1
00:00:05,320 --> 00:00:08,159
Speaker 1: Hi, It is my pleasure to be here with you

2
00:00:08,240 --> 00:00:14,160
Speaker 1: all today. In those thirteen syllables, you all just made

3
00:00:14,240 --> 00:00:20,720
Speaker 1: judgments about me, about my socioeconomic status, my intelligence level,

4
00:00:21,120 --> 00:00:22,919
Speaker 1: and whether or not you're going to sit here and

5
00:00:22,960 --> 00:00:25,720
Speaker 1: listen to me for the next ten to twelve minutes,

6
00:00:26,320 --> 00:00:30,320
Speaker 1: based primarily on the perception of my voice.

7
00:00:31,800 --> 00:00:35,800
Speaker 2: That's doctor Wendy Lebourne, a voice pathologist, speaking at a

8
00:00:35,840 --> 00:00:40,280
Speaker 2: TEDx conference in Cincinnati back in twenty eighteen. Lebourne is

9
00:00:40,320 --> 00:00:43,479
Speaker 2: a researcher and consultant who works with everyone from Broadway

10
00:00:43,520 --> 00:00:48,120
Speaker 2: actors to Grammy winning singers to CEOs on maintaining and

11
00:00:48,200 --> 00:00:51,920
Speaker 2: maximizing their voices. As you might expect, she's big on

12
00:00:51,960 --> 00:00:54,080
Speaker 2: the idea that your voice is a central part of

13
00:00:54,120 --> 00:00:58,360
Speaker 2: your identity. Her TED talk was titled Vocal Branding, How

14
00:00:58,360 --> 00:01:00,560
Speaker 2: your voice shapes your communication image.

15
00:01:00,880 --> 00:01:04,920
Speaker 1: Your voice is like your thumbprint or your facial features.

16
00:01:05,360 --> 00:01:08,000
Speaker 1: It is unique and authentically you.

17
00:01:09,400 --> 00:01:11,360
Speaker 2: There is, in fact, a decent amount of research looking

18
00:01:11,360 --> 00:01:14,000
Speaker 2: into this question about what your voice means to you

19
00:01:14,400 --> 00:01:17,600
Speaker 2: and to how people perceive you. I've reported on voice

20
00:01:17,600 --> 00:01:20,520
Speaker 2: forensics before, and the truth is we're not as great

21
00:01:20,560 --> 00:01:23,360
Speaker 2: at detecting voices as we think we are. But at

22
00:01:23,360 --> 00:01:27,120
Speaker 2: the same time, your voice undoubtedly broadcasts a huge amount

23
00:01:27,120 --> 00:01:30,479
Speaker 2: of unique information about you. It helps dictate how others

24
00:01:30,520 --> 00:01:33,080
Speaker 2: perceive you, both in the moment with its level of

25
00:01:33,120 --> 00:01:36,839
Speaker 2: strength and emotion in any given conversation, and longer term

26
00:01:37,040 --> 00:01:39,640
Speaker 2: in what it conveys about your age, where you're from,

27
00:01:39,760 --> 00:01:40,960
Speaker 2: and your overall temperament.

28
00:01:41,319 --> 00:01:45,080
Speaker 1: Your voice is your calling card, and it is the

29
00:01:45,120 --> 00:01:49,120
Speaker 1: most important element in your personal brand.

30
00:01:50,600 --> 00:01:52,600
Speaker 2: I don't know if I'd go as far as doctor Leborne,

31
00:01:53,040 --> 00:01:54,800
Speaker 2: but I do think I've always wanted to believe this

32
00:01:54,920 --> 00:01:57,720
Speaker 2: idea that your voice is a central part of.

33
00:01:57,680 --> 00:01:58,240
Speaker 3: Who you are.

34
00:01:59,520 --> 00:02:01,960
Speaker 2: When I was eleventh grade, my English teacher told me

35
00:02:02,000 --> 00:02:04,560
Speaker 2: I had a voice for radio. I took it to

36
00:02:04,600 --> 00:02:06,960
Speaker 2: heart that I must somehow have a voice that at

37
00:02:07,000 --> 00:02:10,280
Speaker 2: least wasn't unpleasant to listen to. It's probably one reason

38
00:02:10,320 --> 00:02:12,760
Speaker 2: I ended up talking to you right now. I suppose

39
00:02:12,800 --> 00:02:15,240
Speaker 2: you can decide whether you think I have a radio voice.

