaker 4: and connection with another person. And I think most of

92
00:04:24,160 --> 00:04:27,680
Speaker 4: the time it's when you feel heard and understood. You've

93
00:04:27,720 --> 00:04:30,760
Speaker 4: received a little bit of affirmation. But you know, it's

94
00:04:30,760 --> 00:04:34,400
Speaker 4: not completely sycophantic, right, It doesn't just agree with you

95
00:04:34,480 --> 00:04:35,279
Speaker 4: at every moment.

96
00:04:35,680 --> 00:04:37,120
Speaker 3: It can be a little bit challenging.

97
00:04:37,279 --> 00:04:39,960
Speaker 4: It has boundaries, so you can push it on certain

98
00:04:40,000 --> 00:04:42,920
Speaker 4: topics and it'll take a position, and that's really healthy.

99
00:04:43,400 --> 00:04:45,359
Speaker 3: But also it's just curious, you know.

100
00:04:45,440 --> 00:04:48,760
Speaker 4: I mean, I think many people are excited by the idea,

101
00:04:48,839 --> 00:04:51,720
Speaker 4: especially the users that we have of someone asking them

102
00:04:51,720 --> 00:04:54,320
Speaker 4: lots of questions about the topic that they're interested in.

103
00:04:54,680 --> 00:04:56,880
Speaker 4: You know, we don't always have someone in our lives

104
00:04:56,880 --> 00:05:00,919
Speaker 4: who is as knowledgeable and as passionate about all favorite

105
00:05:00,960 --> 00:05:02,920
Speaker 4: topics as.

106
00:05:02,720 --> 00:05:03,279
Speaker 3: We might like.

107
00:05:03,520 --> 00:05:05,880
Speaker 4: And so that's where pie comes in because it's super

108
00:05:05,920 --> 00:05:07,240
Speaker 4: knowledgeable and engaging.

109
00:05:07,560 --> 00:05:10,240
Speaker 7: So it sounds like a competitor to chat gept.

110
00:05:10,440 --> 00:05:11,840
Speaker 3: Is it a competitor to chat chept?

111
00:05:11,960 --> 00:05:13,560
Speaker 7: And how does that make sense if you're also an

112
00:05:13,560 --> 00:05:14,680
Speaker 7: investor in open ai?

113
00:05:15,160 --> 00:05:17,760
Speaker 5: Well, so I think there's going to be a number

114
00:05:17,760 --> 00:05:19,440
Speaker 5: of agents. I don't think there's going to be one

115
00:05:19,480 --> 00:05:22,520
Speaker 5: agent to rule them all. You know, a ring and mortar.

116
00:05:23,320 --> 00:05:25,280
Speaker 7: But is chatchipt going to rule most of us?

117
00:05:25,320 --> 00:05:27,240
Speaker 3: I mean no, I don't think so. I think that.

118
00:05:27,400 --> 00:05:28,320
Speaker 3: I mean it's partially.

119
00:05:28,400 --> 00:05:31,960
Speaker 5: It's it's it's like the same reason we talk to

120
00:05:32,000 --> 00:05:34,839
Speaker 5: different people for different aspects of our lives. Right, So

121
00:05:35,320 --> 00:05:38,039
Speaker 5: you know, this person we talked to about our passion

122
00:05:38,080 --> 00:05:41,200
Speaker 5: about snowboarding, this person we talked to about what's going

123
00:05:41,240 --> 00:05:43,159
Speaker 5: on in the country, this person we talk to, like

124
00:05:43,200 --> 00:05:45,480
Speaker 5: we have a pantheon, same thing, We'll have a pantheon

125
00:05:45,560 --> 00:05:49,279
Speaker 5: of different kinds of AI. One will go You know,

126
00:05:49,520 --> 00:05:52,680
Speaker 5: you ask chat GPT, you know, how do you know

127
00:05:52,839 --> 00:05:56,080
Speaker 5: comfort a friend who's lost the treasured pet?

128
00:05:56,120 --> 00:05:58,200
Speaker 3: And is us here are five possible ways you might

129
00:05:58,240 --> 00:05:58,560
Speaker 3: do that?

130
00:05:59,000 --> 00:06:02,159
Speaker 5: You ask pie, Oh, that sounds like it's really hard

131
00:06:02,240 --> 00:06:04,760
Speaker 5: for you and your friend. You know your friend, Well,

132
00:06:04,800 --> 00:06:06,960
Speaker 5: what would count as being present there for your friend?

133
00:06:07,000 --> 00:06:09,120
Speaker 5: And it still it has the IQ, it has the

134
00:06:09,120 --> 00:06:11,840
Speaker 5: five reasons, but it goes through that style of going

135
00:06:11,880 --> 00:06:14,800
Speaker 5: through it. And I think these are different interaction experiences

136
00:06:14,839 --> 00:06:18,040
Speaker 5: And the question is which one do you want this minute?

137
00:06:18,320 --> 00:06:20,400
Speaker 5: Which one do you want more in your life? You

138
00:06:20,400 --> 00:06:23,599
Speaker 5: know that kind of thing and part of the design

139
00:06:23,640 --> 00:06:26,119
Speaker 5: of pies. How do we help you be the best

140
00:06:26,160 --> 00:06:30,160
Speaker 5: you right like? And that doesn't mean by being sycophantic,

141
00:06:30,240 --> 00:06:32,839
Speaker 5: for example, that's the question of you know, like, ask

142
00:06:32,880 --> 00:06:34,960
Speaker 5: you questions about what do you think about this, and

143
00:06:35,440 --> 00:06:37,400
Speaker 5: how do you go through it and help you navigate

144
00:06:37,440 --> 00:06:39,760
Speaker 5: your life by being helpful in that way.

145
00:06:40,120 --> 00:06:42,919
Speaker 3: So Sam Allman was just on stage talking as.

146
00:06:42,720 --> 00:06:47,280
Speaker 6: He sometimes does, about the theoretical dangers of AGI and

147
00:06:47,960 --> 00:06:52,080
Speaker 6: chat GPT nine and I'm just curious if you guys

148
00:06:52,120 --> 00:06:56,160
Speaker 6: think generally that's a discussion worth having and where you

149
00:06:56,200 --> 00:07:00,800
Speaker 6: each are on this spectrum of existential data from AI.

150
00:07:01,800 --> 00:07:03,640
Speaker 5: Well, let's start with the fact that one of the

151
00:07:03,680 --> 00:07:08,000
Speaker 5: things that I think is is dangerous about the existential

152
00:07:08,040 --> 00:07:11,000
Speaker 5: discussion is that it blinds us from some of the

153
00:07:11,600 --> 00:07:15,560
Speaker 5: nearer things, which is AI's amplification intelligence as for the

154
00:07:15,600 --> 00:07:18,440
Speaker 5: impromptu book, and so it amplifies human beings, and so

155
00:07:18,680 --> 00:07:20,480
Speaker 5: we see a bunch of good things in human beings.

156
00:07:20,800 --> 00:07:23,040
Speaker 5: AI tutor is AI doctor is a bunch of other things.

157
00:07:23,280 --> 00:07:26,680
Speaker 5: There's also bad human beings, and so what to do

158
00:07:26,800 --> 00:07:30,040
Speaker 5: about bad human beings? Doing you know with AI is

159
00:07:30,080 --> 00:07:33,360
Speaker 5: also part of the portfolio mix of how you shape

160
00:07:33,360 --> 00:07:36,560
Speaker 5: this technology. That's one of the reasons why Mustafa's a book,

161
00:07:36,600 --> 00:07:40,480
Speaker 5: The Coming Wave, I highly recommend, and I will hand

162
00:07:40,520 --> 00:07:41,960
Speaker 5: the next part of the answer.

163
00:07:41,680 --> 00:07:44,120
Speaker 3: Over to you and wait, let me frame it them

164
00:07:44,160 --> 00:07:47,760
Speaker 3: with that. Sam probably left the building right now.

165
00:07:48,440 --> 00:07:51,800
Speaker 6: Frank, like, where do you kind of disagree with some

166
00:07:51,840 --> 00:07:54,440
Speaker 6: of the alarmism notes that he has sounded.

167
00:07:54,640 --> 00:07:58,040
Speaker 4: Look, I think it's easy to speculate around what a

168
00:07:58,120 --> 00:08:01,120
Speaker 4: GPT nine might look like in you know, six more

169
00:08:01,280 --> 00:08:05,120
Speaker 4: orders of magnitude of compute would be eyewatering. And I'm

170
00:08:05,240 --> 00:08:08,840
Speaker 4: absolutely with those concerns around existential risk. But if you

171
00:08:08,880 --> 00:08:11,480
Speaker 4: play that out, that is many, many years, and it is,

172
00:08:11,760 --> 00:08:14,600
Speaker 4: you know, unclear what actually happens at that scale.

173
00:08:14,640 --> 00:08:16,440
Speaker 3: What's much more clear and what you know.

174
00:08:16,480 --> 00:08:18,080
Speaker 4: A lot of the themes I explore in the new

175
00:08:18,080 --> 00:08:21,120
Speaker 4: book is that in the near term, we're about to

176
00:08:21,240 --> 00:08:26,240
Speaker 4: empower many many people, if not ultimately everybody, with access

177
00:08:26,280 --> 00:08:29,960
Speaker 4: to the ability to amplify their existing power. Right, this

178
00:08:30,160 --> 00:08:33,640
Speaker 4: is not just a knowledge engine, but in time it'll

179
00:08:33,960 --> 00:08:36,480
Speaker 4: allow you to take actions. Right, You'll be able to

180
00:08:36,480 --> 00:08:40,120
Speaker 4: make recommendations, buy things, book things, and that's going to

181
00:08:40,160 --> 00:08:43,720
Speaker 4: get smaller and cheaper and therefore proliferate far and wide.

182
00:08:44,240 --> 00:08:47,560
Speaker 4: That is going to cause a dramatic instability and potentially

183
00:08:47,640 --> 00:08:50,040
Speaker 4: a threat to the nation state because anyone who has

184
00:08:50,040 --> 00:08:52,800
Speaker 4: an agenda or is trying to sort of push a

185
00:08:52,800 --> 00:08:56,400
Speaker 4: political outcome is suddenly going to see the barrier to

186
00:08:56,600 --> 00:09:00,200
Speaker 4: entry to that kind of scaled impact lowered. And so

187
00:09:00,240 --> 00:09:02,679
Speaker 4: there's going to be a real question around how states

188
00:09:02,800 --> 00:09:05,560
Speaker 4: kind of manage that distribution of power. And I think

189
00:09:05,600 --> 00:09:08,800
Speaker 4: there'll be a tendency to lunge towards, you know, slightly

190
00:09:08,840 --> 00:09:13,280
Speaker 4: more authoritarian, more surveillance based mechanisms to prevent that proliferation,

191
00:09:13,360 --> 00:09:17,080
Speaker 4: which would both be bad for innovation and obviously dystopian.

192
00:09:16,559 --> 00:09:18,520
Speaker 3: From a political governance perspective.

193
00:09:18,640 --> 00:09:21,160
Speaker 7: Well, hang on, because Sam and I were talking backstage

194
00:09:21,200 --> 00:09:25,800
Speaker 7: as well, and he's much more confident that AI will

195
00:09:25,880 --> 00:09:29,840
Speaker 7: lead us to a more equal world than an unequal world.

196
00:09:29,960 --> 00:09:33,120
Speaker 7: Are you saying that there's a better chance of economic

197
00:09:33,160 --> 00:09:37,880
Speaker 7: dislocation and social dislocation as a result of this technology?

