1
00:00:15,564 --> 00:00:28,084
Speaker 1: Pushkin. Welcome back to Risky Business, our show about making

2
00:00:28,124 --> 00:00:30,564
Speaker 1: better decisions. I'm Maria Kanikova.

3
00:00:30,484 --> 00:00:31,444
Speaker 2: And I'm Nate Silver.

4
00:00:32,804 --> 00:00:35,204
Speaker 1: Today on the show, we're going to be getting into

5
00:00:35,284 --> 00:00:38,284
Speaker 1: it on AI. There has been a lot of news

6
00:00:38,324 --> 00:00:40,844
Speaker 1: on the AI front in the last few weeks, some

7
00:00:41,004 --> 00:00:44,244
Speaker 1: of it coming from the Trump administration and some coming

8
00:00:44,284 --> 00:00:47,964
Speaker 1: from overseas with China and Deep Seek.

9
00:00:48,244 --> 00:00:51,284
Speaker 2: Maria thinks AI is overrated. I think AI is properly rated,

10
00:00:51,804 --> 00:00:52,884
Speaker 2: as you'll see.

11
00:00:54,204 --> 00:00:56,564
Speaker 1: And then we're going to get into a listener question

12
00:00:56,684 --> 00:01:01,404
Speaker 1: about poker and how to beat your local cash can.

13
00:01:15,684 --> 00:01:19,924
Speaker 2: Let's start with our friend artificial intelligence.

14
00:01:20,204 --> 00:01:23,204
Speaker 1: Yeah, so I think we're talking about a few different things, right,

15
00:01:23,564 --> 00:01:26,044
Speaker 1: So we have the Trump initiative on AI. So we

16
00:01:26,124 --> 00:01:31,804
Speaker 1: have a few things that he did, including things that

17
00:01:31,884 --> 00:01:36,404
Speaker 1: he took away, right the executive Order that took away

18
00:01:36,484 --> 00:01:39,044
Speaker 1: certain restrictions on AI that had been put in place

19
00:01:39,844 --> 00:01:43,524
Speaker 1: by Biden. But then also we have you know, this

20
00:01:43,684 --> 00:01:48,524
Speaker 1: big funding initiative into AA by the US government, Stargate,

21
00:01:48,644 --> 00:01:51,724
Speaker 1: and then we also have AI coming out of China

22
00:01:52,164 --> 00:01:55,044
Speaker 1: Deep Seek, which has freaked everyone the fuck out. I

23
00:01:55,044 --> 00:01:58,764
Speaker 1: think that's the scientific way of putting it and has

24
00:01:58,804 --> 00:02:03,644
Speaker 1: made markets, crash video stocks, a lot of other stocks Nasdaq.

25
00:02:03,924 --> 00:02:07,764
Speaker 1: You know, people have not been happy to see the

26
00:02:07,804 --> 00:02:10,844
Speaker 1: success of deep Seek, So there's a lot to talk

27
00:02:10,844 --> 00:02:11,404
Speaker 1: about today.

28
00:02:12,644 --> 00:02:14,964
Speaker 2: Where do you want to start, Maria.

29
00:02:15,084 --> 00:02:17,804
Speaker 1: Well, do we want to start with? I think if

30
00:02:17,884 --> 00:02:21,204
Speaker 1: it makes sense to start with deep seek because it actually,

31
00:02:21,844 --> 00:02:24,284
Speaker 1: you know, I think that the two are very related,

32
00:02:24,604 --> 00:02:27,244
Speaker 1: right because we have what the US government is and

33
00:02:27,444 --> 00:02:29,924
Speaker 1: is not doing, and one of the reasons that right

34
00:02:29,924 --> 00:02:33,364
Speaker 1: now things are freaking out is because of deep Seek.

35
00:02:33,404 --> 00:02:35,644
Speaker 1: So I think we can start with that, see what

36
00:02:35,684 --> 00:02:37,804
Speaker 1: the implications are, what the responses can be, and then

37
00:02:38,004 --> 00:02:40,484
Speaker 1: kind of see what the US has done so far

38
00:02:40,604 --> 00:02:42,724
Speaker 1: to see if it's in line with what we think

39
00:02:42,924 --> 00:02:44,404
Speaker 1: the correct strategy should be. Yeah.

40
00:02:44,404 --> 00:02:48,604
Speaker 2: Look, until about a week and a half ago, when

41
00:02:48,644 --> 00:02:52,204
Speaker 2: people started to notice Deep Seek their model are one.

42
00:02:52,244 --> 00:02:56,124
Speaker 2: In particular, the conventional wisdom was that, like America is

43
00:02:56,244 --> 00:02:59,204
Speaker 2: way ahead in the AI race. I should say with

44
00:02:59,284 --> 00:03:04,284
Speaker 2: respect to large language models, machine learning transformers, right, you know,

45
00:03:04,404 --> 00:03:07,444
Speaker 2: driverless cars is a different enterprise, and drones and things

46
00:03:07,484 --> 00:03:11,204
Speaker 2: like that. But in terms some LM's large language models

47
00:03:11,244 --> 00:03:16,004
Speaker 2: chatcypt like things. Then you know, us probably one, two, three,

48
00:03:16,084 --> 00:03:19,884
Speaker 2: four in the rankings and so this, Yeah, this has

49
00:03:20,004 --> 00:03:23,764
Speaker 2: interesting geopolitical implications. Yeah.

50
00:03:23,804 --> 00:03:26,564
Speaker 1: And the one of the reasons, just to step back,

51
00:03:27,164 --> 00:03:30,164
Speaker 1: that the US was assumed to be ahead was because

52
00:03:30,404 --> 00:03:33,004
Speaker 1: the United States had made it more difficult for foreign

53
00:03:33,044 --> 00:03:37,564
Speaker 1: governments to acquire like China, especially to acquire the chips

54
00:03:37,644 --> 00:03:43,044
Speaker 1: that are necessary to build these large AI models, the resources,

55
00:03:43,044 --> 00:03:45,964
Speaker 1: et cetera, et cetera. And so one of the disconcerting

56
00:03:46,004 --> 00:03:49,524
Speaker 1: things to the United States was that when Deepseak announced

57
00:03:49,524 --> 00:03:52,044
Speaker 1: as results, it also announced that it basically was able

58
00:03:52,084 --> 00:03:55,484
Speaker 1: to do this at one tenth of the cost and

59
00:03:55,604 --> 00:03:58,764
Speaker 1: resources that other models had used. So this was like

60
00:03:58,804 --> 00:04:01,804
Speaker 1: an oh shit, you know, even if we restrict access

61
00:04:01,804 --> 00:04:04,724
Speaker 1: to all of these other things, they're still able to

62
00:04:04,724 --> 00:04:07,644
Speaker 1: do this. Now, I will say, and other people have

63
00:04:07,724 --> 00:04:10,764
Speaker 1: pointed this out, I don't actually know how much we

64
00:04:10,804 --> 00:04:13,204
Speaker 1: can trust the numbers and figures, right, this is just

65
00:04:13,324 --> 00:04:16,044
Speaker 1: a claim. We don't know what the training materials were,

66
00:04:16,244 --> 00:04:18,644
Speaker 1: we don't know what the development costs actually were. We

67
00:04:18,724 --> 00:04:21,204
Speaker 1: just know what they claim that they were. So I

68
00:04:21,244 --> 00:04:23,164
Speaker 1: think that this is something that we should put an

69
00:04:23,204 --> 00:04:26,524
Speaker 1: asterisks next to because it is important to realize, right

70
00:04:26,964 --> 00:04:30,724
Speaker 1: that if you can't if you can't actually brace the

71
00:04:30,804 --> 00:04:34,444
Speaker 1: information and verify it, and you have to take it

72
00:04:35,004 --> 00:04:38,244
Speaker 1: on faith. That's never a good way to take information. Right,

73
00:04:38,244 --> 00:04:40,364
Speaker 1: That's one of the things we say over and over

74
00:04:40,604 --> 00:04:43,644
Speaker 1: on risky business when you make decisions, try not to

75
00:04:43,684 --> 00:04:46,324
Speaker 1: take things on faith. Try to verify, right, That's that's

76
00:04:46,404 --> 00:04:47,924
Speaker 1: much better. Yeah.

77
00:04:47,964 --> 00:04:53,524
Speaker 2: So this is basically built by a Chinese hedge fund, right,

78
00:04:53,564 --> 00:04:57,484
Speaker 2: which is well which is well capitalized. And you know,

79
00:04:57,564 --> 00:05:00,204
Speaker 2: there are a lot of smart computer engineers in China,

80
00:05:00,284 --> 00:05:02,164
Speaker 2: and I think they were all working on this project.

81
00:05:02,204 --> 00:05:05,084
Speaker 2: So so kind of the last step of training. I mean,

82
00:05:05,124 --> 00:05:08,364
Speaker 2: it's it's a little bit like you know Rosie who

83
00:05:08,484 --> 00:05:09,244
Speaker 2: Rosie Ruiz?

84
00:05:09,844 --> 00:05:11,324
Speaker 3: I do know who Rosie is is?

85
00:05:11,764 --> 00:05:14,084
Speaker 2: Yeah, if you've covered like fraudsters, right, And I don't

86
00:05:14,084 --> 00:05:15,764
Speaker 2: mean to say it's a fraud, but it's like, so

87
00:05:15,884 --> 00:05:18,044
Speaker 2: she runs what was the New York Marathon? Right, who?

88
00:05:18,164 --> 00:05:19,644
Speaker 2: Like it was a fraud.

89
00:05:19,684 --> 00:05:22,604
Speaker 1: By the way, So Rosie Ruiz actually figures in my

90
00:05:22,684 --> 00:05:29,124
Speaker 1: next book on cheating. So Rosie Ruiz one quote unquote

91
00:05:29,124 --> 00:05:33,604
Speaker 1: I'm putting this in quotes. The Boston Marathon women's time

92
00:05:33,884 --> 00:05:38,724
Speaker 1: with this incredible story was amazing, wonderful. You know, people

93
00:05:38,844 --> 00:05:40,684
Speaker 1: loved it, and then it turned out that she took

94
00:05:40,724 --> 00:05:44,164
Speaker 1: the subway for a huge portion of the race, and

95
00:05:44,804 --> 00:05:46,764
Speaker 1: but she almost got away with it, which is the

96
00:05:47,244 --> 00:05:50,404
Speaker 1: really fucked up thing. The reason she got caught was

97
00:05:50,404 --> 00:05:52,844
Speaker 1: because the subway car that she happened to get on

98
00:05:53,324 --> 00:05:56,044
Speaker 1: had a reporter who was a photographer who had been

99
00:05:56,084 --> 00:06:00,804
Speaker 1: covering the marat on and was like, wait, what's going on.

100
00:06:01,284 --> 00:06:04,604
Speaker 1: So it actually took them multiple days to figure out

101
00:06:04,684 --> 00:06:07,964
Speaker 1: that Rosy Ruiz did not actually win, did not actually

102
00:06:08,044 --> 00:06:10,564
Speaker 1: run the time that she shran. By the way, this

103
00:06:10,644 --> 00:06:14,164
Speaker 1: was back in nineteen eighty so right the technology was different.

104
00:06:14,644 --> 00:06:17,764
Speaker 1: People were not getting tracked as closely as they are

105
00:06:17,844 --> 00:06:18,244
Speaker 1: right now.

106
00:06:18,444 --> 00:06:19,884
Speaker 3: But spoiler alert for.

107
00:06:20,044 --> 00:06:24,684
Speaker 1: You know, or big fun thing that I get into

108
00:06:24,724 --> 00:06:27,164
Speaker 1: in my book, it's possible even today to pull a

109
00:06:27,284 --> 00:06:27,964
Speaker 1: Rosie release.

110
00:06:28,404 --> 00:06:30,724
Speaker 3: So that's for later. But Nate, why are we talking

111
00:06:30,724 --> 00:06:31,964
Speaker 3: about Rosie.

