1
00:00:02,480 --> 00:00:03,560
Speaker 1: Azo Media.

2
00:00:05,480 --> 00:00:07,640
Speaker 2: Hello and welcome to the second of this week's Better

3
00:00:07,680 --> 00:00:08,680
Speaker 2: Offline Monologues.

4
00:00:08,800 --> 00:00:10,119
Speaker 3: I'm your host ed Zetron.

5
00:00:18,120 --> 00:00:20,480
Speaker 2: It's been a funny few weeks watching the little spurts

6
00:00:20,480 --> 00:00:22,800
Speaker 2: of air coming out of the AI bubble. It's unclear

7
00:00:22,840 --> 00:00:26,000
Speaker 2: when it will burst, what will burst it, really even.

8
00:00:25,800 --> 00:00:26,400
Speaker 3: What's going on.

9
00:00:26,520 --> 00:00:28,440
Speaker 2: Half the time, they just all seem to be running

10
00:00:28,440 --> 00:00:31,080
Speaker 2: around like a billion mister beans. But I get the

11
00:00:31,120 --> 00:00:34,239
Speaker 2: sense that everything will accelerate dramatically based on one or

12
00:00:34,240 --> 00:00:37,680
Speaker 2: two big events, things that just knock the confidence of everyone.

13
00:00:38,240 --> 00:00:40,920
Speaker 2: Maybe it's an AI company dying, or a funding ground

14
00:00:40,960 --> 00:00:44,600
Speaker 2: not coming together, or maybe it's a hyperscalar cutting Capex

15
00:00:45,040 --> 00:00:48,839
Speaker 2: Goldin SAX analyst Rich Pirotovsky said last week that he

16
00:00:48,920 --> 00:00:52,840
Speaker 2: believed that everybody was spending simply to remain competitive. Fucking

17
00:00:52,880 --> 00:00:55,520
Speaker 2: love that, don't you, and added that the first hyperscala

18
00:00:55,560 --> 00:00:57,640
Speaker 2: to signal that it can slow the pace of spending

19
00:00:57,680 --> 00:01:00,680
Speaker 2: would likely see its share price rewarded. If I had

20
00:01:00,720 --> 00:01:03,000
Speaker 2: to bet, I think it's between Meta and Microsoft, the

21
00:01:03,080 --> 00:01:05,600
Speaker 2: latter of which has made numerous noises about using cheaper

22
00:01:05,600 --> 00:01:08,320
Speaker 2: AI models like deep Seek and its co pilot cowork

23
00:01:08,400 --> 00:01:12,040
Speaker 2: product Yeah and Also, I should add the Sacha and

24
00:01:12,080 --> 00:01:14,520
Speaker 2: Adela keeps doing interviews where he's like, yeah, no, one

25
00:01:14,560 --> 00:01:17,800
Speaker 2: company should have all that power. We shouldn't rely on

26
00:01:17,800 --> 00:01:20,600
Speaker 2: one lab, one lab being able to be shut down

27
00:01:20,640 --> 00:01:21,840
Speaker 2: at any times bad.

28
00:01:21,959 --> 00:01:22,360
Speaker 3: We can't.

29
00:01:22,600 --> 00:01:25,440
Speaker 2: It's really funny watching this guy change his chew and

30
00:01:25,520 --> 00:01:27,640
Speaker 2: considering how a year and a half ago, a year

31
00:01:27,680 --> 00:01:30,920
Speaker 2: ago he was rock hard for open AI, and six

32
00:01:30,959 --> 00:01:33,920
Speaker 2: months after that rock harder for Anthropic. It's almost as

33
00:01:33,959 --> 00:01:36,319
Speaker 2: if he's just doing random shit based on what he

34
00:01:36,400 --> 00:01:40,640
Speaker 2: thinks might work, you know, like a fucking cargo cult. Anyway,

35
00:01:40,800 --> 00:01:43,360
Speaker 2: as I've mentioned over the last few months, both Anthropic

36
00:01:43,400 --> 00:01:46,400
Speaker 2: and open Ai both started to charge their enterprise customers

37
00:01:46,480 --> 00:01:48,480
Speaker 2: companies with over one hundred and fifty users for the

38
00:01:48,480 --> 00:01:51,080
Speaker 2: actual cost of the AI tooken spend as of the

39
00:01:51,120 --> 00:01:55,000
Speaker 2: middle of Q one, twenty twenty six. To explain, when

40
00:01:55,040 --> 00:01:57,480
Speaker 2: you use a regular chat, GPT or claud account, you

41
00:01:57,560 --> 00:02:00,240
Speaker 2: burn tokens, each one about three quarters of a word,

42
00:02:00,640 --> 00:02:03,360
Speaker 2: up to a five hour or weekly limit, depending on

43
00:02:03,400 --> 00:02:05,560
Speaker 2: what kind of model you're using. So with Anthropic on

44
00:02:05,560 --> 00:02:08,919
Speaker 2: their clause subscription Opus four point eight and Fable when

45
00:02:08,919 --> 00:02:12,320
Speaker 2: it was around, had a specific limit to themselves. But anyway,

46
00:02:12,560 --> 00:02:14,920
Speaker 2: just the more powerful the model powerful being meshed by

47
00:02:14,960 --> 00:02:17,280
Speaker 2: the companies, of course, the more it burns and the

48
00:02:17,360 --> 00:02:20,120
Speaker 2: less you can use it. On a regular subscription you're

49
00:02:20,120 --> 00:02:22,680
Speaker 2: able to burn and I am not kidding eight thousand

50
00:02:22,680 --> 00:02:25,640
Speaker 2: dollars worth of tokens a month on Claude and fourteen

51
00:02:25,760 --> 00:02:30,440
Speaker 2: thousand dollars a month on open Aiy's chet GPT codec subscription.

52
00:02:30,360 --> 00:02:31,880
Speaker 3: All for two hundred bucks a month.