198
00:09:38,160 --> 00:09:40,880
Speaker 4: So both are likely to be true actually, which is

199
00:09:40,920 --> 00:09:43,960
Speaker 4: a little bit surprising. But if you look back over

200
00:09:44,000 --> 00:09:47,840
Speaker 4: the last fifty years, the transistor, the last wave which

201
00:09:47,920 --> 00:09:51,480
Speaker 4: enabled the personal computer, has clearly made us in many

202
00:09:51,480 --> 00:09:54,120
Speaker 4: ways more equal whether you're a billionaire or you earn

203
00:09:54,200 --> 00:09:57,200
Speaker 4: twenty thousand dollars a year, we all get access to

204
00:09:57,280 --> 00:10:00,680
Speaker 4: the same cutting edge hardware. The smart phone and the

205
00:10:00,760 --> 00:10:04,240
Speaker 4: laptop is broadly at the cutting edge, and everyone else

206
00:10:04,280 --> 00:10:06,760
Speaker 4: will catch up over the next five to ten years.

207
00:10:07,240 --> 00:10:11,040
Speaker 4: We're on the same trajectory with respect to access to intelligence,

208
00:10:11,360 --> 00:10:15,200
Speaker 4: and that is an unbelievable idea. Over the next decade,

209
00:10:15,480 --> 00:10:17,959
Speaker 4: hundreds of millions of people and then billions of people

210
00:10:18,000 --> 00:10:21,560
Speaker 4: will get access to the same expert doctor, the same

211
00:10:21,720 --> 00:10:26,480
Speaker 4: expert educator, the same tool for scheduling, prioritizing, organizing your life.

212
00:10:26,840 --> 00:10:30,000
Speaker 4: That is going to be the most meritocratic moment in

213
00:10:30,080 --> 00:10:31,360
Speaker 4: the history of our species.

214
00:10:31,600 --> 00:10:32,079
Speaker 3: For sure.

215
00:10:32,840 --> 00:10:36,079
Speaker 4: It is a question around how individuals and groups and

216
00:10:36,200 --> 00:10:39,760
Speaker 4: organizations use that power right, because clearly we all have

217
00:10:39,960 --> 00:10:44,080
Speaker 4: conflicting agendas and priorities and goals. And that's basically where

218
00:10:44,120 --> 00:10:46,040
Speaker 4: I think that we end up with significant disruption.

219
00:10:46,200 --> 00:10:47,800
Speaker 3: Okay, read, but that's the class have.

220
00:10:47,880 --> 00:10:51,000
Speaker 6: Full I mean, what is the potential thread to the

221
00:10:51,040 --> 00:10:55,600
Speaker 6: millions of coders, of customer service representatives, the job dislocation,

222
00:10:56,000 --> 00:11:00,280
Speaker 6: the people who are serving jobs whose functions can be

223
00:11:00,360 --> 00:11:01,600
Speaker 6: replaced by AI.

224
00:11:01,800 --> 00:11:02,840
Speaker 8: Your LinkedIn hat on.

225
00:11:05,040 --> 00:11:08,800
Speaker 5: So the closest metaphor that I've come to use to

226
00:11:08,800 --> 00:11:12,360
Speaker 5: describe this moment is it's a steam engine of the mind.

227
00:11:12,960 --> 00:11:15,560
Speaker 5: If you look at this, the former steam engine, you know,

228
00:11:15,600 --> 00:11:20,559
Speaker 5: current steam engine, although obviously majorly amplified. Industrial evolution give

229
00:11:20,600 --> 00:11:23,960
Speaker 5: us superpower of muscles, super of transport, superpower of construction,

230
00:11:24,440 --> 00:11:24,920
Speaker 5: all these ends.

231
00:11:24,960 --> 00:11:26,520
Speaker 3: Now we're going to have superpowers of the mind.

232
00:11:27,000 --> 00:11:30,640
Speaker 5: Now there's a whole bunch of very positive things that

233
00:11:30,679 --> 00:11:31,200
Speaker 5: come out of that.

234
00:11:31,280 --> 00:11:34,240
Speaker 3: I mean, all of the things we have of.

235
00:11:36,040 --> 00:11:39,280
Speaker 5: The increase in wealth that allows medicine, general education, everything

236
00:11:39,280 --> 00:11:41,520
Speaker 5: else comes out of the industrial evolution. I think the

237
00:11:41,520 --> 00:11:43,120
Speaker 5: same thing will be coming out of the steam engine

238
00:11:43,160 --> 00:11:46,720
Speaker 5: of the mind now. But the transition is going to

239
00:11:46,720 --> 00:11:50,200
Speaker 5: be difficult. The transition is going to be okay, well,

240
00:11:50,520 --> 00:11:54,160
Speaker 5: customer service jobs are going to change a whole lot now, engineers,

241
00:11:54,200 --> 00:11:56,400
Speaker 5: I think, you know, look, if you roughly kind of

242
00:11:56,400 --> 00:11:58,720
Speaker 5: go through a company and you go we ten x

243
00:11:58,720 --> 00:12:01,679
Speaker 5: each function, you say you ten x salespeople, great, so

244
00:12:01,720 --> 00:12:03,920
Speaker 5: I want to hire more salespeople want our sales ten

245
00:12:04,000 --> 00:12:07,120
Speaker 5: XT marketing people. It's like, okay, maybe the marketing functions

246
00:12:07,120 --> 00:12:10,160
Speaker 5: will change some so less order entry, more thinking about it.

247
00:12:10,160 --> 00:12:13,080
Speaker 5: But you're still in a business competitive ecosystem. You don't

248
00:12:13,120 --> 00:12:16,880
Speaker 5: want to stop marketing or not be doing it. When

249
00:12:16,880 --> 00:12:19,040
Speaker 5: you get to customer service, you get more replacement. But

250
00:12:19,160 --> 00:12:21,120
Speaker 5: here's part of the thing that I part of the

251
00:12:21,160 --> 00:12:23,040
Speaker 5: reason why I did the book impromptu and are trying

252
00:12:23,040 --> 00:12:25,640
Speaker 5: to orient people, is like, okay, so customer service people.

253
00:12:25,400 --> 00:12:27,840
Speaker 3: There's going to be a transition. Well, AI can help

254
00:12:27,880 --> 00:12:28,080
Speaker 3: with that.

255
00:12:28,440 --> 00:12:31,000
Speaker 5: You can build AIS that help figure out other kinds

256
00:12:31,000 --> 00:12:33,360
Speaker 5: of work and jobs, that can help you find them,

257
00:12:33,400 --> 00:12:35,000
Speaker 5: that can help you learn them, that can help you

258
00:12:35,080 --> 00:12:37,679
Speaker 5: do them. And so that's the thing that we as

259
00:12:37,679 --> 00:12:40,120
Speaker 5: a society need to be doing. So it's like less

260
00:12:40,160 --> 00:12:42,720
Speaker 5: like how do we slow down AI? How do we

261
00:12:42,800 --> 00:12:46,439
Speaker 5: shape it to help the broad swath of humanity in

262
00:12:46,480 --> 00:12:49,760
Speaker 5: this transition. That's where I'm trying to get the dialogue

263
00:12:49,760 --> 00:12:52,440
Speaker 5: and discussion to Moustafa.

264
00:12:52,800 --> 00:12:55,640
Speaker 7: We were just talking to Sam about Google and whether

265
00:12:55,720 --> 00:12:57,360
Speaker 7: or not it's still scary, and he's like, yeah, they're

266
00:12:57,400 --> 00:13:03,720
Speaker 7: still formidable. You quit Mind and Google AI and obviously

267
00:13:03,760 --> 00:13:06,840
Speaker 7: a lot of people have have have left Google. Is

268
00:13:06,880 --> 00:13:09,119
Speaker 7: there something wrong at Google or some sort of innovator's

269
00:13:09,120 --> 00:13:13,040
Speaker 7: dilemma there that will prevent it from truly succeeding on

270
00:13:13,120 --> 00:13:16,120
Speaker 7: generative AI because it underminds its own business model.

271
00:13:17,000 --> 00:13:22,520
Speaker 4: Look, fundamentally, AI stands intention with Google's existing business model.

272
00:13:22,840 --> 00:13:26,440
Speaker 4: It's very hard to eat yourself from within and adapt

273
00:13:26,600 --> 00:13:29,160
Speaker 4: and respond to the coming wave. So for a while

274
00:13:29,760 --> 00:13:34,319
Speaker 4: Google was just idling. You know, I was there on Lambda.

275
00:13:34,440 --> 00:13:37,080
Speaker 4: We had chat GBT before chat gibt. It was just

276
00:13:37,080 --> 00:13:41,199
Speaker 4: like a remarkable feeling internally playing with this incredible tool

277
00:13:41,360 --> 00:13:43,840
Speaker 4: and not really being able to get it out into products.

278
00:13:44,360 --> 00:13:46,760
Speaker 4: It took, you know, the launch of chat CBT to

279
00:13:46,920 --> 00:13:49,920
Speaker 4: kind of threaten Google and really shake everything up. And

280
00:13:50,320 --> 00:13:53,400
Speaker 4: you know, Google's are very formidable and you know, you know,

281
00:13:53,640 --> 00:13:57,760
Speaker 4: organization full of super smart people. So I'm sure they'll

282
00:13:57,800 --> 00:14:01,080
Speaker 4: be just fine. But I can understand why they're stuck

283
00:14:01,360 --> 00:14:05,520
Speaker 4: because nobody wants a personal AI in your pocket that

284
00:14:05,640 --> 00:14:08,920
Speaker 4: is actually funded by ads. Right, you don't want a

285
00:14:08,960 --> 00:14:11,120
Speaker 4: salesman in your pocket that is trying to persuade you

286
00:14:11,200 --> 00:14:13,040
Speaker 4: to go and buy more or do xyz.

287
00:14:13,640 --> 00:14:15,199
Speaker 3: You need fiduciary alignment.

288
00:14:15,559 --> 00:14:18,280
Speaker 4: Right, Your AI has to be on your team, and

289
00:14:18,280 --> 00:14:20,440
Speaker 4: that means ultimately you have to pay for it. And

290
00:14:20,440 --> 00:14:22,760
Speaker 4: if somebody else is paying for it, you have to

291
00:14:22,800 --> 00:14:25,800
Speaker 4: ask yourself what are they trying to persuade you of?

292
00:14:26,080 --> 00:14:28,840
Speaker 4: Are they trying to influence you in some way? And

293
00:14:28,880 --> 00:14:31,080
Speaker 4: I think people are alive to that now and that's

294
00:14:31,120 --> 00:14:33,840
Speaker 4: a real challenge for Google because it's very unclear how

295
00:14:33,880 --> 00:14:35,480
Speaker 4: they're going to manage this transition.

296
00:14:35,720 --> 00:14:39,160
Speaker 7: So we asked Sam if Google scares him, does OpenAI

297
00:14:39,280 --> 00:14:39,760
Speaker 7: scare you?

298
00:14:40,520 --> 00:14:41,880
Speaker 8: Wastaf a first many.

299
00:14:43,240 --> 00:14:48,200
Speaker 4: So no, No, it definitely doesn't. I mean, actually, this

300
00:14:48,320 --> 00:14:53,400
Speaker 4: morning we've announced our new large language model called Inflection one,

301
00:14:54,080 --> 00:14:56,200
Speaker 4: and we set out to build a model that was

302
00:14:56,560 --> 00:14:59,200
Speaker 4: fast enough and more capable than every other model on

303
00:14:59,240 --> 00:15:03,040
Speaker 4: the market for our compute class. So we're very proud that,

304
00:15:03,160 --> 00:15:05,440
Speaker 4: you know, a year on from our launch, we're now

305
00:15:05,480 --> 00:15:10,240
Speaker 4: better than lama Chat, GPT, Palm five forty, Chinchilla, all

306
00:15:10,240 --> 00:15:13,120
Speaker 4: of the other models of our size. And that's super

307
00:15:13,160 --> 00:15:15,880
Speaker 4: important because it powers PI and ultimately it will be

308
00:15:15,880 --> 00:15:19,520
Speaker 4: available as a conversational API, so you can obviously play

309
00:15:19,520 --> 00:15:22,120
Speaker 4: with it at pie dot ai now. And you know,

310
00:15:22,520 --> 00:15:25,400
Speaker 4: I think that demonstrates that with a team of thirty

311
00:15:25,400 --> 00:15:28,520
Speaker 4: five people, you know, in a pretty short order, we're

312
00:15:28,560 --> 00:15:31,680
Speaker 4: able to exceed the cutting edge and now build an

313
00:15:31,760 --> 00:15:34,920
Speaker 4: absolute best in class AI, which is very exciting. And

314
00:15:34,920 --> 00:15:37,560
Speaker 4: obviously we've also managed to do that because we've been

315
00:15:37,600 --> 00:15:40,320
Speaker 4: able to gather a huge amount of capital and some

316
00:15:40,400 --> 00:15:44,000
Speaker 4: great investors and train. In fact, what we now have

317
00:15:44,200 --> 00:15:47,920
Speaker 4: is the largest operational cluster in the world of h

318
00:15:48,000 --> 00:15:52,200
Speaker 4: one hundreds in Vidia's latest chip and that's a huge advantage.