112
00:06:32,004 --> 00:06:35,324
Speaker 2: Because it's a little bit like if I don't say

113
00:06:35,324 --> 00:06:38,364
Speaker 2: there's any actual fraud, I really don't trust anything coming

114
00:06:38,404 --> 00:06:42,284
Speaker 2: out of mainland China. Yeah, but like but yeah, so

115
00:06:42,324 --> 00:06:44,484
Speaker 2: it's a little bit like if you run and around

116
00:06:44,524 --> 00:06:47,444
Speaker 2: the twenty fourth mile and then run two really good

117
00:06:47,484 --> 00:06:49,564
Speaker 2: closing miles, it's still not the same, and it's like

118
00:06:49,644 --> 00:06:52,284
Speaker 2: not like an accurate representation to say that just the

119
00:06:52,284 --> 00:06:54,444
Speaker 2: cost of this training run when you have all these

120
00:06:54,484 --> 00:06:57,084
Speaker 2: resources behind it, and it's kind of like the last step.

121
00:06:58,484 --> 00:07:01,884
Speaker 2: But clearly it's more efficient in terms of you know,

122
00:07:01,924 --> 00:07:06,964
Speaker 2: when you run a request, put in a query, how

123
00:07:07,004 --> 00:07:09,204
Speaker 2: many computer cycles is a burning through? I'm using very

124
00:07:09,244 --> 00:07:14,244
Speaker 2: nigh technical language here, right, You can actually host deep seek.

125
00:07:14,844 --> 00:07:16,404
Speaker 2: You keep going to call deep stack. There are a

126
00:07:16,404 --> 00:07:19,844
Speaker 2: lot of deep stack poker terms. You can host deep

127
00:07:19,884 --> 00:07:24,004
Speaker 2: seek at on a desktop. Right. If you want to

128
00:07:24,004 --> 00:07:26,604
Speaker 2: actually have it say things about Tanaman Square, for example,

129
00:07:26,684 --> 00:07:29,084
Speaker 2: then you need to run your native instance because the

130
00:07:29,124 --> 00:07:32,004
Speaker 2: official Chinese hosted web version doesn't like to you know,

131
00:07:32,044 --> 00:07:33,684
Speaker 2: doesn't like to talk about certain things.

132
00:07:33,964 --> 00:07:36,444
Speaker 1: I would say, Maria, you don't say that, you don't

133
00:07:36,484 --> 00:07:37,724
Speaker 1: say so.

134
00:07:38,084 --> 00:07:39,964
Speaker 2: Then there are a lot of debates though about what's

135
00:07:40,004 --> 00:07:42,844
Speaker 2: it mean if it turns out that I mean so.

136
00:07:42,884 --> 00:07:45,244
Speaker 2: One category to bat is like does the US lead

137
00:07:45,724 --> 00:07:49,084
Speaker 2: lose its lead versus China. That's one kind of whole bucket, right.

138
00:07:49,644 --> 00:07:51,884
Speaker 2: Another bucket is like, what's it mean if you now,

139
00:07:51,964 --> 00:07:55,044
Speaker 2: like run an AI lab on a desktop and they're

140
00:07:55,044 --> 00:07:56,764
Speaker 2: only going to get faster how do we regulate this?

141
00:07:56,804 --> 00:07:58,804
Speaker 2: And then there's a whole bunch of economic stuff about,

142
00:07:58,844 --> 00:08:02,484
Speaker 2: like you know, what's this mean for the price of

143
00:08:03,204 --> 00:08:04,124
Speaker 2: different assets?

144
00:08:04,204 --> 00:08:04,364
Speaker 1: Right?

145
00:08:04,444 --> 00:08:09,284
Speaker 2: So Nvidiah, for example, is the largest manufacturer of semiconductor,

146
00:08:09,644 --> 00:08:15,604
Speaker 2: is a Taiwanese company, you know, so if compute is cheaper,

147
00:08:15,724 --> 00:08:17,804
Speaker 2: is that good for them or bad for them? It's

148
00:08:17,804 --> 00:08:20,364
Speaker 2: not necessarily straightforward.

149
00:08:19,764 --> 00:08:24,004
Speaker 1: Right, Yeah, it's absolutely not straightforward. And also, you know,

150
00:08:24,124 --> 00:08:26,964
Speaker 1: one of the one of the other potential things that

151
00:08:27,044 --> 00:08:30,004
Speaker 1: happened with deep seek we don't know is a process

152
00:08:30,244 --> 00:08:33,084
Speaker 1: of training known as distillations. So that's a slightly more

153
00:08:33,404 --> 00:08:36,724
Speaker 1: technical term, but what it actually means is that you

154
00:08:36,884 --> 00:08:40,244
Speaker 1: are training your model on the outputs of other models, right,

155
00:08:40,444 --> 00:08:43,164
Speaker 1: so you can actually pass the benchmarks and be able

156
00:08:43,204 --> 00:08:46,004
Speaker 1: to trade up more quickly because you are using Okay, so.

157
00:08:45,964 --> 00:08:49,444
Speaker 2: Maybe now we are having more of a Rosie Ruize situation.

158
00:08:49,004 --> 00:08:51,004
Speaker 1: And if that happens, then then it is more of

159
00:08:51,004 --> 00:08:54,564
Speaker 1: a Rosie Rueze situation and just as a like as

160
00:08:54,564 --> 00:08:57,404
Speaker 1: a flag, that's not legal, right, Technically, you're not supposed

161
00:08:57,404 --> 00:08:59,924
Speaker 1: to do that. But if you do do it and

162
00:08:59,964 --> 00:09:01,284
Speaker 1: you're able to then cover.

163
00:09:01,204 --> 00:09:05,284
Speaker 2: These outputs, the respect for intellectual property varies by countries.

164
00:09:05,324 --> 00:09:07,484
Speaker 2: I don't want to be I don't want to bad

165
00:09:07,564 --> 00:09:10,604
Speaker 2: mouth the culture for not respecting electric property rights. But yeah,

166
00:09:10,604 --> 00:09:12,884
Speaker 2: if you're if you're you know, if you can just

167
00:09:12,924 --> 00:09:16,644
Speaker 2: copy off chat, GPT or in for what its weights are,

168
00:09:16,844 --> 00:09:19,124
Speaker 2: then that you know, I mean, that's that's a that's

169
00:09:19,124 --> 00:09:19,644
Speaker 2: an issue.

170
00:09:20,564 --> 00:09:24,164
Speaker 1: Yeah, it absolutely is. But let's let's flip that actually

171
00:09:24,204 --> 00:09:27,044
Speaker 1: around a little bit to talk about some of the

172
00:09:27,044 --> 00:09:32,124
Speaker 1: potential positives of this, which you know there we're obviously

173
00:09:32,124 --> 00:09:36,284
Speaker 1: seeing you know, potential uh red flags and negatives.

174
00:09:36,764 --> 00:09:38,244
Speaker 3: But what about I mean.

175
00:09:38,364 --> 00:09:41,524
Speaker 1: I actually think that it's not a bad thing that

176
00:09:41,604 --> 00:09:45,044
Speaker 1: deep seek is open source, right, so people can try

177
00:09:45,044 --> 00:09:47,604
Speaker 1: to figure out, Okay, what is it actually doing, how

178
00:09:47,684 --> 00:09:48,564
Speaker 1: is it actually doing?

179
00:09:48,604 --> 00:09:50,844
Speaker 3: It? Isn't that? Isn't that good?

180
00:09:51,124 --> 00:09:52,684
Speaker 1: And I know that, and I know that one of

181
00:09:52,724 --> 00:09:55,804
Speaker 1: the things that you know that people don't often like

182
00:09:55,924 --> 00:09:59,044
Speaker 1: is open source. But there's like open source can go

183
00:09:59,164 --> 00:10:02,524
Speaker 1: either way, right, open source bad if it's US and

184
00:10:02,604 --> 00:10:05,924
Speaker 1: China's getting its secrets, but you know, open source good

185
00:10:05,964 --> 00:10:09,044
Speaker 1: in other respects. And I think open source is one

186
00:10:09,084 --> 00:10:11,764
Speaker 1: of the few ways that we can actually peek inside

187
00:10:11,764 --> 00:10:14,244
Speaker 1: the black box. And I would be much more concerned

188
00:10:14,284 --> 00:10:16,324
Speaker 1: if it wasn't even open source, right, if we had

189
00:10:16,324 --> 00:10:17,844
Speaker 1: no idea how it was trained, if we didn't know

190
00:10:17,844 --> 00:10:19,444
Speaker 1: about the funding, if we didn't know it of this

191
00:10:19,764 --> 00:10:22,684
Speaker 1: and it was an open source right, if all of

192
00:10:22,684 --> 00:10:26,604
Speaker 1: those things were compelled, I'd be more worried. Since it

193
00:10:26,724 --> 00:10:29,204
Speaker 1: is open source, I actually think that could potentially be

194
00:10:29,284 --> 00:10:31,964
Speaker 1: a good thing in terms of knowledge sharing and trying

195
00:10:32,004 --> 00:10:34,964
Speaker 1: to figure out how do we make these processes more efficient?

196
00:10:35,004 --> 00:10:38,244
Speaker 1: And Meta has actually created a workforce that is studying

197
00:10:38,284 --> 00:10:41,084
Speaker 1: the deep sea processes and figuring out, okay, how can

198
00:10:41,124 --> 00:10:43,444
Speaker 1: we use these and do we want to change the

199
00:10:43,484 --> 00:10:47,364
Speaker 1: way that we're developing some of our llms, some of

200
00:10:47,404 --> 00:10:50,684
Speaker 1: our chat you know, models to mimic their processes so

201
00:10:50,724 --> 00:10:55,924
Speaker 1: that we become more efficient. And if this process actually

202
00:10:55,964 --> 00:10:59,324
Speaker 1: means that we use fewer resources, that's obviously a net

203
00:10:59,364 --> 00:11:03,004
Speaker 1: positive for the environment. However, if it means proliferation of

204
00:11:03,044 --> 00:11:05,684
Speaker 1: ais everywhere and it's cheaper than could actually be a

205
00:11:05,724 --> 00:11:08,324
Speaker 1: net negative. Which is all to say that this is

206
00:11:08,364 --> 00:11:11,684
Speaker 1: really complicated. This is not black and white, and when

207
00:11:11,684 --> 00:11:13,804
Speaker 1: you're making these sources of decisions, you have to assign

208
00:11:13,804 --> 00:11:16,084
Speaker 1: weights to all of these different outcomes, all of these

209
00:11:16,124 --> 00:11:19,124
Speaker 1: different probabilities, and frankly, I don't have the I don't

210
00:11:19,124 --> 00:11:21,444
Speaker 1: think anyone has the expertise to do that because it's

211
00:11:21,484 --> 00:11:23,884
Speaker 1: such a new world. I certainly don't have the expertise,

212
00:11:23,924 --> 00:11:26,844
Speaker 1: but I don't think anyone can see the future clearly

213
00:11:26,964 --> 00:11:28,724
Speaker 1: enough to figure out, you know, how do we wait this?

214
00:11:28,844 --> 00:11:31,764
Speaker 1: Because there's a lot of uncertainty around this.

215
00:11:32,084 --> 00:11:38,564
Speaker 2: So the hardcore AI concern people, the doomers, tend to

216
00:11:38,564 --> 00:11:42,444
Speaker 2: be anti open source. Open AI was founded as open AI,

217
00:11:42,684 --> 00:11:45,764
Speaker 2: but no longer kind of abides by that mission, and

218
00:11:45,804 --> 00:11:49,764
Speaker 2: I think so I think their view. Let me try

219
00:11:49,764 --> 00:11:53,644
Speaker 2: to give you the summary slash kind of Steelman version

220
00:11:53,684 --> 00:11:59,204
Speaker 2: of it. Right, these think, okay, this period where AI

221
00:11:59,404 --> 00:12:02,844
Speaker 2: is being birth, where artificial general intelligence ranging up to

222
00:12:02,964 --> 00:12:07,204
Speaker 2: artificial superintelligence. The first means it can do most things

223
00:12:07,244 --> 00:12:10,364
Speaker 2: at a human level or human life well plus right,

224
00:12:10,484 --> 00:12:13,404
Speaker 2: and super intelligence means it achieves breakthroughs that no human

225
00:12:13,524 --> 00:12:17,804
Speaker 2: can across the broad range of fields. Basically, right, it

226
00:12:17,884 --> 00:12:20,444
Speaker 2: thinks this process of giving birth to these models is

227
00:12:20,484 --> 00:12:24,044
Speaker 2: going to be very dangerous, and therefore you want to

228
00:12:24,084 --> 00:12:26,324
Speaker 2: have it in the hands of as few people as

229
00:12:26,324 --> 00:12:29,644
Speaker 2: possible and trusted people as possible. I think at one

230
00:12:29,644 --> 00:12:32,724
Speaker 2: point Sam Malton was trusted by this community. You know,

231
00:12:32,804 --> 00:12:36,324
Speaker 2: Anthropic is a bunch of people who left open AI

232
00:12:36,444 --> 00:12:38,364
Speaker 2: because they thought open a I was moving too fast.

