1
00:00:02,840 --> 00:00:07,600
Speaker 1: Zone Media. Hello and welcome to Better Offline. I'm your

2
00:00:07,600 --> 00:00:22,119
Speaker 1: host and chief romance officer, ed Ze Tron. In the

3
00:00:22,200 --> 00:00:25,200
Speaker 1: last episode, I dug into the fundamental weaknesses and open

4
00:00:25,239 --> 00:00:28,280
Speaker 1: AI the supposed leader in the genital of AI boom,

5
00:00:28,400 --> 00:00:30,440
Speaker 1: and today I'm going to get into a much larger,

6
00:00:30,720 --> 00:00:34,200
Speaker 1: more systemic, more terminal problem and the signs that things

7
00:00:34,240 --> 00:00:37,320
Speaker 1: are really really falling apart. And as ever, I will

8
00:00:37,320 --> 00:00:39,920
Speaker 1: have links to everything I'm talking about in the episode notes,

9
00:00:40,080 --> 00:00:42,000
Speaker 1: so you know I'm not making it up, which one

10
00:00:42,040 --> 00:00:44,600
Speaker 1: person suggested I did once and it bothered me a

11
00:00:44,640 --> 00:00:48,479
Speaker 1: great deal. But back to the actual stuff. The problems

12
00:00:48,479 --> 00:00:50,640
Speaker 1: that open ai is facing are those faced by the

13
00:00:50,800 --> 00:00:54,320
Speaker 1: entire generative AI industry. One's born of their sole focused

14
00:00:54,360 --> 00:00:57,560
Speaker 1: on the transformer based architecture underlying large language models like

15
00:00:57,680 --> 00:01:01,640
Speaker 1: chat GPT open A issues besides the fact that they're

16
00:01:01,640 --> 00:01:04,400
Speaker 1: in a terrible business as discussed in the last episode,

17
00:01:04,600 --> 00:01:07,280
Speaker 1: is that generative AI, and by extension, the model GPT

18
00:01:07,360 --> 00:01:10,720
Speaker 1: in the product Chat GPT, doesn't really solve complex problems

19
00:01:10,760 --> 00:01:13,720
Speaker 1: that would justify the massive costs behind it. It is

20
00:01:13,760 --> 00:01:16,800
Speaker 1: these massive intractable challenges that are a result of these

21
00:01:16,800 --> 00:01:19,880
Speaker 1: models being probabilistic, meaning that they don't know anything. They're

22
00:01:19,880 --> 00:01:22,840
Speaker 1: just generating an answer based on maths and training data,

23
00:01:23,200 --> 00:01:25,280
Speaker 1: something that model developers are running out of at an

24
00:01:25,280 --> 00:01:29,880
Speaker 1: incredible pace. Hallucinations, which occur when models authoritatively state something

25
00:01:29,920 --> 00:01:31,679
Speaker 1: that isn't true, or, in the case of an image

26
00:01:31,720 --> 00:01:35,000
Speaker 1: or a video, makes something that just looks wrong. Well,

27
00:01:35,600 --> 00:01:38,560
Speaker 1: they're impossible to resolve without new branches of maths, and

28
00:01:38,600 --> 00:01:40,560
Speaker 1: while you might be able to reduce or mitigate them,

29
00:01:40,600 --> 00:01:43,399
Speaker 1: their existence makes it hard for business critical applications to

30
00:01:43,440 --> 00:01:46,040
Speaker 1: truly rely on this kind of AI. I don't even

31
00:01:46,080 --> 00:01:48,920
Speaker 1: know if i'd call it an AI, but regardless, we

32
00:01:49,000 --> 00:01:52,880
Speaker 1: go forward, and even tech's most dominant players can't seem

33
00:01:52,960 --> 00:01:57,000
Speaker 1: to turn generative AI into any kind of real business line.

34
00:01:57,080 --> 00:02:00,000
Speaker 1: The Information reported in early September that customers of Micross

35
00:02:00,280 --> 00:02:03,360
Speaker 1: three sixty five Suite are barely adopting its AI powered

36
00:02:03,440 --> 00:02:06,960
Speaker 1: copilot products, with somewhere between zero point one percent and

37
00:02:07,040 --> 00:02:09,959
Speaker 1: one percent of the four hundred and forty million paying

38
00:02:10,040 --> 00:02:12,799
Speaker 1: people who pay for Microsoft three sixty five, which is

39
00:02:12,800 --> 00:02:15,200
Speaker 1: about thirty to fifty dollars a person, by the way,

40
00:02:15,720 --> 00:02:18,800
Speaker 1: are willing to pay for AI, and just to be clear,

41
00:02:18,840 --> 00:02:21,840
Speaker 1: I muddied that little. It's thirty to fifty bucks per

42
00:02:22,040 --> 00:02:24,600
Speaker 1: person per head to add this stuff. I'll get into

43
00:02:24,639 --> 00:02:27,440
Speaker 1: it in a minute. One firm, according to the information,

44
00:02:27,840 --> 00:02:30,440
Speaker 1: was testing the AI features and was quoted as saying

45
00:02:30,480 --> 00:02:33,000
Speaker 1: that most people don't find it that valuable right now,

46
00:02:33,080 --> 00:02:35,840
Speaker 1: and others are saying that many businesses haven't seen breakthroughs

47
00:02:35,840 --> 00:02:39,280
Speaker 1: in productivity or other benefits, and that they're not sure

48
00:02:39,320 --> 00:02:42,840
Speaker 1: that they will. In an internal presentation provided to be

49
00:02:42,919 --> 00:02:46,120
Speaker 1: by a source, users of Microsoft SharePoint copilot complained that

50
00:02:46,160 --> 00:02:50,520
Speaker 1: Microsoft chatbot kept getting questions wrong, sometimes failing to provide

51
00:02:50,520 --> 00:02:53,800
Speaker 1: references even for correct answers, with another complaining that the

52
00:02:53,840 --> 00:02:57,080
Speaker 1: copilot was and I quote using content not connected as

53
00:02:57,120 --> 00:02:59,919
Speaker 1: a document resource to answer questions. And by the way,

54
00:03:00,040 --> 00:03:02,119
Speaker 1: the whole point of share point is that it's your

55
00:03:02,200 --> 00:03:06,400
Speaker 1: data informing everything. I assume it was drawing from its

56
00:03:06,440 --> 00:03:11,040
Speaker 1: training data or perhaps the internet anyway, genuinely not useful.

57
00:03:11,560 --> 00:03:14,960
Speaker 1: And you'd think that with these new services that don't

58
00:03:14,960 --> 00:03:18,560
Speaker 1: seem that useful, that are questionably useful, that Microsoft would

59
00:03:18,520 --> 00:03:21,800
Speaker 1: be doing people a deal right wrong. How much is

60
00:03:21,800 --> 00:03:26,119
Speaker 1: Microsoft charging for these services? Thirty dollars a seat per

61
00:03:26,200 --> 00:03:29,040
Speaker 1: person on top of what you are already paying, are

62
00:03:29,080 --> 00:03:31,880
Speaker 1: as much as fifty dollars a month extra for specialist

63
00:03:31,880 --> 00:03:36,000
Speaker 1: products like co pilots for sales, Microsoft is effectively asking

64
00:03:36,040 --> 00:03:39,280
Speaker 1: customers to double their spend, and by the way, that's

65
00:03:39,320 --> 00:03:42,200
Speaker 1: with an annual commitment for products that don't seem to

66
00:03:42,240 --> 00:03:45,600
Speaker 1: be that helpful. And really, that is kind of the

67
00:03:45,720 --> 00:03:49,080
Speaker 1: state of generative AI, the literal leader and productivity in

68
00:03:49,120 --> 00:03:51,920
Speaker 1: business software, cannot seem to find a product that will

69
00:03:51,960 --> 00:03:55,440
Speaker 1: make people more productive and that they will then pay for.

70
00:03:55,800 --> 00:03:58,240
Speaker 1: And it's in part because the results are kind of mediocre,

71
00:03:58,600 --> 00:04:01,000
Speaker 1: and also that the costs are so burdensome that there's

72
00:04:01,000 --> 00:04:05,040
Speaker 1: no way for Microsoft to avoid charging a premium. And really,

73
00:04:05,080 --> 00:04:07,720
Speaker 1: if Microsoft needs to charge this much, it's either because

74
00:04:07,720 --> 00:04:10,320
Speaker 1: Sachin Adela is really desperate to hit half a trillion

75
00:04:10,360 --> 00:04:13,520
Speaker 1: dollars in revenue by twenty thirty, or that the costs

76
00:04:13,520 --> 00:04:16,000
Speaker 1: are too high to charge much less. Maybe it's a

77
00:04:16,040 --> 00:04:19,359
Speaker 1: little bit of both. And this all only serves to

78
00:04:19,400 --> 00:04:22,400
Speaker 1: shed further light on just the mediocrity of generative AI

79
00:04:22,640 --> 00:04:26,200
Speaker 1: and how limited large language models are. And all of this,

80
00:04:26,279 --> 00:04:29,640
Speaker 1: by the way, is existentially threatening to Open AI because

81
00:04:30,200 --> 00:04:33,880
Speaker 1: they've coasted to one hundred and fifty seven billion dollar valuation,

82
00:04:34,240 --> 00:04:38,120
Speaker 1: almost entirely based on hype. And so it's that company's

83
00:04:38,120 --> 00:04:40,080
Speaker 1: always tried to tell us that the future of AI

84
00:04:40,120 --> 00:04:42,240
Speaker 1: will blow us away, that the next generation of large

85
00:04:42,279 --> 00:04:45,080
Speaker 1: language models are eminent and they're going to be incredible.

86
00:04:45,400 --> 00:04:48,360
Speaker 1: And the artificial general intelligence, where machines can reason and

87
00:04:48,440 --> 00:04:52,520
Speaker 1: act beyond human capabilities, that's just around the corner. And

88
00:04:52,560 --> 00:04:54,520
Speaker 1: by the way, all of that is in part thanks

89
00:04:54,560 --> 00:04:57,000
Speaker 1: to the media sloping it down and just assuming that

90
00:04:57,040 --> 00:05:00,280
Speaker 1: they get it right. But until now that that's all

91
00:05:00,279 --> 00:05:03,279
Speaker 1: they've really had to do. But I think we're finally

92
00:05:03,320 --> 00:05:06,480
Speaker 1: getting the rubber meeting the road with this. I previously

93
00:05:06,480 --> 00:05:08,919
Speaker 1: said one of the pale horses of the AI apocalypse

94
00:05:08,960 --> 00:05:11,839
Speaker 1: is when a big stupid magic trick was necessary, a

95
00:05:11,880 --> 00:05:14,160
Speaker 1: product that someone shoves out the door in hopes it

96
00:05:14,200 --> 00:05:17,400
Speaker 1: will impress people and keep them believing in the magical future.

97
00:05:18,400 --> 00:05:20,600
Speaker 1: And you'd think that they'd have something really good right

98
00:05:20,640 --> 00:05:23,400
Speaker 1: now because open ai just raised all this money and

99
00:05:23,440 --> 00:05:28,880
Speaker 1: the practical applications just are obviously not there, except well,

100
00:05:29,240 --> 00:05:31,000
Speaker 1: you know, no, no, no, no, this is open AI.

101
00:05:31,080 --> 00:05:33,920
Speaker 1: They wouldn't make a big, stupid mistake, would they. I mean,

102
00:05:34,279 --> 00:05:36,000
Speaker 1: one of the things I always tell clients of mine

103
00:05:36,000 --> 00:05:37,760
Speaker 1: in pr is not to shove a product out the

104
00:05:37,800 --> 00:05:40,080
Speaker 1: door before it's ready, and to also make sure it's

105
00:05:40,160 --> 00:05:43,760
Speaker 1: really obvious why people should pay for it. Otherwise you're

106
00:05:43,800 --> 00:05:46,400
Speaker 1: just kind of launching something into the ether and hope

107
00:05:46,400 --> 00:05:49,880
Speaker 1: people will find a reason to sell it for you.