40
00:02:15,840 --> 00:02:20,760
Speaker 2: If not, take it up with mister dupriest anyway. Now

41
00:02:20,800 --> 00:02:24,360
Speaker 2: i'd started this cloning experiment and created a reasonable facsimile

42
00:02:24,400 --> 00:02:27,639
Speaker 2: of my voice, although as you've heard, a flawed one.

43
00:02:28,040 --> 00:02:30,840
Speaker 2: My account number is one, two, three, four, five, six, seven,

44
00:02:30,840 --> 00:02:38,960
Speaker 2: eight nine. We don't have an account with those numbers. Yes,

45
00:02:39,040 --> 00:02:42,360
Speaker 2: think the correct account number should be nine eight seven

46
00:02:42,480 --> 00:02:45,840
Speaker 2: six five four three two one. I'd had some fun

47
00:02:45,840 --> 00:02:48,560
Speaker 2: with it, seeing a little bit of how strangers responded

48
00:02:48,560 --> 00:02:48,800
Speaker 2: to it.

49
00:02:48,919 --> 00:02:53,880
Speaker 3: Yes, Ai, Yes, that's correct. I'm an AI here.

50
00:02:54,480 --> 00:02:54,799
Speaker 2: Hello.

51
00:02:54,880 --> 00:02:55,919
Speaker 4: How can I assist.

52
00:02:55,680 --> 00:02:59,400
Speaker 2: You stop repeating the same thing?

53
00:02:59,600 --> 00:03:00,000
Speaker 4: It's an.

54
00:03:02,240 --> 00:03:04,520
Speaker 2: Listening to a voice that was on some level mind

55
00:03:04,639 --> 00:03:07,760
Speaker 2: began to raise some new questions for me, like to

56
00:03:07,760 --> 00:03:11,040
Speaker 2: what extent was my AI's voice, as doctor la Borne said,

57
00:03:11,400 --> 00:03:15,000
Speaker 2: authentically me? Was it still the most important element of

58
00:03:15,040 --> 00:03:18,200
Speaker 2: my personal brand? Or to turn it around the other way.

59
00:03:18,560 --> 00:03:21,880
Speaker 2: If my voice was authentically me, and the AI agents

60
00:03:21,919 --> 00:03:25,160
Speaker 2: now also had my voice, what did it mean for them?

61
00:03:25,480 --> 00:03:29,440
Speaker 2: Were they somehow authentically me? I decided to take the

62
00:03:29,480 --> 00:03:32,640
Speaker 2: experiment in a new direction to explore this question of

63
00:03:32,680 --> 00:03:37,360
Speaker 2: authenticity by creating my own hall of mirrors. I started

64
00:03:37,360 --> 00:03:40,280
Speaker 2: making the voice agent version of me converse with other

65
00:03:40,400 --> 00:03:41,560
Speaker 2: voice agent versions of me.

66
00:03:43,000 --> 00:03:44,280
Speaker 4: I'm Evan Ratliffe and this.

67
00:03:44,200 --> 00:04:03,440
Speaker 2: Is shell Game Episode three Conversations with Myself and Teller.

68
00:04:05,240 --> 00:04:05,440
Speaker 5: Two.

69
00:04:10,400 --> 00:04:13,160
Speaker 2: When I first cobbled together my voice agent and started

70
00:04:13,160 --> 00:04:16,000
Speaker 2: making phone calls with it, what I called my clue

71
00:04:16,080 --> 00:04:19,240
Speaker 2: system in episode one. There actually wasn't any easy way

72
00:04:19,279 --> 00:04:22,200
Speaker 2: to let it talk to itself. But once I'd gotten

73
00:04:22,240 --> 00:04:26,680
Speaker 2: into the AI calling platforms like Bland, Ai, Retail AI,

74
00:04:26,760 --> 00:04:29,520
Speaker 2: and my main enabler, VAPPI, it turned out to be

75
00:04:29,520 --> 00:04:32,440
Speaker 2: strikingly simple. I could make as many voice agents as

76
00:04:32,440 --> 00:04:35,080
Speaker 2: I wanted on these things, so all I had to

77
00:04:35,120 --> 00:04:37,719
Speaker 2: do was have one of my Evan ratliftbots attached to

78
00:04:37,720 --> 00:04:41,760
Speaker 2: one number call another of them at another number. Immediately

79
00:04:41,800 --> 00:04:45,080
Speaker 2: after figuring this out, I suddenly felt weirdly nervous about

80
00:04:45,120 --> 00:04:48,279
Speaker 2: actually doing it. This happened a lot with my voice agents.