53
00:02:32,080 --> 00:02:35,720
Speaker 2: Pretty good deal, right, you know? It's like the discount

54
00:02:35,760 --> 00:02:38,520
Speaker 2: prices sketch from Tim and Eric that I reference every

55
00:02:38,560 --> 00:02:41,760
Speaker 2: so often just to see if anyone emails me. And

56
00:02:41,800 --> 00:02:44,639
Speaker 2: so Yeah, after years of being able to burn thousands

57
00:02:44,680 --> 00:02:46,959
Speaker 2: of dollars of tokens for two hundred dollars a month,

58
00:02:47,080 --> 00:02:50,240
Speaker 2: enterprises are suddenly having to pay the actual costs. The

59
00:02:50,280 --> 00:02:53,520
Speaker 2: result is multiple companies capping their workers token spend. As

60
00:02:53,520 --> 00:02:55,720
Speaker 2: I've reported, both T Mobile and Brecks have done so,

61
00:02:55,800 --> 00:02:58,760
Speaker 2: and others have reported that Uber, Meta, Walmart, Coinbase, and

62
00:02:58,800 --> 00:03:01,880
Speaker 2: Cisco have all done so. Meta notably had a token

63
00:03:01,960 --> 00:03:06,480
Speaker 2: maxing leaderboard. Very funny again, just changing direction based on

64
00:03:06,600 --> 00:03:10,160
Speaker 2: random signals. None of these people have a plan, none

65
00:03:10,160 --> 00:03:12,440
Speaker 2: of them. None of these companies have a plan. Nobody

66
00:03:12,520 --> 00:03:15,240
Speaker 2: integrating AI has a plan. If they did, they would

67
00:03:15,240 --> 00:03:17,440
Speaker 2: have been like, Yeah, let's make sure we know how

68
00:03:17,480 --> 00:03:19,800
Speaker 2: much this is gonna cost, or let's make sure we

69
00:03:19,880 --> 00:03:21,799
Speaker 2: know whether this is good or not, whether we can

70
00:03:22,160 --> 00:03:26,040
Speaker 2: measure the roy you know, anything that would suggest anyone

71
00:03:26,120 --> 00:03:30,160
Speaker 2: knew what the hell they're doing, but they don't. And

72
00:03:30,320 --> 00:03:33,240
Speaker 2: four o four media reports that Management Consultancy Firm and

73
00:03:33,600 --> 00:03:37,520
Speaker 2: Evil Corporation Accenture, which is an insane way of saying

74
00:03:37,520 --> 00:03:39,640
Speaker 2: that I'm just going to keep has seen what it

75
00:03:39,720 --> 00:03:42,960
Speaker 2: calls soaring token spend based on leaked audio recordings, with

76
00:03:43,680 --> 00:03:46,200
Speaker 2: much of that token spend being driven by non engineers

77
00:03:46,200 --> 00:03:49,920
Speaker 2: doing things like combining PDFs into presentation slides. I love

78
00:03:50,000 --> 00:03:53,320
Speaker 2: this shit is the easiest stuff in the world made

79
00:03:53,600 --> 00:03:57,640
Speaker 2: slightly easier but probably worse, costing way more. You could

80
00:03:57,720 --> 00:03:59,640
Speaker 2: definitely hire an extra, but you could just hire a

81
00:03:59,640 --> 00:04:02,160
Speaker 2: contract to do this. You could iira a bloke at

82
00:04:02,200 --> 00:04:04,080
Speaker 2: forty five bucks an hour to do that for you.

83
00:04:04,320 --> 00:04:07,880
Speaker 2: Maybe the happiest man alive probably cost your less thought too.

84
00:04:07,960 --> 00:04:11,760
Speaker 2: I know, Ugh, God, this stuff is exhausting. You ever

85
00:04:11,840 --> 00:04:14,520
Speaker 2: sit and wonder what you could do with the rest

86
00:04:14,560 --> 00:04:17,919
Speaker 2: of that money I could have, Oh no, not like

87
00:04:18,080 --> 00:04:20,599
Speaker 2: millions of dollars a month. I could have like a

88
00:04:20,640 --> 00:04:23,640
Speaker 2: week's worth of diet coke. Anyway, it turns out that

89
00:04:23,720 --> 00:04:26,680
Speaker 2: telling your workers that AI can do basically anything and

90
00:04:26,760 --> 00:04:28,839
Speaker 2: to use it as much as possible starts costing you

91
00:04:28,839 --> 00:04:31,240
Speaker 2: a lot of money when you actually pay what it costs.

92
00:04:31,640 --> 00:04:34,839
Speaker 2: As I've said before, we're barely three months into enterprises

93
00:04:34,839 --> 00:04:37,480
Speaker 2: paying the actual cost of AI, and they're already screeching

94
00:04:37,520 --> 00:04:39,880
Speaker 2: like they're being pecked to death by birds. And that

95
00:04:40,040 --> 00:04:42,640
Speaker 2: joke is a courtesy of Kisei Kugawa, whos said that

96
00:04:42,720 --> 00:04:44,880
Speaker 2: to me about a year ago when we talked about

97
00:04:44,880 --> 00:04:47,080
Speaker 2: the amount of money that Anthropic was allowing people to

98
00:04:47,080 --> 00:04:50,760
Speaker 2: burn on Claude subscriptions. Meanwhile, the threat of open source

99
00:04:50,800 --> 00:04:53,240
Speaker 2: models appears to be getting more serious. I've heard from

100
00:04:53,320 --> 00:04:57,120
Speaker 2: numerous sources that Chinese AI lab z poos GLM five

101
00:04:57,120 --> 00:05:00,599
Speaker 2: point two coding model is competitive with Anthropic Sopus four

102
00:05:00,600 --> 00:05:02,880
Speaker 2: point eight at somewhere between a quarter and a sixth

103
00:05:02,920 --> 00:05:04,880
Speaker 2: of the price, but that seems to vary based on

104
00:05:04,960 --> 00:05:08,000
Speaker 2: the cost of the task. As far as whether GLM

105
00:05:08,080 --> 00:05:10,919
Speaker 2: five point two is actually competitive goes it appears that

106
00:05:10,960 --> 00:05:13,440
Speaker 2: the hype is somewhat real, but it's kind of you

107
00:05:13,520 --> 00:05:16,400
Speaker 2: kind of had to check your saucing. Open source AI

108
00:05:16,440 --> 00:05:19,400
Speaker 2: coding company Cliin ran benchmarks on a natural bug from

109
00:05:19,400 --> 00:05:22,080
Speaker 2: its repo and found that while GLM used twice as

110
00:05:22,120 --> 00:05:24,200
Speaker 2: many tokens to fix it, it also did so at

111
00:05:24,200 --> 00:05:27,080
Speaker 2: a cheaper cost forty one cents versus OPUS four point

112
00:05:27,080 --> 00:05:30,680
Speaker 2: eights eighty one cents. Opus was quicker, taking one point

113
00:05:30,720 --> 00:05:34,279
Speaker 2: six minutes versus glm's four point seven minutes, but one

114
00:05:34,320 --> 00:05:37,440
Speaker 2: thing that stood out was the GLM to quote client

115
00:05:37,680 --> 00:05:40,440
Speaker 2: cleaned up dead code and verified the build compiled before

116
00:05:40,480 --> 00:05:43,719
Speaker 2: completing something that Opus, the more expensive frontier model run

117
00:05:43,720 --> 00:05:46,360
Speaker 2: by the company that uses billions of dollars, apparently didn't

118
00:05:46,360 --> 00:05:49,279
Speaker 2: bother to do. It's just one bug, but GLM five

119
00:05:49,279 --> 00:05:51,120
Speaker 2: point two seems to score well on many of the

120
00:05:51,160 --> 00:05:55,480
Speaker 2: major benchmarks, coming third on cybersecurity firms semgreps cyber benchmarks,

121
00:05:55,480 --> 00:05:58,360
Speaker 2: with the top two being semgp's own harness, so their

122
00:05:58,400 --> 00:05:59,960
Speaker 2: own thing that they put on top of the models

123
00:06:00,080 --> 00:06:03,160
Speaker 2: running Opus four point eight and GPT five point five,

124
00:06:03,680 --> 00:06:07,360
Speaker 2: followed by GLM five point two, which is interesting.