108
00:05:51,120 --> 00:05:53,599
Speaker 1: And yeah, that's exactly what they did. It happened. On

109
00:05:53,640 --> 00:05:56,800
Speaker 1: September twelfth. Open Ai launched OH one, which had been

110
00:05:56,880 --> 00:05:59,719
Speaker 1: code named Strawberry, with all of the excitement as a

111
00:05:59,720 --> 00:06:04,120
Speaker 1: trip to the Proctologist. Across a series of tweets, CEO

112
00:06:04,240 --> 00:06:07,320
Speaker 1: Samuel and described one as open ayes most capable and

113
00:06:07,360 --> 00:06:11,000
Speaker 1: aligned models, yet then immediately conceded that O one was

114
00:06:11,040 --> 00:06:14,440
Speaker 1: still flawed, still limited, and it still seems more impressive

115
00:06:14,480 --> 00:06:16,599
Speaker 1: on its first use than it does after you spend

116
00:06:16,600 --> 00:06:20,200
Speaker 1: more time with it. Oh my god, he admitted. He

117
00:06:20,240 --> 00:06:23,279
Speaker 1: then promised it would deliver more accurate results when performing

118
00:06:23,279 --> 00:06:26,040
Speaker 1: the kinds of activities where there's a definitive right answer,

119
00:06:26,160 --> 00:06:30,159
Speaker 1: like coding maths or answering science questions. One might think

120
00:06:30,200 --> 00:06:32,320
Speaker 1: that he'd walk in with I don't know, like a

121
00:06:32,360 --> 00:06:35,000
Speaker 1: product built on top of one or like an use

122
00:06:35,080 --> 00:06:37,880
Speaker 1: case or thing that would make the audience go, wow,

123
00:06:38,000 --> 00:06:41,720
Speaker 1: I could build something with this. He didn't. I don't

124
00:06:41,760 --> 00:06:44,119
Speaker 1: think he wants to try. I don't think he hasn't

125
00:06:44,120 --> 00:06:46,520
Speaker 1: had to try that hard. So far people have been

126
00:06:46,560 --> 00:06:50,040
Speaker 1: sloping down his slop happily. This boy may not have

127
00:06:50,080 --> 00:06:54,279
Speaker 1: any tricks left. But let's talk about how O one works.

128
00:06:54,279 --> 00:06:55,839
Speaker 1: And I'm going to introduce you to a bunch of

129
00:06:55,880 --> 00:06:58,080
Speaker 1: new concepts here, but I promise I won't get too

130
00:06:58,120 --> 00:07:00,760
Speaker 1: deep into the weeds. And I really want you to

131
00:07:00,800 --> 00:07:04,360
Speaker 1: know how these machines work. It's critical for critiquing these companies.

132
00:07:04,640 --> 00:07:06,880
Speaker 1: And the big way they take advantage of you is

133
00:07:06,920 --> 00:07:08,719
Speaker 1: that they claim all of this is black magic, that

134
00:07:08,760 --> 00:07:11,920
Speaker 1: you could never possibly understand it. You absolutely can. And

135
00:07:12,000 --> 00:07:13,880
Speaker 1: if you want their explanation, I'm going to have it

136
00:07:13,920 --> 00:07:18,880
Speaker 1: in the show notes. Okay. When presented with a problem,

137
00:07:19,000 --> 00:07:21,920
Speaker 1: OH one breaks it down into individual steps that hopefully

138
00:07:22,000 --> 00:07:24,400
Speaker 1: would lead to a correct answer in a process called

139
00:07:24,480 --> 00:07:28,040
Speaker 1: chain of thought. Again, these things are not thinking. They're

140
00:07:28,040 --> 00:07:30,960
Speaker 1: not thinking, but this is the term. It's also a

141
00:07:31,000 --> 00:07:33,920
Speaker 1: little easier if you think of OH one as two

142
00:07:34,000 --> 00:07:37,480
Speaker 1: parts of one model. On each step, one part of

143
00:07:37,520 --> 00:07:40,560
Speaker 1: the model applies something called reinforcement learning with the other one,

144
00:07:40,600 --> 00:07:44,320
Speaker 1: which is the model actually outputting things rewarded or punished

145
00:07:44,360 --> 00:07:47,320
Speaker 1: based on the perceived correctness of their progress. And this

146
00:07:47,520 --> 00:07:49,680
Speaker 1: is what is called reasoning, by the way, even though

147
00:07:49,680 --> 00:07:53,320
Speaker 1: it really doesn't match human reasoning at all, and then

148
00:07:53,440 --> 00:07:56,360
Speaker 1: based on the reward of the punishment, it generates a

149
00:07:56,400 --> 00:08:00,560
Speaker 1: final answer from this chain of thought consideration. This is

150
00:08:00,640 --> 00:08:03,000
Speaker 1: different to how other large language models work in the

151
00:08:03,040 --> 00:08:06,600
Speaker 1: sense that the model is generating outputs than actually looking

152
00:08:06,640 --> 00:08:09,280
Speaker 1: back at them then ignoring or approving what it thinks

153
00:08:09,280 --> 00:08:11,320
Speaker 1: are good steps to get to an answer, rather than

154
00:08:11,440 --> 00:08:14,640
Speaker 1: just generating one and saying here's the answer. This may

155
00:08:14,680 --> 00:08:17,200
Speaker 1: seem like a big breakthrough or even another step towards

156
00:08:17,280 --> 00:08:21,000
Speaker 1: artificial general intelligence, and it isn't, And you can tell

157
00:08:21,040 --> 00:08:23,119
Speaker 1: that by the fact that open ai opt to release

158
00:08:23,160 --> 00:08:25,800
Speaker 1: O one as its own standalone product rather than something

159
00:08:25,840 --> 00:08:29,840
Speaker 1: built into GPT. It's also telling that the examples demonstrated

160
00:08:29,880 --> 00:08:32,800
Speaker 1: by open AI, like maths and science problems, are the

161
00:08:32,800 --> 00:08:34,839
Speaker 1: ones where the answer can be known ahead of time

162
00:08:35,040 --> 00:08:37,720
Speaker 1: and a solution is either correct or false, thus allowing

163
00:08:37,760 --> 00:08:39,880
Speaker 1: the model to guide the chain of thought through each

164
00:08:39,920 --> 00:08:43,679
Speaker 1: step towards that answer, rather than actually having to produce

165
00:08:43,720 --> 00:08:47,480
Speaker 1: something where they might not necessarily be one. Open ai

166
00:08:47,600 --> 00:08:50,920
Speaker 1: didn't show the one model trying to tackle complex problems

167
00:08:51,400 --> 00:08:54,520
Speaker 1: such as high end mathematical equations or otherwise where the

168
00:08:54,520 --> 00:08:57,640
Speaker 1: solution isn't known in advance by its own admission. Open

169
00:08:57,679 --> 00:09:00,320
Speaker 1: AI has heard reports that one is actually more prone

170
00:09:00,320 --> 00:09:03,000
Speaker 1: to hallucinations than GPT four H, and the model is

171
00:09:03,080 --> 00:09:05,199
Speaker 1: less inclined to admit when it doesn't have the answer

172
00:09:05,280 --> 00:09:08,240
Speaker 1: to a question when compared to other previous models. This

173
00:09:08,400 --> 00:09:10,960
Speaker 1: is because despite there being a model that checks the

174
00:09:11,000 --> 00:09:14,080
Speaker 1: work of the model, the work checking part of the

175
00:09:14,120 --> 00:09:17,320
Speaker 1: model is still capable of hallucinations. It's kind of like

176
00:09:18,000 --> 00:09:20,280
Speaker 1: a kid being taught something by a teacher who just

177
00:09:20,320 --> 00:09:23,880
Speaker 1: occasionally gets things horribly wrong. That child, though they may

178
00:09:24,200 --> 00:09:28,280
Speaker 1: mostly get right answers, will learn bad things. Now, learning

179
00:09:28,320 --> 00:09:31,520
Speaker 1: here isn't really what's happening, but the output of the

180
00:09:31,640 --> 00:09:35,280
Speaker 1: end will be informed by a model that makes the loocinations.

181
00:09:35,280 --> 00:09:37,560
Speaker 1: It's like, I don't know, got a town full of dogs.

182
00:09:37,600 --> 00:09:39,720
Speaker 1: You get a bunch of baboons in to get rid

183
00:09:39,760 --> 00:09:41,599
Speaker 1: of the dogs. The baboons succeed and getting rid of

184
00:09:41,640 --> 00:09:43,199
Speaker 1: the dogs Now you just got a bunch of baboons,

185
00:09:43,200 --> 00:09:46,800
Speaker 1: so you get in, aren't no robots? Robots destroy the baboons.

186
00:09:46,800 --> 00:09:48,960
Speaker 1: At this point, you've got robots. If the robots are autonomous,

187
00:09:48,960 --> 00:09:50,839
Speaker 1: they start taking over the town. So they need to

188
00:09:50,880 --> 00:09:52,720
Speaker 1: find a bigger robot to take over the town from

189
00:09:52,800 --> 00:09:56,080
Speaker 1: the robots. Now you've just got an escalating problem where

190
00:09:56,080 --> 00:09:58,520
Speaker 1: things are only going to get worse. And if you

191
00:09:58,559 --> 00:10:01,760
Speaker 1: work open ai and that sounds accurate, please email me anyway.

192
00:10:02,600 --> 00:10:05,360
Speaker 1: According to open ai, OH one also, thanks to this

193
00:10:05,559 --> 00:10:08,679
Speaker 1: chain of thought process, feels more convincing to human users

194
00:10:08,679 --> 00:10:12,120
Speaker 1: because it provides more detailed answers, and thus people are

195
00:10:12,120 --> 00:10:16,000
Speaker 1: more inclined to trust the outputs even when they're completely wrong. Now,

196
00:10:16,000 --> 00:10:18,320
Speaker 1: if you think I'm being overly hard on open ai,

197
00:10:18,480 --> 00:10:20,920
Speaker 1: consider the ways in which the company is marketed. One

198
00:10:21,760 --> 00:10:25,760
Speaker 1: open ai described O one's reinforcement training as thinking and

199
00:10:25,880 --> 00:10:28,720
Speaker 1: reasoning when it's making guesses and then guessing on the

200
00:10:28,760 --> 00:10:31,320
Speaker 1: correctness of these guesses at each step. Where the end

201
00:10:31,360 --> 00:10:34,000
Speaker 1: destination is often something that can be known in advance.

202
00:10:34,760 --> 00:10:38,920
Speaker 1: Generative AI does not know anything. These are still probabilistic models.

203
00:10:39,760 --> 00:10:42,840
Speaker 1: This thing is not thinking at all. There is no reasoning.

204
00:10:43,480 --> 00:10:45,480
Speaker 1: It's got a model, reading a model, giving a model

205
00:10:45,480 --> 00:10:48,800
Speaker 1: answers from it's a mess, and it's an insult to people,

206
00:10:48,960 --> 00:10:52,200
Speaker 1: actual human beings who, when they think, are acting based

207
00:10:52,240 --> 00:10:56,120
Speaker 1: on many, many complex factors, their experience, their knowledge, the

208
00:10:56,200 --> 00:11:00,119
Speaker 1: knowledge they've accumulated over years of experiences, their brain chemistry,

209
00:11:00,559 --> 00:11:03,480
Speaker 1: so on and so forth. Well, we may to guess

210
00:11:03,520 --> 00:11:07,000
Speaker 1: about the correctness of each thing we're guessing at, and

211
00:11:07,040 --> 00:11:09,880
Speaker 1: we may reason through a complex problem. All of this

212
00:11:09,960 --> 00:11:13,120
Speaker 1: is based on something concrete. When we get something wrong,

213
00:11:13,160 --> 00:11:18,120
Speaker 1: it's based on actual experience versus training data and probabilistic models.

214
00:11:19,240 --> 00:11:22,559
Speaker 1: This shit is not thinking at all, and by god

215
00:11:22,679 --> 00:11:26,240
Speaker 1: is it expensive. Pricing for one preview, which is the

216
00:11:26,240 --> 00:11:29,560
Speaker 1: first model, is fifteen dollars per million input tokens and

217
00:11:29,600 --> 00:11:32,840
Speaker 1: sixteen per million output tokens. In essence, it's three times

218
00:11:32,880 --> 00:11:35,800
Speaker 1: as expensive as their most expensive model GPT four O

219
00:11:36,160 --> 00:11:39,280
Speaker 1: for input and four times is expensive for output. And

220
00:11:39,320 --> 00:11:42,840
Speaker 1: then there's a hidden cost. Data scientist Max Wolf reported

221
00:11:42,840 --> 00:11:45,520
Speaker 1: the open ayes reasoning tokens the output it uses to

222
00:11:45,520 --> 00:11:47,880
Speaker 1: get you to the final answer where it says, Okay,

223
00:11:47,960 --> 00:11:51,600
Speaker 1: I need to find out the solution to this problem.