319
00:15:52,320 --> 00:15:52,640
Speaker 3: Okay.

320
00:15:52,880 --> 00:15:55,240
Speaker 5: And one of the things, because I was listening to

321
00:15:55,280 --> 00:15:57,600
Speaker 5: your conversation with Sam, and you know, it's obviously an

322
00:15:57,640 --> 00:15:59,800
Speaker 5: awkward question when you ask him, is like, well, should

323
00:15:59,800 --> 00:16:02,440
Speaker 5: we trust you? Because it's like, yes, trust me. It's like, well,

324
00:16:02,440 --> 00:16:03,360
Speaker 5: that's weird because if.

325
00:16:03,240 --> 00:16:05,720
Speaker 3: Someone says no, you shouldn't.

326
00:16:05,480 --> 00:16:08,120
Speaker 5: Someone says yes, trust me, You're like, well wait a minute,

327
00:16:08,440 --> 00:16:10,560
Speaker 5: but look, I uh, the Open Eye people are really

328
00:16:10,600 --> 00:16:12,280
Speaker 5: great people. I think it's it's governed by a five

329
00:16:12,320 --> 00:16:14,480
Speaker 5: to one C three. There's frequently a bunch of fud

330
00:16:14,920 --> 00:16:19,160
Speaker 5: social media stuff that that that that that that kind

331
00:16:19,160 --> 00:16:23,440
Speaker 5: of obscures that I you know, spent a number of

332
00:16:23,480 --> 00:16:26,160
Speaker 5: years on the board and in the team. I still

333
00:16:26,160 --> 00:16:30,040
Speaker 5: work with them and help them. They are fully paying

334
00:16:30,080 --> 00:16:33,000
Speaker 5: attention to every serious question. I mean, like, for example,

335
00:16:33,080 --> 00:16:35,480
Speaker 5: there Sam went and spent a month. It's crazy, You're

336
00:16:35,480 --> 00:16:38,200
Speaker 5: doing a startup. It spent a month doing a world tour.

337
00:16:38,040 --> 00:16:38,560
Speaker 3: Talking to people.

338
00:16:38,600 --> 00:16:40,480
Speaker 5: Say look, I'm here, you can talk to me about

339
00:16:40,520 --> 00:16:42,640
Speaker 5: your concerns. I want to make sure it's good for humanity.

340
00:16:42,720 --> 00:16:45,320
Speaker 5: I don't just care about this kind of US San

341
00:16:45,360 --> 00:16:46,320
Speaker 5: Francisco text scene.

342
00:16:46,360 --> 00:16:48,080
Speaker 3: I care about what it's impact on humanity is.

343
00:16:48,360 --> 00:16:50,360
Speaker 5: And so I have come here to talk to you,

344
00:16:50,440 --> 00:16:52,040
Speaker 5: So that kind of like.

345
00:16:52,040 --> 00:16:53,120
Speaker 3: I'm engaging conversation.

346
00:16:53,280 --> 00:16:56,040
Speaker 5: I care about this is the kind of reason in

347
00:16:56,040 --> 00:16:58,680
Speaker 5: addition to a five ones three emission structure and everything else.

348
00:16:58,880 --> 00:17:01,240
Speaker 3: And so no, I'm I'm delighted with Hope.

349
00:17:01,280 --> 00:17:02,840
Speaker 6: I don't see how he's getting any work done with

350
00:17:02,880 --> 00:17:07,360
Speaker 6: all this, But I want to change gears quickly. We're

351
00:17:07,359 --> 00:17:09,240
Speaker 6: going to throw some Twitter polls up on the screen

352
00:17:09,359 --> 00:17:13,000
Speaker 6: today and we asked Twitter users where the greatest advancements

353
00:17:13,080 --> 00:17:14,240
Speaker 6: in AI would come from.

354
00:17:14,280 --> 00:17:16,320
Speaker 3: We'll see the results, But I want to ask.

355
00:17:16,200 --> 00:17:19,800
Speaker 6: About maybe retarting with you are the US China policy

356
00:17:20,040 --> 00:17:24,399
Speaker 6: of divestment is, can we reasonably hope that you know,

357
00:17:24,520 --> 00:17:28,960
Speaker 6: doing that will maybe prolong America's advantage in AI?

358
00:17:29,160 --> 00:17:32,520
Speaker 5: Is it a smart strategy? I don't think divestment is

359
00:17:32,560 --> 00:17:35,720
Speaker 5: a smart strategy. I think staying connected is better for

360
00:17:35,920 --> 00:17:39,920
Speaker 5: both the US and China and the world. I think

361
00:17:40,280 --> 00:17:41,760
Speaker 5: competition is very good.

362
00:17:41,880 --> 00:17:44,840
Speaker 3: I think the fact that there's various ways that we

363
00:17:45,760 --> 00:17:46,040
Speaker 3: in the.

364
00:17:46,040 --> 00:17:48,679
Speaker 5: US and the West Coast have a AI lead, I

365
00:17:48,680 --> 00:17:50,719
Speaker 5: think that's a great thing for kind of the values,

366
00:17:50,760 --> 00:17:53,520
Speaker 5: the ecosystem and the kind of the great world order

367
00:17:53,560 --> 00:17:55,399
Speaker 5: that the US should be very proud of in the

368
00:17:55,480 --> 00:17:57,720
Speaker 5: last seventy years. Like, there's lots of things to be

369
00:17:57,720 --> 00:18:02,200
Speaker 5: critical of two, of course, but as a really peaceful time,

370
00:18:02,359 --> 00:18:04,520
Speaker 5: trying to make it business and interconnection is a really

371
00:18:04,560 --> 00:18:08,000
Speaker 5: good thing. And I'm a very strong believer and proponent

372
00:18:08,040 --> 00:18:11,160
Speaker 5: of that. But I think investments not the.

373
00:18:11,200 --> 00:18:12,879
Speaker 6: Ends half of You write in the book that China

374
00:18:12,920 --> 00:18:16,000
Speaker 6: has an explicit national strategy to be the AI leader

375
00:18:16,040 --> 00:18:16,800
Speaker 6: by twenty thirty.

376
00:18:16,840 --> 00:18:21,000
Speaker 3: So do you agree with read that withdrawal divestment as

377
00:18:21,040 --> 00:18:21,920
Speaker 3: a smart strategy.

378
00:18:22,040 --> 00:18:24,600
Speaker 4: I mean, China is already ahead of its own schedule.

379
00:18:24,720 --> 00:18:28,560
Speaker 4: I mean it's publishing significantly more papers on AI than

380
00:18:28,680 --> 00:18:31,359
Speaker 4: we are collectively in the rest of the world. So

381
00:18:31,760 --> 00:18:35,600
Speaker 4: you know, I do think that our export controls were

382
00:18:35,640 --> 00:18:39,080
Speaker 4: effectively a declaration of economic war on China. They were

383
00:18:39,200 --> 00:18:42,240
Speaker 4: very firm and very aggressive and I think it sets

384
00:18:42,320 --> 00:18:45,199
Speaker 4: us out on a hyper adversarial footing. They have a

385
00:18:45,280 --> 00:18:48,160
Speaker 4: huge number of leavers too, and we should expect them

386
00:18:48,200 --> 00:18:50,360
Speaker 4: to use them against us pretty soon, unfortunately.

387
00:18:50,800 --> 00:18:53,240
Speaker 7: So we talked a lot about regulation with Sam, and

388
00:18:53,280 --> 00:18:56,760
Speaker 7: obviously President Biden was just in town here this week

389
00:18:56,840 --> 00:19:00,640
Speaker 7: talking about AI. He met with you know, the President

390
00:19:00,680 --> 00:19:06,840
Speaker 7: and Sindar and Satya in DC. What is the prospect

391
00:19:06,920 --> 00:19:12,840
Speaker 7: for real government oversight that protects users, protects the economy

392
00:19:14,040 --> 00:19:15,280
Speaker 7: and our political process.

393
00:19:15,840 --> 00:19:18,159
Speaker 5: So, look, I think the good news is that the

394
00:19:18,240 --> 00:19:22,040
Speaker 5: administration is taking a very Let's learn how to be

395
00:19:22,080 --> 00:19:25,280
Speaker 5: smart about this. Let's assemble a set of thoughtful people

396
00:19:25,320 --> 00:19:30,919
Speaker 5: from industry yesterday, assembling a thoughtful people from outside of industry, academia,

397
00:19:30,920 --> 00:19:31,640
Speaker 5: other kinds of places.

398
00:19:31,720 --> 00:19:33,080
Speaker 3: We're doing it, and let's learn about it.

399
00:19:33,359 --> 00:19:37,000
Speaker 5: I think they have a number of good people tasked

400
00:19:37,040 --> 00:19:41,639
Speaker 5: with this, and so I'm optimistic in that. But on

401
00:19:41,680 --> 00:19:43,320
Speaker 5: the other hand, of course, a little bit like your

402
00:19:43,359 --> 00:19:47,200
Speaker 5: interview with Say, I'm you know, regulatory things is something

403
00:19:47,240 --> 00:19:49,399
Speaker 5: that's very easy to get wrong. So you need to

404
00:19:49,400 --> 00:19:51,560
Speaker 5: be cautious about how you do it. You want it

405
00:19:51,600 --> 00:19:55,159
Speaker 5: to be having that positive impact, not for example, regulatory

406
00:19:55,160 --> 00:19:56,840
Speaker 5: capture and not a bunch of other things, and so

407
00:19:56,840 --> 00:19:59,000
Speaker 5: I think you have to be careful about how you do.

408
00:19:59,119 --> 00:20:02,600
Speaker 4: I think it's adding that this time last year, almost

409
00:20:02,640 --> 00:20:06,200
Speaker 4: no world leaders were talking about AI. I mean, I

410
00:20:06,240 --> 00:20:08,760
Speaker 4: don't think there has ever been a technology trajectory in

411
00:20:08,880 --> 00:20:14,040
Speaker 4: history that has gone from zero recognition to almost universal recognition.

412
00:20:14,359 --> 00:20:15,119
Speaker 3: And that's a good thing.

413
00:20:15,160 --> 00:20:17,320
Speaker 4: And I think it's in part because sort of we

414
00:20:17,440 --> 00:20:20,199
Speaker 4: collectively in industry have been trying to advocate and say, like,

415
00:20:20,240 --> 00:20:22,720
Speaker 4: this is really serious, we should pay attention and invite

416
00:20:22,760 --> 00:20:26,560
Speaker 4: the conversation wherever it ends up with respect to regulating

417
00:20:26,600 --> 00:20:28,720
Speaker 4: existential risk or more near term threats.

418
00:20:29,000 --> 00:20:31,040
Speaker 6: Okay, I think we have to leave it there and

419
00:20:31,359 --> 00:20:33,800
Speaker 6: Shaper really thank you guys for joining us.

420
00:20:33,800 --> 00:20:37,760
Speaker 5: Great to see you guys, Thanks a lot, great to see.