125
00:06:07,640 --> 00:06:09,760
Speaker 3: It's very interesting if.

126
00:06:09,960 --> 00:06:11,960
Speaker 2: I was Glamy Samuel when I'd be looking at this

127
00:06:12,000 --> 00:06:14,960
Speaker 2: and getting a little worried. I'd genuinely be looking at

128
00:06:14,960 --> 00:06:17,160
Speaker 2: them and even saying, oh, why don't I train my own?

129
00:06:17,960 --> 00:06:20,400
Speaker 2: But you know, that might actually be a bad idea,

130
00:06:20,440 --> 00:06:23,880
Speaker 2: even though it's fully possible, because GLM five point two

131
00:06:23,920 --> 00:06:26,200
Speaker 2: is a true open weights model. It's released under something

132
00:06:26,240 --> 00:06:29,960
Speaker 2: called the unrestricted MIT license, which means you can anyone,

133
00:06:30,080 --> 00:06:32,800
Speaker 2: by the way company, a regular person whomever, can self

134
00:06:32,839 --> 00:06:34,640
Speaker 2: host on, fine tune it, or use it in any

135
00:06:34,680 --> 00:06:37,760
Speaker 2: way they wish. And yeah, I mean in theory, any

136
00:06:37,760 --> 00:06:39,880
Speaker 2: company that wants to use LMS could either run their

137
00:06:39,920 --> 00:06:42,040
Speaker 2: own version of it locally or be on a rig

138
00:06:42,120 --> 00:06:44,159
Speaker 2: that cost tens of thousands of dollars, or pay an

139
00:06:44,200 --> 00:06:46,400
Speaker 2: inference provider like base ten to spin up and train

140
00:06:46,480 --> 00:06:49,880
Speaker 2: their own version. Now, I must be clear that there's

141
00:06:50,040 --> 00:06:52,479
Speaker 2: not a ton of evidence that running GLM five point

142
00:06:52,560 --> 00:06:55,080
Speaker 2: two is profitable for the inference provider. In fact, there's

143
00:06:55,279 --> 00:06:57,919
Speaker 2: really note I genuinely have looked, nor do we have

144
00:06:57,960 --> 00:07:00,320
Speaker 2: any significant signs that customers are moving away way on

145
00:07:00,440 --> 00:07:04,000
Speaker 2: mass from open AI and anthropics models. People generally like

146
00:07:04,160 --> 00:07:07,960
Speaker 2: name brands, and they even though Anthropic shit is constantly broken.

147
00:07:08,040 --> 00:07:12,440
Speaker 2: Go look at their stability page, stability page, uptime page.

148
00:07:12,440 --> 00:07:14,160
Speaker 2: I don't know what you call that. Email me if

149
00:07:14,160 --> 00:07:15,800
Speaker 2: you know what they call that, or I could just

150
00:07:15,840 --> 00:07:16,400
Speaker 2: look after this.

151
00:07:16,440 --> 00:07:19,520
Speaker 3: I'm going to keep going though. Nevertheless, people want name brands.

152
00:07:19,520 --> 00:07:22,520
Speaker 2: People like paying for a company that ostensibly works on

153
00:07:22,560 --> 00:07:23,760
Speaker 2: this and keeps it updated.

154
00:07:24,000 --> 00:07:25,120
Speaker 3: How true is that? Who knows.

155
00:07:25,160 --> 00:07:27,640
Speaker 2: I don't think people are buying lllms because of particularly

156
00:07:27,640 --> 00:07:31,240
Speaker 2: sensible reasons anyway. But what might change things would be Microsoft,

157
00:07:31,320 --> 00:07:34,200
Speaker 2: Amazon or Google offering GLM five point two via their

158
00:07:34,280 --> 00:07:38,080
Speaker 2: foundry Bedrock or Vertex AI platforms. And even then it's

159
00:07:38,400 --> 00:07:41,000
Speaker 2: how does left customers are excited about AI or just

160
00:07:41,120 --> 00:07:44,119
Speaker 2: paying open AI and Anthropic to use buzzy models? Really

161
00:07:44,160 --> 00:07:48,040
Speaker 2: isn't clear. We should know by now. All that being said,

162
00:07:49,120 --> 00:07:51,560
Speaker 2: now does feel like the perfect time for a shift

163
00:07:51,560 --> 00:07:53,840
Speaker 2: towards open source models, se if only because everyone is

164
00:07:53,880 --> 00:07:57,320
Speaker 2: crapping themselves about costs. Everyone is really freaking out, And

165
00:07:57,360 --> 00:07:58,760
Speaker 2: I think that this is going to be the time

166
00:07:58,800 --> 00:08:01,720
Speaker 2: when we start separate the wheat from the chaff, the

167
00:08:01,760 --> 00:08:04,880
Speaker 2: people that actually like this stuff versus the people who

168
00:08:04,960 --> 00:08:07,720
Speaker 2: just feel good because they're doing AI and they don't

169
00:08:07,720 --> 00:08:09,920
Speaker 2: know what else to do with their time, and they

170
00:08:10,040 --> 00:08:11,600
Speaker 2: like forcing people to use it.

171
00:08:12,040 --> 00:08:23,480
Speaker 4: Now I'll get to who actually does in a minute.

172
00:08:25,080 --> 00:08:28,600
Speaker 2: Open Routers ranking show a dramatic move away from Frontier labs.

173
00:08:28,680 --> 00:08:30,960
Speaker 2: The top ten most popular models are all open source

174
00:08:31,000 --> 00:08:33,760
Speaker 2: safe for Opus four point seven at six, Opus four

175
00:08:33,760 --> 00:08:36,440
Speaker 2: point eight at eight, and Sonnet four point six at ten,

176
00:08:36,640 --> 00:08:39,160
Speaker 2: as well as a mysterious new model called al Alpha,

177
00:08:39,280 --> 00:08:41,360
Speaker 2: which is currently free to use, which is probably why

178
00:08:41,400 --> 00:08:44,720
Speaker 2: it's trending. To be clear, anthropics models still dominate the

179
00:08:44,720 --> 00:08:47,240
Speaker 2: top models by task across the board, and that's based

180
00:08:47,240 --> 00:08:51,160
Speaker 2: on the spend on the task. I hate to give

181
00:08:51,200 --> 00:08:53,720
Speaker 2: him any credit though, I really don't like doing this,

182
00:08:53,840 --> 00:08:56,840
Speaker 2: but Scott Gallaway recently made the point that China could

183
00:08:56,880 --> 00:08:59,960
Speaker 2: thrive by AI dumping high quality open source models into

184
00:09:00,040 --> 00:09:03,079
Speaker 2: the ecosystem as a means of destabilizing the major model providers.