224
00:11:51,640 --> 00:11:55,080
Speaker 1: So here are the thirty steps I've gone through. Yeah,

225
00:11:55,400 --> 00:11:58,880
Speaker 1: those are actually generated using the most expensive tokens, the

226
00:11:58,920 --> 00:12:01,439
Speaker 1: output tokens. So the more it has to think, the

227
00:12:01,480 --> 00:12:04,520
Speaker 1: more expensive it gets. All of the things it generates

228
00:12:04,520 --> 00:12:08,040
Speaker 1: to consider an answer are also charged for, which means

229
00:12:08,040 --> 00:12:10,160
Speaker 1: the more complex it is, the more expensive it's going

230
00:12:10,240 --> 00:12:13,800
Speaker 1: to be worse. Still, if you integrate this model, open

231
00:12:13,840 --> 00:12:18,440
Speaker 1: ai does not show you what it's reasoning. All of

232
00:12:18,480 --> 00:12:21,839
Speaker 1: that calculation happens in the background, and they still charge

233
00:12:21,880 --> 00:12:25,240
Speaker 1: you for it. You just don't know how much every

234
00:12:25,440 --> 00:12:29,720
Speaker 1: oh one step is charged to you in an indeterminate way,

235
00:12:30,000 --> 00:12:32,600
Speaker 1: and open ai claims that they can't show you because

236
00:12:32,600 --> 00:12:38,400
Speaker 1: of competitive reasons. Ugh, nasty company, really greasy and they're

237
00:12:38,400 --> 00:12:42,680
Speaker 1: still gonna burn Okay, okay, though it's different GPT four.

238
00:12:42,720 --> 00:12:45,880
Speaker 1: Oh and it's really expensive. But is it better? Of

239
00:12:45,920 --> 00:12:49,240
Speaker 1: course it must be better, right, right? It sounds great.

240
00:12:49,440 --> 00:12:52,000
Speaker 1: It's thinking, right, it's reasoning right. No, No, it's not,

241
00:12:52,360 --> 00:12:57,559
Speaker 1: it's not. It's worse. This crap's worse. Let's talk about accuracy.

242
00:12:57,760 --> 00:13:00,000
Speaker 1: On how can news the reddit styles I owned by

243
00:13:00,000 --> 00:13:03,160
Speaker 1: am Horman's former alum y Combinator, one person complained about

244
00:13:03,160 --> 00:13:06,240
Speaker 1: O one hallucinating libraries and functions when presented with a

245
00:13:06,360 --> 00:13:09,079
Speaker 1: programming task, and making mistakes when asked questions where the

246
00:13:09,080 --> 00:13:13,160
Speaker 1: answer isn't readily available in the Internet. On Twitter, Henrik Nyberg,

247
00:13:13,320 --> 00:13:15,839
Speaker 1: a startup founder and former game developer, asked OH one

248
00:13:15,880 --> 00:13:18,600
Speaker 1: to write a Python program that multiplied two numbers, then

249
00:13:18,600 --> 00:13:21,959
Speaker 1: calculated the expected output of said program. While OH one

250
00:13:22,080 --> 00:13:24,560
Speaker 1: correctly wrote the code, although said code could have been

251
00:13:24,559 --> 00:13:29,880
Speaker 1: more succinct, the actual result was wildly incorrect. Karthig Cannon, himself,

252
00:13:29,920 --> 00:13:32,440
Speaker 1: a founder of an AI company, tried a programming task

253
00:13:32,520 --> 00:13:35,199
Speaker 1: on O one where it also hallucinated a non existent

254
00:13:35,200 --> 00:13:39,920
Speaker 1: command for the API he was using. Another person, Sashi Yanshin,

255
00:13:40,160 --> 00:13:41,640
Speaker 1: tried to play a game of chess with O one,

256
00:13:41,679 --> 00:13:44,360
Speaker 1: and it hallucinated an entire piece onto the board and

257
00:13:44,400 --> 00:13:46,880
Speaker 1: then it lost. And because I'm a little shit, I

258
00:13:46,920 --> 00:13:49,079
Speaker 1: also tried asking one to listen a number of states

259
00:13:49,080 --> 00:13:51,920
Speaker 1: with A in the name. After contemplating for eighteen seconds,

260
00:13:51,960 --> 00:13:55,240
Speaker 1: it provided the names of thirty seven states, including Mississippi,

261
00:13:55,440 --> 00:13:56,960
Speaker 1: you know, the classic state with an A in it.

262
00:13:57,360 --> 00:13:59,679
Speaker 1: By the way, there are thirty six states that have

263
00:14:00,080 --> 00:14:02,520
Speaker 1: in them, just in case you're curious. I then asked

264
00:14:02,520 --> 00:14:04,240
Speaker 1: for a list of states with the letter W in

265
00:14:04,240 --> 00:14:05,880
Speaker 1: the name, and then it sat and it thought for

266
00:14:05,920 --> 00:14:09,440
Speaker 1: eleven seconds, and then included North Carolina and North Dakota.

267
00:14:10,360 --> 00:14:13,000
Speaker 1: Great stuff. By the way, I also asked tho one

268
00:14:13,000 --> 00:14:14,840
Speaker 1: to count the number of times the letter R appears

269
00:14:14,840 --> 00:14:17,480
Speaker 1: in the word strawberry, which is the pre release code

270
00:14:17,559 --> 00:14:20,400
Speaker 1: name for this. It's said too, I would have hard

271
00:14:20,400 --> 00:14:23,080
Speaker 1: coded that one. Personally. You can't give me that kind

272
00:14:23,120 --> 00:14:26,040
Speaker 1: of joy now. Open AI claims that I one performs

273
00:14:26,040 --> 00:14:30,120
Speaker 1: similarly to PhD students on challenging benchmark tasks in physics, chemistry,

274
00:14:30,120 --> 00:14:33,560
Speaker 1: and biology, just not in geography, it seems, or basic

275
00:14:33,640 --> 00:14:38,520
Speaker 1: elementary level English, or maths or programming. Also, I mean

276
00:14:38,720 --> 00:14:42,160
Speaker 1: for the PhD listeners. I've met a few PhD people

277
00:14:42,160 --> 00:14:44,800
Speaker 1: who authoritatively state things that are completely untrue that they

278
00:14:44,800 --> 00:14:47,480
Speaker 1: know nothing about. This is not a broad stroke thing,

279
00:14:47,560 --> 00:14:51,200
Speaker 1: but I get the sense that it's true anyway. This

280
00:14:51,400 --> 00:14:53,960
Speaker 1: is I should know, the big stupid magic trick I

281
00:14:54,000 --> 00:14:57,120
Speaker 1: predicted in the past. Open AI is shoving strawberry out

282
00:14:57,120 --> 00:14:58,840
Speaker 1: the door as a means of proving to investors and

283
00:14:58,880 --> 00:15:01,800
Speaker 1: the greater public. But they've still got it that the

284
00:15:01,840 --> 00:15:05,480
Speaker 1: AI revolution is still here, that this thing is thinking,

285
00:15:06,280 --> 00:15:08,520
Speaker 1: and what they actually have is a clunky, on exciting

286
00:15:08,560 --> 00:15:10,760
Speaker 1: and expensive model that doesn't really seem to have any

287
00:15:10,800 --> 00:15:14,600
Speaker 1: measurable improvement. Okay, I'm sorry. It has a measurable improvement.

288
00:15:15,000 --> 00:15:17,240
Speaker 1: You can measure it on the weird rigged test they

289
00:15:17,240 --> 00:15:19,600
Speaker 1: do for all of these things. And the thing is,

290
00:15:20,840 --> 00:15:24,200
Speaker 1: at this point, you'd think that even Apple, when they

291
00:15:24,200 --> 00:15:26,040
Speaker 1: pulled together a new thing, even when they had the

292
00:15:26,160 --> 00:15:28,240
Speaker 1: first Apple Watch and it was not obvious why you

293
00:15:28,240 --> 00:15:30,440
Speaker 1: had to own it, they still had apps that were

294
00:15:30,440 --> 00:15:33,000
Speaker 1: connected to it. They still had things you could point

295
00:15:33,000 --> 00:15:36,120
Speaker 1: out and go, oh, that's cool, I've got four square

296
00:15:36,120 --> 00:15:38,359
Speaker 1: on this four square on there at the time. Nevertheless,

297
00:15:38,440 --> 00:15:41,360
Speaker 1: they had apps to show. I just feel like open

298
00:15:41,400 --> 00:15:45,680
Speaker 1: Ai has this deep contempt for Silicon Valley and for

299
00:15:45,800 --> 00:15:48,720
Speaker 1: the world at large. They don't even have it in

300
00:15:48,760 --> 00:15:51,640
Speaker 1: them to be like, Okay, we have this new model,

301
00:15:51,800 --> 00:15:53,960
Speaker 1: and here is the new thing we built with it,

302
00:15:54,280 --> 00:15:57,480
Speaker 1: and this thing does this and now you will see

303
00:15:57,480 --> 00:16:01,600
Speaker 1: how important this company is. Instead, we get this crap.

304
00:16:02,680 --> 00:16:06,160
Speaker 1: We just get this very boring crap. And sure, I'm

305
00:16:06,160 --> 00:16:09,280
Speaker 1: sure someone technical is going to email me and say, ed, wow,

306
00:16:09,400 --> 00:16:11,760
Speaker 1: chain thought reasoning. There are other companies that have been

307
00:16:11,760 --> 00:16:14,880
Speaker 1: doing it already. Anthropic already had something like this, and

308
00:16:14,960 --> 00:16:18,120
Speaker 1: even then they didn't do shit with it. Where's the product, man,

309
00:16:18,320 --> 00:16:22,400
Speaker 1: where's the thing I meant to care about? Why should

310
00:16:22,440 --> 00:16:25,880
Speaker 1: anybody give a shit about this? Well, sam Altman is

311
00:16:25,960 --> 00:16:29,480
Speaker 1: likely trying to trump up the reasoning abilities of O one. Well,

312
00:16:29,520 --> 00:16:32,160
Speaker 1: people you know, such as the people bankrolling him, will

313
00:16:32,200 --> 00:16:34,440
Speaker 1: actually see he's a ten to twenty second waiting time

314
00:16:34,480 --> 00:16:37,280
Speaker 1: for an answer which may or may not be correct.

315
00:16:37,640 --> 00:16:39,840
Speaker 1: But you have a bit more detail, which isn't even

316
00:16:39,880 --> 00:16:43,960
Speaker 1: the reasoning happening because open AI hides that bit. Nobody

317
00:16:43,960 --> 00:16:47,640
Speaker 1: gives a shit about better answers anymore. They want generative

318
00:16:47,680 --> 00:16:50,080
Speaker 1: AI to do something new, and I don't think the

319
00:16:50,080 --> 00:16:52,600
Speaker 1: open AI has any idea how to make that happen.

320
00:16:53,080 --> 00:16:56,320
Speaker 1: Sam Orman's limp shitty attempts to anthropomorphize I one by

321
00:16:56,360 --> 00:16:59,520
Speaker 1: making it think can use reasoning obvious attempts to suggest

322
00:16:59,520 --> 00:17:01,960
Speaker 1: that this is on how part of the path through AGI.

323
00:17:02,160 --> 00:17:06,200
Speaker 1: But even the most staunch AI advocates, well, they can't

324
00:17:06,200 --> 00:17:08,920
Speaker 1: seem to get excited about this. In fact, I kind

325
00:17:08,920 --> 00:17:10,800
Speaker 1: of argue that O one shows that open AI is

326
00:17:10,840 --> 00:17:14,119
Speaker 1: desperate and out of ideas. Now if you don't have

327
00:17:14,160 --> 00:17:17,440
Speaker 1: any ideas, though, the following advertisements will be more than

328
00:17:17,480 --> 00:17:20,280
Speaker 1: happy to fill your empty little brain with new ideas

329
00:17:20,359 --> 00:17:23,280
Speaker 1: that involve giving someone money or downloading something. And I

330
00:17:23,400 --> 00:17:28,800
Speaker 1: must implore you to just accept everything that follows. I

331
00:17:28,880 --> 00:17:30,720
Speaker 1: don't endorse any of it because I don't know what

332
00:17:30,760 --> 00:17:45,440
Speaker 1: it's going to be, but you must. And we're back.