233
00:12:38,444 --> 00:12:42,844
Speaker 2: Right and then and then Google Gemini. You know, Google

234
00:12:42,844 --> 00:12:47,084
Speaker 2: again had been fairly conservative about how it moved on AI.

235
00:12:47,284 --> 00:12:49,444
Speaker 2: Didn't want to you know, it's a huge enterprise, doesn't

236
00:12:49,444 --> 00:12:51,244
Speaker 2: want to risk that and other things and kind of

237
00:12:51,444 --> 00:12:56,164
Speaker 2: a different culture I would say at Google, then Facebook

238
00:12:56,204 --> 00:13:00,684
Speaker 2: or Meta, the reason why their models were open source

239
00:13:00,844 --> 00:13:04,324
Speaker 2: is because, like they weren't competitive. This is what people

240
00:13:04,364 --> 00:13:07,004
Speaker 2: would say in the AIXPEC community, right, is because they

241
00:13:07,044 --> 00:13:11,124
Speaker 2: weren't as good as these, you know, clearly as Anthropic

242
00:13:11,124 --> 00:13:13,484
Speaker 2: which has clawed, or open ai or clearly the two

243
00:13:13,524 --> 00:13:16,964
Speaker 2: leading ones. And then and then you know, Google Gemini

244
00:13:17,084 --> 00:13:20,404
Speaker 2: had some problem with like drawing like woke Nazis and

245
00:13:20,404 --> 00:13:23,364
Speaker 2: stuff like that, which I you know, I think it's

246
00:13:23,404 --> 00:13:25,804
Speaker 2: not as good as the other two by a fair bit.

247
00:13:25,844 --> 00:13:27,524
Speaker 2: But like, but that was that was considered in the

248
00:13:27,524 --> 00:13:31,484
Speaker 2: pecking order, right that Microsoft was excuse me, that Meta

249
00:13:31,844 --> 00:13:35,244
Speaker 2: too many m's was fourth, and therefore they open source

250
00:13:35,284 --> 00:13:37,124
Speaker 2: it to kind of say, okay, well we're gonna like

251
00:13:37,444 --> 00:13:41,684
Speaker 2: in some sense it's it's uh, not quite like sabotage.

252
00:13:42,124 --> 00:13:43,564
Speaker 2: You're like, well, fuck you, You're not going to get

253
00:13:43,564 --> 00:13:46,484
Speaker 2: these same returns to us, and will have applications people

254
00:13:46,484 --> 00:13:48,404
Speaker 2: who want to pay for it for free or whatever. Right,

255
00:13:49,364 --> 00:13:52,524
Speaker 2: So the fact that it's open source is interesting. I mean,

256
00:13:52,524 --> 00:13:58,324
Speaker 2: you know, China clearly is willing to do things, uh

257
00:13:58,724 --> 00:14:05,004
Speaker 2: to undermine the American even if so. I mean, the

258
00:14:05,004 --> 00:14:07,764
Speaker 2: other big one in the news is of course TikTok,

259
00:14:07,764 --> 00:14:10,444
Speaker 2: which is owned by bike Dance. Right, the US Congress

260
00:14:10,444 --> 00:14:13,004
Speaker 2: passes a law upheld by the courts. It says you

261
00:14:13,084 --> 00:14:17,564
Speaker 2: have to sell to an American company or at least

262
00:14:17,644 --> 00:14:19,404
Speaker 2: a country not listed on like the in there's like

263
00:14:19,404 --> 00:14:22,204
Speaker 2: literally like an enemy's list of countries, and China is

264
00:14:22,284 --> 00:14:27,644
Speaker 2: on it, right, And bike Dan says, we have this

265
00:14:27,684 --> 00:14:29,604
Speaker 2: really valuable enterprise, but we'll just turn it off, right,

266
00:14:29,684 --> 00:14:32,284
Speaker 2: If you make us sell it, then we'll just turn off. Well, clearly,

267
00:14:32,324 --> 00:14:34,284
Speaker 2: I mean obviously it would be a forced sale and

268
00:14:34,324 --> 00:14:36,644
Speaker 2: you wouldn't have quite the market clearing price, but like

269
00:14:36,924 --> 00:14:39,124
Speaker 2: you know, the value is a lot more than zero,

270
00:14:39,284 --> 00:14:41,004
Speaker 2: and they're willing to So you know, there's a little

271
00:14:41,004 --> 00:14:43,524
Speaker 2: bit of suspicion that like this is just meant to

272
00:14:43,724 --> 00:14:47,964
Speaker 2: like undermine America's lead in AI and not make a

273
00:14:47,964 --> 00:14:50,124
Speaker 2: whole lot of profit for China. That's kind of like

274
00:14:50,164 --> 00:14:53,204
Speaker 2: addition by subtraction, right, you know, And people say, hey,

275
00:14:53,204 --> 00:14:56,364
Speaker 2: these people are idealistic. It's not the same capitalist system there.

276
00:14:56,404 --> 00:15:00,564
Speaker 2: And even in the US sometimes the founder aren't that greedy.

277
00:15:00,564 --> 00:15:02,204
Speaker 2: They just want to make a really cool product. So

278
00:15:02,644 --> 00:15:04,444
Speaker 2: there might be some of that too, But it's in

279
00:15:04,484 --> 00:15:07,724
Speaker 2: line with previous Chinese strategy, I suppose.

280
00:15:08,484 --> 00:15:11,404
Speaker 1: Yeah, it's actually interesting that you that you mentioned TikTok

281
00:15:11,444 --> 00:15:14,804
Speaker 1: because this is another kind of strategic element of this

282
00:15:15,084 --> 00:15:19,964
Speaker 1: that you know, we're we being the US has moved

283
00:15:20,004 --> 00:15:21,204
Speaker 1: to band TikTok right now?

284
00:15:23,044 --> 00:15:26,764
Speaker 2: Sorry, what USA?

285
00:15:27,204 --> 00:15:33,644
Speaker 3: You are mooting for the US? Nice and nice? All right?

286
00:15:33,764 --> 00:15:38,804
Speaker 1: Tem U s I is uh was moving to band TikTok.

287
00:15:38,844 --> 00:15:40,404
Speaker 1: We have no idea what's going to happen with that now.

288
00:15:40,604 --> 00:15:43,444
Speaker 1: But in some ways, you know, and deep seek is

289
00:15:43,564 --> 00:15:46,724
Speaker 1: just like la la la la la. Right, There's there's

290
00:15:46,764 --> 00:15:49,844
Speaker 1: no movement in that direction, and instead Trump is talking

291
00:15:49,884 --> 00:15:53,644
Speaker 1: about tariffs and other things and and trying to kind

292
00:15:53,644 --> 00:15:55,364
Speaker 1: of get at it that way. And I think that

293
00:15:55,444 --> 00:15:58,084
Speaker 1: it's a very interesting dichotomy where like, if you're worried

294
00:15:58,084 --> 00:16:01,044
Speaker 1: about TikTok, like, shouldn't you be worried about the strategic

295
00:16:01,404 --> 00:16:05,044
Speaker 1: and security risks of you know, of a company that

296
00:16:06,084 --> 00:16:08,844
Speaker 1: runs you know that all of the idol Ai models

297
00:16:08,884 --> 00:16:12,044
Speaker 1: are run into right, Like that's Chinese owned, Chinese developed,

298
00:16:12,324 --> 00:16:15,524
Speaker 1: Like do you if you're going to be consistent like that?

299
00:16:15,524 --> 00:16:17,724
Speaker 1: That seems to be a much greater risk than TikTok

300
00:16:17,804 --> 00:16:18,764
Speaker 1: to be perfectly honest.

301
00:16:21,644 --> 00:16:23,524
Speaker 2: And we'll be right back after this break.

302
00:16:33,204 --> 00:16:36,844
Speaker 1: So what do we, like, what can we take from

303
00:16:36,884 --> 00:16:39,644
Speaker 1: this and from what's happened in the last week. How

304
00:16:39,684 --> 00:16:45,364
Speaker 1: does that mesh with the types of endeavors that that

305
00:16:46,284 --> 00:16:50,004
Speaker 1: Trump and team have already put forward. Now, one of

306
00:16:50,044 --> 00:16:52,404
Speaker 1: the things, so I started off by saying that they've

307
00:16:52,444 --> 00:16:56,844
Speaker 1: rescented the executive order. The executive order that Biden had

308
00:16:57,284 --> 00:16:59,884
Speaker 1: on Ai did have to do with open source, right,

309
00:17:00,164 --> 00:17:03,124
Speaker 1: and it did have to say it was actually very

310
00:17:03,124 --> 00:17:06,724
Speaker 1: skeptical of open source as well, because it wanted, you know,

311
00:17:07,444 --> 00:17:10,564
Speaker 1: full reporting, knowing everything that was going on, but let's

312
00:17:10,644 --> 00:17:15,444
Speaker 1: keep that information from foreign governments. Now that's out right,

313
00:17:15,844 --> 00:17:19,204
Speaker 1: So that's been rescinded. So what's and that's one of

314
00:17:19,204 --> 00:17:21,524
Speaker 1: the main issues that we're seeing with deep seek. So

315
00:17:21,844 --> 00:17:23,404
Speaker 1: so what do we think about that? What do we

316
00:17:23,404 --> 00:17:27,404
Speaker 1: think about the government approach? Is it misguided or is

317
00:17:27,444 --> 00:17:29,684
Speaker 1: it on the right track? And if we want to

318
00:17:29,844 --> 00:17:33,844
Speaker 1: I think all of us, no matter if you're AAAI ORNAI,

319
00:17:34,204 --> 00:17:36,564
Speaker 1: I think everyone wants to minimize pe doom, like I

320
00:17:36,564 --> 00:17:39,724
Speaker 1: would hope that no one wants the world to be destroyed.

321
00:17:40,804 --> 00:17:44,244
Speaker 2: So I don't think anybody thought that like this Biden

322
00:17:44,324 --> 00:17:48,684
Speaker 2: executive order is going to stop pe doom by itself.