185
00:09:04,080 --> 00:09:06,600
Speaker 2: And I hate to say that's what's happening. It's like

186
00:09:06,640 --> 00:09:10,199
Speaker 2: a fucking broken clock. I guess. Zepu and Minimax are

187
00:09:10,200 --> 00:09:13,559
Speaker 2: both partially government funded by the Chinese government and also

188
00:09:13,679 --> 00:09:16,439
Speaker 2: both wofully unprofitable, but neither do so at the scale

189
00:09:16,480 --> 00:09:19,360
Speaker 2: of anthropic or open aim. While both do the kfabe

190
00:09:19,400 --> 00:09:23,520
Speaker 2: discussion of AGI and autonomous compute and all that bullshit,

191
00:09:23,720 --> 00:09:26,520
Speaker 2: they seem far more focused on creating open weight models

192
00:09:26,559 --> 00:09:28,960
Speaker 2: to compete with them. And just for the differentiation here,

193
00:09:29,160 --> 00:09:31,080
Speaker 2: open source means you can share some of it. Open

194
00:09:31,080 --> 00:09:33,400
Speaker 2: weights means you actually can deddle with the model itself.

195
00:09:33,720 --> 00:09:36,080
Speaker 2: That's a flattening. Someone's gonna email. They're gonna be mad

196
00:09:36,120 --> 00:09:36,360
Speaker 2: at me.

197
00:09:36,920 --> 00:09:37,480
Speaker 3: I don't care.

198
00:09:38,320 --> 00:09:41,000
Speaker 2: Sorry, Like it gets the point across, all right, just

199
00:09:41,280 --> 00:09:42,480
Speaker 2: right on the subreddit if.

200
00:09:42,400 --> 00:09:43,560
Speaker 3: You're mad anyway.

201
00:09:43,720 --> 00:09:46,200
Speaker 2: In truth, I think the future of lms will end

202
00:09:46,280 --> 00:09:50,160
Speaker 2: up being locally run models and expensive specialist hardware. The

203
00:09:50,280 --> 00:09:53,439
Speaker 2: underlying economics of cloud based GPU compute do not make

204
00:09:53,480 --> 00:09:55,840
Speaker 2: sense for anybody. You need to make sure the GPUs

205
00:09:55,840 --> 00:09:59,360
Speaker 2: are saturated. You have to buy in advance, otherwise you

206
00:09:59,400 --> 00:10:01,680
Speaker 2: won't like, you just won't be able to get the capacity.

207
00:10:01,720 --> 00:10:03,480
Speaker 2: And if you have less people than you expect, you

208
00:10:03,480 --> 00:10:04,600
Speaker 2: will lose money for sure.

209
00:10:04,600 --> 00:10:05,199
Speaker 3: And if you have.

210
00:10:05,480 --> 00:10:07,760
Speaker 2: More people than you expect, you have to get more compute,

211
00:10:07,760 --> 00:10:10,000
Speaker 2: which means you'll also lose money. So as long as

212
00:10:10,040 --> 00:10:13,240
Speaker 2: you can goldilocks this bullshit, maybe you'll eke out the

213
00:10:13,240 --> 00:10:17,080
Speaker 2: world's worst gross margins in history, or you'll just lose money,

214
00:10:17,120 --> 00:10:18,120
Speaker 2: which is what everyone's doing.

215
00:10:18,559 --> 00:10:19,800
Speaker 3: But I think there may be.

216
00:10:19,800 --> 00:10:22,720
Speaker 2: A point at which ZEPU, minimacs or even American model

217
00:10:22,720 --> 00:10:26,640
Speaker 2: developers start focusing more on services driven custom deployments and

218
00:10:26,720 --> 00:10:30,520
Speaker 2: become boring, ugly licensed driven monstrosities like Oracle.

219
00:10:31,000 --> 00:10:31,760
Speaker 3: And that's the thing.

220
00:10:32,520 --> 00:10:36,320
Speaker 2: I've talked to quite a few people recently and they

221
00:10:36,360 --> 00:10:39,520
Speaker 2: are all saying the same thing. They've even in Vidia

222
00:10:39,679 --> 00:10:42,760
Speaker 2: is moving into this realm and they're selling these giant

223
00:10:42,880 --> 00:10:46,880
Speaker 2: hundred grand machines. I think that's kind of interesting. And

224
00:10:46,960 --> 00:10:49,920
Speaker 2: like I said, that GPU compute, it requires a certain

225
00:10:49,920 --> 00:10:52,280
Speaker 2: critical mass of customer demand that I don't believe will

226
00:10:52,320 --> 00:10:55,480
Speaker 2: exist once the cargo culture of the AI bubble passes,

227
00:10:55,520 --> 00:10:59,000
Speaker 2: because GPU commitments are done a few years in advance,

228
00:10:59,080 --> 00:11:01,840
Speaker 2: and any drop Paul Lull kills the margins of even

229
00:11:01,880 --> 00:11:04,720
Speaker 2: a cheap to run model. Remember Baseten just they're an

230
00:11:04,720 --> 00:11:07,200
Speaker 2: influence provider. They just raise one and a half billion dollars.

231
00:11:07,720 --> 00:11:11,719
Speaker 2: They're not profitable. I mean, ZEPU isn't profitable, and they

232
00:11:11,760 --> 00:11:14,120
Speaker 2: do the same thing open ai does with their stuff

233
00:11:14,120 --> 00:11:16,960
Speaker 2: where they say, oh yeah, yeah, our cost of revenue,

234
00:11:16,960 --> 00:11:19,080
Speaker 2: it's lower than our revenue. However, when you add the

235
00:11:19,120 --> 00:11:22,280
Speaker 2: sales and marketing, it's actually it's actually unprofitable as well.