333
00:17:47,000 --> 00:17:49,080
Speaker 1: So I think now is a good time to get

334
00:17:49,119 --> 00:17:52,080
Speaker 1: back to the root of the generative AI problem. Generative

335
00:17:52,080 --> 00:17:54,960
Speaker 1: AI is being sold to you on multiple lives that

336
00:17:55,040 --> 00:17:58,760
Speaker 1: it's AI, it's actually artificial intelligence, it's going to get better,

337
00:17:59,119 --> 00:18:02,560
Speaker 1: that this will become artificial general intelligence, that this will

338
00:18:02,600 --> 00:18:06,480
Speaker 1: become the thinking computer, and all of this is inevitable.

339
00:18:07,640 --> 00:18:10,879
Speaker 1: Putting aside terms like performance, as they're largely us as

340
00:18:10,960 --> 00:18:13,920
Speaker 1: a means of generating things accurately or faster, rather than

341
00:18:13,920 --> 00:18:17,520
Speaker 1: being good at anything. Large language models have effectively platowed

342
00:18:18,040 --> 00:18:21,120
Speaker 1: more powerful never seems to mean does more, and more

343
00:18:21,160 --> 00:18:24,760
Speaker 1: powerful often means more expensive to run or more expensive

344
00:18:24,760 --> 00:18:27,080
Speaker 1: for you as the user to access, meaning that you've

345
00:18:27,160 --> 00:18:29,320
Speaker 1: just made something that doesn't do more and does cost

346
00:18:29,359 --> 00:18:32,520
Speaker 1: more to run. If the combined forces of every venture

347
00:18:32,560 --> 00:18:34,879
Speaker 1: capitalist and big tech hyperscaler have yet to come up

348
00:18:34,920 --> 00:18:36,800
Speaker 1: with a meaningful use case that lots of people will

349
00:18:36,800 --> 00:18:39,920
Speaker 1: actually pay for. I just don't see one coming. Large

350
00:18:39,960 --> 00:18:42,320
Speaker 1: language models and yes, that's where all of these billions

351
00:18:42,359 --> 00:18:44,919
Speaker 1: of dollars are going. Are not going to magically sprout

352
00:18:44,920 --> 00:18:47,840
Speaker 1: new capabilities a big Tech and open Ai burn another

353
00:18:47,880 --> 00:18:51,200
Speaker 1: one hundred and fifty billion dollars. And yes that number

354
00:18:51,240 --> 00:18:54,000
Speaker 1: isn't hyperbole. It's actually pretty close to the amount being

355
00:18:54,040 --> 00:18:56,800
Speaker 1: plowed into these companies when you include things like investments

356
00:18:56,800 --> 00:18:59,840
Speaker 1: in companies like Anthropic and open Ai. And they're genuinely

357
00:18:59,840 --> 00:19:02,120
Speaker 1: and sane amount of capex from the likes of Google,

358
00:19:02,160 --> 00:19:06,240
Speaker 1: Amazon and Microsoft going into expanding data centers and buying GPUs.

359
00:19:07,040 --> 00:19:09,760
Speaker 1: Nobody seems to be trying to make these things more efficient,

360
00:19:09,960 --> 00:19:12,199
Speaker 1: or at the very least nobody's succeeded in doing so,

361
00:19:12,359 --> 00:19:14,639
Speaker 1: because I think if they had, they'd be shouting it

362
00:19:14,640 --> 00:19:18,320
Speaker 1: from the rooftops and as on the side. By the way,

363
00:19:18,920 --> 00:19:21,680
Speaker 1: the biggest sign that no one's actually making money from

364
00:19:21,680 --> 00:19:23,760
Speaker 1: this is that no one's talking about how much money

365
00:19:23,760 --> 00:19:28,520
Speaker 1: they're making. Microsoft and all of these companies they love

366
00:19:28,640 --> 00:19:32,680
Speaker 1: talking about making profit. They love doing that. Beyond earnings.

367
00:19:32,720 --> 00:19:36,160
Speaker 1: They love talking about it instead whenever they're asked to go, oh, hey,

368
00:19:36,160 --> 00:19:38,520
Speaker 1: I will do some things in the future, I need

369
00:19:38,560 --> 00:19:39,879
Speaker 1: to take a phone CALLT and then they kind of

370
00:19:39,920 --> 00:19:43,040
Speaker 1: disappear from the room. Amy heard CFO of Microsoft classic

371
00:19:43,040 --> 00:19:48,280
Speaker 1: bullshit artists dancing around Yeah, oh net revenue increase checking

372
00:19:48,280 --> 00:19:52,720
Speaker 1: a watch. It's just really sad. It's really sad because

373
00:19:52,720 --> 00:19:55,320
Speaker 1: what we have here is a shared delusion, a shared

374
00:19:55,320 --> 00:19:58,399
Speaker 1: delusion about a dead end technology that runs on copyright theft,

375
00:19:58,560 --> 00:20:01,040
Speaker 1: one that requires a CONTINUALUS supply of capital to keep

376
00:20:01,119 --> 00:20:03,720
Speaker 1: running as it provides services that are at best in

377
00:20:03,960 --> 00:20:06,080
Speaker 1: essential sold to US dressed up as a kind of

378
00:20:06,119 --> 00:20:09,159
Speaker 1: automation that does not exist, and it doesn't provide, costing

379
00:20:09,200 --> 00:20:11,840
Speaker 1: billions and millions of dollars and continuing to do so

380
00:20:12,200 --> 00:20:16,000
Speaker 1: im perpetuity. Generative AI doesn't run on money or cloud

381
00:20:16,040 --> 00:20:18,399
Speaker 1: credit so much as it does on faith. And the

382
00:20:18,440 --> 00:20:21,240
Speaker 1: problem is that faith, like investor capital, is actually a

383
00:20:21,280 --> 00:20:24,760
Speaker 1: finite resource. And that's where I bring you one of

384
00:20:24,800 --> 00:20:28,320
Speaker 1: my biggest anxieties about this industry, because I think we're

385
00:20:28,359 --> 00:20:30,919
Speaker 1: in the midst of a SUBPRIMEI crisis where thousands of

386
00:20:30,960 --> 00:20:34,880
Speaker 1: companies have integrated this stuff into their software at prices

387
00:20:34,920 --> 00:20:38,240
Speaker 1: that are far from stable and even further from profitable

388
00:20:38,240 --> 00:20:41,240
Speaker 1: for the services providing them. This concern, by the way,

389
00:20:41,320 --> 00:20:44,199
Speaker 1: isn't unfounded. At the latest open ai dev Day, they

390
00:20:44,240 --> 00:20:46,960
Speaker 1: said that they'd slash prices for their APIs by ninety

391
00:20:47,000 --> 00:20:50,080
Speaker 1: nine percent over the previous two years, largely as tech crunchies.

392
00:20:50,119 --> 00:20:53,320
Speaker 1: MAXT theorized due to price pressure from Meta and Google,

393
00:20:53,480 --> 00:20:55,840
Speaker 1: both of whom want to take that API access for

394
00:20:56,480 --> 00:21:00,880
Speaker 1: I assume some reason. Anyway, almost every AI powered startup

395
00:21:01,000 --> 00:21:03,960
Speaker 1: uses large language model features is based on some combination

396
00:21:04,040 --> 00:21:07,360
Speaker 1: of GPT or chlaud so open Ai or Anthropics models.

397
00:21:07,800 --> 00:21:10,680
Speaker 1: These models are built by two companies that are deeply unprofitable.

398
00:21:10,720 --> 00:21:13,720
Speaker 1: Open Ai they can lose five billion this year, Anthropic

399
00:21:13,880 --> 00:21:16,280
Speaker 1: is on course to lose two point seven billion this

400
00:21:16,440 --> 00:21:20,000
Speaker 1: year on much less revenue, and they all have pricing

401
00:21:20,040 --> 00:21:22,240
Speaker 1: design to get more customers through the door than make

402
00:21:22,280 --> 00:21:25,800
Speaker 1: any kind of profit. Open Ai, as mentioned, is subsidized

403
00:21:25,800 --> 00:21:28,639
Speaker 1: by Microsoft, both in cloud credits they received in the

404
00:21:28,760 --> 00:21:32,280
Speaker 1: twenty twenty three investment and the preferential pricing Microsoft offers

405
00:21:32,280 --> 00:21:34,880
Speaker 1: for their cloud services about a quarter of the price

406
00:21:34,920 --> 00:21:38,280
Speaker 1: of what everyone else pays. And these companies willow open

407
00:21:38,320 --> 00:21:41,520
Speaker 1: Ai and Anthropic. Their pricing is entirely dependent on the

408
00:21:41,520 --> 00:21:43,640
Speaker 1: support of big tech in the case of open ai,

409
00:21:43,760 --> 00:21:47,399
Speaker 1: Microsoft's continued support. In the case of Anthropic, Amazon and

410
00:21:47,440 --> 00:21:51,240
Speaker 1: Google both as investors and service providers. Based on how

411
00:21:51,320 --> 00:21:54,680
Speaker 1: unprofitable these companies are. I hypothesized that if open ai

412
00:21:54,800 --> 00:21:57,880
Speaker 1: or Anthropic charge prices closer to their actual costs, they'll

413
00:21:57,880 --> 00:21:59,679
Speaker 1: be a ten to one hundred times increase in the

414
00:21:59,680 --> 00:22:02,199
Speaker 1: price of API calls, though it's impossible to say how

415
00:22:02,280 --> 00:22:06,639
Speaker 1: much without the actual numbers of direct burn from these companies. However,

416
00:22:06,760 --> 00:22:10,320
Speaker 1: Let's consider for a moment that the numbers reported by

417
00:22:10,320 --> 00:22:13,680
Speaker 1: the Information Estimate the open AI's server costs with Microsoft

418
00:22:13,680 --> 00:22:16,000
Speaker 1: will be four billion dollars in twenty twenty four, which

419
00:22:16,000 --> 00:22:18,119
Speaker 1: I add are over two and a half times cheaper

420
00:22:18,160 --> 00:22:21,959
Speaker 1: than what Microsoft charges others. It's like about four dollars

421
00:22:21,960 --> 00:22:24,280
Speaker 1: in something and they pay them out a dollar something

422
00:22:24,359 --> 00:22:27,600
Speaker 1: per GP per hour. And then consider, after knowing that

423
00:22:27,640 --> 00:22:30,919
Speaker 1: they're getting this massive discount, that open ai still loses

424
00:22:30,920 --> 00:22:34,160
Speaker 1: over five billion dollars a year. Open ai is more

425
00:22:34,160 --> 00:22:36,480
Speaker 1: than likely charging only a small percentage of what it

426
00:22:36,680 --> 00:22:39,160
Speaker 1: likely costs to run its models, and can only continue

427
00:22:39,160 --> 00:22:41,359
Speaker 1: to do so if it's able to continually raise more

428
00:22:41,440 --> 00:22:44,720
Speaker 1: venture funding than has ever been raised ever and continue

429
00:22:44,720 --> 00:22:48,040
Speaker 1: to receive preferential pricing from Microsoft, a company that recently

430
00:22:48,080 --> 00:22:50,920
Speaker 1: mentioned that it considers open Ai a competitor and has

431
00:22:51,000 --> 00:22:55,040
Speaker 1: complete access to its IP and research. While I can't

432
00:22:55,080 --> 00:22:57,399
Speaker 1: say for certain, I would think it's reasonable to believe

433
00:22:57,440 --> 00:23:00,560
Speaker 1: that Anthropic receives a similarly preferential pricing package from both

434
00:23:00,600 --> 00:23:03,960
Speaker 1: Amazon Web Services and Google Cloud. Both of those companies,

435
00:23:03,960 --> 00:23:06,960
Speaker 1: by the way, put billions into them. Assuming that Microsoft

436
00:23:07,040 --> 00:23:09,359
Speaker 1: gave open ai ten billion dollars of cloud credits and

437
00:23:09,400 --> 00:23:11,760
Speaker 1: it spent four billion on server costs and let's say

438
00:23:12,680 --> 00:23:15,440
Speaker 1: two three billion dollars on training costs, that are both

439
00:23:15,480 --> 00:23:18,360
Speaker 1: short to increase. With new models, open Ai will either

440
00:23:18,400 --> 00:23:20,520
Speaker 1: need more credits will have to pay actual cash to

441
00:23:20,560 --> 00:23:24,560
Speaker 1: Microsoft sometime in twenty twenty five, and Microsoft did participate

442
00:23:24,560 --> 00:23:26,720
Speaker 1: in the latest round, by the way, but it's not

443
00:23:26,760 --> 00:23:29,720
Speaker 1: obvious how much, and it was much less than last time,

444
00:23:29,760 --> 00:23:32,440
Speaker 1: which was I believe ten billion, mostly in cloud credits.