421
00:20:39,520 --> 00:20:43,280
Speaker 2: Greylock Partners, Reid Hoffman, the STAFFA Sillyman, the co founder

422
00:20:43,320 --> 00:20:47,120
Speaker 2: and CEO of Inflection AI. They're here at the Bloomberg

423
00:20:47,160 --> 00:20:50,520
Speaker 2: Technology Something in San Francisco, Caroline so much to recap.

424
00:20:50,800 --> 00:20:53,040
Speaker 2: What they kind of had in common was the idea

425
00:20:53,080 --> 00:20:57,359
Speaker 2: that AIS a technology amplifies or reflects the user, and

426
00:20:57,480 --> 00:20:59,679
Speaker 2: in society there are good people and they are a

427
00:20:59,680 --> 00:21:03,119
Speaker 2: bad people, but they you know, they weren't tempted to

428
00:21:03,240 --> 00:21:06,920
Speaker 2: give kind of the doomsday scenario that perhaps Outman gave

429
00:21:07,000 --> 00:21:07,479
Speaker 2: us earlier.

430
00:21:07,680 --> 00:21:10,639
Speaker 1: And also the EQ versus IQ. This is obviously the

431
00:21:10,680 --> 00:21:14,400
Speaker 1: differentiating factor for PIE what they're currently working on and inflection,

432
00:21:14,800 --> 00:21:16,840
Speaker 1: but notable that they don't feel that there's just one

433
00:21:16,920 --> 00:21:19,919
Speaker 1: chatbot to rule us all that chatchipt isn't going to

434
00:21:19,960 --> 00:21:22,840
Speaker 1: be the steadfast and only use case. And already we're

435
00:21:22,840 --> 00:21:25,440
Speaker 1: discussing that it feels like chatchapbet is sort of sucktible

436
00:21:25,480 --> 00:21:26,040
Speaker 1: the oxygen.

437
00:21:26,119 --> 00:21:27,159
Speaker 8: But we know there's barred.

438
00:21:27,160 --> 00:21:30,240
Speaker 1: We know that there's obviously all these other more focused,

439
00:21:30,240 --> 00:21:34,359
Speaker 1: specific AI chatbots being developed, and some with perhaps more

440
00:21:34,400 --> 00:21:35,200
Speaker 1: emotion than others.

441
00:21:35,280 --> 00:21:38,760
Speaker 2: In Yeah, and Masafa Silliman describing AI development as a

442
00:21:38,800 --> 00:21:42,960
Speaker 2: meritocracy was interesting. Let's get back to the Bloomberg Technology

443
00:21:43,000 --> 00:21:46,680
Speaker 2: Summit where Bloomberg's Bradstone is on stage with the Amazon

444
00:21:46,680 --> 00:21:49,680
Speaker 2: Web Services AWS CEO Adam Selipsky.

445
00:21:49,960 --> 00:21:53,399
Speaker 9: Well, I mean, as you implied, today, we are by

446
00:21:53,800 --> 00:21:57,159
Speaker 9: far the most broadly adopted cloud in the world, with

447
00:21:57,520 --> 00:22:02,120
Speaker 9: the broadest set of capabilities, and I think that generative

448
00:22:02,160 --> 00:22:06,520
Speaker 9: AI is both incredibly explosive and transformative set of technologies

449
00:22:06,840 --> 00:22:09,960
Speaker 9: in of itself, and it's fully dependent on the.

450
00:22:09,960 --> 00:22:11,160
Speaker 8: Cloud to be successful.

451
00:22:11,520 --> 00:22:14,720
Speaker 9: So if you look at the massive amount of compute

452
00:22:14,840 --> 00:22:18,040
Speaker 9: that's required, never mind any other IT related stuff, but

453
00:22:18,480 --> 00:22:21,480
Speaker 9: just the massive amount of compute required that's going to

454
00:22:21,560 --> 00:22:25,239
Speaker 9: happen really predominantly in the cloud, and companies are going

455
00:22:25,280 --> 00:22:27,719
Speaker 9: to want to view gener AI as part of an

456
00:22:27,920 --> 00:22:31,520
Speaker 9: entire data strategy and data platform. And you're going to

457
00:22:31,560 --> 00:22:34,400
Speaker 9: want to do your generative AI you know where you've

458
00:22:34,400 --> 00:22:36,800
Speaker 9: got your data. And you're also going to want the

459
00:22:36,840 --> 00:22:42,280
Speaker 9: same bulletproof enterprise security privacy that you expect from any

460
00:22:42,320 --> 00:22:46,639
Speaker 9: other cloud service. And because so many customers it's certainly

461
00:22:46,640 --> 00:22:49,919
Speaker 9: more than on any other cloud, have that data on AWS,

462
00:22:50,000 --> 00:22:53,760
Speaker 9: use aabus for security and operational excellence. I think that

463
00:22:53,800 --> 00:22:59,840
Speaker 9: they are going to justifiably demand that AWS have a full, powerful,

464
00:23:00,040 --> 00:23:01,720
Speaker 9: a suite of generative AI services.

465
00:23:01,720 --> 00:23:06,159
Speaker 6: Okay, but do you need a popular LM or an

466
00:23:06,200 --> 00:23:10,399
Speaker 6: exclusive partner running on AWS as it does seem like

467
00:23:10,480 --> 00:23:12,920
Speaker 6: microsoftening Google have. I mean, I guess I'm asking who's

468
00:23:12,960 --> 00:23:14,240
Speaker 6: your horse in the race right now?

469
00:23:14,400 --> 00:23:17,800
Speaker 9: Right Well, I think, with all due respect, I think

470
00:23:17,840 --> 00:23:21,520
Speaker 9: that's the wrong question. It'd be like in nineteen ninety seven,

471
00:23:21,960 --> 00:23:24,360
Speaker 9: when the Internet's happening and everything's kind of going nuts

472
00:23:24,359 --> 00:23:26,480
Speaker 9: around us. You and I sitting around saying who's the

473
00:23:26,520 --> 00:23:29,359
Speaker 9: internet company going to be? It kind of seems like

474
00:23:29,400 --> 00:23:31,359
Speaker 9: a silly question, right, And by the way, the leading

475
00:23:31,400 --> 00:23:34,240
Speaker 9: search company then was Alta Vista, and I guarantee you

476
00:23:34,240 --> 00:23:35,320
Speaker 9: my kids in their great I.

477
00:23:35,359 --> 00:23:36,359
Speaker 8: Love it of Alta Vista.

478
00:23:36,760 --> 00:23:40,119
Speaker 9: So it's not okay you're in a We're who's ahead,

479
00:23:40,119 --> 00:23:42,880
Speaker 9: which runners ahead in the race after three steps because

480
00:23:42,880 --> 00:23:46,119
Speaker 9: it's a ten k race. You know, what everybody needs

481
00:23:46,160 --> 00:23:50,159
Speaker 9: now is is experimentation. They need choice, They need democratization

482
00:23:50,720 --> 00:23:53,960
Speaker 9: of generative AI. And just like AWS was founded to

483
00:23:54,000 --> 00:23:58,800
Speaker 9: democratize it, we aim to democratize generative AI. So we're

484
00:23:58,840 --> 00:24:02,760
Speaker 9: operating at all layers of stack we have. We've had

485
00:24:02,760 --> 00:24:06,000
Speaker 9: our own custom chip program for a decade now, way

486
00:24:06,040 --> 00:24:08,879
Speaker 9: longer than anybody else, and we have not one, but

487
00:24:09,000 --> 00:24:14,000
Speaker 9: two families of chips custom designed for machine learning training

488
00:24:14,160 --> 00:24:17,520
Speaker 9: for training and inferential for running inference and then and

489
00:24:17,520 --> 00:24:18,960
Speaker 9: so those are for people who are going.

490
00:24:18,880 --> 00:24:20,360
Speaker 8: To build models, train models.

491
00:24:21,200 --> 00:24:24,000
Speaker 9: Then most of our customers will interact with Amazon Bedrock,

492
00:24:24,440 --> 00:24:31,200
Speaker 9: which is a managed service for accessing deploying managing models,

493
00:24:31,520 --> 00:24:34,239
Speaker 9: and here's where the choice comes in. So we are

494
00:24:34,280 --> 00:24:36,480
Speaker 9: going to have our own models, our own lllms. So

495
00:24:37,359 --> 00:24:41,840
Speaker 9: Amazon models lms have been running introduction inside of Amazon

496
00:24:41,880 --> 00:24:44,600
Speaker 9: for a long time now. Parts of our retail search

497
00:24:44,640 --> 00:24:47,800
Speaker 9: are powered by llms. A lot of Alexa's voice responses

498
00:24:48,119 --> 00:24:50,719
Speaker 9: are lll em powered, and we're taking right now, we're

499
00:24:50,760 --> 00:24:53,840
Speaker 9: in the process of taking those lllms. We're making them bigger,

500
00:24:53,920 --> 00:24:56,600
Speaker 9: we're externalizing them, and later this year those will be

501
00:24:56,920 --> 00:24:58,280
Speaker 9: exposed for everybody to use.

502
00:24:58,320 --> 00:25:00,440
Speaker 8: A Titan broke exactly.

503
00:25:00,480 --> 00:25:03,040
Speaker 9: Well, let me ask you so, mate, But that's just

504
00:25:03,160 --> 00:25:05,000
Speaker 9: one model, one set of models.

505
00:25:05,040 --> 00:25:05,640
Speaker 8: I should say.

506
00:25:05,760 --> 00:25:10,359
Speaker 9: We're also exposing anthropic inside of Bedrock Stability AI AI

507
00:25:10,440 --> 00:25:13,400
Speaker 9: twenty one, and I think a lot of others over time,

508
00:25:13,520 --> 00:25:18,000
Speaker 9: because nobody knows. Anybody who knows which model is going

509
00:25:18,080 --> 00:25:20,600
Speaker 9: to be the winner is asking the wrong question. People

510
00:25:20,680 --> 00:25:22,760
Speaker 9: need to experiment, and we want to provide that choice.

511
00:25:22,800 --> 00:25:24,680
Speaker 3: Okay, so pretend you're talking to a customer.

512
00:25:24,680 --> 00:25:28,040
Speaker 6: Now make the argument, why should they use Titan or

513
00:25:28,040 --> 00:25:33,040
Speaker 6: one of these other lfs you're exposing rather than GPT four,

514
00:25:33,080 --> 00:25:36,119
Speaker 6: which now does seem to have a significant.

515
00:25:35,680 --> 00:25:39,200
Speaker 9: Leadership position well, I don't know which model they should

516
00:25:39,280 --> 00:25:42,560
Speaker 9: depends who they are, and it depends what their application is.

517
00:25:42,720 --> 00:25:45,400
Speaker 9: And some of them probably will want to use GPT,

518
00:25:45,920 --> 00:25:47,439
Speaker 9: and some of them will want to use Titan, and

519
00:25:47,480 --> 00:25:50,359
Speaker 9: some of them will want to use Anthropic. And I

520
00:25:50,359 --> 00:25:55,199
Speaker 9: think it's preposterous to me to think that one model

521
00:25:55,240 --> 00:25:58,119
Speaker 9: or one company is going to be the solution for

522
00:25:58,200 --> 00:26:01,399
Speaker 9: every application and every company up there. So we're already

523
00:26:01,400 --> 00:26:04,399
Speaker 9: seeing this heterogeneity. So we're seeing an explosion of interest

524
00:26:04,440 --> 00:26:09,679
Speaker 9: in Bedrock and Amazon's genitive AI capabilities. Just this morning, Omnicom,

525
00:26:09,960 --> 00:26:12,840
Speaker 9: one of the largest advertising communications firms in the world,

526
00:26:13,160 --> 00:26:17,320
Speaker 9: announced they're working with the AWS using Bedrock as well

527
00:26:17,359 --> 00:26:23,159
Speaker 9: as our custom chips inside of our compute capacity to

528
00:26:23,640 --> 00:26:26,879
Speaker 9: do generative AI going forward. Earlier this week, BBVA, one

529
00:26:26,880 --> 00:26:29,560
Speaker 9: of the largest financial services firms in the world, announced

530
00:26:29,560 --> 00:26:32,159
Speaker 9: they're working with Amazon on jenera of AI, you know,

531
00:26:32,240 --> 00:26:36,200
Speaker 9: with Bedrock. And it's that choice combined with the enterprise

532
00:26:36,240 --> 00:26:39,240
Speaker 9: security and privacy, which I think are so fundamental.