323
00:17:48,804 --> 00:17:51,764
Speaker 2: The California law that was vetoed by Gavin Newsom probably

324
00:17:52,124 --> 00:17:54,324
Speaker 2: was considered a bigger deal. It was a state law,

325
00:17:54,364 --> 00:17:56,364
Speaker 2: but they're all based. If they want to operate in California,

326
00:17:56,364 --> 00:17:58,484
Speaker 2: then they were subject to it, right, Like that might

327
00:17:58,524 --> 00:18:00,684
Speaker 2: have been a bigger deal. But like the general pattern

328
00:18:00,724 --> 00:18:03,004
Speaker 2: here with the California law failing, with this thing being rescinded,

329
00:18:05,044 --> 00:18:10,724
Speaker 2: with Sam Altman being lesson less shall we say, concerned

330
00:18:11,564 --> 00:18:14,284
Speaker 2: about safety, right, we're going to get this AI raise

331
00:18:14,404 --> 00:18:18,444
Speaker 2: and this idea that you could stop it by having

332
00:18:19,604 --> 00:18:21,884
Speaker 2: only you know, three operators until you reach some point

333
00:18:21,884 --> 00:18:26,964
Speaker 2: where safety was achieved is not going to happen clearly, right,

334
00:18:28,244 --> 00:18:30,924
Speaker 2: you know, Like, if you read any other technology, you

335
00:18:30,924 --> 00:18:33,084
Speaker 2: would say, because one concern I have about AI or

336
00:18:33,204 --> 00:18:36,564
Speaker 2: about this at the newsletter this week. You know, when

337
00:18:36,564 --> 00:18:39,004
Speaker 2: you have these very big, powerful companies that have this

338
00:18:39,164 --> 00:18:42,604
Speaker 2: lead in computing power and engineering talent, right, usually that's

339
00:18:42,604 --> 00:18:44,244
Speaker 2: not how it works, Right, you have the next big thing,

340
00:18:44,284 --> 00:18:46,764
Speaker 2: and the next big thing is created by new companies

341
00:18:46,804 --> 00:18:50,364
Speaker 2: because the old companies are nostodgy and bloated and it's

342
00:18:50,364 --> 00:18:52,844
Speaker 2: not their mission in the first place, and they're not cool,

343
00:18:52,884 --> 00:18:55,764
Speaker 2: so don't attract young talent. Right, So to some accept

344
00:18:55,764 --> 00:18:59,204
Speaker 2: the fact that, like, you know, it can be disrupted

345
00:18:59,284 --> 00:19:04,284
Speaker 2: might lesson the worry about hegemonic concerns over AI and

346
00:19:04,324 --> 00:19:07,044
Speaker 2: the kind of paternalistic slash people get very rich off

347
00:19:07,044 --> 00:19:09,684
Speaker 2: it concerns about AI. But like, but yeah, I mean, look,

348
00:19:10,324 --> 00:19:15,004
Speaker 2: I think we're a long way from achieving artificial superintelligence,

349
00:19:15,044 --> 00:19:17,844
Speaker 2: which is the super human capabilities, right, But there are

350
00:19:17,884 --> 00:19:21,084
Speaker 2: ways that AIS can be dangerous far short of this.

351
00:19:21,364 --> 00:19:21,564
Speaker 1: Right.

352
00:19:21,684 --> 00:19:25,684
Speaker 2: Take ching you how to like mix chemical compounds or

353
00:19:25,804 --> 00:19:28,684
Speaker 2: build a pipe bomb or things like that, right, or

354
00:19:28,844 --> 00:19:31,724
Speaker 2: can aid in a bit like sue with cidal thoughts

355
00:19:31,764 --> 00:19:33,524
Speaker 2: and like in those use cases are going to be.

356
00:19:33,524 --> 00:19:35,404
Speaker 3: Like they're already happening.

357
00:19:35,604 --> 00:19:39,404
Speaker 1: They're already happening. We know that the guy who blew

358
00:19:39,484 --> 00:19:41,364
Speaker 1: up the cyber truck in front of the Trump Hotel

359
00:19:42,444 --> 00:19:45,364
Speaker 1: used chat GPT to figure out how to do it,

360
00:19:45,604 --> 00:19:47,924
Speaker 1: which was actually probably one of the reasons it wasn't

361
00:19:47,964 --> 00:19:51,724
Speaker 1: more destructive because the instructions were not very good. So

362
00:19:51,764 --> 00:19:53,844
Speaker 1: I think I think we should be we should be

363
00:19:53,844 --> 00:19:58,404
Speaker 1: grateful for the limitations of AI models for now. But yeah, no,

364
00:19:58,524 --> 00:20:01,684
Speaker 1: there are I think I think that they're to me.

365
00:20:02,684 --> 00:20:04,884
Speaker 1: That's actually one of the one of the more interesting

366
00:20:04,924 --> 00:20:07,404
Speaker 1: points is that you know, we're worried about p doo,

367
00:20:07,604 --> 00:20:10,524
Speaker 1: about you know, super intelligence, all these things, but I

368
00:20:10,564 --> 00:20:12,964
Speaker 1: think that in the shorter term and potentially in the

369
00:20:13,004 --> 00:20:15,364
Speaker 1: longer term, we need to be more worried about stupidity,

370
00:20:15,844 --> 00:20:20,044
Speaker 1: right about the fact that uh that, like, they aren't

371
00:20:20,084 --> 00:20:25,524
Speaker 1: super intelligent, right, and and people can misuse them, and

372
00:20:25,564 --> 00:20:29,564
Speaker 1: they can give flawed outputs. And as they become more

373
00:20:29,644 --> 00:20:34,604
Speaker 1: and more dominant mainstream used in searches, you know, used

374
00:20:34,644 --> 00:20:38,044
Speaker 1: in used in day to day stuff, but also used

375
00:20:38,084 --> 00:20:41,644
Speaker 1: higher up for people who you know, want something to

376
00:20:41,644 --> 00:20:44,764
Speaker 1: summarize research for them, et cetera, et cetera, that the

377
00:20:45,364 --> 00:20:48,804
Speaker 1: problem is going to be kind of much more mundane,

378
00:20:49,044 --> 00:20:51,124
Speaker 1: right that it gives you that, it gives you bad information,

379
00:20:51,164 --> 00:20:55,844
Speaker 1: it gives you bad instructions, It allides over something. It

380
00:20:55,844 --> 00:20:59,484
Speaker 1: doesn't It doesn't quite synthesize something correctly. I think that

381
00:20:59,484 --> 00:21:04,004
Speaker 1: that is actually the more pressing problem and is not

382
00:21:04,164 --> 00:21:06,444
Speaker 1: P doom boom We're going to blow up, but is

383
00:21:06,484 --> 00:21:10,644
Speaker 1: a kind of a smaller P dooms. It's already lowercase,

384
00:21:10,684 --> 00:21:15,004
Speaker 1: but like subscript fe doom on a day to day basis,

385
00:21:16,004 --> 00:21:17,244
Speaker 1: depending on who uses.

386
00:21:17,044 --> 00:21:19,764
Speaker 2: It n for what. I've kind of flipped on this

387
00:21:19,804 --> 00:21:22,564
Speaker 2: a little bit where I I think the hallucinations are

388
00:21:22,604 --> 00:21:25,204
Speaker 2: an overrated problem. I think what the models are doing

389
00:21:25,284 --> 00:21:30,764
Speaker 2: is is very impressive. They have fewer hallucinations than before.

390
00:21:32,324 --> 00:21:34,764
Speaker 2: And you know, I use them a lot just for

391
00:21:34,884 --> 00:21:37,964
Speaker 2: like research and problem solving a little bit of programming

392
00:21:38,004 --> 00:21:40,364
Speaker 2: and things like that. And I think, you know, the

393
00:21:40,404 --> 00:21:42,124
Speaker 2: one I've used the most is O one, which is

394
00:21:42,124 --> 00:21:47,724
Speaker 2: the latest public build of chatgypt or open AI. It

395
00:21:47,804 --> 00:21:51,164
Speaker 2: just can do higher level shit pretty well. You know,

396
00:21:51,964 --> 00:21:55,684
Speaker 2: it also can catch itself in midstream. They put some

397
00:21:55,764 --> 00:21:59,444
Speaker 2: routine in where it like well, typically a large language model.

398
00:21:59,684 --> 00:22:01,404
Speaker 2: The reason why it seems like it's printing one word

399
00:22:01,404 --> 00:22:03,244
Speaker 2: at a time. Is because like that's actually kind of

400
00:22:03,324 --> 00:22:08,644
Speaker 2: like how it works, right. It literally it literally kind

401
00:22:08,644 --> 00:22:10,844
Speaker 2: of goes to quench, It compresses its whole tex string,

402
00:22:10,844 --> 00:22:13,484
Speaker 2: puts it through transformer. Right. But then you know then

403
00:22:13,564 --> 00:22:15,844
Speaker 2: kind of it is recurs on how it puts it out.

404
00:22:16,004 --> 00:22:18,004
Speaker 2: It doesn't think of it all at once, right. But

405
00:22:18,084 --> 00:22:20,844
Speaker 2: now they've trained it to actually go back and check

406
00:22:20,884 --> 00:22:26,604
Speaker 2: its output to check for hallucinations, and it catches them

407
00:22:26,884 --> 00:22:29,924
Speaker 2: a lot of the time, but not a lot of

408
00:22:29,964 --> 00:22:30,284
Speaker 2: the time.

409
00:22:30,364 --> 00:22:30,524
Speaker 3: Right.

410
00:22:30,564 --> 00:22:34,084
Speaker 1: So I actually just did some test runs prior to

411
00:22:34,364 --> 00:22:36,684
Speaker 1: taping this so that I could see what was going

412
00:22:36,724 --> 00:22:40,084
Speaker 1: on right now. And when it's in my area of expertise,

413
00:22:40,284 --> 00:22:44,324
Speaker 1: I catch a lot of inaccuracies, not just hallucinations, but

414
00:22:44,404 --> 00:22:48,124
Speaker 1: things that are almost right but kind of misunderstood the point, right,

415
00:22:48,164 --> 00:22:51,644
Speaker 1: which which is actually which could be even more problematic.

416
00:22:51,724 --> 00:22:54,044
Speaker 1: So I had to do some test runs in psychology

417
00:22:54,084 --> 00:22:55,124
Speaker 1: where first of all.

418
00:22:55,004 --> 00:22:55,804
Speaker 3: It did hallucinate.

419
00:22:55,844 --> 00:22:59,124
Speaker 1: It still is making up studies, making up data that

420
00:22:59,164 --> 00:23:01,284
Speaker 1: does not actually exist. When I try to, I'm like, oh,

421
00:23:01,284 --> 00:23:03,484
Speaker 1: this is really cool. I want to look it up. No,

422
00:23:03,604 --> 00:23:07,284
Speaker 1: it doesn't exist. But it also miss like it actually

423
00:23:07,284 --> 00:23:10,564
Speaker 1: misinterprets findings and doesn't all always get it right. This

424
00:23:10,724 --> 00:23:12,884
Speaker 1: is my expertise, right, I have a PhD at it,

425
00:23:13,004 --> 00:23:15,324
Speaker 1: so I can figure this out and be like, you

426
00:23:15,324 --> 00:23:16,244
Speaker 1: know what this is.

427
00:23:16,404 --> 00:23:17,964
Speaker 3: It's pretty good, but not.

428
00:23:18,084 --> 00:23:21,244
Speaker 1: Actually like you don't want to rely on this and

429
00:23:21,284 --> 00:23:25,404
Speaker 1: this is just outright wrong. But because it's overall it

430
00:23:25,444 --> 00:23:27,884
Speaker 1: seems pretty good, I don't catch it when it's not

431
00:23:27,964 --> 00:23:30,084
Speaker 1: my area of expertise, and I just assume that it's

432
00:23:30,124 --> 00:23:32,684
Speaker 1: pretty good and that pretty is doing a lot of

433
00:23:32,684 --> 00:23:33,364
Speaker 1: heavy lifting.

434
00:23:33,884 --> 00:23:35,924
Speaker 2: What about if you read a New York Times or

435
00:23:35,964 --> 00:23:40,284
Speaker 2: Washington Post article on poker or sports betting, right or

436
00:23:40,404 --> 00:23:42,164
Speaker 2: some feel bad, but.

437
00:23:42,124 --> 00:23:45,124
Speaker 1: It doesn't actually hallucinate. It's not going to really it's

438
00:23:45,124 --> 00:23:47,404
Speaker 1: not going to tell me about findings that don't exist

439
00:23:47,524 --> 00:23:48,284
Speaker 1: to prove its point.

440
00:23:48,684 --> 00:23:50,324
Speaker 2: It can tell a lot of white lies though, and

441
00:23:50,404 --> 00:23:52,484
Speaker 2: misrepresent and be fundamentally dishonest.

442
00:23:53,044 --> 00:23:56,084
Speaker 1: Right, Well, sure you have. You have a problem of

443
00:23:56,404 --> 00:23:59,644
Speaker 1: reporting bias always, But there's a problem when you think

444
00:23:59,684 --> 00:24:02,684
Speaker 1: something is factual information and it's not right. When I'm

445
00:24:02,684 --> 00:24:04,444
Speaker 1: reading an op ed, I know it's an op ed

446
00:24:04,484 --> 00:24:07,404
Speaker 1: when I'm reading an article, but when i'm when I

447
00:24:07,444 --> 00:24:09,164
Speaker 1: want this for facts.

448
00:24:09,444 --> 00:24:11,524
Speaker 2: But I want I want to know, not the use case.

449
00:24:11,564 --> 00:24:13,484
Speaker 2: Though you can't put everything in the microwave oven.