236
00:11:22,400 --> 00:11:24,680
Speaker 2: Oh also we lose hundreds of millions of dollars because

237
00:11:24,720 --> 00:11:28,480
Speaker 2: of training. Other than that, we're super profitable. Yeah, it's

238
00:11:28,559 --> 00:11:32,480
Speaker 2: really fucking stupid industry. And look, even if GLM five

239
00:11:32,520 --> 00:11:34,400
Speaker 2: point two can do things at half the cost of

240
00:11:34,440 --> 00:11:37,000
Speaker 2: OPAHS four point eight, that's still half of what in

241
00:11:37,040 --> 00:11:39,880
Speaker 2: some cases is hundreds of thousands or millions of dollars

242
00:11:39,920 --> 00:11:42,880
Speaker 2: a month. I've heard tell from multiple source is that

243
00:11:42,880 --> 00:11:46,440
Speaker 2: that hardware demand is growing and that customers at big

244
00:11:46,440 --> 00:11:49,479
Speaker 2: companies are starting to look into those high end hardware solutions.

245
00:11:50,040 --> 00:11:53,800
Speaker 2: And I think that that is the only future for

246
00:11:53,880 --> 00:11:56,040
Speaker 2: this stuff, and I think it would reduce the cost

247
00:11:56,080 --> 00:11:59,160
Speaker 2: significantly for the end user, probably keep this shit off

248
00:11:59,200 --> 00:12:04,920
Speaker 2: the internet, massively marginalize the damage that LM based code

249
00:12:04,960 --> 00:12:07,520
Speaker 2: can do by focusing on people actually give enough of

250
00:12:07,559 --> 00:12:10,600
Speaker 2: a shit to buy the buy the actual machines. And

251
00:12:10,960 --> 00:12:13,480
Speaker 2: ah No, I feel like an actual serious fiscal commitment

252
00:12:13,520 --> 00:12:16,280
Speaker 2: makes companies take things a little bit more seriously, unless,

253
00:12:16,320 --> 00:12:19,040
Speaker 2: of course, your Microsoft, Google and Amazon and meta buying GPUs,

254
00:12:19,040 --> 00:12:21,840
Speaker 2: at which case you gives a shit. But I mean,

255
00:12:21,880 --> 00:12:25,280
Speaker 2: if this takes off, it is it's curtains vazousha. It's

256
00:12:25,320 --> 00:12:27,800
Speaker 2: it's a bad time for Open AI and Anthropic unless

257
00:12:27,840 --> 00:12:31,160
Speaker 2: they too start doing these hardware driven things. This is

258
00:12:31,240 --> 00:12:33,240
Speaker 2: just a guess. This is just a guess they would

259
00:12:33,240 --> 00:12:36,480
Speaker 2: ever think about that. But I think it's possible. I

260
00:12:36,480 --> 00:12:39,160
Speaker 2: think it's really the only economically viable end for this

261
00:12:39,520 --> 00:12:41,400
Speaker 2: unless there are just going to be companies that lose

262
00:12:41,640 --> 00:12:44,240
Speaker 2: billions of dollars a year for no apparent reason. I

263
00:12:44,320 --> 00:12:46,679
Speaker 2: just don't think that's gonna happen. It's not enough money

264
00:12:46,679 --> 00:12:50,120
Speaker 2: to do it. And goddamn, please go and listen to

265
00:12:50,200 --> 00:12:53,480
Speaker 2: the Open AI will not get bailed out AI is

266
00:12:53,559 --> 00:12:56,240
Speaker 2: not too big to Fail podcast They did.

267
00:12:56,320 --> 00:12:57,320
Speaker 3: I'm tired of.

268
00:12:57,240 --> 00:12:59,959
Speaker 2: Getting Please don't please stop emailing me.

269
00:13:00,240 --> 00:13:01,520
Speaker 3: I love hearing from you.

270
00:13:01,559 --> 00:13:03,040
Speaker 2: I really I would love to hear from you all

271
00:13:03,080 --> 00:13:06,360
Speaker 2: more other than the people who email me saying I

272
00:13:06,400 --> 00:13:08,640
Speaker 2: think the GPUs all for surveillance.

273
00:13:08,880 --> 00:13:11,000
Speaker 1: I think it's all that. I think it's a big

274
00:13:11,040 --> 00:13:16,320
Speaker 1: going s that's called Oracle. Oracle's first client was the CIA.

275
00:13:16,440 --> 00:13:20,080
Speaker 1: Oracle has a giant government gpu array. They have several

276
00:13:20,080 --> 00:13:20,320
Speaker 1: of them.

277
00:13:20,440 --> 00:13:23,600
Speaker 2: Jesus Christ, that's already happening, and they're not doing surveillance

278
00:13:23,600 --> 00:13:25,000
Speaker 2: on that. They did that with Maven, they did it

279
00:13:25,040 --> 00:13:28,080
Speaker 2: with the promise software. Jesus fucking Christ. Sorry, I understand

280
00:13:28,120 --> 00:13:31,040
Speaker 2: that I sound agitated here. It's just that if the

281
00:13:31,080 --> 00:13:34,360
Speaker 2: only choice you can make when looking at something is

282
00:13:34,360 --> 00:13:36,920
Speaker 2: to say, well, bad thing will definitely happen, that's not

283
00:13:37,360 --> 00:13:42,160
Speaker 2: actually an intellectual pursuit. It's just crapping yourself. It's just saying, well,

284
00:13:42,160 --> 00:13:44,839
Speaker 2: everything's bad, so everything stays bad. I actually want to

285
00:13:44,840 --> 00:13:46,880
Speaker 2: bring you a message of hope. I don't think these

286
00:13:46,880 --> 00:13:49,480
Speaker 2: companies get bailed down. I don't think they get anything.

287
00:13:49,880 --> 00:13:52,520
Speaker 2: I think that they get allowed to die. I think

288
00:13:52,640 --> 00:13:55,920
Speaker 2: partner assets might be scooped up by Amazon, Google and Microsoft.

289
00:13:56,080 --> 00:13:58,960
Speaker 2: I think the government might take a chunk out of them.

290
00:13:59,160 --> 00:14:02,640
Speaker 2: Perhaps they get some sort of bridge financing. But I

291
00:14:02,800 --> 00:14:05,040
Speaker 2: actually don't know if the government could even extend that

292
00:14:05,200 --> 00:14:10,080
Speaker 2: because AI is very unpopular with regular people. Regular people

293
00:14:10,200 --> 00:14:12,880
Speaker 2: do not like this, and they hate data centers, and

294
00:14:12,920 --> 00:14:17,400
Speaker 2: they have intelligently connected data centers, these horrible AI companies,

295
00:14:18,280 --> 00:14:20,120
Speaker 2: and so in the end, I tell you please stop

296
00:14:20,640 --> 00:14:23,840
Speaker 2: emailing me. The surveillance thing. Hey, it's not true. You're wrong.