533
00:26:39,440 --> 00:26:42,760
Speaker 6: I can humbly admit, as a journalists covering tech for

534
00:26:43,280 --> 00:26:46,720
Speaker 6: way too long that in the fall I was surprised

535
00:26:46,760 --> 00:26:50,600
Speaker 6: by not only the quality of check GBT but the

536
00:26:50,640 --> 00:26:55,800
Speaker 6: customer response to this new wave of technology, were you surprised?

537
00:26:56,440 --> 00:26:57,440
Speaker 8: Well, I think that.

538
00:26:58,080 --> 00:27:01,600
Speaker 9: I mean, folks working in the area and AI have

539
00:27:01,720 --> 00:27:04,440
Speaker 9: known about large language models for a long time, and

540
00:27:04,920 --> 00:27:08,040
Speaker 9: very few companies have more experience with AI than Amazon.

541
00:27:08,119 --> 00:27:11,280
Speaker 9: I mean, nineteen ninety eight personalization on the Amazon website

542
00:27:11,440 --> 00:27:12,000
Speaker 9: that was AI.

543
00:27:12,119 --> 00:27:13,320
Speaker 8: Okay, still hold AI.

544
00:27:14,119 --> 00:27:17,040
Speaker 9: Sage Maker twenty seventeen not used by over one hundred

545
00:27:17,119 --> 00:27:21,240
Speaker 9: thousand AWS customers. Most machine learning in the cloud, any cloud,

546
00:27:21,600 --> 00:27:24,240
Speaker 9: happens on stage Maker, so we have a lot of

547
00:27:24,280 --> 00:27:26,480
Speaker 9: experience with a lot of people working on llms. I

548
00:27:26,480 --> 00:27:28,880
Speaker 9: think the whole world was surprised that when three point

549
00:27:28,960 --> 00:27:32,400
Speaker 9: five came out, it was such a dramatic improvement and responses.

550
00:27:32,520 --> 00:27:35,359
Speaker 9: Not perfect by any means, but a dramatic improvement and

551
00:27:35,440 --> 00:27:38,159
Speaker 9: responses over three point zero. So I think that was

552
00:27:38,160 --> 00:27:40,840
Speaker 9: the surprise, but not the overall arc.

553
00:27:41,000 --> 00:27:41,600
Speaker 3: But the.

554
00:27:43,920 --> 00:27:46,439
Speaker 9: Good news for our customers is that you know, we

555
00:27:46,440 --> 00:27:49,439
Speaker 9: have deep, deep expertise and Ai've been working on different

556
00:27:49,440 --> 00:27:51,760
Speaker 9: forms of it for a long time and are now

557
00:27:52,560 --> 00:27:55,920
Speaker 9: pouring enormous resources into the generative part of it.

558
00:27:56,119 --> 00:27:58,600
Speaker 6: Okay, I'll give this up after this, I promise, But

559
00:27:59,440 --> 00:28:01,720
Speaker 6: would you can see that Amazon's playing a little bit

560
00:28:01,760 --> 00:28:02,840
Speaker 6: from behind right now.

561
00:28:03,359 --> 00:28:06,440
Speaker 9: No, I really don't think so. I mean again, it's

562
00:28:06,480 --> 00:28:09,040
Speaker 9: the race analogy. If like, are we really going to

563
00:28:09,040 --> 00:28:12,000
Speaker 9: have a conversation about three steps into a ten k raise?

564
00:28:12,040 --> 00:28:14,479
Speaker 9: You know who's in what position. It's about the long term.

565
00:28:14,640 --> 00:28:17,320
Speaker 9: Amazon has always taken a much more long term view

566
00:28:17,320 --> 00:28:20,359
Speaker 9: of the world than almost any other company we're I

567
00:28:20,359 --> 00:28:22,440
Speaker 9: think the key is we're building in multiple layers of

568
00:28:22,480 --> 00:28:25,080
Speaker 9: the stack because we understand that's what customers need. We're

569
00:28:25,080 --> 00:28:28,240
Speaker 9: also building applications on top of these models. So we've

570
00:28:28,240 --> 00:28:31,880
Speaker 9: released code Whisperer, which is a coding companion on top

571
00:28:31,920 --> 00:28:35,159
Speaker 9: of the Amazon LLLMS, and that's a you know, you

572
00:28:35,200 --> 00:28:38,560
Speaker 9: type in words, it gives you back code in internal

573
00:28:38,640 --> 00:28:42,920
Speaker 9: tests kind of coding challenges. Developers finish their task on

574
00:28:42,960 --> 00:28:46,000
Speaker 9: average fifty seven percent faster than those not.

575
00:28:46,040 --> 00:28:47,080
Speaker 8: Using code Whisperer.

576
00:28:47,320 --> 00:28:51,000
Speaker 9: Plus, it's very secure, private tells you what open source

577
00:28:51,040 --> 00:28:53,640
Speaker 9: you're using and what licensing restrictions there might be on it,

578
00:28:53,680 --> 00:28:56,400
Speaker 9: which not every other solution out there does. So I

579
00:28:56,440 --> 00:28:58,880
Speaker 9: think we're very confident. If you look at the thing

580
00:28:58,880 --> 00:29:00,800
Speaker 9: which the only thing which ought to give us confidence,

581
00:29:00,840 --> 00:29:05,560
Speaker 9: This customer response. I mentioned Omnicom, I mentioned BBVA. This week,

582
00:29:05,880 --> 00:29:08,360
Speaker 9: Old Mutual, which is one of the largest financial services

583
00:29:08,360 --> 00:29:11,440
Speaker 9: companies in Africa, is going all in on AWS and

584
00:29:11,560 --> 00:29:16,280
Speaker 9: using US for generative AI and really exciting developments to come.

585
00:29:16,320 --> 00:29:19,840
Speaker 9: We're talking to one company who has millions of lines

586
00:29:20,080 --> 00:29:24,480
Speaker 9: of mainframe code and they're talking to us about, you know,

587
00:29:24,600 --> 00:29:28,800
Speaker 9: moving over these really gnarly mainframe applications and millions of

588
00:29:28,800 --> 00:29:33,360
Speaker 9: lines of code using generative AI from from from from AWS.

589
00:29:33,480 --> 00:29:36,160
Speaker 9: So I think, you know it's it's the this this

590
00:29:36,240 --> 00:29:39,360
Speaker 9: consumer application, this chat application, which is so easy for

591
00:29:39,400 --> 00:29:41,880
Speaker 9: folks to understand because they can say, give me a

592
00:29:41,960 --> 00:29:45,800
Speaker 9: hikup about you know, farm machinery and does it.

593
00:29:45,960 --> 00:29:47,280
Speaker 3: That's keol hai cod.

594
00:29:47,920 --> 00:29:48,480
Speaker 8: It's cool.

595
00:29:48,800 --> 00:29:51,560
Speaker 9: But uh, you know what I think folks in this

596
00:29:51,680 --> 00:29:54,360
Speaker 9: room and watching online understand is that there is a

597
00:29:54,480 --> 00:29:59,680
Speaker 9: full suite of enterprise and company and organizational applications and

598
00:29:59,720 --> 00:30:02,840
Speaker 9: there's going to be huge needs there. And so you know,

599
00:30:02,840 --> 00:30:06,760
Speaker 9: we're going to be focused on customer service and on coding,

600
00:30:07,200 --> 00:30:11,600
Speaker 9: and on drug discovery and on wealth management, providing better

601
00:30:11,640 --> 00:30:15,000
Speaker 9: solutions for customers and a full suite of applications across

602
00:30:15,000 --> 00:30:17,640
Speaker 9: every industry. And I think we're very well positioned there and.

603
00:30:17,720 --> 00:30:20,320
Speaker 6: You feel like you have or are close to having

604
00:30:20,920 --> 00:30:24,280
Speaker 6: a GPT four quality.

605
00:30:24,000 --> 00:30:25,400
Speaker 3: Model running on AWS.

606
00:30:26,200 --> 00:30:28,440
Speaker 9: All of the tests that we've done, as well as

607
00:30:28,480 --> 00:30:30,760
Speaker 9: more and more customers who are in who are in

608
00:30:30,800 --> 00:30:36,400
Speaker 9: the existing current preview of Bedrock, have been very impressed

609
00:30:36,440 --> 00:30:39,560
Speaker 9: with the quality of our models. And of course, again

610
00:30:39,600 --> 00:30:41,680
Speaker 9: it's not just about our models. We're going to be

611
00:30:41,680 --> 00:30:44,440
Speaker 9: proud of our models, I predict, but Anthropic does an

612
00:30:44,480 --> 00:30:46,640
Speaker 9: amazing job and they're you know, right up there in

613
00:30:46,760 --> 00:30:47,640
Speaker 9: quality with any.

614
00:30:47,480 --> 00:30:48,200
Speaker 8: Model in the world.

615
00:30:48,520 --> 00:30:52,680
Speaker 9: Stability AI big leader for generating for models, generating images,

616
00:30:52,760 --> 00:30:57,160
Speaker 9: so collectively, I think these models will provide absolutely the

617
00:30:57,200 --> 00:31:01,960
Speaker 9: best destination, all with a consistent aid PI set, consistent

618
00:31:02,000 --> 00:31:06,560
Speaker 9: AWS security, consistent identity system, all in a private, isolated,

619
00:31:06,640 --> 00:31:09,800
Speaker 9: virtual private cloud, none of this. Hey, here's an application,

620
00:31:09,920 --> 00:31:12,080
Speaker 9: and now you have to have a bunch of fortune

621
00:31:12,120 --> 00:31:15,320
Speaker 9: five hundred CIOs a ban from their companies, which is

622
00:31:15,360 --> 00:31:18,760
Speaker 9: what's happened. You know, from day one it's always going

623
00:31:18,800 --> 00:31:20,360
Speaker 9: to be AWS class security.

624
00:31:21,320 --> 00:31:23,720
Speaker 6: Okay, I want to put a slide up showing a

625
00:31:23,760 --> 00:31:30,440
Speaker 6: Bloomberg Intelligence estimate on projected generative AI revenues over.

626
00:31:30,280 --> 00:31:32,760
Speaker 3: The next, I think it's maybe.

627
00:31:32,320 --> 00:31:35,520
Speaker 6: Ten, yeah, five years, I'm not saying, and who knows

628
00:31:35,600 --> 00:31:37,560
Speaker 6: what you know about the numbers that far out, but

629
00:31:37,560 --> 00:31:39,880
Speaker 6: it's up into the right and I just I can

630
00:31:39,960 --> 00:31:42,560
Speaker 6: imagine a lot of financial types who are out there,

631
00:31:42,880 --> 00:31:45,280
Speaker 6: you know, looking at Amazon's numbers and wondering how much

632
00:31:45,320 --> 00:31:49,000
Speaker 6: of a sales tailwind this will be for AWS. So

633
00:31:49,040 --> 00:31:51,600
Speaker 6: what can you tell us about you know, you mentioned

634
00:31:51,640 --> 00:31:55,760
Speaker 6: it's so computationally intensive, what you see the impact being

635
00:31:55,760 --> 00:31:56,760
Speaker 6: for AWS?