450
00:24:13,764 --> 00:24:16,084
Speaker 1: Of course I want to put one of the things

451
00:24:16,724 --> 00:24:18,844
Speaker 1: if I want to learn about a field. So let

452
00:24:18,884 --> 00:24:21,044
Speaker 1: me give you another example. I didn't just use psychology.

453
00:24:21,364 --> 00:24:25,564
Speaker 1: I asked about I wanted to know about a specific

454
00:24:25,604 --> 00:24:27,164
Speaker 1: type of company. I'm not going to give you kind

455
00:24:27,164 --> 00:24:30,964
Speaker 1: of the search that was making a specific type of

456
00:24:30,964 --> 00:24:33,964
Speaker 1: crypto investment. So I asked, like, what's a list of

457
00:24:33,964 --> 00:24:35,604
Speaker 1: companies that's done it? And gave me a list. I

458
00:24:35,604 --> 00:24:38,084
Speaker 1: was like, great, now will you tell me what the

459
00:24:38,084 --> 00:24:40,924
Speaker 1: specific investments are? And it's like, oh, sorry, we don't

460
00:24:40,924 --> 00:24:42,884
Speaker 1: actually know, Like we don't know that these companies have

461
00:24:42,924 --> 00:24:45,884
Speaker 1: made any investments. We just gave you a list, And

462
00:24:46,284 --> 00:24:47,244
Speaker 1: I was like, okay.

463
00:24:47,124 --> 00:24:52,484
Speaker 2: You're using it wrong. You there are a lot of

464
00:24:52,484 --> 00:25:00,364
Speaker 2: situations in life where precision isn't that important, right, you know?

465
00:25:00,564 --> 00:25:04,284
Speaker 2: I mean, for example, I was as you made no listeners.

466
00:25:04,324 --> 00:25:06,604
Speaker 2: I was in Korea and Japan recently, and like something

467
00:25:06,604 --> 00:25:09,364
Speaker 2: I've been lazy about. I never taught myself how to

468
00:25:09,564 --> 00:25:14,164
Speaker 2: distinguished Japanese, Chinese and Korean characters, and so I like

469
00:25:14,244 --> 00:25:17,924
Speaker 2: told to bet I'm on the flight to Tokyo. I'm like, hey,

470
00:25:17,924 --> 00:25:20,804
Speaker 2: give me a little very quick summary and give me

471
00:25:20,804 --> 00:25:22,084
Speaker 2: a pop quiz and like you learn it in like

472
00:25:22,084 --> 00:25:24,044
Speaker 2: ten or fifteen minutes, right, And they're like, you know,

473
00:25:24,044 --> 00:25:27,124
Speaker 2: I don't have to like distinguish those characters one hundred accuracy.

474
00:25:27,164 --> 00:25:30,444
Speaker 2: But it's like basically pretty good, and like I can

475
00:25:30,484 --> 00:25:34,844
Speaker 2: tailor exactly how how that resources geared toward me.

476
00:25:36,124 --> 00:25:38,204
Speaker 1: I'm not saying this is useless. I'm just saying that

477
00:25:38,244 --> 00:25:41,004
Speaker 1: it's right now. It is being used for the cases

478
00:25:41,044 --> 00:25:42,844
Speaker 1: that I've told you about because it's at the top

479
00:25:42,884 --> 00:25:46,964
Speaker 1: of your such when you do a Google search, like

480
00:25:47,644 --> 00:25:49,604
Speaker 1: that's the first thing that comes up as the.

481
00:25:49,724 --> 00:25:54,524
Speaker 2: I think it's very bad branding decision by by Google, right.

482
00:25:54,564 --> 00:25:56,644
Speaker 2: I do think it's a bad because like and Google's

483
00:25:56,764 --> 00:26:00,044
Speaker 2: I'm sorry, it's not as good as opening Eye show anthropic.

484
00:26:00,084 --> 00:26:03,964
Speaker 2: It's not it's not sorry Google, and like it undermines

485
00:26:04,044 --> 00:26:06,204
Speaker 2: Google's kind of lead in search. And I think Google

486
00:26:06,204 --> 00:26:08,244
Speaker 2: has handled this a lot of things in this space

487
00:26:08,364 --> 00:26:10,764
Speaker 2: very badly. Although I mean, you know, it was Google

488
00:26:10,804 --> 00:26:13,164
Speaker 2: engineers who came up with transformer paper and they and

489
00:26:13,204 --> 00:26:17,244
Speaker 2: they and they, like you know, still hire lots of

490
00:26:17,244 --> 00:26:18,724
Speaker 2: great pale up but they've kind of become like this

491
00:26:18,804 --> 00:26:23,124
Speaker 2: feeder system to like the hip or cooler AI companies,

492
00:26:23,444 --> 00:26:26,124
Speaker 2: I think, But like anyway I think people are I

493
00:26:26,164 --> 00:26:30,044
Speaker 2: think people are way too hipster about this, Like this

494
00:26:30,084 --> 00:26:33,444
Speaker 2: is the most quickly adopted technology in the history of

495
00:26:33,444 --> 00:26:34,844
Speaker 2: the world by some mess.

496
00:26:34,684 --> 00:26:37,644
Speaker 1: Absolutely, and I'm not like I actually like, I think

497
00:26:37,684 --> 00:26:40,284
Speaker 1: that it has a lot of potential. I want it

498
00:26:40,324 --> 00:26:43,444
Speaker 1: to do better, right, Like I like I want them

499
00:26:43,444 --> 00:26:44,284
Speaker 1: to fix this shit.

500
00:26:44,764 --> 00:26:44,964
Speaker 3: Right.

501
00:26:45,044 --> 00:26:47,404
Speaker 2: Look, we are poker players, Maria. We should be used

502
00:26:47,444 --> 00:26:51,724
Speaker 2: to being able to accept information as shifting your prior

503
00:26:51,804 --> 00:26:54,684
Speaker 2: or shifting your view, but like not being definitive, right,

504
00:26:54,724 --> 00:26:57,124
Speaker 2: And that's a journalist. We're both journalists too, Like you know,

505
00:26:57,244 --> 00:26:59,004
Speaker 2: when a source tells you something, you've vet.

506
00:26:59,004 --> 00:27:03,004
Speaker 1: It absolutely, But it actually adds more work for me

507
00:27:03,444 --> 00:27:05,444
Speaker 1: because I have to go through it and try to

508
00:27:05,484 --> 00:27:07,644
Speaker 1: figure out what can I rely on, what can't I

509
00:27:07,724 --> 00:27:11,844
Speaker 1: rely on. Now, one listener who who has shared his

510
00:27:11,924 --> 00:27:15,524
Speaker 1: experience on UH on social media as well, and I

511
00:27:15,564 --> 00:27:18,884
Speaker 1: know you wrote you kind of referenced it in your

512
00:27:18,884 --> 00:27:22,524
Speaker 1: newsletter this week, Kevin Ruse had emailed me about poker

513
00:27:22,564 --> 00:27:26,364
Speaker 1: training and said that he was using chat GPT and

514
00:27:26,444 --> 00:27:28,764
Speaker 1: AIS to help him with poker. And I said, don't

515
00:27:28,764 --> 00:27:31,324
Speaker 1: do that because it's going to tell you the wrong thing.

516
00:27:32,284 --> 00:27:36,364
Speaker 1: And he did it. He still did it and he said, oh,

517
00:27:36,364 --> 00:27:38,484
Speaker 1: I want to tournament. It was it was good, it

518
00:27:38,564 --> 00:27:41,524
Speaker 1: was helpful. So I actually had chat GPT do some

519
00:27:41,604 --> 00:27:42,564
Speaker 1: poker training for me.

520
00:27:42,884 --> 00:27:43,964
Speaker 3: It's not good.

521
00:27:44,404 --> 00:27:46,844
Speaker 1: It gives you incorrect advice if you don't know, if

522
00:27:46,844 --> 00:27:49,164
Speaker 1: you're a novice, and if you're using this, you might

523
00:27:49,204 --> 00:27:52,284
Speaker 1: get lucky, right, and it were all works out, but

524
00:27:52,404 --> 00:27:55,644
Speaker 1: let's go to poker, like, try to use chat GPT

525
00:27:55,684 --> 00:27:58,404
Speaker 1: to teach you poker strategy. It is not going to

526
00:27:58,404 --> 00:28:01,604
Speaker 1: teach you strategy. Did you did you and the better

527
00:28:01,644 --> 00:28:01,924
Speaker 1: you are?

528
00:28:03,804 --> 00:28:04,004
Speaker 3: Yeah?

529
00:28:04,444 --> 00:28:06,564
Speaker 2: Yeah, I just I did the same and I thought

530
00:28:06,564 --> 00:28:09,724
Speaker 2: it was pretty good. It was it misses things that

531
00:28:09,804 --> 00:28:12,124
Speaker 2: like so I actually did this last night. I mean

532
00:28:12,164 --> 00:28:14,444
Speaker 2: it gets you know, I so what do you? What

533
00:28:14,444 --> 00:28:14,844
Speaker 2: do you actually?

534
00:28:15,244 --> 00:28:19,284
Speaker 3: But here's the thing, though, are you yellous? You're you're able.

535
00:28:19,044 --> 00:28:22,124
Speaker 1: To distinguish what it's missing and what is getting pretty well.

536
00:28:22,364 --> 00:28:25,724
Speaker 1: If you're using this as the tool to train yourself

537
00:28:26,244 --> 00:28:30,684
Speaker 1: and you don't have any background knowledge, that is the problem, right,

538
00:28:30,724 --> 00:28:33,244
Speaker 1: you need it to be you need it to teach

539
00:28:33,244 --> 00:28:36,644
Speaker 1: you correctly. It's much more difficult as someone who started

540
00:28:36,684 --> 00:28:40,804
Speaker 1: poker from zero as an adult. Right, let me tell you, Like,

541
00:28:40,924 --> 00:28:42,804
Speaker 1: one of the most important things I learned was that

542
00:28:42,844 --> 00:28:46,004
Speaker 1: it's much easier to teach someone from zero because I

543
00:28:46,004 --> 00:28:47,204
Speaker 1: didn't have any bad habits.

544
00:28:47,284 --> 00:28:47,444
Speaker 3: Right.

545
00:28:47,484 --> 00:28:51,004
Speaker 1: If I had instead learned from chat GPT and those

546
00:28:51,084 --> 00:28:53,844
Speaker 1: were kind of the habits and the thought processes that

547
00:28:53,884 --> 00:28:56,564
Speaker 1: I acquired, and some of them were just wrong or

548
00:28:56,604 --> 00:28:59,484
Speaker 1: didn't teach me how to think correctly through things, I'd

549
00:28:59,484 --> 00:29:01,084
Speaker 1: be a really shitty poker player.

550
00:29:01,684 --> 00:29:03,604
Speaker 2: Yeah, so let me give you some example of how

551
00:29:03,644 --> 00:29:05,444
Speaker 2: I use chat GPT.

552
00:29:05,804 --> 00:29:05,964
Speaker 1: Right.

553
00:29:07,244 --> 00:29:11,924
Speaker 2: You know, one is kind of as a research assistant,

554
00:29:12,004 --> 00:29:14,324
Speaker 2: but like once you already know something about a topic, right,

555
00:29:14,324 --> 00:29:15,964
Speaker 2: Like I'm not getting a first brief, but I'm like

556
00:29:16,044 --> 00:29:18,804
Speaker 2: quarrying it where I'm saying, Okay, I talked to this person.

557
00:29:19,924 --> 00:29:24,044
Speaker 2: Here's a description of how an AI thing works, or

558
00:29:24,084 --> 00:29:26,644
Speaker 2: a crypto thing works, or concept and finance works. Right,

559
00:29:27,004 --> 00:29:30,684
Speaker 2: will you vet this for me? What critiques might you have? Right?

560
00:29:31,604 --> 00:29:33,524
Speaker 2: It's maybe not quite as good as talking to like

561
00:29:33,564 --> 00:29:35,684
Speaker 2: an expert, but I find like you often get a

562
00:29:35,724 --> 00:29:37,404
Speaker 2: lot of value from that, and again it's not the

563
00:29:37,444 --> 00:29:38,924
Speaker 2: last step in the process.

564
00:29:39,004 --> 00:29:39,124
Speaker 1: Right.