445
00:23:33,400 --> 00:23:35,960
Speaker 1: While it might be possible that Microsoft, Amazon and Google

446
00:23:35,960 --> 00:23:39,200
Speaker 1: extend their preferred pricing indefinitely, the question is whether these

447
00:23:39,240 --> 00:23:42,159
Speaker 1: transactions are profitable for them in any way. As we

448
00:23:42,200 --> 00:23:45,600
Speaker 1: saw following Microsoft's most recent quarterly earnings, there's growing investor

449
00:23:45,640 --> 00:23:48,800
Speaker 1: concern over how capex is being spent and the amount

450
00:23:48,800 --> 00:23:51,560
Speaker 1: that's being required to build the infrastructure for generative AI,

451
00:23:51,760 --> 00:23:55,200
Speaker 1: with many voicing skepticism about the potential profitability of the technology,

452
00:23:55,200 --> 00:23:58,240
Speaker 1: including Jim Cavello of Gold and Sex. And what we

453
00:23:58,359 --> 00:24:01,920
Speaker 1: really don't know is how on profit generative AIS for hyperscalers,

454
00:24:02,119 --> 00:24:04,760
Speaker 1: because they baked those costs into other parts of their ownings.

455
00:24:05,280 --> 00:24:07,439
Speaker 1: What we can't know for sure, I imagine this stuff

456
00:24:07,480 --> 00:24:10,120
Speaker 1: is if this stuff was in any way profitable, they'd

457
00:24:10,119 --> 00:24:12,880
Speaker 1: be talking about it all the time. They would never

458
00:24:12,920 --> 00:24:15,879
Speaker 1: shut up. This would be their new golden goose, and

459
00:24:15,920 --> 00:24:18,720
Speaker 1: they're not. In fact, the most concrete information we have

460
00:24:18,760 --> 00:24:21,400
Speaker 1: about open AI's balance sheet comes from leaked reports, well

461
00:24:21,440 --> 00:24:23,920
Speaker 1: sourced reporters at places like The New York Times, and

462
00:24:23,960 --> 00:24:27,080
Speaker 1: the information and invested prospectuses that found a wider audience

463
00:24:27,080 --> 00:24:30,520
Speaker 1: than Altman perhaps would have liked. So you may remember

464
00:24:30,520 --> 00:24:32,400
Speaker 1: from a few months ago that the markets have become

465
00:24:32,440 --> 00:24:35,320
Speaker 1: a little skeptical of the generative AI boom and Nvidia

466
00:24:35,520 --> 00:24:38,520
Speaker 1: CEO Jensen Huang had no real answers about AI's return

467
00:24:38,560 --> 00:24:41,639
Speaker 1: and investment from his latest earnings, which led to a

468
00:24:41,840 --> 00:24:44,480
Speaker 1: historic two hundred and seventy nine billion dollar drop in

469
00:24:44,600 --> 00:24:47,680
Speaker 1: Nvidia's market cap in a single day. This, by the way,

470
00:24:47,760 --> 00:24:50,720
Speaker 1: was the largest route in US market histories. The total

471
00:24:50,800 --> 00:24:53,960
Speaker 1: value lost is equivalent of nearly five Layman Brothers Its

472
00:24:54,000 --> 00:24:57,240
Speaker 1: peak value. They've recovered some of it, but nevertheless, that's

473
00:24:57,280 --> 00:24:59,440
Speaker 1: what we in the business called are not so good.

474
00:25:00,160 --> 00:25:02,800
Speaker 1: At the beginning of August, Microsoft, Amazon, and Google all

475
00:25:02,800 --> 00:25:04,919
Speaker 1: took a similar beating for the markets for their massive

476
00:25:04,920 --> 00:25:07,719
Speaker 1: capital expenditures related to AI, and all three of them

477
00:25:07,720 --> 00:25:09,920
Speaker 1: will face the wheel next quarter in a couple weeks

478
00:25:09,920 --> 00:25:12,280
Speaker 1: in fact, if they can't show a significant increase in

479
00:25:12,280 --> 00:25:14,439
Speaker 1: revenue from the combined one hundred and fifty billion or

480
00:25:14,560 --> 00:25:17,080
Speaker 1: more in capex that they put into new data centers

481
00:25:17,080 --> 00:25:20,639
Speaker 1: in the Nvidio GPUs. What's important to remember here is

482
00:25:20,680 --> 00:25:23,919
Speaker 1: that other than AI, bigtech really doesn't have any other ideas.

483
00:25:24,119 --> 00:25:26,919
Speaker 1: There are no more hypergrowth markets left, and as firms

484
00:25:26,960 --> 00:25:29,840
Speaker 1: like Microsoft and Amazon begin to show signs of declining growth,

485
00:25:29,960 --> 00:25:32,159
Speaker 1: so too does their desperation to show the markets that

486
00:25:32,160 --> 00:25:35,960
Speaker 1: they've still got it. Google, a company almost entirely sustained

487
00:25:35,960 --> 00:25:39,040
Speaker 1: by multiple at risk monopolies in search and advertising, also

488
00:25:39,119 --> 00:25:40,960
Speaker 1: needs something new and sexy to wave in front of

489
00:25:40,960 --> 00:25:43,680
Speaker 1: the street. Except none of this is working because the

490
00:25:43,720 --> 00:25:46,040
Speaker 1: products aren't that useful, and it appears most of its

491
00:25:46,080 --> 00:25:49,320
Speaker 1: revenue comes from companies trying out AI and then realizing

492
00:25:49,359 --> 00:25:52,359
Speaker 1: it wasn't worth it. And if you think back to

493
00:25:52,400 --> 00:25:55,040
Speaker 1: what I was saying about open aised cloud costs, they're

494
00:25:55,080 --> 00:25:59,119
Speaker 1: making what eight hundred to a billion on this? How

495
00:25:59,200 --> 00:26:03,080
Speaker 1: much does Google make? Probably much less considering their multiple

496
00:26:03,119 --> 00:26:06,359
Speaker 1: stories about people not really caring about Gemini. But at

497
00:26:06,359 --> 00:26:09,840
Speaker 1: this point there are really two eventualities. Big Tech realizes

498
00:26:09,880 --> 00:26:12,080
Speaker 1: that they've gotten in way too deep in this, and

499
00:26:12,160 --> 00:26:13,960
Speaker 1: out of the deep fear of pissing off the street,

500
00:26:14,160 --> 00:26:17,679
Speaker 1: chooses to reduce capital expenditures related to AI, or the

501
00:26:17,720 --> 00:26:20,480
Speaker 1: second one, Big Tech, desperate to find a new growth hog,

502
00:26:20,800 --> 00:26:24,520
Speaker 1: decides instead to cut costs to sustain their stupid fucking ideas,

503
00:26:24,840 --> 00:26:27,760
Speaker 1: laying off workers and reallocating capital from other operations as

504
00:26:27,800 --> 00:26:31,840
Speaker 1: a means of sustaining this death march nowhere, it's unclear

505
00:26:31,840 --> 00:26:34,960
Speaker 1: which will happen if Big Tech accepts that generative AI

506
00:26:35,040 --> 00:26:37,120
Speaker 1: is in the future. I don't really have anything else

507
00:26:37,160 --> 00:26:39,400
Speaker 1: to waive at Wall Street, but they could do their

508
00:26:39,400 --> 00:26:42,199
Speaker 1: own version of from twenty twenty two. Meta did this

509
00:26:42,280 --> 00:26:45,960
Speaker 1: Year of Efficiency thing, which involved reducing capital expenditures and

510
00:26:46,040 --> 00:26:49,159
Speaker 1: laying off thousands of people while also promising to slow

511
00:26:49,280 --> 00:26:52,720
Speaker 1: down a little with investment. This, by the way, is

512
00:26:52,720 --> 00:26:55,160
Speaker 1: the most likely path for Amazon and Google, who, while

513
00:26:55,200 --> 00:26:57,520
Speaker 1: desperate to make Wall Street happy, they still kind of

514
00:26:57,600 --> 00:27:02,800
Speaker 1: have their profitable monopolies now at least. Nevertheless, there really

515
00:27:02,840 --> 00:27:05,200
Speaker 1: needs to be some kind of revenue growth from AI

516
00:27:05,240 --> 00:27:07,520
Speaker 1: in the next few quarters. That has to be material.

517
00:27:07,920 --> 00:27:09,840
Speaker 1: It can't just be this thing about AI being a

518
00:27:09,880 --> 00:27:13,920
Speaker 1: maturing market or how annualized run rates have improved, and

519
00:27:13,960 --> 00:27:17,720
Speaker 1: said material contribution will have to be magnitudes higher if

520
00:27:17,760 --> 00:27:20,520
Speaker 1: capex has increased along with it. I just don't think

521
00:27:20,520 --> 00:27:23,760
Speaker 1: it's going to be there, whether it's Q four twenty

522
00:27:23,800 --> 00:27:26,040
Speaker 1: twenty four or Q one twenty twenty five, or maybe

523
00:27:26,080 --> 00:27:28,680
Speaker 1: a little later. Wall Street's going to punish big tech

524
00:27:28,760 --> 00:27:31,960
Speaker 1: for this the sin of lust, and the punishment is

525
00:27:31,960 --> 00:27:34,480
Speaker 1: going to be to savage these companies, even more harshly

526
00:27:34,520 --> 00:27:38,920
Speaker 1: than Nvidia, which, despite Jensen Huang's bluster and empty platitudes

527
00:27:39,000 --> 00:27:41,679
Speaker 1: is pretty much the only company that's actually making money

528
00:27:41,680 --> 00:27:44,200
Speaker 1: on AI, and that's because you do need their chips

529
00:27:44,200 --> 00:27:47,480
Speaker 1: to do all this. But I worry more than anything

530
00:27:47,560 --> 00:27:51,600
Speaker 1: that option two is more possible. I think these companies

531
00:27:51,600 --> 00:27:54,200
Speaker 1: are really capable of committing to AI as the future

532
00:27:54,359 --> 00:27:56,919
Speaker 1: and the cultures are so disconnected from the creation of

533
00:27:57,000 --> 00:28:01,000
Speaker 1: actual value or like software or solving problems that actual

534
00:28:01,000 --> 00:28:04,440
Speaker 1: people face, that they're willingly start laying people off if

535
00:28:04,480 --> 00:28:08,720
Speaker 1: it means bankrolling these operations. I really really worry about that.

536
00:28:08,840 --> 00:28:11,120
Speaker 1: By the way, the mass layoffs that could come from

537
00:28:11,119 --> 00:28:13,600
Speaker 1: this will be horrifying, because otherwise it's just going to

538
00:28:13,600 --> 00:28:16,480
Speaker 1: be feeding profit into this, and at this point they're

539
00:28:16,480 --> 00:28:20,240
Speaker 1: feeding in pretty much all their profits. And all of this,

540
00:28:20,320 --> 00:28:22,160
Speaker 1: by the way, could have been stopped if the media

541
00:28:22,200 --> 00:28:25,280
Speaker 1: had actually held the leaders of tech companies accountable. This

542
00:28:25,400 --> 00:28:28,360
Speaker 1: narrative was sold through the same con as the previous

543
00:28:28,440 --> 00:28:31,160
Speaker 1: hype cycles, and the media assumed that these companies would

544
00:28:31,160 --> 00:28:32,879
Speaker 1: just work it out like they did with crypto and

545
00:28:32,920 --> 00:28:35,720
Speaker 1: the metaverse, despite the fact that it was blatantly obvious

546
00:28:35,760 --> 00:28:38,800
Speaker 1: that they wouldn't work this out. You think I'm a duma, Well,

547
00:28:38,840 --> 00:28:42,440
Speaker 1: ask to me this, what's the plan, what does generative

548
00:28:42,480 --> 00:28:45,720
Speaker 1: AI do next? If your answer is that they'll work

549
00:28:45,760 --> 00:28:47,760
Speaker 1: it out or that they have something behind the scenes

550
00:28:47,800 --> 00:28:52,640
Speaker 1: that is incredible, you're an absolute mark. You're a participant

551
00:28:52,640 --> 00:28:56,840
Speaker 1: in a marketing scheme. It's time to wake up. It

552
00:28:56,960 --> 00:28:59,400
Speaker 1: is time to wake up to how stupid this is.