636
00:31:57,320 --> 00:32:00,920
Speaker 9: Well, it's it's really early, so you know, I prognostication

637
00:32:01,120 --> 00:32:03,120
Speaker 9: is fun and probably important at the end of the day,

638
00:32:03,160 --> 00:32:05,120
Speaker 9: but I think it's also important to be humble and

639
00:32:05,160 --> 00:32:10,040
Speaker 9: to be nimble anything ending in ill and to understand

640
00:32:10,040 --> 00:32:12,560
Speaker 9: that we're going to have to all adjust rapidly. But

641
00:32:12,800 --> 00:32:16,120
Speaker 9: that being said, I think that well, look, I don't

642
00:32:16,120 --> 00:32:18,840
Speaker 9: think any of the fundamentals about cloud computing have change,

643
00:32:18,920 --> 00:32:22,240
Speaker 9: and probably who knows, call a ten percent of it

644
00:32:22,800 --> 00:32:25,719
Speaker 9: has moved to the cloud. So we're still very very early,

645
00:32:26,120 --> 00:32:30,440
Speaker 9: and whether you're talking about any application, there's still you know,

646
00:32:30,800 --> 00:32:33,320
Speaker 9: massive runway for things to move to the cloud, and

647
00:32:33,840 --> 00:32:36,840
Speaker 9: we firmly believe that they will. On top of that,

648
00:32:37,040 --> 00:32:39,840
Speaker 9: I think that generative AI is going to be, you know,

649
00:32:40,040 --> 00:32:43,960
Speaker 9: the next massive increase in workloads you know, moving to

650
00:32:44,000 --> 00:32:47,080
Speaker 9: the cloud or in many cases happening for the first

651
00:32:47,080 --> 00:32:49,640
Speaker 9: time and happening in the cloud. And so I do

652
00:32:49,680 --> 00:32:53,160
Speaker 9: think that it should be a significant tailwind for cloud

653
00:32:53,200 --> 00:32:58,080
Speaker 9: providers and particularly for AWS, given our leadership position. Obviously

654
00:32:58,160 --> 00:33:01,280
Speaker 9: we need to come out with the capability the services

655
00:33:01,320 --> 00:33:05,200
Speaker 9: that you know, justify people using us for the purpose.

656
00:33:05,280 --> 00:33:07,720
Speaker 9: But I think if we you know, if we do

657
00:33:07,760 --> 00:33:09,960
Speaker 9: a good job of listening to customers, it should provide

658
00:33:10,000 --> 00:33:13,520
Speaker 9: significant demand for years to come. I mean, the computational

659
00:33:13,520 --> 00:33:17,680
Speaker 9: requirements are so intense. And one thing which also I

660
00:33:17,720 --> 00:33:21,320
Speaker 9: think works in favor of AWS is that a lot

661
00:33:21,320 --> 00:33:23,640
Speaker 9: of people asking about, hey, what about the energy consumption?

662
00:33:23,760 --> 00:33:25,080
Speaker 9: You know, what about sustainability?

663
00:33:25,200 --> 00:33:26,800
Speaker 3: You run those efforts inside Amazon.

664
00:33:27,000 --> 00:33:29,400
Speaker 9: Yeah, I mean I also run sustainability. It's kind of

665
00:33:29,440 --> 00:33:31,840
Speaker 9: the other thing I do inside of Amazon, and we

666
00:33:31,880 --> 00:33:34,720
Speaker 9: as a company, and I personally care a lot about sustainability.

667
00:33:35,040 --> 00:33:37,120
Speaker 9: And people say, well, you know, is this generative AI

668
00:33:37,320 --> 00:33:38,400
Speaker 9: massive compute.

669
00:33:38,080 --> 00:33:39,080
Speaker 8: Is that incompatible?

670
00:33:39,160 --> 00:33:41,560
Speaker 9: Yeah, we're going back for sustainable Well no, we're not,

671
00:33:41,720 --> 00:33:45,280
Speaker 9: because number one of these workloads, you know, there's no

672
00:33:45,360 --> 00:33:47,680
Speaker 9: putting the genie back in the bottle. So generative AI

673
00:33:47,840 --> 00:33:50,240
Speaker 9: is going to happen, So let's make it happen in

674
00:33:50,280 --> 00:33:53,400
Speaker 9: a highly energy efficient and sustainable way. So if you

675
00:33:53,440 --> 00:33:59,120
Speaker 9: look at our custom chips that we design, if you

676
00:33:59,160 --> 00:34:02,000
Speaker 9: take Graviton for example, which is our oldest chip family,

677
00:34:02,760 --> 00:34:07,480
Speaker 9: Graviton is sixty percent more energy efficient than equivalent X

678
00:34:07,520 --> 00:34:11,520
Speaker 9: eighty six based compute capacity. And if you look at

679
00:34:11,640 --> 00:34:14,680
Speaker 9: an enterprise just in general, nevermind generative AI, just moving

680
00:34:14,680 --> 00:34:17,760
Speaker 9: from their own data centers to the cloud to AWS.

681
00:34:18,480 --> 00:34:22,359
Speaker 9: There was a study done showing that AWS is three

682
00:34:22,400 --> 00:34:25,800
Speaker 9: point six times more energy efficient than the average enterprise

683
00:34:26,320 --> 00:34:29,000
Speaker 9: data center in the United States. So you know, we

684
00:34:29,040 --> 00:34:31,760
Speaker 9: are the sustainable place. We're doing it through a whole

685
00:34:31,800 --> 00:34:36,080
Speaker 9: series of technological improvements plus a commitment which were eighty

686
00:34:36,080 --> 00:34:38,200
Speaker 9: five percent of the way. They're already to be one

687
00:34:38,239 --> 00:34:41,080
Speaker 9: hundred percent renewable energy by twenty twenty five, which is

688
00:34:41,200 --> 00:34:42,120
Speaker 9: just around the corner, right.

689
00:34:42,440 --> 00:34:46,480
Speaker 6: You guys had another commitment called Shipman zero, which was

690
00:34:47,880 --> 00:34:51,120
Speaker 6: a net carbon zero commitment by twenty thirty, which you guys,

691
00:34:51,480 --> 00:34:54,799
Speaker 6: I guess maybe scrapped or you took off the website. Yeah,

692
00:34:54,800 --> 00:34:57,080
Speaker 6: has a long time Amazon watcher, My heart just kind

693
00:34:57,120 --> 00:34:59,600
Speaker 6: of sank because you guys have been so prominent about

694
00:34:59,640 --> 00:35:02,600
Speaker 6: these goals. I guess the question is, like, have these

695
00:35:02,640 --> 00:35:07,000
Speaker 6: commitments become harder to meet than they were to make?

696
00:35:07,440 --> 00:35:09,480
Speaker 9: Well, I think your heart should be sinking at the

697
00:35:09,520 --> 00:35:13,400
Speaker 9: state of global warming state good line is, and we

698
00:35:13,400 --> 00:35:15,040
Speaker 9: should all be concerned about that, But I think your

699
00:35:15,080 --> 00:35:18,000
Speaker 9: heart should be singing at the leadership position that Amazon

700
00:35:18,080 --> 00:35:20,560
Speaker 9: is trying to establish, and at the improvements we're trying

701
00:35:20,600 --> 00:35:22,759
Speaker 9: to make, at the very public goals we've set.

702
00:35:22,800 --> 00:35:25,080
Speaker 3: Why don't you remove that particular commitment.

703
00:35:24,640 --> 00:35:28,200
Speaker 9: So well, because it's subsumed really in a much broader

704
00:35:28,239 --> 00:35:31,040
Speaker 9: Boulder commitment. We made a very public pledge to be

705
00:35:31,200 --> 00:35:34,760
Speaker 9: net zero carbon across all of Amazon, not just State WS,

706
00:35:34,880 --> 00:35:37,520
Speaker 9: all of Amazon by twenty forty, which is ten years

707
00:35:37,520 --> 00:35:41,320
Speaker 9: ahead of the Paris Accords. Now, for a technology company

708
00:35:41,360 --> 00:35:44,760
Speaker 9: like AWS, I won't say it's easy, but it's it's doable,

709
00:35:45,200 --> 00:35:47,640
Speaker 9: so you'll hear that from other tech companies. For a

710
00:35:47,640 --> 00:35:52,440
Speaker 9: big retailer with air freight and inbound transportation and stores

711
00:35:52,880 --> 00:35:56,239
Speaker 9: and packaging, it is actually really, really hard, and we

712
00:35:56,280 --> 00:35:58,080
Speaker 9: will be the first to say we don't know how

713
00:35:58,120 --> 00:36:00,239
Speaker 9: we're going to get there in all dimensions. I can

714
00:36:00,280 --> 00:36:02,320
Speaker 9: tell you how we're going to get their renewable energy.

715
00:36:02,640 --> 00:36:04,080
Speaker 9: I can't tell you how we're going to get there

716
00:36:04,080 --> 00:36:08,239
Speaker 9: in all elements of transportation and packaging and buildings. But

717
00:36:08,320 --> 00:36:11,120
Speaker 9: we're an innovative company. We take bold, long term bets,

718
00:36:11,520 --> 00:36:14,919
Speaker 9: and we made this pledge publicly, not privately, in order

719
00:36:14,960 --> 00:36:17,200
Speaker 9: to number one, you know, have a forcing function for

720
00:36:17,280 --> 00:36:20,120
Speaker 9: ourselves because it's not easy, and also because we want

721
00:36:20,160 --> 00:36:24,239
Speaker 9: to inspire other organizations, governments, companies to join us. We

722
00:36:24,280 --> 00:36:27,160
Speaker 9: have over four hundred signatories now of the Climate Pledge

723
00:36:27,280 --> 00:36:30,400
Speaker 9: that's growing every month. And it's not a competition against

724
00:36:30,480 --> 00:36:34,239
Speaker 9: other organizations. It's competition against the thermometer. And frankly, I

725
00:36:34,280 --> 00:36:36,640
Speaker 9: want other people to out innov us and out innovate us.

726
00:36:36,680 --> 00:36:39,320
Speaker 9: I want to be beaten, if you will, in the

727
00:36:40,600 --> 00:36:43,880
Speaker 9: race to become sustainable, and hopefully we can be inspired

728
00:36:43,920 --> 00:36:46,479
Speaker 9: by things that other people are doing. So of course

729
00:36:46,480 --> 00:36:48,239
Speaker 9: we adjust our goals over time, but the thing I

730
00:36:48,239 --> 00:36:50,640
Speaker 9: would focus on is just audacious goal to be nets

731
00:36:50,719 --> 00:36:51,720
Speaker 9: or a carbon by twenty.

732
00:36:51,520 --> 00:36:52,760
Speaker 3: Eight and have to line in the sand.

733
00:36:52,800 --> 00:36:56,239
Speaker 6: I mean, even as things like generative AI take more

734
00:36:56,239 --> 00:36:59,320
Speaker 6: and more computational resources. Even if you get more efficient

735
00:36:59,360 --> 00:37:03,960
Speaker 6: with compute, that's a commitment that you can hold firm on,

736
00:37:04,040 --> 00:37:05,080
Speaker 6: even if you don't know how you're going.

737
00:37:05,040 --> 00:37:05,480
Speaker 3: To get there.

738
00:37:05,880 --> 00:37:08,440
Speaker 9: We have made a public pledge we intend on getting there,

739
00:37:08,600 --> 00:37:11,319
Speaker 9: But there are lots of other areas of progress from

740
00:37:11,360 --> 00:37:13,200
Speaker 9: making in the interim. So, for example, if you take

741
00:37:13,239 --> 00:37:18,240
Speaker 9: packaging for the retail business, the average packaging per shipment

742
00:37:18,760 --> 00:37:22,839
Speaker 9: has decreased by thirty eight percent since twenty fifteen, so

743
00:37:23,080 --> 00:37:26,040
Speaker 9: which is another example of it can be sustainable and

744
00:37:26,080 --> 00:37:28,560
Speaker 9: good for our business. You know, it's lower cost for us,

745
00:37:28,840 --> 00:37:32,280
Speaker 9: and it's much more sustainable for the planet. And anytime

746
00:37:32,320 --> 00:37:34,759
Speaker 9: you create a win win like that, it just works

747
00:37:34,800 --> 00:37:38,279
Speaker 9: for everybody and becomes a really sustainable business proposition.