565
00:29:39,124 --> 00:29:41,804
Speaker 2: You can also use it for creative inspiration. Give me

566
00:29:41,924 --> 00:29:44,444
Speaker 2: ten potential headlines from this. You can use to fill

567
00:29:44,484 --> 00:29:46,004
Speaker 2: in missing words, because it thinks in terms of a

568
00:29:46,004 --> 00:29:48,724
Speaker 2: big matrix, right, so like what's an analogy that I

569
00:29:48,764 --> 00:29:51,844
Speaker 2: can think of? What's this word or concept that I'm missing?

570
00:29:51,884 --> 00:29:55,044
Speaker 2: Invent the name for this thing or that thing. You

571
00:29:55,044 --> 00:29:57,364
Speaker 2: can use it to kind of squeeze quantitative data out

572
00:29:57,364 --> 00:29:59,764
Speaker 2: of qualitative information, like I asked it, for example, to

573
00:30:00,564 --> 00:30:02,604
Speaker 2: an identify straft to this again, right, but to like

574
00:30:03,524 --> 00:30:07,564
Speaker 2: vet my estimate of how liberal or conservative different eras

575
00:30:07,564 --> 00:30:10,204
Speaker 2: in American history we're on a negative to to positive

576
00:30:10,244 --> 00:30:13,324
Speaker 2: ten scale. It can make ranking lists of different kinds.

577
00:30:13,324 --> 00:30:15,324
Speaker 2: I mean, it's just like there are so many use

578
00:30:15,324 --> 00:30:19,964
Speaker 2: cases for it, and like people just want us to

579
00:30:19,964 --> 00:30:23,084
Speaker 2: like cheat on papers or use it as a substitute

580
00:30:23,084 --> 00:30:24,804
Speaker 2: for like Wikipedia or something, which you're not the best

581
00:30:24,924 --> 00:30:26,964
Speaker 2: use cases for it. And if you ask chet ChiPT,

582
00:30:27,124 --> 00:30:28,724
Speaker 2: it will tell you that those are not the best

583
00:30:29,164 --> 00:30:30,044
Speaker 2: use cases for it.

584
00:30:30,164 --> 00:30:30,324
Speaker 1: Right.

585
00:30:30,404 --> 00:30:32,964
Speaker 2: The queriable nature of it and the fact that it

586
00:30:33,044 --> 00:30:35,564
Speaker 2: reorganizes this information in a way that for many purposes

587
00:30:35,564 --> 00:30:38,684
Speaker 2: but not all purposes, is much more approachable accessible, is

588
00:30:38,804 --> 00:30:41,204
Speaker 2: it's I don't know, it's I think it's a miraculous technology.

589
00:30:41,444 --> 00:30:44,684
Speaker 2: I mean, you know you had woken up. If I

590
00:30:44,724 --> 00:30:50,004
Speaker 2: had fallen to a coma years in twenty fifteen and

591
00:30:50,284 --> 00:30:55,604
Speaker 2: woken up, I missed the whole first Trump administration, I'm like, oh,

592
00:30:55,604 --> 00:31:00,284
Speaker 2: Trump's prison, Oh he was already President's surprise then, like

593
00:31:00,644 --> 00:31:02,844
Speaker 2: you would be fucking blown away by this shit, right,

594
00:31:02,844 --> 00:31:04,964
Speaker 2: and be like, oh my fucking god. Right, just like

595
00:31:05,004 --> 00:31:07,764
Speaker 2: passing the Turing test. I mean, there are definitions to

596
00:31:07,764 --> 00:31:08,924
Speaker 2: meet it's but whether it's a good test, whether it

597
00:31:09,004 --> 00:31:13,324
Speaker 2: actually is, But like it basically is like human esque

598
00:31:13,444 --> 00:31:18,564
Speaker 2: intelligence in some ways, inferior in some cases superior over

599
00:31:18,644 --> 00:31:20,844
Speaker 2: a large domain of fields. Just the way it can

600
00:31:21,004 --> 00:31:26,164
Speaker 2: like parse this very open ended, fuzzy logic of techt strings.

601
00:31:26,204 --> 00:31:27,804
Speaker 2: I mean, I just think it's kind of an amazing

602
00:31:28,644 --> 00:31:32,404
Speaker 2: It's amazically a robust in some ways, right, sure that

603
00:31:32,444 --> 00:31:35,204
Speaker 2: you can misspell things you get, I mean, it's amazingly

604
00:31:35,284 --> 00:31:37,684
Speaker 2: robust in a way that like it's hard to think

605
00:31:37,684 --> 00:31:43,564
Speaker 2: of other technologies that compare to it exactly, and you know,

606
00:31:43,804 --> 00:31:49,004
Speaker 2: and solved using very simple underlying math, right, I mean,

607
00:31:49,044 --> 00:31:51,124
Speaker 2: the code for deep seek is something like a few

608
00:31:51,204 --> 00:31:54,444
Speaker 2: hundred lines of code long. My fucking election model is

609
00:31:54,524 --> 00:31:58,084
Speaker 2: longer than that. Right, It's like it's like it's kind

610
00:31:58,084 --> 00:31:58,724
Speaker 2: of a miracle.

611
00:31:59,084 --> 00:32:04,924
Speaker 1: Absolutely absolutely I agree with all of that. Where but

612
00:32:05,244 --> 00:32:08,324
Speaker 1: now I think to push it a little bit further. Obviously,

613
00:32:08,364 --> 00:32:12,564
Speaker 1: this miracle, as we know, it's coming at a big cost, right,

614
00:32:12,924 --> 00:32:17,444
Speaker 1: environmental cost. It also P doom. Right, we've talked about

615
00:32:17,524 --> 00:32:21,164
Speaker 1: that potential risk is the miracle of you know, giving

616
00:32:21,164 --> 00:32:25,164
Speaker 1: you a good analogy worth it at this cost. That's

617
00:32:25,204 --> 00:32:26,164
Speaker 1: I think those are kind.

618
00:32:26,084 --> 00:32:28,524
Speaker 2: Of don't give me this environmental.

619
00:32:28,324 --> 00:32:31,284
Speaker 1: I'm not talking I'm talking about P doom and environmental.

620
00:32:31,284 --> 00:32:34,164
Speaker 1: It's not crap, Nate. We talked about this before, like

621
00:32:34,604 --> 00:32:35,444
Speaker 1: a lot of energy.

622
00:32:35,524 --> 00:32:37,324
Speaker 2: Okay, look at this. This is something that's supposed to scare me.

623
00:32:37,524 --> 00:32:41,564
Speaker 2: Four impress dot org. The energy consumption for training chat

624
00:32:41,604 --> 00:32:45,244
Speaker 2: EPT leading model is even more staggering, equated to that

625
00:32:45,284 --> 00:32:47,724
Speaker 2: of an American hassole for more than seven hundred years.

626
00:32:47,764 --> 00:32:51,324
Speaker 2: So basically, to train this leading model only took seven

627
00:32:51,404 --> 00:32:56,244
Speaker 2: hundred households worth like one subdivision of some fucking neighborhood

628
00:32:56,284 --> 00:33:00,044
Speaker 2: in Tulsa. Right, it's not very much like don't you

629
00:33:00,084 --> 00:33:02,444
Speaker 2: know you undermind the argument for ped doom. We all

630
00:33:02,484 --> 00:33:05,084
Speaker 2: die if like I mean this, you know, obviously if

631
00:33:05,764 --> 00:33:07,804
Speaker 2: the models get hungrier and hungrier. But by the way,

632
00:33:07,804 --> 00:33:11,844
Speaker 2: the deep seek thing should be good news for the environment.

633
00:33:11,964 --> 00:33:14,124
Speaker 1: Well that's why I said, we don't know, because on

634
00:33:14,164 --> 00:33:17,644
Speaker 1: the one hand, maybe depending on what the actual resources are.

635
00:33:17,844 --> 00:33:19,764
Speaker 1: On the other hand, if it means that every single

636
00:33:19,764 --> 00:33:22,164
Speaker 1: person is now running these smaller things, or not every

637
00:33:22,164 --> 00:33:25,364
Speaker 1: single person, but if it makes it more likely that

638
00:33:25,484 --> 00:33:28,204
Speaker 1: more of these are what's the net impact, right, If

639
00:33:28,244 --> 00:33:30,964
Speaker 1: it's one tenth but you actually have one hundred times

640
00:33:31,204 --> 00:33:33,724
Speaker 1: more people using it as a result, then it's a

641
00:33:33,764 --> 00:33:37,324
Speaker 1: net obviously negative impact instead of net positive. These are

642
00:33:37,364 --> 00:33:39,764
Speaker 1: all open questions, right, And like I said, I am

643
00:33:39,844 --> 00:33:42,524
Speaker 1: not an AI skeptic. I think it's really cool. I

644
00:33:42,564 --> 00:33:45,044
Speaker 1: think there are lots of really interesting things here. I

645
00:33:45,124 --> 00:33:47,044
Speaker 1: just think that there are other you know, you can't

646
00:33:47,084 --> 00:33:49,364
Speaker 1: also be like a raw raw cheer later none of

647
00:33:49,364 --> 00:33:52,004
Speaker 1: this matters. Of course, it does matter. I think all

648
00:33:52,044 --> 00:33:52,844
Speaker 1: of these things matter.

649
00:33:53,204 --> 00:33:56,884
Speaker 2: I just think people are over indexing to like I

650
00:33:56,884 --> 00:33:59,724
Speaker 2: don't know. I mean, have you taken wes Wimo Maria

651
00:33:59,724 --> 00:34:01,284
Speaker 2: and by the way if you're not aware, this is

652
00:34:01,324 --> 00:34:06,604
Speaker 2: a self driving car company which is available in San Francisco, Phoenix,

653
00:34:06,644 --> 00:34:08,044
Speaker 2: and maybe one or two other places.

654
00:34:08,124 --> 00:34:08,284
Speaker 1: La.

655
00:34:08,364 --> 00:34:10,204
Speaker 2: I think have you taken a Weimo in any.

656
00:34:10,044 --> 00:34:12,324
Speaker 3: Of those places, Maria, I have not name.

657
00:34:12,764 --> 00:34:15,244
Speaker 2: It's fucking blade wrinner. I'm telling you, it's a very

658
00:34:15,284 --> 00:34:17,724
Speaker 2: good experience. It's a much smoother ride than i'd say

659
00:34:17,764 --> 00:34:21,364
Speaker 2: ninety five percent of ubers. They have like space sage music,

660
00:34:21,404 --> 00:34:24,124
Speaker 2: and you feel like you're in the fucking future. And

661
00:34:24,324 --> 00:34:29,324
Speaker 2: I would almost guarantee you that driver less cars are

662
00:34:29,324 --> 00:34:32,204
Speaker 2: going to be a very popular technology.

663
00:34:32,324 --> 00:34:33,164
Speaker 3: I don't know. I don't know.

664
00:34:33,804 --> 00:34:36,364
Speaker 1: I watched I watched that. I watched that episode of

665
00:34:36,404 --> 00:34:39,044
Speaker 1: Silicon Valley where he's in the driver less car and

666
00:34:39,164 --> 00:34:42,044
Speaker 1: en step on a on a boat somewhere in the

667
00:34:42,084 --> 00:34:47,404
Speaker 1: middle of the ocean. So I'm a little obviously TV

668
00:34:47,484 --> 00:34:48,364
Speaker 1: show comedy.

669
00:34:48,444 --> 00:34:51,364
Speaker 3: But but you know, you never you never know how

670
00:34:51,364 --> 00:34:53,164
Speaker 3: the experience will will end up.

671
00:34:53,284 --> 00:34:55,404
Speaker 1: But I'm going to San Francisco next week, Nate, so

672
00:34:55,604 --> 00:34:57,644
Speaker 1: you know, maybe I'll take my first Weimo.

673
00:34:57,644 --> 00:35:00,164
Speaker 2: Take a Wimo. It's it's it's like a ninety fifth

674
00:35:00,244 --> 00:35:02,764
Speaker 2: percentile Uber driver.