553
00:29:00,080 --> 00:29:02,920
Speaker 1: And I'm sure some of you will say, oh, oh,

554
00:29:03,000 --> 00:29:05,360
Speaker 1: you're going to look so stupid in six months. People

555
00:29:05,360 --> 00:29:07,320
Speaker 1: were telling me that six months ago. And I still

556
00:29:07,320 --> 00:29:09,280
Speaker 1: don't look stupid other than the ways they do, and

557
00:29:09,280 --> 00:29:13,000
Speaker 1: they're unrelated to the podcast. But let's get back to

558
00:29:13,040 --> 00:29:15,640
Speaker 1: the real problem, and let's get back to the really

559
00:29:15,680 --> 00:29:19,200
Speaker 1: worrying stuff, because I believe that the very least Microsoft

560
00:29:19,280 --> 00:29:22,080
Speaker 1: will begin reducing costs in other areas of its business

561
00:29:22,200 --> 00:29:24,800
Speaker 1: as a means of sustaining the AI boom. In an

562
00:29:24,800 --> 00:29:27,040
Speaker 1: email shared with me by a source from earlier this year,

563
00:29:27,320 --> 00:29:30,040
Speaker 1: Microsoft's senior leadership team requested in a plan that was

564
00:29:30,120 --> 00:29:34,640
Speaker 1: eventually scrapped, reducing power requirements from multiple areas within the

565
00:29:34,680 --> 00:29:37,400
Speaker 1: company as a means of freeing up power for GPUs,

566
00:29:37,560 --> 00:29:40,760
Speaker 1: including moving other services compute to other countries as a

567
00:29:40,760 --> 00:29:44,800
Speaker 1: means of freeing upseid capacity, specifically for AI. On the

568
00:29:44,840 --> 00:29:47,880
Speaker 1: Microsoft section of anonymous social network Blind, where you're required

569
00:29:47,880 --> 00:29:49,640
Speaker 1: to verify that you have a corporate email of the

570
00:29:49,680 --> 00:29:52,959
Speaker 1: company in question. One Microsoft worker complained in mid December

571
00:29:53,000 --> 00:29:56,520
Speaker 1: twenty twenty three that AI was taking their money, saying

572
00:29:56,520 --> 00:29:58,800
Speaker 1: that the cost of AI is so much that it

573
00:29:58,840 --> 00:30:01,160
Speaker 1: is eating up pay raises and that things will not

574
00:30:01,200 --> 00:30:04,760
Speaker 1: get better. In mid July twenty twenty four, another shared

575
00:30:04,800 --> 00:30:06,720
Speaker 1: their anxiety about how it was apparent to them that

576
00:30:06,720 --> 00:30:09,600
Speaker 1: Microsoft had and I quote a borderline addiction to cut

577
00:30:09,640 --> 00:30:11,760
Speaker 1: costs in order to fund in Video's stock price with

578
00:30:11,800 --> 00:30:14,520
Speaker 1: operational cash flows, and that doing so had and I

579
00:30:14,640 --> 00:30:19,080
Speaker 1: quote damaged Microsoft's culture deeply. Another added that they believe

580
00:30:19,120 --> 00:30:22,600
Speaker 1: that copilot is going to ruin Microsoft's FY twenty five,

581
00:30:22,640 --> 00:30:24,840
Speaker 1: referring of course to their financial year twenty twenty five,

582
00:30:25,240 --> 00:30:28,080
Speaker 1: adding that the f y twenty five copilot focus is

583
00:30:28,120 --> 00:30:31,240
Speaker 1: going to massively fall in f y twenty five, and

584
00:30:31,280 --> 00:30:34,120
Speaker 1: they knew of big co pilot deals in their country

585
00:30:34,120 --> 00:30:36,960
Speaker 1: that have less than twenty percent usage after almost a

586
00:30:37,040 --> 00:30:41,000
Speaker 1: year of integration, adding that corpor is too much and

587
00:30:41,000 --> 00:30:44,080
Speaker 1: that Microsoft's huge AI investments are not going to be realized.

588
00:30:44,960 --> 00:30:47,920
Speaker 1: While Blind is anonymous, it's kind of hard to ignore

589
00:30:47,960 --> 00:30:49,880
Speaker 1: the fact that there are many, many posts that tell

590
00:30:50,120 --> 00:30:52,880
Speaker 1: a tale of a kind of cultural cancer in Microsoft,

591
00:30:52,880 --> 00:30:56,520
Speaker 1: with disconnected senior leadership, the only funds projects if they

592
00:30:56,520 --> 00:31:00,280
Speaker 1: have AI takeed onto the side. Many posts the men

593
00:31:00,360 --> 00:31:03,400
Speaker 1: satching the Della's words salard approach and complain of a

594
00:31:03,480 --> 00:31:06,440
Speaker 1: lack of bonuses or upward mobility, and an organization focused

595
00:31:06,480 --> 00:31:09,600
Speaker 1: on chasing an AI boom that may not exist. And

596
00:31:09,640 --> 00:31:12,120
Speaker 1: at the very least, there's a deep cultural sadness there

597
00:31:12,120 --> 00:31:14,680
Speaker 1: with the many posts I've seen oscillating between I don't

598
00:31:14,760 --> 00:31:17,160
Speaker 1: like working at Microsoft and I don't know where we're

599
00:31:17,160 --> 00:31:20,200
Speaker 1: putting so much into AI, and then someone replying with

600
00:31:20,320 --> 00:31:22,920
Speaker 1: get used to it, SAJA doesn't get a shit, and

601
00:31:22,960 --> 00:31:25,680
Speaker 1: it all feels so ridiculous because there's so many signs

602
00:31:25,680 --> 00:31:28,640
Speaker 1: that these products don't have a product market fit. At

603
00:31:28,640 --> 00:31:30,520
Speaker 1: the start of this episode, I mentioned an article from

604
00:31:30,520 --> 00:31:33,880
Speaker 1: the Information about a lack of adoption of Microsoft's AI features.

605
00:31:34,440 --> 00:31:37,000
Speaker 1: Buried within that one was a particularly worrying thought about

606
00:31:37,000 --> 00:31:40,000
Speaker 1: the actual utilization of their data centers for this AI

607
00:31:40,400 --> 00:31:43,640
Speaker 1: and it said, and I quote around March of this year,

608
00:31:43,840 --> 00:31:46,400
Speaker 1: Microsoft had set aside enough server capacity in its data

609
00:31:46,400 --> 00:31:49,160
Speaker 1: centers for three sixty five copilot to handle daily users

610
00:31:49,160 --> 00:31:52,160
Speaker 1: of the AI system in the low millions. According to

611
00:31:52,200 --> 00:31:55,040
Speaker 1: someone with direct knowledge of those plans, it couldn't be

612
00:31:55,120 --> 00:31:57,120
Speaker 1: learned how much of that capacity was used at the time.

613
00:31:57,920 --> 00:32:01,600
Speaker 1: Based on the information's estimates, Elsewhere, Microsoft has somewhere between

614
00:32:01,640 --> 00:32:04,200
Speaker 1: four hundred thousand and four million years of its office

615
00:32:04,240 --> 00:32:07,280
Speaker 1: Copilot features, meaning that there's a decent chance that Microsoft

616
00:32:07,360 --> 00:32:10,680
Speaker 1: has built out capacity that isn't getting used. Now. One

617
00:32:10,720 --> 00:32:12,840
Speaker 1: could argue that it's building with the belief that the

618
00:32:12,840 --> 00:32:15,800
Speaker 1: product category will grow. But here's another idea. What if

619
00:32:15,840 --> 00:32:20,080
Speaker 1: it doesn't. Huh ah, what do you think? What if?

620
00:32:20,120 --> 00:32:23,120
Speaker 1: And this is crazy, Microsoft, Google and Amazon built out

621
00:32:23,160 --> 00:32:26,080
Speaker 1: these massive data sentences to capture demand that may never arrive.

622
00:32:27,440 --> 00:32:30,240
Speaker 1: I realized that sound a little crazy saying this, But

623
00:32:30,280 --> 00:32:31,800
Speaker 1: back in March I made the point that I could

624
00:32:31,800 --> 00:32:34,680
Speaker 1: find no companies that had integrated generative AI in a

625
00:32:34,720 --> 00:32:37,520
Speaker 1: way that was truly benefited their bottom line, And just

626
00:32:37,600 --> 00:32:41,360
Speaker 1: under six months later, I'm still looking. The best that

627
00:32:41,400 --> 00:32:43,640
Speaker 1: I can find is that big companies appear to have

628
00:32:44,280 --> 00:32:48,720
Speaker 1: done is stapled AI onto existing products and hoping that

629
00:32:48,720 --> 00:32:52,200
Speaker 1: that helps them shift them something that does not seem

630
00:32:52,240 --> 00:32:54,960
Speaker 1: to be working either. It doesn't work for Microsoft, doesn't

631
00:32:54,960 --> 00:32:56,800
Speaker 1: work for Box. It does seem to be working anywhere

632
00:32:57,160 --> 00:32:59,400
Speaker 1: as I'm not sure any of these AI upgrades give

633
00:32:59,400 --> 00:33:02,880
Speaker 1: any kind of significant business value. Now. While there may

634
00:33:02,920 --> 00:33:05,760
Speaker 1: be companies integrating AI that are driving some degree of

635
00:33:05,800 --> 00:33:09,160
Speaker 1: spend on Microsoft as your Amazon Web Services and Google Cloud,

636
00:33:10,120 --> 00:33:12,600
Speaker 1: I don't know how much it is, considering the last

637
00:33:12,680 --> 00:33:15,080
Speaker 1: episode saying about how open ai was only making about

638
00:33:15,080 --> 00:33:18,680
Speaker 1: a billion dollars licensing out their models, and I hypothesize

639
00:33:18,680 --> 00:33:21,200
Speaker 1: that any of this demand is driven by investor sentiment,

640
00:33:21,240 --> 00:33:24,280
Speaker 1: because companies are right now everywhere in the economy being

641
00:33:24,360 --> 00:33:28,280
Speaker 1: pushed to invest in AI without really knowing if it

642
00:33:28,360 --> 00:33:31,000
Speaker 1: will work, or whether it's useful, or whether their users

643
00:33:31,000 --> 00:33:35,080
Speaker 1: will like it. Nevertheless, these companies have spent a great

644
00:33:35,080 --> 00:33:37,760
Speaker 1: deal of time and money baking generative AI features into

645
00:33:37,800 --> 00:33:40,840
Speaker 1: their products, and I think they're going to face one

646
00:33:40,880 --> 00:33:45,600
Speaker 1: of a few different scenarios. Scenario the first, after developing

647
00:33:45,600 --> 00:33:47,960
Speaker 1: and launching these features, these companies are going to find

648
00:33:47,960 --> 00:33:50,320
Speaker 1: customers don't want to pay for them, as Microsoft's finding

649
00:33:50,320 --> 00:33:52,840
Speaker 1: with three sixty five Copilot and if they can't find

650
00:33:52,880 --> 00:33:54,880
Speaker 1: a way to make them pay for it. Now, they're

651
00:33:54,880 --> 00:33:57,480
Speaker 1: going to be really hard pressed when nobody's telling them

652
00:33:57,480 --> 00:34:00,800
Speaker 1: to get in on AI. And there's the same scenario.

653
00:34:01,120 --> 00:34:03,800
Speaker 1: After developing and launching these features, these companies can't find

654
00:34:03,840 --> 00:34:05,200
Speaker 1: a way to get users to pay for them, or

655
00:34:05,280 --> 00:34:08,600
Speaker 1: at least pay extra for them, which means that everyone

656
00:34:08,680 --> 00:34:10,520
Speaker 1: is going to have to bake the same thing into

657
00:34:10,560 --> 00:34:12,840
Speaker 1: their products. Everyone's going to have to do this because

658
00:34:13,520 --> 00:34:15,760
Speaker 1: none of these companies are able to function without copying

659
00:34:15,760 --> 00:34:19,279
Speaker 1: their competitors, which will turn Generative AI into a kind

660
00:34:19,280 --> 00:34:22,960
Speaker 1: of parasite. Now, just to broaden out what I mean here,

661
00:34:23,920 --> 00:34:27,760
Speaker 1: I looked across most of the software as a service

662
00:34:27,840 --> 00:34:31,480
Speaker 1: industry and a previous newsletter and I was looking and

663
00:34:31,560 --> 00:34:33,359
Speaker 1: most of them are doing much the same thing. It's

664
00:34:33,400 --> 00:34:38,719
Speaker 1: document summarization, document search, generation of staff, so emails and

665
00:34:38,760 --> 00:34:42,640
Speaker 1: the like, and summarization summarization can be emails can be documents.