297
00:14:24,240 --> 00:14:27,840
Speaker 2: You sound you sound like someone that wants to be

298
00:14:27,920 --> 00:14:29,960
Speaker 2: paranoid but doesn't want to do the reading well enough

299
00:14:29,960 --> 00:14:31,920
Speaker 2: to be paranoid. If you want to be paranoid, go

300
00:14:31,920 --> 00:14:34,200
Speaker 2: read the promised software. Go look at Oracle like Go

301
00:14:34,240 --> 00:14:36,160
Speaker 2: look at the actual history of these companies if you

302
00:14:36,200 --> 00:14:37,960
Speaker 2: want to be freaked out by something.

303
00:14:38,000 --> 00:14:40,760
Speaker 3: But in truth, the AI industry is just.

304
00:14:40,720 --> 00:14:45,280
Speaker 2: A direction the zeg Grigor of capital bumbles about and

305
00:14:45,720 --> 00:14:48,960
Speaker 2: is only kept alive by a belief system that's mostly

306
00:14:49,040 --> 00:14:52,200
Speaker 2: growth focused and doesn't really understand how to run a

307
00:14:52,200 --> 00:14:56,440
Speaker 2: business anymore or even build innovation or technology. And it's

308
00:14:56,480 --> 00:14:59,680
Speaker 2: blatantly obvious that lms are at best a classic risk

309
00:14:59,760 --> 00:15:03,040
Speaker 2: versus reward software tool for the real sickos that know

310
00:15:03,080 --> 00:15:05,440
Speaker 2: what they're getting into. But the problem is it's been

311
00:15:05,520 --> 00:15:09,040
Speaker 2: sold as this mass market panacea for basically any problem.

312
00:15:09,480 --> 00:15:13,680
Speaker 2: And I'll say that my thesis remains undeterred. I believe

313
00:15:13,680 --> 00:15:16,640
Speaker 2: a great deal of AI usage is cargo cult shit

314
00:15:16,800 --> 00:15:19,320
Speaker 2: driven by executives that don't do any work and people

315
00:15:19,320 --> 00:15:21,600
Speaker 2: that feel pressured into adopting AI by their peers on

316
00:15:21,640 --> 00:15:24,920
Speaker 2: the media, or of course, people that spend twelve to

317
00:15:25,120 --> 00:15:29,080
Speaker 2: eighteen hours a day on Twitter who just follow the algorithm.

318
00:15:29,080 --> 00:15:32,080
Speaker 2: I call it the Dunce's Casino because you just crank

319
00:15:32,200 --> 00:15:34,320
Speaker 2: the thing and a bunch of insane people come out.

320
00:15:34,560 --> 00:15:36,720
Speaker 2: And all of these people come out and they're like, if.

321
00:15:36,560 --> 00:15:38,920
Speaker 3: You don't use agents, you're going to die.

322
00:15:38,960 --> 00:15:41,080
Speaker 2: They're going to kill your wife, They're going to eat

323
00:15:41,120 --> 00:15:41,640
Speaker 2: your dog.

324
00:15:41,800 --> 00:15:44,360
Speaker 1: If you don't use loops, Boris Journey's going to blow

325
00:15:44,440 --> 00:15:46,320
Speaker 1: up your house. Ah.

326
00:15:46,760 --> 00:15:50,480
Speaker 2: Twitter is driving people insane using AI, but not because

327
00:15:50,520 --> 00:15:53,840
Speaker 2: of groc or anything, but because it just fire hoses

328
00:15:54,040 --> 00:15:58,240
Speaker 2: nutters into your feed who will constantly make you feel

329
00:15:58,320 --> 00:16:01,480
Speaker 2: terrible for not picking up the next AI thing. It's

330
00:16:01,520 --> 00:16:05,320
Speaker 2: this insane. It's I keep saying cargo cult because that's

331
00:16:05,360 --> 00:16:08,000
Speaker 2: what it is. It's people just going around being like, yeah,

332
00:16:08,040 --> 00:16:09,240
Speaker 2: you know, this is the big thing now.

333
00:16:09,360 --> 00:16:11,200
Speaker 3: We all like that, right, we love it, We love it.

334
00:16:11,200 --> 00:16:11,600
Speaker 3: It's good.

335
00:16:11,600 --> 00:16:13,800
Speaker 2: If we all follow this, the great prophecy will come

336
00:16:13,800 --> 00:16:17,760
Speaker 2: true and everyone then everyone will make money somewhere somehow.

337
00:16:17,880 --> 00:16:33,400
Speaker 2: I think it's sickness. It's a genuine sickness. So I

338
00:16:33,400 --> 00:16:35,600
Speaker 2: went a little Tucker cos and there, what's going on?

339
00:16:35,960 --> 00:16:37,080
Speaker 2: Why are we using lms.

340
00:16:37,760 --> 00:16:39,360
Speaker 3: What are they going to do to us? What are

341
00:16:39,360 --> 00:16:40,600
Speaker 3: they going to do to your family?

342
00:16:41,280 --> 00:16:43,560
Speaker 2: But anyway, back to the podcast, back to the monologue,

343
00:16:43,560 --> 00:16:47,200
Speaker 2: I apologize for out libbing. I must be clear as well.

344
00:16:47,520 --> 00:16:49,800
Speaker 2: I do love hearing from you because I keep getting

345
00:16:49,800 --> 00:16:52,000
Speaker 2: emails from people that actually use AI, and I got

346
00:16:52,000 --> 00:16:54,600
Speaker 2: one just before I started recording this from a bloke

347
00:16:54,640 --> 00:16:58,080
Speaker 2: who said, and I quote, he's used it to deliver

348
00:16:58,080 --> 00:17:01,760
Speaker 2: a good work and seeing it tried to submit absolute cramp.

349
00:17:02,160 --> 00:17:04,760
Speaker 2: And that really is the AI industry. Hey, you know,

350
00:17:04,960 --> 00:17:07,440
Speaker 2: sometime it works, and when it does, it's all right,

351
00:17:07,480 --> 00:17:11,040
Speaker 2: and when it doesn't is dogshit. Yay, great, I'm so

352
00:17:11,160 --> 00:17:14,560
Speaker 2: glad we sunk a trillion plus dollars into this. Look,

353
00:17:14,600 --> 00:17:17,640
Speaker 2: these are specialist tools that are being forced to solve

354
00:17:17,720 --> 00:17:21,040
Speaker 2: generalist problems, and they're probabilistic models that fail to do

355
00:17:21,119 --> 00:17:24,840
Speaker 2: deterministic things, and they're really just an eternal financial black

356
00:17:24,840 --> 00:17:27,600
Speaker 2: hole for anyone trying to scale them into a Google

357
00:17:27,680 --> 00:17:31,480
Speaker 2: search or a meta sized business. And it's never going

358
00:17:31,520 --> 00:17:34,840
Speaker 2: to work like that. Just before this, as well, I

359
00:17:34,880 --> 00:17:39,119
Speaker 2: saw that Broadcom and open Ai announced their Jalipino chip

360
00:17:39,400 --> 00:17:42,720
Speaker 2: is giant dinner place sized motherfucker.