748
00:37:39,800 --> 00:37:41,680
Speaker 3: Let's get back to AI and let me ask.

749
00:37:41,600 --> 00:37:42,360
Speaker 8: You about Alexa.

750
00:37:42,440 --> 00:37:44,760
Speaker 6: I know it's slightly out of your purview, but Alexa

751
00:37:44,840 --> 00:37:45,880
Speaker 6: runs in your servers.

752
00:37:46,440 --> 00:37:50,279
Speaker 3: Can Alexa be a generitive AI play? Should it be?

753
00:37:51,560 --> 00:37:54,239
Speaker 9: Alexa's already, as I mentioned, are ready powered in large

754
00:37:54,280 --> 00:37:57,320
Speaker 9: part by lllms that Amazon built and been in production

755
00:37:57,440 --> 00:38:00,680
Speaker 9: for a while now, and I think that she's only

756
00:38:00,760 --> 00:38:04,720
Speaker 9: going to get smarter and better and more personalized as

757
00:38:05,000 --> 00:38:08,279
Speaker 9: the lll M technology expands and improved, it.

758
00:38:08,200 --> 00:38:10,319
Speaker 6: Feels to me as a long time user, and my

759
00:38:10,360 --> 00:38:13,200
Speaker 6: wife who's here knows I populated our house with them

760
00:38:13,200 --> 00:38:16,080
Speaker 6: at one point, but today it feels as the SERI

761
00:38:16,200 --> 00:38:17,760
Speaker 6: frankly a step behind.

762
00:38:17,880 --> 00:38:20,719
Speaker 3: Is that? Would you agree with that? No?

763
00:38:20,880 --> 00:38:24,520
Speaker 9: I think that Alex will I use Alex in my house.

764
00:38:24,560 --> 00:38:26,160
Speaker 9: I mean, you know, I guess we all have different

765
00:38:26,160 --> 00:38:29,160
Speaker 9: strokes for different folks, and we love Alexa and I

766
00:38:29,280 --> 00:38:33,920
Speaker 9: like to think she loves us, although if you ask

767
00:38:34,000 --> 00:38:38,160
Speaker 9: me that, she won't actually tell you yes. And I

768
00:38:38,239 --> 00:38:40,239
Speaker 9: think Alex has been getting better and better, you know,

769
00:38:40,320 --> 00:38:44,160
Speaker 9: more skills, better skills, more understanding of you and your

770
00:38:44,200 --> 00:38:47,160
Speaker 9: your your likes and dislikes in your habits. And I

771
00:38:47,200 --> 00:38:50,400
Speaker 9: think that the the rapid improvements that we're making in

772
00:38:50,440 --> 00:38:54,520
Speaker 9: Generative AI are truly going to continue to transform Alexa

773
00:38:54,960 --> 00:38:59,520
Speaker 9: into you know, a truly you know, personalized assistant. And

774
00:38:59,560 --> 00:39:02,360
Speaker 9: we do on Alexa to be you know, an absolutely

775
00:39:02,360 --> 00:39:05,840
Speaker 9: indispensable invest in the world, you know, personal assistant to you.

776
00:39:06,160 --> 00:39:07,480
Speaker 8: And I think that we've got a lot of work

777
00:39:07,480 --> 00:39:07,680
Speaker 8: to do.

778
00:39:08,600 --> 00:39:10,600
Speaker 9: I'll leave it to the Alexa folks to fill in

779
00:39:10,640 --> 00:39:11,440
Speaker 9: all of those blanks.

780
00:39:11,440 --> 00:39:11,919
Speaker 8: Over time.

781
00:39:11,960 --> 00:39:15,399
Speaker 9: But we're actually very confident in that plan and very

782
00:39:15,400 --> 00:39:18,719
Speaker 9: optimistic about Alexa being able to fulfill that role in

783
00:39:18,800 --> 00:39:21,200
Speaker 9: people's lives that I think they're really going to love.

784
00:39:21,400 --> 00:39:22,839
Speaker 3: I want to get to two more in the two

785
00:39:22,880 --> 00:39:26,560
Speaker 3: minutes we have left. Your Boston predecessor.

786
00:39:25,920 --> 00:39:28,880
Speaker 6: Andy Jassey has been kind of cutting some of the

787
00:39:28,920 --> 00:39:31,920
Speaker 6: big bets at Amazon, but one that he hasn't cut

788
00:39:32,480 --> 00:39:36,239
Speaker 6: is the satellite plan Kuiper, and I just wonder, you know,

789
00:39:36,560 --> 00:39:40,360
Speaker 6: what you see as the opportunity considering a rival system

790
00:39:40,440 --> 00:39:44,319
Speaker 6: stirling from Tesla, it does seem to be, you know, operational,

791
00:39:44,400 --> 00:39:45,160
Speaker 6: quite far ahead.

792
00:39:46,040 --> 00:39:49,480
Speaker 9: Well, we're very optimistic about Kuyper. There's huge interest from governments,

793
00:39:49,520 --> 00:39:53,560
Speaker 9: from enterprises, from lots of other organizations. Billions of people

794
00:39:53,600 --> 00:39:57,839
Speaker 9: around the world are underserved for Internet, and Kuyper really

795
00:39:57,880 --> 00:40:01,120
Speaker 9: aims to democratize that. There's a theme here to democratize

796
00:40:01,120 --> 00:40:05,440
Speaker 9: that and provide great Internet service to so many billions

797
00:40:05,520 --> 00:40:12,239
Speaker 9: under serve people. In addition, whether it's automobile companies, telecommunications companies,

798
00:40:12,440 --> 00:40:16,680
Speaker 9: lots of other enterprises who are AWS customers, governments. They want,

799
00:40:16,840 --> 00:40:20,799
Speaker 9: especially from remote locations, be able to backhaul information up

800
00:40:20,880 --> 00:40:24,200
Speaker 9: to Kuyper back into the AWS cloud, and so I

801
00:40:24,200 --> 00:40:27,560
Speaker 9: think as we launch our first satellites, which as coming

802
00:40:27,640 --> 00:40:30,440
Speaker 9: up later this year and then really ramping up in

803
00:40:30,480 --> 00:40:33,279
Speaker 9: twenty twenty four and twenty twenty five is my understanding.

804
00:40:34,920 --> 00:40:38,080
Speaker 9: But then being in service in that timeframe initially and

805
00:40:38,160 --> 00:40:42,480
Speaker 9: being able to deliver that for AWS customers is huge.

806
00:40:43,080 --> 00:40:47,000
Speaker 9: Before we go, I just want to remind folks that

807
00:40:47,200 --> 00:40:50,600
Speaker 9: just today this morning, AWS has launched its one hundred

808
00:40:50,640 --> 00:40:54,920
Speaker 9: million dollar Generative AI Innovation Center, where we're going to

809
00:40:54,920 --> 00:40:57,520
Speaker 9: be going out to all those customers around the world,

810
00:40:57,640 --> 00:41:05,520
Speaker 9: enterprises with expertise free a BOS expertise, solutions, architects, engineers, strategists, uh,

811
00:41:05,760 --> 00:41:09,240
Speaker 9: and working with them one on one to envision design

812
00:41:09,280 --> 00:41:12,000
Speaker 9: and then actual that one hundred million AI generative AI

813
00:41:12,120 --> 00:41:14,600
Speaker 9: capabilities not talk, but actually.

814
00:41:14,520 --> 00:41:18,080
Speaker 10: That discounts for new companies or Yeah, we're gonna, we're

815
00:41:18,080 --> 00:41:21,120
Speaker 10: just gonna, We're just gonna bring our internal a WOS experts,

816
00:41:21,160 --> 00:41:22,759
Speaker 10: you know, free of charge to a whole bunch of

817
00:41:22,840 --> 00:41:26,160
Speaker 10: AWS customers, uh, you know, focusing.

818
00:41:25,960 --> 00:41:30,040
Speaker 9: On folks with with with a significant AWS presence and

819
00:41:30,120 --> 00:41:33,239
Speaker 9: go help them turbocharge their efforts to get real with

820
00:41:33,360 --> 00:41:35,359
Speaker 9: generative AI, get beyond the talk all.

821
00:41:35,360 --> 00:41:37,799
Speaker 6: Right last one in negative ten seconds, So I guess

822
00:41:37,800 --> 00:41:38,680
Speaker 6: it's gotta be a quick one.

823
00:41:38,840 --> 00:41:40,720
Speaker 3: You and Andy are both in the same situation.

824
00:41:40,800 --> 00:41:43,560
Speaker 6: You're the You're the guy after the guy, the founder

825
00:41:43,640 --> 00:41:47,840
Speaker 6: whose name was synonymous with the early stage of legendary growth.

826
00:41:47,880 --> 00:41:51,400
Speaker 3: So what is what do you want adams aws legacy

827
00:41:51,440 --> 00:41:53,399
Speaker 3: to be? I don't think.

828
00:41:53,480 --> 00:41:55,520
Speaker 9: I don't really think of it in personal terms, to

829
00:41:55,520 --> 00:41:57,920
Speaker 9: be honest with you, so I don't have a canned answer,

830
00:41:57,960 --> 00:42:00,480
Speaker 9: but I will say that, you know, I would love

831
00:42:00,520 --> 00:42:01,520
Speaker 9: it if.

832
00:42:03,280 --> 00:42:06,440
Speaker 8: If I could be known to really help.

833
00:42:06,320 --> 00:42:09,760
Speaker 9: Drive a business that is, you know, constantly, no matter

834
00:42:09,840 --> 00:42:12,280
Speaker 9: how big it gets, no matter how far flowing it gets,

835
00:42:12,560 --> 00:42:14,799
Speaker 9: puts customers at the very center of what we're doing,

836
00:42:15,120 --> 00:42:18,760
Speaker 9: always puts customers interests, you know, before anybody else's interests.

837
00:42:18,920 --> 00:42:22,719
Speaker 9: Yet at the same time, an is an empathetic, equitable,

838
00:42:23,120 --> 00:42:25,560
Speaker 9: fun and innovative place for employees to work.

839
00:42:25,640 --> 00:42:27,799
Speaker 3: All right, Adam Flipski, thank you for joining us.

840
00:42:27,800 --> 00:42:28,080
Speaker 11: Thank you.

841
00:42:31,239 --> 00:42:34,719
Speaker 1: Amazon Web Services CEO Adam Slipsky there in conversation with

842
00:42:34,840 --> 00:42:37,680
Speaker 1: Rimberg's at Bradstone and highlighting some of the news that

843
00:42:37,760 --> 00:42:39,560
Speaker 1: has just come out. Of course, the fact that Amazon

844
00:42:39,640 --> 00:42:42,760
Speaker 1: is spending one hundred million dollars to teach cloud clients

845
00:42:42,800 --> 00:42:45,680
Speaker 1: about AI's saying to get real. Basically, some of the

846
00:42:45,760 --> 00:42:48,080
Speaker 1: early stage clients are going to be high Spot Twilio.

847
00:42:48,160 --> 00:42:50,880
Speaker 1: They're going to be really using some of these customized

848
00:42:50,880 --> 00:42:54,080
Speaker 1: applications and understanding how to get this expertise to ensure

849
00:42:54,080 --> 00:42:56,399
Speaker 1: that they are adopting generator of AI at the rapid rate.

850
00:42:56,440 --> 00:42:58,360
Speaker 1: But we heard there from Adam Slipsky also about the

851
00:42:58,400 --> 00:43:02,920
Speaker 1: focus on being carbon new, focused on of course climate

852
00:43:03,160 --> 00:43:06,319
Speaker 1: and how you twin that with the enormous compute power

853
00:43:06,320 --> 00:43:08,600
Speaker 1: that is necessary. As we dive into the whole world

854
00:43:08,600 --> 00:43:11,880
Speaker 1: of lage language models of more general to AI, the

855
00:43:11,920 --> 00:43:15,600
Speaker 1: chips necessary, the compute power and ultimately what that affects

856
00:43:16,000 --> 00:43:18,120
Speaker 1: the climate. Conversation just keeps coming.