666
00:34:43,000 --> 00:34:46,040
Speaker 1: For the most part, that's what everyone is doing. The

667
00:34:46,160 --> 00:34:49,759
Speaker 1: problem is that everyone doing the same thing means that

668
00:34:49,920 --> 00:34:52,160
Speaker 1: no one can really make money off of it. And

669
00:34:52,400 --> 00:34:54,960
Speaker 1: Jim Cavello out of Gold and Sex made the same

670
00:34:55,280 --> 00:34:58,279
Speaker 1: worrying we had the same thought as me, which makes

671
00:34:58,640 --> 00:35:01,120
Speaker 1: probably him smarter than me. I shouldn't think about that

672
00:35:01,160 --> 00:35:05,600
Speaker 1: too much anyway. I mentioned previously in the last episode

673
00:35:05,640 --> 00:35:08,480
Speaker 1: the commoditization effect of these large language models, and I

674
00:35:08,520 --> 00:35:10,879
Speaker 1: think there's going to be a further commoditization of these

675
00:35:10,880 --> 00:35:15,359
Speaker 1: effects themselves, of these features. If everyone summarizes email, now

676
00:35:15,400 --> 00:35:17,280
Speaker 1: you have to do it too, because otherwise the customer

677
00:35:17,280 --> 00:35:19,840
Speaker 1: can go, there's another feature. I'm going to pay for

678
00:35:19,840 --> 00:35:22,440
Speaker 1: this one because it's got more stuff in it, except

679
00:35:22,520 --> 00:35:26,799
Speaker 1: the feature in questions more expensive. It's very worrying, But

680
00:35:26,920 --> 00:35:29,400
Speaker 1: in general, what I fear is a kind of cascade effect.

681
00:35:30,520 --> 00:35:32,520
Speaker 1: I believe that a lot of businesses right now are

682
00:35:32,560 --> 00:35:35,520
Speaker 1: trying AI, and once those trials end, and Gartner predicts

683
00:35:35,520 --> 00:35:38,279
Speaker 1: that thirty percent of GENERAVIAI projects will be abandoned after

684
00:35:38,320 --> 00:35:40,279
Speaker 1: the proof of concept by the end of twenty twenty five,

685
00:35:40,600 --> 00:35:42,479
Speaker 1: these companies are going to stop paying for the extra

686
00:35:42,560 --> 00:35:46,720
Speaker 1: features or stop integrating GENERATEVII into their products. If this happens,

687
00:35:46,840 --> 00:35:49,600
Speaker 1: it will reduce the already kind of shitty revenue flowing

688
00:35:49,600 --> 00:35:52,440
Speaker 1: to the hyperscalers providing cloud computer or access to models

689
00:35:52,440 --> 00:35:56,000
Speaker 1: for generat AI, which in turn could create more price

690
00:35:56,040 --> 00:35:59,600
Speaker 1: pressure on these companies. They're already negative margins sour. At

691
00:35:59,600 --> 00:36:02,440
Speaker 1: that point, Open AI and Anthropic will almost certainly have

692
00:36:02,560 --> 00:36:06,279
Speaker 1: to raise prices. And what's fun is they're already not

693
00:36:06,400 --> 00:36:09,359
Speaker 1: making that much money from this. So we're in this

694
00:36:09,400 --> 00:36:13,319
Speaker 1: weird situation where it isn't obvious which it's going to be.

695
00:36:13,640 --> 00:36:15,400
Speaker 1: Is it that they're going to have to raise prices,

696
00:36:15,480 --> 00:36:17,319
Speaker 1: or that no one wants to pay them, or some

697
00:36:17,360 --> 00:36:20,239
Speaker 1: combination of both. It's also important to note that the

698
00:36:20,320 --> 00:36:23,120
Speaker 1: hyperscalers are also terrified of pissing off Wall Street. I

699
00:36:23,200 --> 00:36:26,440
Speaker 1: really mean that one of them will eventually blink. And

700
00:36:26,480 --> 00:36:29,160
Speaker 1: while they could theoretically do the layoffs and cost cutting

701
00:36:29,200 --> 00:36:32,720
Speaker 1: measures I've mentioned, these are short term solutions that don't

702
00:36:32,800 --> 00:36:37,000
Speaker 1: really work against burning billions tens of billions, like more

703
00:36:37,000 --> 00:36:40,480
Speaker 1: than half more than fifty billion a year for each

704
00:36:40,520 --> 00:36:43,719
Speaker 1: of them. How are you going to cut enough to

705
00:36:43,800 --> 00:36:48,720
Speaker 1: bankroll that? But in any case, putting aside the amount

706
00:36:48,719 --> 00:36:51,560
Speaker 1: of money they're having to invest, it might be time

707
00:36:51,600 --> 00:36:54,080
Speaker 1: to accept that there really isn't money here in Generative AI.

708
00:36:54,520 --> 00:36:56,279
Speaker 1: It might be time to stop and take stock of

709
00:36:56,320 --> 00:36:58,520
Speaker 1: the fact that we're in the midst of what our

710
00:36:58,600 --> 00:37:03,160
Speaker 1: third delusional epop our third stupid idea that everyone claims

711
00:37:03,239 --> 00:37:08,360
Speaker 1: the future, But unlike cryptocurrency, in the metaverse, everyone seems

712
00:37:08,360 --> 00:37:10,919
Speaker 1: to have joined this pie, and everyone's decided to burn

713
00:37:10,960 --> 00:37:15,759
Speaker 1: as much money as he mainly possible on this unsustainable, unreliable, unprofitable,

714
00:37:15,880 --> 00:37:21,360
Speaker 1: environmentally destructive bullshit sold to customers and businesses as artificial

715
00:37:21,360 --> 00:37:25,239
Speaker 1: intelligence to law may everything without ever having a path

716
00:37:25,280 --> 00:37:27,319
Speaker 1: to do so. Because that's the thing, none of this

717
00:37:27,440 --> 00:37:31,560
Speaker 1: is even AI. This is an automation. It's generation generation

718
00:37:31,640 --> 00:37:35,040
Speaker 1: in different hats, and it burns the world around us

719
00:37:35,080 --> 00:37:39,879
Speaker 1: to provide it. But you know, I don't think the

720
00:37:39,920 --> 00:37:42,120
Speaker 1: following is going to burn the world. In fact, I

721
00:37:42,160 --> 00:37:45,319
Speaker 1: think it could really make your life better. And I

722
00:37:45,480 --> 00:37:50,360
Speaker 1: need you to directly and vacifiously engage with the following

723
00:37:50,400 --> 00:38:07,600
Speaker 1: advertisements and we're back. See might ask why does this

724
00:38:07,680 --> 00:38:11,520
Speaker 1: keep happening? Why do we keep getting these stupid movements?

725
00:38:12,360 --> 00:38:14,840
Speaker 1: Why did they tell us that cryptocurrency was the future.

726
00:38:14,840 --> 00:38:16,760
Speaker 1: Why did they tell us the metaverse was the future?

727
00:38:16,760 --> 00:38:18,799
Speaker 1: Why are they telling us that generative AI is the

728
00:38:18,840 --> 00:38:21,800
Speaker 1: future when none of these things from the very beginning

729
00:38:22,400 --> 00:38:25,879
Speaker 1: looked like the future. There were signs from GPT through

730
00:38:25,920 --> 00:38:28,200
Speaker 1: like oh cool, you can generate entire things, and like

731
00:38:28,239 --> 00:38:31,920
Speaker 1: a minute, wow, that's crazy. But past that point, past

732
00:38:31,960 --> 00:38:33,960
Speaker 1: that moment of oh you can do that, I guess

733
00:38:34,560 --> 00:38:38,759
Speaker 1: what was there? And why does this keep happening. It's

734
00:38:38,800 --> 00:38:42,120
Speaker 1: the natural result of a tech industry that's become entirely

735
00:38:42,160 --> 00:38:45,920
Speaker 1: focused on making each customer more valuable rather than providing

736
00:38:45,920 --> 00:38:48,400
Speaker 1: more value to the customer in exchange for I don't know,

737
00:38:48,640 --> 00:38:51,960
Speaker 1: money or attention. The products you're being sold today almost

738
00:38:52,040 --> 00:38:55,080
Speaker 1: certainly tried to wed you to a particular ecosystem when

739
00:38:55,120 --> 00:38:58,800
Speaker 1: owned by Microsoft, Apple, Amazon or Google as a consumer

740
00:38:58,840 --> 00:39:01,880
Speaker 1: at least and internally increase the burden of lead things

741
00:39:01,880 --> 00:39:05,840
Speaker 1: said ecosystem. Imagine trying to move all of your subscribe

742
00:39:05,880 --> 00:39:09,480
Speaker 1: and save shit off of Amazon. Imagine trying I mean

743
00:39:09,600 --> 00:39:13,359
Speaker 1: moving iOS to Android or It's not that easy. And

744
00:39:13,400 --> 00:39:18,120
Speaker 1: that's by design. Everything is about further monetization, about increasing

745
00:39:18,120 --> 00:39:20,160
Speaker 1: the dollar per head value of each customer, be it

746
00:39:20,200 --> 00:39:22,359
Speaker 1: through keeping them doing stuff on the platform, to show

747
00:39:22,360 --> 00:39:25,200
Speaker 1: them more advertising, upselling them new features that are only

748
00:39:25,280 --> 00:39:28,399
Speaker 1: kind of useful or previously we're free, or creating some

749
00:39:28,440 --> 00:39:31,520
Speaker 1: new monopoly or oligopoly where only those with the massive

750
00:39:31,520 --> 00:39:34,319
Speaker 1: war chests a big tech can really play, and very

751
00:39:34,400 --> 00:39:36,759
Speaker 1: very little about this is about delivering any kind of

752
00:39:36,760 --> 00:39:39,640
Speaker 1: real value or utility or thing that you, the customer

753
00:39:39,719 --> 00:39:43,640
Speaker 1: might like. Generative AI might not be super useful, but

754
00:39:43,680 --> 00:39:46,200
Speaker 1: it's really easy to integrate into stuff and make new

755
00:39:46,280 --> 00:39:49,279
Speaker 1: things happen, creating all sorts of new things that the

756
00:39:49,320 --> 00:39:52,279
Speaker 1: company could theoretically charge for, both for a customer and

757
00:39:52,320 --> 00:39:55,880
Speaker 1: an enterprise customer. Sam Molton was smart enough to realize

758
00:39:55,880 --> 00:39:58,120
Speaker 1: that the tech industry needed a new thing, a new

759
00:39:58,160 --> 00:40:00,520
Speaker 1: technology that everybody could take a piece of and sell.

760
00:40:00,680 --> 00:40:03,359
Speaker 1: And while he might not really understand technology, all one

761
00:40:03,480 --> 00:40:07,480
Speaker 1: understands growth and the lust that the economy has for growth,

762
00:40:08,040 --> 00:40:11,920
Speaker 1: and he's productized Transformer based architecture is something that everybody

763
00:40:11,920 --> 00:40:14,600
Speaker 1: could sell, a magical tool that could plug into things

764
00:40:14,640 --> 00:40:17,680
Speaker 1: and kind of connect to an ephemeral concept like AI.

765
00:40:19,760 --> 00:40:22,480
Speaker 1: The problem is that the desperation to integrate genera if

766
00:40:22,480 --> 00:40:25,480
Speaker 1: AI everywhere has shown a pretty nasty light on how

767
00:40:25,520 --> 00:40:28,840
Speaker 1: disconnected these companies are from actual consumer needs or even

768
00:40:29,400 --> 00:40:33,920
Speaker 1: running good companies. Like really, I'm not even being facetious.

769
00:40:34,840 --> 00:40:37,360
Speaker 1: I would genuinely like it if this stuff was useful.

770
00:40:37,520 --> 00:40:40,640
Speaker 1: I like useful things. There would be ethical concerns about

771
00:40:40,640 --> 00:40:43,640
Speaker 1: the copyright theft and such, but I would at least

772
00:40:43,719 --> 00:40:45,920
Speaker 1: tip my hat to them if I could find something,

773
00:40:46,480 --> 00:40:50,040
Speaker 1: anything that I looked at and could say, Wow, that's

774
00:40:50,080 --> 00:40:52,680
Speaker 1: really useful in my daily life. I got nothing, and

775
00:40:52,719 --> 00:40:56,280
Speaker 1: I've really looked. You can email me easy that's echoes

776
00:40:56,320 --> 00:40:59,560
Speaker 1: Abetter offline dot com if you have one. But I've

777
00:40:59,640 --> 00:41:03,440
Speaker 1: yet impressed by one of those emails, So please try harder.