361
00:17:42,359 --> 00:17:44,720
Speaker 3: Dinner place, dinner plate. I'm not fixing that.

362
00:17:45,640 --> 00:17:49,200
Speaker 2: And it's funny because everyone's doing the cargo culture around it.

363
00:17:49,240 --> 00:17:51,160
Speaker 2: They're all like, oh, well, it's going to bring down

364
00:17:51,200 --> 00:17:53,880
Speaker 2: inference costs. If they did that, that would fix everything.

365
00:17:54,119 --> 00:17:58,000
Speaker 2: If Kalapino fixes everything, then everything will work out. There's

366
00:17:58,000 --> 00:18:01,280
Speaker 2: no proof that that's the case. Actually, here's the thing.

367
00:18:01,440 --> 00:18:03,960
Speaker 2: If you want if you want to like, here's a

368
00:18:04,000 --> 00:18:07,399
Speaker 2: specific request if you have an information subscription. At some

369
00:18:07,520 --> 00:18:09,479
Speaker 2: point in the last year, I know there was an

370
00:18:09,520 --> 00:18:13,600
Speaker 2: information story that suggested that the Broadcom's open Ai chip

371
00:18:13,680 --> 00:18:15,720
Speaker 2: would have modest gains. If you bring that to me,

372
00:18:16,119 --> 00:18:19,040
Speaker 2: I will fuck it. I will venmo you five dollars

373
00:18:19,520 --> 00:18:22,760
Speaker 2: if you can bring me that story. Seriously, I need

374
00:18:22,760 --> 00:18:27,080
Speaker 2: that story because I'm pretty sure that if Broadcom's chip

375
00:18:27,160 --> 00:18:29,440
Speaker 2: was going to change everything for open Ai, open Ai

376
00:18:29,520 --> 00:18:32,920
Speaker 2: would sink every fucking dollar they had into it. Open

377
00:18:32,960 --> 00:18:37,280
Speaker 2: Ai would dispense within video. Instead, it's like, oh, yeah,

378
00:18:37,320 --> 00:18:39,199
Speaker 2: we're getting Microsoft to pay for some of it, and

379
00:18:39,200 --> 00:18:40,840
Speaker 2: it'll be there in twenty twenty seven.

380
00:18:40,920 --> 00:18:42,680
Speaker 3: Maybe. Uh oh no.

381
00:18:43,400 --> 00:18:45,960
Speaker 2: Sounds like something you pre ordered on Kickstarter and you

382
00:18:46,040 --> 00:18:47,840
Speaker 2: got it and it kind of sucks and it turns

383
00:18:47,840 --> 00:18:50,840
Speaker 2: out the video is AI generated, which I did just

384
00:18:50,880 --> 00:18:53,879
Speaker 2: see with this thing where someone had done a Kickstarter

385
00:18:53,920 --> 00:18:55,960
Speaker 2: where it's a yeah, you can build things using real

386
00:18:56,000 --> 00:18:58,400
Speaker 2: principles and you pour your own cement, but it's really

387
00:18:58,440 --> 00:19:02,400
Speaker 2: small and the whole thing is AI generated. Fucking how

388
00:19:02,640 --> 00:19:04,320
Speaker 2: this industry really is miserable?

389
00:19:04,359 --> 00:19:04,719
Speaker 3: Isn't it?

390
00:19:05,160 --> 00:19:08,880
Speaker 2: All the people every other era, other than of course

391
00:19:08,920 --> 00:19:11,280
Speaker 2: crypto and the metaverse, when something cool happened. Even in

392
00:19:11,320 --> 00:19:15,360
Speaker 2: the augmented reality era, there were still people doing goofy,

393
00:19:15,440 --> 00:19:18,960
Speaker 2: funny shit. There were still people doing really interesting things

394
00:19:18,960 --> 00:19:21,840
Speaker 2: with computer vision. There were people doing interesting things with

395
00:19:21,920 --> 00:19:24,360
Speaker 2: depth sensing. The fact you can measure stuff on your

396
00:19:24,520 --> 00:19:26,400
Speaker 2: iPhone I'm pretty sure you can do on Android too,

397
00:19:26,680 --> 00:19:29,200
Speaker 2: is because of depth sensing cameras and augmented reality.

398
00:19:29,440 --> 00:19:30,040
Speaker 3: It's cool.

399
00:19:30,440 --> 00:19:33,120
Speaker 2: The snap spectacles are embarrassing, and if you get them,

400
00:19:33,200 --> 00:19:37,040
Speaker 2: I assume that people will just come and beat you up.

401
00:19:37,240 --> 00:19:40,600
Speaker 2: I mean it triggers a deep bully instinct. I think

402
00:19:40,680 --> 00:19:44,960
Speaker 2: in ninety percent of human beings, you look like Roy Orbison,

403
00:19:45,720 --> 00:19:48,639
Speaker 2: but so much worse and with so much less talent.

404
00:19:49,400 --> 00:19:52,359
Speaker 2: Hell no, this whole era pisses me off. That's the

405
00:19:52,400 --> 00:19:54,080
Speaker 2: long and shure of it. I think you've probably got

406
00:19:54,119 --> 00:19:56,520
Speaker 2: that by this point. I and apologize that I've been

407
00:19:56,560 --> 00:19:58,520
Speaker 2: a little all over the place of this monologue. But

408
00:19:58,800 --> 00:20:01,199
Speaker 2: it has been a long few months. I've still got

409
00:20:01,200 --> 00:20:04,400
Speaker 2: plenty of energy, but this has definitely been a tough

410
00:20:04,480 --> 00:20:08,040
Speaker 2: few But it's hard to tell what happens next. I

411
00:20:08,080 --> 00:20:09,800
Speaker 2: know it's hard. I know you asked me when it

412
00:20:09,800 --> 00:20:12,639
Speaker 2: will burst. I truly don't know. But I'm gonna do

413
00:20:12,720 --> 00:20:15,119
Speaker 2: my best every week to inform you, keep you up

414
00:20:15,160 --> 00:20:17,400
Speaker 2: to date, and to try and explain what the fuck

415
00:20:17,440 --> 00:20:20,040
Speaker 2: is going on, because quite frankly, I don't think most

416
00:20:20,080 --> 00:20:22,960
Speaker 2: people have any idea, and I think that that's actually

417
00:20:23,280 --> 00:20:26,240
Speaker 2: the story of the moment, because when you look around

418
00:20:26,280 --> 00:20:30,240
Speaker 2: the tech industry right now, does this strike you as

419
00:20:30,240 --> 00:20:34,960
Speaker 2: an industry full of leaders or followers? Meta is apparently

420
00:20:35,040 --> 00:20:39,480
Speaker 2: working on a polymarket competitor after releasing some new three

421
00:20:39,560 --> 00:20:44,239
Speaker 2: hundred dollars fucking glasses that only perverts will wear. What

422
00:20:44,440 --> 00:20:47,600
Speaker 2: Microsoft a year ago was saying open source was bad,

423
00:20:47,760 --> 00:20:51,400
Speaker 2: now it's good. Six months ago, Anthropic was good. Now

424
00:20:51,440 --> 00:20:54,200
Speaker 2: most of a Sulliman says they're too expensive and they're

425
00:20:54,200 --> 00:20:57,200
Speaker 2: going to stop working with them. This is an industry

426
00:20:57,200 --> 00:20:59,760
Speaker 2: full of people working things out as they go along.