675
00:35:03,324 --> 00:35:06,324
Speaker 1: All right, Well, well, on that positive note, Shall we

676
00:35:06,444 --> 00:35:08,484
Speaker 1: talk a little bit more poker and switch to a

677
00:35:08,524 --> 00:35:16,844
Speaker 1: listener question? Okay, fine, we'll be back right after this.

678
00:35:27,604 --> 00:35:32,684
Speaker 1: All right, So we had a poker related listener question

679
00:35:33,004 --> 00:35:38,364
Speaker 1: that I think, Nate, you are probably more equipped to

680
00:35:38,404 --> 00:35:41,404
Speaker 1: answer in the sense that you play cash home games

681
00:35:41,444 --> 00:35:43,964
Speaker 1: and I don't. By the way, I'm really sorry if

682
00:35:44,004 --> 00:35:47,964
Speaker 1: you can hear some knock knock noises in the background.

683
00:35:48,724 --> 00:35:52,804
Speaker 1: Apparently the apartment above mine has just started construction. It

684
00:35:52,884 --> 00:35:56,444
Speaker 1: actually just happened as we started taping this podcast. This

685
00:35:56,604 --> 00:35:59,404
Speaker 1: is the first hammering I have heard. But of course

686
00:35:59,484 --> 00:36:02,724
Speaker 1: you get to experience it alongside me, because I love

687
00:36:02,724 --> 00:36:04,804
Speaker 1: our listeners and I want to share all of my

688
00:36:04,924 --> 00:36:09,204
Speaker 1: experiences with them. So Nate, here's the listener question. I

689
00:36:09,284 --> 00:36:12,204
Speaker 1: have a neighborhood poker night with my friends. Everyone plays

690
00:36:12,204 --> 00:36:14,924
Speaker 1: really loose and passive, lots of calling, not much raising.

691
00:36:15,364 --> 00:36:18,004
Speaker 1: How do I win against real amateurs like that? What

692
00:36:18,084 --> 00:36:20,404
Speaker 1: are the most common and easy to detect tells by

693
00:36:20,444 --> 00:36:23,644
Speaker 1: amateurs like this? So part of this I can answer too, right,

694
00:36:23,644 --> 00:36:26,164
Speaker 1: because this happens in tournaments as well. But let's start

695
00:36:26,164 --> 00:36:29,844
Speaker 1: with what you think. Since you play in home games,

696
00:36:29,924 --> 00:36:32,964
Speaker 1: you know this is something that you find fun and

697
00:36:33,564 --> 00:36:35,764
Speaker 1: not an experience that I often have.

698
00:36:37,564 --> 00:36:39,884
Speaker 2: So it depends this comes from listener.

699
00:36:40,044 --> 00:36:40,204
Speaker 1: Hugh.

700
00:36:40,524 --> 00:36:46,924
Speaker 2: H u g h. I guess that's the way you spill, Hugh.

701
00:36:48,684 --> 00:36:49,764
Speaker 3: You does sound British.

702
00:36:49,844 --> 00:36:52,884
Speaker 1: I'm sorry, it's just a name that I automatically associate

703
00:36:52,924 --> 00:36:53,924
Speaker 1: with you with being English.

704
00:36:54,444 --> 00:36:57,844
Speaker 2: We have Okay, So what are basics for loose home?

705
00:36:57,884 --> 00:37:00,364
Speaker 2: I mean depends on if you're talking about like really

706
00:37:00,404 --> 00:37:02,804
Speaker 2: bad players. He seems to be.

707
00:37:03,044 --> 00:37:05,364
Speaker 3: He seems to be, so.

708
00:37:06,164 --> 00:37:08,764
Speaker 2: You know, the basics are you actually don't want to

709
00:37:08,764 --> 00:37:10,884
Speaker 2: play like everyone else beating, you know, don't be so

710
00:37:11,084 --> 00:37:16,724
Speaker 2: loose and passive. Right, you need particularly out of position

711
00:37:17,044 --> 00:37:21,804
Speaker 2: hands that can make the nuts right, so suited hands

712
00:37:21,844 --> 00:37:26,804
Speaker 2: particularly you know asex suited, Broadway suited hands ten nine

713
00:37:26,844 --> 00:37:29,444
Speaker 2: suited kind of an above right. So number one, hand

714
00:37:29,484 --> 00:37:34,884
Speaker 2: selection becomes more more oriented toward strong hands that can

715
00:37:34,884 --> 00:37:39,404
Speaker 2: make big you know, straights and flushes and better. That's

716
00:37:39,484 --> 00:37:42,404
Speaker 2: part one, right, Number two, you're gonna want to like

717
00:37:42,924 --> 00:37:46,924
Speaker 2: increase your bet sizing maybe quite a bit. Theory says

718
00:37:46,924 --> 00:37:48,524
Speaker 2: in a cash game that you're supposed to raise to

719
00:37:48,564 --> 00:37:50,764
Speaker 2: maybe two and a half x a big blind here

720
00:37:50,804 --> 00:37:53,564
Speaker 2: you can go four or five if they're already limpers.

721
00:37:53,684 --> 00:37:56,444
Speaker 2: You can raise even more than that. Right, There are

722
00:37:56,444 --> 00:37:58,284
Speaker 2: some games where the standard open might be to like

723
00:37:58,644 --> 00:38:00,204
Speaker 2: ten x or things like that. But like, let me

724
00:38:00,244 --> 00:38:02,644
Speaker 2: maybe let me even back up a little further, right,

725
00:38:02,764 --> 00:38:05,684
Speaker 2: I actually have dealt poker games to total rank amateurs

726
00:38:05,764 --> 00:38:07,484
Speaker 2: or like literally five to ten people. I have never

727
00:38:07,524 --> 00:38:08,764
Speaker 2: played booker before.

728
00:38:08,844 --> 00:38:09,044
Speaker 1: Right.

729
00:38:11,404 --> 00:38:14,404
Speaker 2: The two things that they most routinely get wrong are

730
00:38:14,644 --> 00:38:18,724
Speaker 2: number one, they call too much, meaning they call and

731
00:38:18,804 --> 00:38:21,124
Speaker 2: play it like a slot machine instead of folding or

732
00:38:21,164 --> 00:38:25,244
Speaker 2: raising more. And number two, they don't understand bet sizing.

733
00:38:25,764 --> 00:38:27,604
Speaker 2: What is the size of your bet relative to the

734
00:38:27,684 --> 00:38:31,844
Speaker 2: size of the pot. Right, If you don't know anything

735
00:38:31,844 --> 00:38:35,884
Speaker 2: about poker, no, nothing at all, then just bet half

736
00:38:35,924 --> 00:38:38,484
Speaker 2: the size of the pot. Keep track of what's in

737
00:38:38,524 --> 00:38:41,084
Speaker 2: the pot and bet half that size. But in general,

738
00:38:41,404 --> 00:38:42,444
Speaker 2: don't do so much calling.

739
00:38:42,484 --> 00:38:42,604
Speaker 1: Right.

740
00:38:42,644 --> 00:38:44,204
Speaker 2: If you have a good hand or a good bluff,

741
00:38:44,444 --> 00:38:45,684
Speaker 2: or even if you just kind of think other people

742
00:38:45,724 --> 00:38:49,964
Speaker 2: are scared, then then do some raising if you think

743
00:38:50,004 --> 00:38:52,084
Speaker 2: there'll be then most amateur players are not going to

744
00:38:52,164 --> 00:38:53,764
Speaker 2: go nuts. I mean it's a little complicate because like

745
00:38:54,204 --> 00:38:56,404
Speaker 2: they might not understand hands strikes, so you might want

746
00:38:56,404 --> 00:38:58,324
Speaker 2: to weig absolute strength a little bit more, but like,

747
00:38:59,324 --> 00:39:00,884
Speaker 2: but don't be afraid to get the money. And when

748
00:39:00,924 --> 00:39:02,644
Speaker 2: you have a good hand, and when other people have

749
00:39:02,684 --> 00:39:06,404
Speaker 2: a good representing a good hand, then it gets a

750
00:39:06,444 --> 00:39:10,564
Speaker 2: little bit complicated. But then you know, you don't have

751
00:39:10,604 --> 00:39:13,764
Speaker 2: to do a whole excess amount of bluff catching. I mean,

752
00:39:13,764 --> 00:39:17,524
Speaker 2: those are the basic And then I'd say, like, in general,

753
00:39:17,564 --> 00:39:20,404
Speaker 2: in these lose cash games, people are very sticky. Now

754
00:39:20,404 --> 00:39:22,924
Speaker 2: we're talking about a slightly higher caliber of players people

755
00:39:22,964 --> 00:39:25,924
Speaker 2: have had played before. Right, people are mostly very sticky

756
00:39:25,964 --> 00:39:29,644
Speaker 2: pre flop and on the flop, and then they will

757
00:39:29,724 --> 00:39:32,484
Speaker 2: start to fold cash can players do like to fold

758
00:39:32,524 --> 00:39:35,404
Speaker 2: on turns and rivers sometimes, Right, So that means that,

759
00:39:35,484 --> 00:39:37,964
Speaker 2: like you know that can affect your whole strategy for

760
00:39:38,004 --> 00:39:39,484
Speaker 2: the whole hand is that you tend not to have

761
00:39:39,524 --> 00:39:41,644
Speaker 2: a lot of full equity preflop in on flops, and

762
00:39:41,684 --> 00:39:43,924
Speaker 2: then it requires multiple barrels sometimes.

763
00:39:44,364 --> 00:39:45,124
Speaker 3: Yeah.

764
00:39:45,204 --> 00:39:48,604
Speaker 1: I think that as someone who is a tournament player,

765
00:39:49,604 --> 00:39:52,964
Speaker 1: there is some advice that I think applies all around,

766
00:39:53,684 --> 00:39:56,964
Speaker 1: which basically goes hand in hand with what you said Nate.

767
00:39:57,804 --> 00:40:00,484
Speaker 1: Number one, You don't want to follow the tendencies of

768
00:40:00,564 --> 00:40:03,084
Speaker 1: the people who are making mistakes, right, So If people

769
00:40:03,124 --> 00:40:05,564
Speaker 1: are too loose, you actually want to tighten up. If

770
00:40:05,604 --> 00:40:09,564
Speaker 1: people are passive, you want to become more aggressive. I

771
00:40:09,604 --> 00:40:11,924
Speaker 1: think that's important. But you should also realize if they're

772
00:40:11,964 --> 00:40:14,724
Speaker 1: going to be sticky, then you should just bet huge, right,

773
00:40:14,844 --> 00:40:17,404
Speaker 1: Like if you are going to if you'd normally bet

774
00:40:17,404 --> 00:40:20,284
Speaker 1: half pot, just bet pot they're going to call anyway, right,

775
00:40:20,724 --> 00:40:23,964
Speaker 1: build massive pots with your good hands. This will also

776
00:40:24,084 --> 00:40:28,484
Speaker 1: enable you to bluff, right, because they will eventually fold. Now,

777
00:40:28,684 --> 00:40:30,684
Speaker 1: something you said, I think this is actually true, not

778
00:40:30,804 --> 00:40:35,044
Speaker 1: just of cash games, but in general in tournaments as well.

779
00:40:35,604 --> 00:40:41,204
Speaker 1: People do tend to overcall flop and overfold turn. So

780
00:40:41,364 --> 00:40:44,444
Speaker 1: that's a great you know, I think building an over

781
00:40:44,484 --> 00:40:47,644
Speaker 1: betting strategy into your game in a game like that

782
00:40:47,764 --> 00:40:50,884
Speaker 1: is really good, right, And sometimes they'll get very sticky

783
00:40:51,084 --> 00:40:52,924
Speaker 1: like I've had. Because the second part of this question

784
00:40:53,044 --> 00:40:55,404
Speaker 1: was tells, which I think is just bad. Do not

785
00:40:55,564 --> 00:40:58,324
Speaker 1: use tells, even though in games like that people probably

786
00:40:58,364 --> 00:41:00,764
Speaker 1: do have tells, especially if you're going to play with

787
00:41:00,804 --> 00:41:02,564
Speaker 1: them over and over. It might be a little different,

788
00:41:02,804 --> 00:41:04,644
Speaker 1: but I just don't think that's great to rely on.