778
00:41:04,880 --> 00:41:07,200
Speaker 1: And the really worrying part is that other than AI,

779
00:41:08,360 --> 00:41:10,440
Speaker 1: many of these companies don't seem to have any other

780
00:41:10,520 --> 00:41:15,320
Speaker 1: new products. What else is there? What are the things

781
00:41:15,320 --> 00:41:18,120
Speaker 1: do they have to grow their companies? No? Really, what

782
00:41:18,400 --> 00:41:21,040
Speaker 1: do they have? The new iPhone? I bought the new iPhone.

783
00:41:21,040 --> 00:41:23,920
Speaker 1: I'm a little pig point cooin coin. I bought the iPhone.

784
00:41:23,960 --> 00:41:26,319
Speaker 1: I bought the new one, and I've bought it every year.

785
00:41:26,360 --> 00:41:28,520
Speaker 1: I am that guy I sell the old one about

786
00:41:28,560 --> 00:41:31,399
Speaker 1: the new one. This is the first year. I think

787
00:41:31,440 --> 00:41:33,600
Speaker 1: from the beginning where I bought it and being like

788
00:41:33,680 --> 00:41:35,640
Speaker 1: why did I do that? Man? What does this do?

789
00:41:36,760 --> 00:41:39,600
Speaker 1: And that's because I think we're hitting a wall. This

790
00:41:39,640 --> 00:41:41,719
Speaker 1: is the rock combubble. I talked about a few months ago.

791
00:41:42,640 --> 00:41:46,160
Speaker 1: They've not got anything. There's nothing, They've got nothing. And

792
00:41:46,200 --> 00:41:50,879
Speaker 1: that really is the problem, because when everything falls, when

793
00:41:50,920 --> 00:41:54,360
Speaker 1: everyone realizes, when the markets look at tech and say, wow,

794
00:41:54,480 --> 00:41:56,400
Speaker 1: you're not going to grow forever. You're not going to

795
00:41:56,480 --> 00:41:58,040
Speaker 1: come up with a new wiz bang that you can

796
00:41:58,080 --> 00:42:01,440
Speaker 1: market to everyone and make billions in returns. You're not

797
00:42:01,480 --> 00:42:05,520
Speaker 1: going to do that. No, they're not going to react

798
00:42:05,520 --> 00:42:08,640
Speaker 1: well at all, because when you take away the massive

799
00:42:08,680 --> 00:42:12,320
Speaker 1: growth that tech has, you have a very annoying industry

800
00:42:12,360 --> 00:42:15,520
Speaker 1: full of annoying young people that will piss off the markets.

801
00:42:15,760 --> 00:42:19,359
Speaker 1: They will piss off those with the money. The tech

802
00:42:19,400 --> 00:42:21,440
Speaker 1: industry has a terrible rep with the government and a

803
00:42:21,520 --> 00:42:25,480
Speaker 1: terrible rep with society. The reevaluation of these companies will

804
00:42:25,520 --> 00:42:30,400
Speaker 1: be merciless, and there are very few friends left, and

805
00:42:30,480 --> 00:42:32,680
Speaker 1: I think there will be a cascade down to the

806
00:42:32,719 --> 00:42:35,399
Speaker 1: other companies in the tech space, just in the same

807
00:42:35,440 --> 00:42:38,280
Speaker 1: way that it will hit workers who will get laid

808
00:42:38,280 --> 00:42:41,320
Speaker 1: off when all of this falls apart. Despite none of

809
00:42:41,360 --> 00:42:44,040
Speaker 1: these people doing anything wrong other than the people up

810
00:42:44,080 --> 00:42:47,800
Speaker 1: top having no creativity, no real innovation, and no understanding

811
00:42:47,840 --> 00:42:52,600
Speaker 1: of real people's problems. I hypothesize a kind of SUBPRIMEI

812
00:42:52,760 --> 00:42:55,799
Speaker 1: crisis is brewing where almost the entire tech industry is

813
00:42:55,840 --> 00:42:59,239
Speaker 1: brought in and a technology sold at this insanely discounted rate,

814
00:42:59,440 --> 00:43:03,320
Speaker 1: heavily sentralized and subsidized by big tech companies like Microsoft, Amazon,

815
00:43:03,360 --> 00:43:07,160
Speaker 1: and Google. At some point, this incredible toxic burn ray

816
00:43:07,600 --> 00:43:09,799
Speaker 1: is going to burn through GENERATIVEAI and it's going to

817
00:43:09,840 --> 00:43:12,319
Speaker 1: catch up with them. And when the price increases come

818
00:43:12,560 --> 00:43:16,280
Speaker 1: or companies realize that these features are not that useful

819
00:43:16,440 --> 00:43:19,919
Speaker 1: and they see the lack of user adoption, they're going

820
00:43:19,960 --> 00:43:23,279
Speaker 1: to start getting nervous. But right now are in the

821
00:43:23,320 --> 00:43:26,239
Speaker 1: piss take section of the economy. Right now we're seeing

822
00:43:26,239 --> 00:43:30,360
Speaker 1: the egregious share like Salesforce charging two dollars a conversation

823
00:43:30,480 --> 00:43:34,720
Speaker 1: for their new agent Force product. But eventually the markets

824
00:43:34,760 --> 00:43:39,319
Speaker 1: will catch up because the money isn't there. And when

825
00:43:39,360 --> 00:43:42,279
Speaker 1: these prices go up, I'm not confident that will have

826
00:43:42,560 --> 00:43:45,920
Speaker 1: much of a generative AI industry left. And that's assuming

827
00:43:45,960 --> 00:43:49,400
Speaker 1: that these companies still have enough money. It's assuming that

828
00:43:49,520 --> 00:43:52,080
Speaker 1: open ai is able to raise another six and a

829
00:43:52,120 --> 00:43:55,040
Speaker 1: half billion dollar round in the next six to eight months.

830
00:43:55,680 --> 00:43:58,440
Speaker 1: How long can they do that? For? How many times?

831
00:43:58,640 --> 00:44:01,120
Speaker 1: How many years? A VSA he's willing to prop up

832
00:44:01,480 --> 00:44:05,480
Speaker 1: open AI. How many years is Microsoft ready to burn

833
00:44:05,760 --> 00:44:09,840
Speaker 1: capital to make what a billion or two on Generative AI?

834
00:44:10,560 --> 00:44:14,440
Speaker 1: This is embarrassing. It's bad business and it's bad product.

835
00:44:15,320 --> 00:44:19,319
Speaker 1: Satch an Adela, Sundarpishi, sam Ortmon, the whole lot of them.

836
00:44:19,360 --> 00:44:22,120
Speaker 1: They should be absolutely fucking ashamed of themselves. They're insult

837
00:44:22,200 --> 00:44:26,000
Speaker 1: to innovation and insult Silicon Valley and insult to their consumers.

838
00:44:27,320 --> 00:44:30,520
Speaker 1: And what happens, you tell me this when the tech industry,

839
00:44:30,640 --> 00:44:33,200
Speaker 1: the entire tech industry, relies on the success of a

840
00:44:33,280 --> 00:44:35,960
Speaker 1: kind of software that only loses money and doesn't create

841
00:44:36,080 --> 00:44:39,440
Speaker 1: much value when it does so. And what happens when

842
00:44:39,440 --> 00:44:42,239
Speaker 1: the heat gets too hot and these products become impossible

843
00:44:42,280 --> 00:44:45,960
Speaker 1: to reconcile with, and that everyone realizes that none of

844
00:44:46,040 --> 00:44:50,360
Speaker 1: these companies have anything else to sell. I really don't know.

845
00:44:51,800 --> 00:44:59,280
Speaker 1: I'm scared. I'm not trying to do fud, doing a fud, fear, uncertainty,

846
00:44:59,320 --> 00:45:01,840
Speaker 1: in doubt, told to spell these things out, but I

847
00:45:01,960 --> 00:45:06,320
Speaker 1: am worried because really the only other alternative to what

848
00:45:06,400 --> 00:45:10,000
Speaker 1: I'm saying is that they magically make this profitable, that

849
00:45:10,120 --> 00:45:12,680
Speaker 1: they just keep doing this until it goes into the green,

850
00:45:13,160 --> 00:45:16,040
Speaker 1: despite no one appearing to know how, despite their not

851
00:45:16,280 --> 00:45:20,040
Speaker 1: being a path there. How willing are you to believe

852
00:45:20,120 --> 00:45:22,520
Speaker 1: them after they've lied to you for so many years?

853
00:45:23,760 --> 00:45:27,960
Speaker 1: How ridiculous is this really? How ridiculous have you been

854
00:45:28,040 --> 00:45:31,239
Speaker 1: thinking this is? How much can you let them coast on?

855
00:45:31,640 --> 00:45:34,040
Speaker 1: They'll work it out? Because they haven't. They haven't worked

856
00:45:34,040 --> 00:45:36,200
Speaker 1: it out for a while. It's been over a decade

857
00:45:36,200 --> 00:45:39,320
Speaker 1: since the last significance consumer tech innovation. It's been a

858
00:45:39,400 --> 00:45:42,040
Speaker 1: ton on the chip side. But what is there for you?

859
00:45:42,160 --> 00:45:44,879
Speaker 1: And I? Not really much? And I don't think there's

860
00:45:44,960 --> 00:45:47,560
Speaker 1: much in this industry either, And I worry that the

861
00:45:47,640 --> 00:45:51,160
Speaker 1: tech industry is building towards a really grotesque reckoning, with

862
00:45:51,280 --> 00:45:54,160
Speaker 1: a total lack of creativity, enabled by an economy that

863
00:45:54,239 --> 00:45:58,480
Speaker 1: rewards growth over innovation, and monopolization over loyalty, and management

864
00:45:58,560 --> 00:46:01,280
Speaker 1: over those who actually build things. The people in control

865
00:46:01,360 --> 00:46:03,880
Speaker 1: of the tech industry are not the ones who built it.

866
00:46:04,320 --> 00:46:07,880
Speaker 1: These people are management consultants. Even Samultman is one of them.

867
00:46:09,000 --> 00:46:12,680
Speaker 1: These people are superficially interesting and superficially smart, just like

868
00:46:12,840 --> 00:46:17,520
Speaker 1: jat GPT. And I worry, I worry so much, So

869
00:46:17,760 --> 00:46:21,160
Speaker 1: promise me, dear listener, then The next time someone tells

870
00:46:21,160 --> 00:46:23,240
Speaker 1: you they'll work it out, that this stuff is the future,

871
00:46:23,600 --> 00:46:26,120
Speaker 1: tell them some of this shit, send them the podcast,

872
00:46:26,280 --> 00:46:28,000
Speaker 1: or just yell at them at the top of your voice.

873
00:46:28,200 --> 00:46:31,240
Speaker 1: Don't even need to use words, but I'm so grateful

874
00:46:31,280 --> 00:46:41,400
Speaker 1: to have you as listeners. Thank you for listening to

875
00:46:41,440 --> 00:46:44,400
Speaker 1: Better Offline. The editor and composer of the Better Offline

876
00:46:44,440 --> 00:46:47,080
Speaker 1: theme song is Matasowski. You can check out more of

877
00:46:47,120 --> 00:46:50,719
Speaker 1: his music and audio projects at Mattasowski dot com, m

878
00:46:50,800 --> 00:46:54,879
Speaker 1: A T T O S O W s ki dot com.

879
00:46:55,640 --> 00:46:57,879
Speaker 1: You can email me at easy at Better Offline dot

880
00:46:57,960 --> 00:47:00,200
Speaker 1: com or visit Better Offline dot com to find more

881
00:47:00,239 --> 00:47:03,560
Speaker 1: podcast links and of course, my newsletter. I also really

882
00:47:03,600 --> 00:47:05,839
Speaker 1: recommend you go to chat dot Where's youread dot at

883
00:47:05,920 --> 00:47:08,359
Speaker 1: to visit the discord, and go to our slash Better

884
00:47:08,400 --> 00:47:11,560
Speaker 1: Offline to check out I'll Reddit. Thank you so much

885
00:47:11,600 --> 00:47:15,400
Speaker 1: for listening. Better Offline is a production of cool Zone Media.

886
00:47:15,600 --> 00:47:18,439
Speaker 1: For more from cool Zone Media, visit our website cool

887
00:47:18,480 --> 00:47:21,799
Speaker 1: Zonemedia dot com, or check us out on the iHeartRadio app,

888
00:47:21,880 --> 00:47:24,280
Speaker 1: Apple Podcasts, or wherever you get your podcasts.