427
00:21:00,080 --> 00:21:02,480
Speaker 2: Except the difference between you and them is that they

428
00:21:02,520 --> 00:21:07,200
Speaker 2: have billions of dollars, billions and dollars, billions and dollars

429
00:21:07,240 --> 00:21:11,040
Speaker 2: I'm keeping it billions of dollars, and also all the

430
00:21:11,080 --> 00:21:14,000
Speaker 2: talent and all the resources and all the ability to

431
00:21:14,080 --> 00:21:16,520
Speaker 2: raise that they could do anything.

432
00:21:16,240 --> 00:21:17,840
Speaker 3: And they choose to do this.

433
00:21:18,800 --> 00:21:22,399
Speaker 2: Remember that even when this era ends, remember that this

434
00:21:22,600 --> 00:21:26,320
Speaker 2: is what they chose to do. This is the intentional

435
00:21:26,640 --> 00:21:30,879
Speaker 2: choice of Google, Microsoft, Meta and Amazon. They could have

436
00:21:30,960 --> 00:21:35,359
Speaker 2: done anything else, anything, And there are real problems they

437
00:21:35,359 --> 00:21:37,639
Speaker 2: could solve. There are problems they could solve with their

438
00:21:37,680 --> 00:21:42,040
Speaker 2: own fucking problems that they've caused with their apps. They

439
00:21:42,080 --> 00:21:45,679
Speaker 2: could make Facebook better or Instagram better, except doing so

440
00:21:45,800 --> 00:21:46,840
Speaker 2: with lower their revenue.

441
00:21:46,960 --> 00:21:47,320
Speaker 3: I don't know.

442
00:21:47,400 --> 00:21:53,240
Speaker 2: Microsoft could redesign the office suite to work to actually

443
00:21:53,240 --> 00:21:56,359
Speaker 2: work to They could make Microsoft teams stable. Google Search

444
00:21:56,400 --> 00:21:58,520
Speaker 2: could work again again. It would lower the growth, It

445
00:21:58,560 --> 00:22:00,600
Speaker 2: would make the growth go down. And I think by

446
00:22:00,680 --> 00:22:02,720
Speaker 2: the time these fuck nuts realize that they have a

447
00:22:02,760 --> 00:22:05,720
Speaker 2: real problem, they'll realize that they've pissed away the only

448
00:22:05,760 --> 00:22:10,479
Speaker 2: opportunity they really had to change direction, because had they

449
00:22:10,520 --> 00:22:12,359
Speaker 2: started this era and just been like, we're out of

450
00:22:12,440 --> 00:22:14,640
Speaker 2: hypergrowth ideas now. They probably wouldn't say it in those

451
00:22:14,680 --> 00:22:17,160
Speaker 2: exact words. They could have said, this is the era

452
00:22:17,280 --> 00:22:20,320
Speaker 2: of experimentation, this is finding our new business lines the

453
00:22:20,359 --> 00:22:23,680
Speaker 2: next Google search and announced it like some world's fair bullshit.

454
00:22:23,920 --> 00:22:26,480
Speaker 2: Oh no, Maybe the markets wouldn't have been quite as happy,

455
00:22:27,080 --> 00:22:29,520
Speaker 2: but they'd be happier that they're going to be at

456
00:22:29,520 --> 00:22:32,760
Speaker 2: the end of this era. And I think in the end,

457
00:22:33,240 --> 00:22:36,160
Speaker 2: it's frustrating to watch because logic and reason have gone

458
00:22:36,160 --> 00:22:40,040
Speaker 2: out the window. But gravity exists and this era will

459
00:22:40,080 --> 00:22:42,760
Speaker 2: come to an end. And next week we've got an

460
00:22:42,840 --> 00:22:45,879
Speaker 2: awesome interview something that's not about AI for once, with

461
00:22:45,960 --> 00:22:49,000
Speaker 2: two writers from The Wall Street Journal about the greasy, nasty,

462
00:22:49,080 --> 00:22:53,160
Speaker 2: freaky world of poly markets influencer marketing involving all sorts

463
00:22:53,200 --> 00:22:56,439
Speaker 2: of weird shit, like ads that include fake bets on

464
00:22:56,560 --> 00:23:00,480
Speaker 2: fake poly market websites, genuinely weird shit, And that'll be

465
00:23:00,480 --> 00:23:04,040
Speaker 2: coming up next Wednesday. I enjoy hearing from you. Please

466
00:23:04,280 --> 00:23:09,040
Speaker 2: email me itron a Easyatbetter Offline dot com. Jesus Christ,

467
00:23:09,040 --> 00:23:11,720
Speaker 2: I messed up my own email. It's been a long

468
00:23:11,800 --> 00:23:15,320
Speaker 2: few weeks, folks, but I'm loving hearing from you. Please

469
00:23:15,320 --> 00:23:18,320
Speaker 2: email me, please jump on the subreddit. I love to

470
00:23:18,359 --> 00:23:22,679
Speaker 2: hear from you. I love feedback. I love feedback. That's reasonable.

471
00:23:22,960 --> 00:23:25,159
Speaker 2: If you send me a typo correction, I'm only going

472
00:23:25,240 --> 00:23:25,800
Speaker 2: to get mad.

473
00:23:26,280 --> 00:23:26,920
Speaker 3: But I do.

474
00:23:26,960 --> 00:23:29,920
Speaker 2: Love hearing from you, even the typo correctors, even the pedants.

475
00:23:30,119 --> 00:23:31,320
Speaker 2: I love having the feedback.

476
00:23:31,320 --> 00:23:33,800
Speaker 3: And I love you all. I love you all. I

477
00:23:33,920 --> 00:23:34,919
Speaker 3: enjoy hearing from you.

478
00:23:35,040 --> 00:23:37,440
Speaker 2: I enjoy being here and speaking to you, and I'll

479
00:23:37,480 --> 00:23:38,480
Speaker 2: speak to you next week.

480
00:23:38,760 --> 00:23:39,639
Speaker 3: Zitron out