789
00:41:05,004 --> 00:41:07,884
Speaker 1: But when I've relied on tells, I've actually made really

790
00:41:07,924 --> 00:41:11,724
Speaker 1: big mistakes. Because I've had I've had a situations where

791
00:41:11,724 --> 00:41:13,764
Speaker 1: I'm like, oh, this person really likes their hand, they

792
00:41:13,844 --> 00:41:16,924
Speaker 1: must be really strong, and I end up folding. And

793
00:41:17,044 --> 00:41:19,484
Speaker 1: they had like ace deucee off suit, but there was

794
00:41:19,484 --> 00:41:21,044
Speaker 1: an ace on the board, and they thought that it

795
00:41:21,084 --> 00:41:24,484
Speaker 1: was just like the nuts, right, because they completely overvalued

796
00:41:24,524 --> 00:41:26,924
Speaker 1: the fact that they had an ace, and so they

797
00:41:26,924 --> 00:41:28,564
Speaker 1: were playing it like they had the nuts and they

798
00:41:28,604 --> 00:41:30,884
Speaker 1: thought they had the nuts, but they really didn't. So

799
00:41:30,924 --> 00:41:34,124
Speaker 1: if people are bad, don't use tells because the strength

800
00:41:34,764 --> 00:41:37,524
Speaker 1: their perceived strength of their hand may not actually be

801
00:41:37,604 --> 00:41:39,564
Speaker 1: the actual strength of their hands.

802
00:41:39,604 --> 00:41:44,204
Speaker 2: I'm more till it's funny because we had like opposite personnelity.

803
00:41:44,204 --> 00:41:46,404
Speaker 2: You're like way more sound theoretically, and I'm kind of

804
00:41:46,444 --> 00:41:51,364
Speaker 2: like psychoanalyzing people a little bit more. And no, look,

805
00:41:51,404 --> 00:41:55,684
Speaker 2: I think the premise are like two categories of tells

806
00:41:55,724 --> 00:42:00,724
Speaker 2: from very inexperienced players that I don't think I're always

807
00:42:00,724 --> 00:42:03,484
Speaker 2: hard to distinguish, but require additional context. Right, what is

808
00:42:03,524 --> 00:42:08,204
Speaker 2: a really bad actor tell right where they just are

809
00:42:08,244 --> 00:42:10,564
Speaker 2: like they watched like poker movies, We're supposed to act

810
00:42:11,084 --> 00:42:12,604
Speaker 2: if you're strong and starting every week and they just

811
00:42:12,644 --> 00:42:16,044
Speaker 2: really over knew it. In like a comical way, right, yeah,

812
00:42:16,044 --> 00:42:19,524
Speaker 2: and then I have seen that, but just sometimes people

813
00:42:19,564 --> 00:42:22,444
Speaker 2: are like extremely There's also there's a.

814
00:42:22,324 --> 00:42:25,684
Speaker 1: Lot of hollywooding actually, and I've seen this in uh

815
00:42:25,804 --> 00:42:28,524
Speaker 1: from amateur where like if you have the nuts right,

816
00:42:28,564 --> 00:42:32,284
Speaker 1: like say you flopped quads right or something like that,

817
00:42:32,564 --> 00:42:37,724
Speaker 1: and then they'll just be like the size and then

818
00:42:37,764 --> 00:42:41,644
Speaker 1: like I guess I call right like that if someone's

819
00:42:41,724 --> 00:42:43,884
Speaker 1: doing that, like holy shit, you're beat.

820
00:42:44,604 --> 00:42:47,964
Speaker 3: Like just there are there are certain situations like.

821
00:42:47,924 --> 00:42:51,444
Speaker 1: That, but I think in general it's it's better to

822
00:42:51,884 --> 00:42:53,004
Speaker 1: we haven't played.

823
00:42:52,764 --> 00:42:55,324
Speaker 3: In Hughes game or Hughes game.

824
00:42:55,964 --> 00:42:59,764
Speaker 1: So I think that just sticking to the advice that

825
00:42:59,764 --> 00:43:04,564
Speaker 1: that we've given, which is, you know, don't be lose passive. Basically,

826
00:43:04,764 --> 00:43:06,324
Speaker 1: you have to tighten up your ranges. You have to

827
00:43:06,324 --> 00:43:11,364
Speaker 1: figure out what those ranges are. And you know, the

828
00:43:11,364 --> 00:43:14,284
Speaker 1: the other thing is you know your bet sizing is

829
00:43:14,324 --> 00:43:16,004
Speaker 1: going to change, because if people are going to be

830
00:43:16,044 --> 00:43:18,924
Speaker 1: calling stations, great, exploit it. If people are going to

831
00:43:18,924 --> 00:43:22,124
Speaker 1: call pre flop anyway, great, make your sizes bigger. Just

832
00:43:22,604 --> 00:43:25,404
Speaker 1: build pots when you have very strong hands.

833
00:43:25,484 --> 00:43:27,204
Speaker 2: I mean the other thing, you know, to close this

834
00:43:27,284 --> 00:43:30,884
Speaker 2: cussing on tells I mean, people can also be very honest, right,

835
00:43:31,004 --> 00:43:35,764
Speaker 2: Like they don't you know, in games where the stakes

836
00:43:35,804 --> 00:43:38,964
Speaker 2: are low relative to people's like net worth, which depends

837
00:43:39,004 --> 00:43:42,444
Speaker 2: on people's net worth, right, then they just don't necessarily

838
00:43:42,524 --> 00:43:48,004
Speaker 2: take a lot of action to like conceal disappointment with

839
00:43:48,084 --> 00:43:52,404
Speaker 2: a bad flop or things like that. You know a

840
00:43:52,444 --> 00:43:53,924
Speaker 2: lot of times all I gotta catch my car like

841
00:43:53,964 --> 00:43:57,564
Speaker 2: that actually often is more than not honest, more in

842
00:43:57,644 --> 00:44:00,004
Speaker 2: cash games and in tournaments. I don't know why, right,

843
00:44:01,244 --> 00:44:04,644
Speaker 2: I think tournaments people like are just like playing their

844
00:44:04,844 --> 00:44:09,604
Speaker 2: A game a bit more Amit's secret of cash right more?

845
00:44:09,604 --> 00:44:12,604
Speaker 2: Playing their A game more often in tournaments at games.

846
00:44:12,524 --> 00:44:15,004
Speaker 1: Yeah, no, I actually I think I think there's something

847
00:44:15,004 --> 00:44:17,084
Speaker 1: too that people do tend to be more honest in

848
00:44:17,124 --> 00:44:19,964
Speaker 1: cash games. I had a hilarious situation at a higher

849
00:44:19,964 --> 00:44:23,724
Speaker 1: stakes cash game where I had raised and I don't

850
00:44:23,724 --> 00:44:25,804
Speaker 1: remember if it was small or big blind had defended

851
00:44:26,844 --> 00:44:32,244
Speaker 1: and anyway, went, you know, check bet on the flop,

852
00:44:33,404 --> 00:44:38,604
Speaker 1: then check and check on the turn and check and no, no,

853
00:44:38,644 --> 00:44:41,324
Speaker 1: not check And I was like looking to see like

854
00:44:41,444 --> 00:44:43,444
Speaker 1: I had nothing what I wanted to bet on the

855
00:44:43,524 --> 00:44:46,164
Speaker 1: river and he just folded. He's like, you definitely have

856
00:44:46,284 --> 00:44:49,444
Speaker 1: me beat, because like I've got nothing and I had nothing, right, Like,

857
00:44:49,484 --> 00:44:51,244
Speaker 1: I don't think I definitely had a beat, and he

858
00:44:51,364 --> 00:44:53,284
Speaker 1: just folded to me, right. I didn't even have to

859
00:44:53,284 --> 00:44:55,084
Speaker 1: think about the sizing or whether I was going to

860
00:44:55,124 --> 00:44:57,444
Speaker 1: bet or any of it. That would never happen in

861
00:44:57,484 --> 00:45:00,124
Speaker 1: a tournament, but it happens in cash games all the time.

862
00:45:00,164 --> 00:45:01,284
Speaker 3: And I still remember this hand.

863
00:45:01,284 --> 00:45:03,964
Speaker 1: I don't play cash very often, so things like that

864
00:45:04,044 --> 00:45:06,364
Speaker 1: stand out, but I've seen people do that and then

865
00:45:06,364 --> 00:45:07,964
Speaker 1: they try to do it in tournaments. Actually, you can

866
00:45:08,004 --> 00:45:10,724
Speaker 1: often spot a cash player a tournament because they will

867
00:45:10,764 --> 00:45:13,244
Speaker 1: sometimes fold out of turn. They'll do things that are

868
00:45:13,284 --> 00:45:16,484
Speaker 1: just like very honestly communicate that they have no more

869
00:45:16,524 --> 00:45:17,204
Speaker 1: interest in.

870
00:45:17,084 --> 00:45:21,364
Speaker 2: This hand and don't be a super knit right, like

871
00:45:21,444 --> 00:45:25,404
Speaker 2: to show people over value, especially in cash games, the

872
00:45:25,484 --> 00:45:27,604
Speaker 2: last thing they saw, right, So like if you have

873
00:45:27,724 --> 00:45:32,444
Speaker 2: like an occasional hand where you get a little out

874
00:45:32,444 --> 00:45:36,004
Speaker 2: of line, right, and again, I think you know usually

875
00:45:36,164 --> 00:45:38,804
Speaker 2: worth picking your spots carefully, and there's some psychology to that.

876
00:45:38,884 --> 00:45:40,964
Speaker 2: And then like if you show, oh, I three bit

877
00:45:41,084 --> 00:45:44,684
Speaker 2: five four suited from the button, which might actually be

878
00:45:44,724 --> 00:45:48,604
Speaker 2: a perfectly fine near gto three bed occasionally right, if

879
00:45:48,644 --> 00:45:52,164
Speaker 2: you turn as straight with that an the other guy folds.

880
00:45:52,204 --> 00:45:53,764
Speaker 2: You definitely want to show that hand right. You want

881
00:45:53,764 --> 00:45:57,604
Speaker 2: to like you're maintaining your reputation because, like a seat

882
00:45:57,644 --> 00:45:59,484
Speaker 2: in a good cash game is a valuable thing and

883
00:45:59,524 --> 00:46:02,644
Speaker 2: people absolutely will notice if you're being a knit. Don't

884
00:46:02,644 --> 00:46:03,044
Speaker 2: be a knit.

885
00:46:03,364 --> 00:46:03,564
Speaker 3: Yep.

886
00:46:03,604 --> 00:46:06,524
Speaker 2: Have fun, play toward the looser end of your GTO range.

887
00:46:06,844 --> 00:46:09,004
Speaker 2: How your GT arrange may actually be pretty tight against

888
00:46:09,684 --> 00:46:12,964
Speaker 2: phishing number fold yep, good luck you, good luck you.

889
00:46:19,764 --> 00:46:21,924
Speaker 2: Let us know what you think of the show. Reach

890
00:46:22,004 --> 00:46:26,484
Speaker 2: out to us at Risky Business at pushkin dot Fm.

891
00:46:26,644 --> 00:46:30,204
Speaker 2: Risky Business is hosted by me Maria Kondikova and byby

892
00:46:30,484 --> 00:46:31,004
Speaker 2: Nate Silver.

893
00:46:31,844 --> 00:46:35,364
Speaker 1: The show is a co production of Pushkin Industries and iHeartMedia.

894
00:46:35,884 --> 00:46:39,444
Speaker 1: This episode was produced by Isabel Carter. Our associate producer

895
00:46:39,524 --> 00:46:43,044
Speaker 1: is Gabriel Hunter Chang. Our executive producer is Jacob Goldstein.

896
00:46:43,484 --> 00:46:45,604
Speaker 2: If you like the show, please rate and review us

897
00:46:45,604 --> 00:46:47,684
Speaker 2: so other people can find us too. And if you

898
00:46:47,724 --> 00:46:50,084
Speaker 2: want to listen to an ad free version, sign up

899
00:46:50,084 --> 00:46:52,484
Speaker 2: for Pushkin Plus. For We're six, ten and nine a

900
00:46:52,524 --> 00:46:55,004
Speaker 2: month you get access to ad free listening. Thanks for

901
00:46:55,044 --> 00:46:55,444
Speaker 2: tuning in.