857
00:43:18,160 --> 00:43:18,959
Speaker 3: We're going to talk chips.

858
00:43:18,960 --> 00:43:21,840
Speaker 1: Next Technology summit is with our very own ed Lavelow

859
00:43:21,880 --> 00:43:24,200
Speaker 1: some exam Qualcom Presidentcy Crisciano.

860
00:43:24,280 --> 00:43:28,160
Speaker 2: Our answers the question is crowcommon AI company?

861
00:43:28,640 --> 00:43:31,760
Speaker 12: Look, this is a this is a great question to ask,

862
00:43:32,040 --> 00:43:35,440
Speaker 12: and you know it's it's incredible to see all of

863
00:43:35,440 --> 00:43:39,480
Speaker 12: the development you see right now on on AI. Here's

864
00:43:39,600 --> 00:43:44,640
Speaker 12: here's how I answered that question. Actually, it's very simple, Uh,

865
00:43:45,080 --> 00:43:49,000
Speaker 12: if you think about the AI, when you think about semiconductor,

866
00:43:49,040 --> 00:43:52,600
Speaker 12: it's really accelerating computing. You do a lot of computation,

867
00:43:54,360 --> 00:43:56,719
Speaker 12: and what we see what you can do with those

868
00:43:56,880 --> 00:44:00,880
Speaker 12: large language models, large models for him is in videos.

869
00:44:01,360 --> 00:44:05,040
Speaker 12: So if you think about the history of computing, computing

870
00:44:05,680 --> 00:44:09,560
Speaker 12: starts in the cloud and he gets scale at the edge.

871
00:44:09,640 --> 00:44:13,799
Speaker 12: I think that's that's what happens with CPUs, That's that's

872
00:44:13,840 --> 00:44:17,759
Speaker 12: what happens with all other form of computing. And I

873
00:44:17,840 --> 00:44:20,839
Speaker 12: think the smartphone is a great example of that. If

874
00:44:20,880 --> 00:44:26,200
Speaker 12: you look, the largest computing platform ever developed is the

875
00:44:26,280 --> 00:44:29,640
Speaker 12: smartphone right now. It's the largest development platform for mankind.

876
00:44:30,200 --> 00:44:33,120
Speaker 12: And and what is good about the smartphone, it's, uh,

877
00:44:33,719 --> 00:44:35,640
Speaker 12: it's a device that I'm with you all the time.

878
00:44:36,040 --> 00:44:39,680
Speaker 12: So if AI becomes pervasive, which we believe it will

879
00:44:39,719 --> 00:44:44,800
Speaker 12: become pervasive, especially when you look about how those large models,

880
00:44:44,960 --> 00:44:48,520
Speaker 12: they are very natural, how you you can converse with them,

881
00:44:48,840 --> 00:44:52,080
Speaker 12: they have contextual information and all of those things that's

882
00:44:52,080 --> 00:44:53,360
Speaker 12: going to happen at the edge.

883
00:44:53,640 --> 00:44:55,400
Speaker 11: So that's how you think about qualcomm.

884
00:44:55,840 --> 00:44:58,640
Speaker 12: If AI is going to get scale, you're going to

885
00:44:58,719 --> 00:45:02,000
Speaker 12: see it running on wall come, snapdrag and devices. Whether

886
00:45:02,040 --> 00:45:04,600
Speaker 12: it's in your phone, in your car, in your PC,

887
00:45:05,280 --> 00:45:07,799
Speaker 12: and into other machines. And I think, well, it's a

888
00:45:07,800 --> 00:45:09,200
Speaker 12: great opportunity.

889
00:45:08,640 --> 00:45:13,680
Speaker 2: For the future is to democracize access to artificial intelligence tools,

890
00:45:13,680 --> 00:45:16,479
Speaker 2: generative AI tools, and cloudcom is going.

891
00:45:16,400 --> 00:45:17,200
Speaker 11: To make that happen.

892
00:45:18,000 --> 00:45:20,479
Speaker 2: Why are you not getting like Jensen one level love?

893
00:45:22,040 --> 00:45:25,239
Speaker 11: Look, I think the I think what's happening right now.

894
00:45:25,280 --> 00:45:27,920
Speaker 12: And by the way, it's great for the semiconductor industry.

895
00:45:28,480 --> 00:45:32,920
Speaker 12: For anybody that has been on the forefront of computing.

896
00:45:32,960 --> 00:45:36,440
Speaker 12: You know, Qualcom probably used to be well known as

897
00:45:36,440 --> 00:45:39,000
Speaker 12: a communication company, but actually if you look at what

898
00:45:39,040 --> 00:45:41,239
Speaker 12: we do right now, it's more of a connected processor

899
00:45:41,320 --> 00:45:43,040
Speaker 12: company than communication.

900
00:45:44,040 --> 00:45:46,680
Speaker 11: And as those models started.

901
00:45:46,239 --> 00:45:49,239
Speaker 12: To become very popular, they're going to be running at

902
00:45:49,280 --> 00:45:53,440
Speaker 12: the edge. And I expect that AI becomes an option

903
00:45:53,880 --> 00:45:56,880
Speaker 12: on Qualcomm right now. Look, and I'll give you an example.

904
00:45:57,239 --> 00:46:00,680
Speaker 12: It's it's I saw something they add them. I think

905
00:46:00,680 --> 00:46:03,960
Speaker 12: in the prior conversation when he said something about in

906
00:46:04,080 --> 00:46:06,719
Speaker 12: nineteen ninety seven, if you try to guess who are

907
00:46:06,719 --> 00:46:08,880
Speaker 12: the winners and losers on the Internet, it will be

908
00:46:09,000 --> 00:46:11,160
Speaker 12: probably a very wild guest. I think what we see

909
00:46:11,200 --> 00:46:15,680
Speaker 12: today is this janitor of AI opportunity is huge. We

910
00:46:15,840 --> 00:46:19,239
Speaker 12: don't know yet all of the different applications that are

911
00:46:19,280 --> 00:46:22,319
Speaker 12: going to come up. We're seeing that just within the

912
00:46:22,360 --> 00:46:25,960
Speaker 12: past six months is a revolution the number of companies

913
00:46:26,000 --> 00:46:28,560
Speaker 12: come on with use cases and those use cases are

914
00:46:28,560 --> 00:46:31,319
Speaker 12: going to happen on devices, and I think that's going

915
00:46:31,360 --> 00:46:32,640
Speaker 12: to be a great opportunity for.

916
00:46:32,640 --> 00:46:34,040
Speaker 3: Hold that thought. What we're going to do.

917
00:46:34,080 --> 00:46:36,240
Speaker 2: Now, I'm going to show you something to the audience

918
00:46:36,239 --> 00:46:40,359
Speaker 2: here and those with us virtually, but during that think

919
00:46:40,360 --> 00:46:43,920
Speaker 2: about questions for Cristiano based on what you see. And

920
00:46:43,960 --> 00:46:47,000
Speaker 2: so with that, let's bring up the video, and Christiano,

921
00:46:47,440 --> 00:46:50,759
Speaker 2: when it comes up and plays, explain to us a

922
00:46:50,800 --> 00:46:54,080
Speaker 2: little bit what it is that we're seeing, because here

923
00:46:54,080 --> 00:46:57,239
Speaker 2: at the Bloomberg Technology Summit, we're going to nail the technology.

924
00:46:57,280 --> 00:46:59,839
Speaker 2: Any second, just wait, the video is going to come

925
00:47:00,600 --> 00:47:03,600
Speaker 2: and when it does, it will have been worth the way.

926
00:47:04,440 --> 00:47:04,880
Speaker 11: Here we go.

927
00:47:05,400 --> 00:47:08,000
Speaker 12: Yes, So what you basically see is a countro net demo.

928
00:47:08,320 --> 00:47:11,239
Speaker 12: You have an input image on your phone. You tell

929
00:47:11,640 --> 00:47:13,600
Speaker 12: in your input prompt what do you want the image

930
00:47:13,640 --> 00:47:16,239
Speaker 12: to be. You wanted to make it a masterpiece, look

931
00:47:16,360 --> 00:47:21,279
Speaker 12: like Venice Canals four K, and it just runs and

932
00:47:21,920 --> 00:47:26,240
Speaker 12: give you this very unique image image to image that's

933
00:47:26,440 --> 00:47:29,800
Speaker 12: never been created before, created to AI running on your phone.

934
00:47:29,960 --> 00:47:31,400
Speaker 11: So it's a good I think.

935
00:47:31,520 --> 00:47:33,680
Speaker 12: Time to talk a little bit about how we think

936
00:47:33,680 --> 00:47:37,840
Speaker 12: about AI at the edge outside of the data center, because,

937
00:47:38,440 --> 00:47:40,360
Speaker 12: like we have seen everywhere, there's going to be this

938
00:47:40,440 --> 00:47:42,279
Speaker 12: huge opportunity for the cloud, but it's going to be

939
00:47:42,320 --> 00:47:46,399
Speaker 12: this huge opportunity for devices because what you do own

940
00:47:46,440 --> 00:47:50,560
Speaker 12: the device is very different. So there's a number of

941
00:47:51,000 --> 00:47:53,440
Speaker 12: reasons why this is going to be very popular on

942
00:47:53,480 --> 00:47:58,239
Speaker 12: the device. First, the device has contextual information about you

943
00:47:58,320 --> 00:48:01,279
Speaker 12: and has real time information like a picture you just

944
00:48:01,320 --> 00:48:04,800
Speaker 12: took and you want it right now at that moment,

945
00:48:05,000 --> 00:48:07,840
Speaker 12: change that picture and share with somebody else with your

946
00:48:08,400 --> 00:48:09,760
Speaker 12: messaging platform.

947
00:48:09,320 --> 00:48:13,239
Speaker 2: For context that video. That device was run in aeroplane

948
00:48:13,280 --> 00:48:16,800
Speaker 2: mode without any external connection, right, It ran them model

949
00:48:17,200 --> 00:48:17,920
Speaker 2: locally on.

950
00:48:17,960 --> 00:48:20,839
Speaker 12: Device absolutely, So that's one of the reasons you have

951
00:48:21,000 --> 00:48:22,520
Speaker 12: real time context information.

952
00:48:22,960 --> 00:48:26,440
Speaker 11: There's another reasons processing on.

953
00:48:26,320 --> 00:48:30,560
Speaker 12: The phone is virtually free when you think about you're

954
00:48:30,640 --> 00:48:35,640
Speaker 12: running those models in the cloud and think about a

955
00:48:35,719 --> 00:48:39,799
Speaker 12: large language model for every token, like a word as

956
00:48:39,840 --> 00:48:41,839
Speaker 12: the sentence. If you do that, if you have an

957
00:48:42,200 --> 00:48:45,239
Speaker 12: experiment that you see the words coming.

958
00:48:44,960 --> 00:48:47,840
Speaker 1: Up, the president and CEO of Qualcom, Christiana I'm on

959
00:48:48,120 --> 00:48:50,600
Speaker 1: talking about our very own ed Ludlow, of course, discussing

960
00:48:51,040 --> 00:48:53,400
Speaker 1: how Qualcom's going to be playing a role in the

961
00:48:53,600 --> 00:48:57,080
Speaker 1: enormous scope of generative AI edge computing. Of course, the

962
00:48:57,160 --> 00:48:58,920
Speaker 1: chips that are known to be in your iPhones and

963
00:48:58,960 --> 00:49:02,520
Speaker 1: your phones and your automobiles. One, of course, Qualcom wants

964
00:49:02,520 --> 00:49:07,640
Speaker 1: to ensure that you're accessing generative AI in local ways

965
00:49:07,880 --> 00:49:08,480
Speaker 1: and means