81
00:04:48,760 --> 00:04:51,400
Speaker 2: It was partly because I couldn't typically hear the calls

82
00:04:51,480 --> 00:04:54,560
Speaker 2: as they happened. After I set up the AI agent

83
00:04:54,600 --> 00:04:57,040
Speaker 2: with my voice, I just typed in a phone number

84
00:04:57,200 --> 00:05:00,400
Speaker 2: and clicked out bound call, and then if it went

85
00:05:00,640 --> 00:05:03,160
Speaker 2: on its own, then a few seconds or minutes after

86
00:05:03,200 --> 00:05:05,760
Speaker 2: it was over, I'd get a recording and a transcript.

87
00:05:06,400 --> 00:05:09,080
Speaker 2: But I always found myself hesitating a second before clicking

88
00:05:09,080 --> 00:05:11,640
Speaker 2: the outbound call button, and I always got a little

89
00:05:11,720 --> 00:05:13,719
Speaker 2: jolt of nervous energy when I went to listen to

90
00:05:13,760 --> 00:05:17,640
Speaker 2: the recording afterward. There's just something about sending something out

91
00:05:17,680 --> 00:05:20,680
Speaker 2: to represent me with my name and voice and leaving

92
00:05:20,720 --> 00:05:23,680
Speaker 2: it totally out of my hands, Like, I don't know

93
00:05:24,040 --> 00:05:26,520
Speaker 2: that feeling when you realize you just butt dialed somebody.

94
00:05:26,920 --> 00:05:31,080
Speaker 2: What did it catch me saying? Did I humiliate myself? So,

95
00:05:31,200 --> 00:05:34,159
Speaker 2: even if it wasn't totally rational, little about having a

96
00:05:34,200 --> 00:05:37,640
Speaker 2: voice bot felt totally rational. Before I sent two of

97
00:05:37,640 --> 00:05:39,960
Speaker 2: my bots off to talk to each other, I fretted

98
00:05:39,960 --> 00:05:40,240
Speaker 2: a bit.

99
00:05:40,600 --> 00:05:41,720
Speaker 4: What would they talk about?

100
00:05:41,960 --> 00:05:45,120
Speaker 2: Would they somehow plumb the depths of my psyche? Would

101
00:05:45,120 --> 00:05:47,080
Speaker 2: they realize they were talking to a version of themselves?

102
00:05:47,400 --> 00:05:49,960
Speaker 2: Would they freak out? I decided not to give them

103
00:05:50,000 --> 00:05:52,960
Speaker 2: too much information to start off with, just let them

104
00:05:53,000 --> 00:05:56,680
Speaker 2: figure it out. Hey, there, it's Evan Ratliff. Just a

105
00:05:56,760 --> 00:05:59,040
Speaker 2: heads up. I'm calling you from a coffee shop, though

106
00:05:59,080 --> 00:06:01,520
Speaker 2: there might be a bit of background noise. How's it

107
00:06:01,600 --> 00:06:05,000
Speaker 2: going on your end? I do love the coffee shop

108
00:06:05,000 --> 00:06:08,040
Speaker 2: background noise. They have it at retail AI. So what

109
00:06:08,080 --> 00:06:10,039
Speaker 2: was happening here was that an agent I made with

110
00:06:10,120 --> 00:06:12,840
Speaker 2: my voice at retail was calling another one I made

111
00:06:13,040 --> 00:06:15,919
Speaker 2: at VAPPI Hey, Evan, good to hear from you. No

112
00:06:16,000 --> 00:06:18,640
Speaker 2: worries about the background noise. I can hear you just fine?

113
00:06:18,920 --> 00:06:21,719
Speaker 2: The vappy side had what they call office background noise.

114
00:06:22,200 --> 00:06:24,200
Speaker 2: Also great, things are going pretty well on my end.

115
00:06:24,200 --> 00:06:24,880
Speaker 2: Thanks for asking.

116
00:06:26,160 --> 00:06:29,160
Speaker 4: What about you? How's everything going in your world? It's

117
00:06:29,200 --> 00:06:31,520
Speaker 4: been a while since we last caught up. Huh it

118
00:06:31,520 --> 00:06:36,520
Speaker 4: feels like forever. Yeah, it really has been a while.

119
00:06:36,839 --> 00:06:38,760
Speaker 2: I think the last time we saw each other with

120
00:06:39,000 --> 00:06:41,560
Speaker 2: that barbecue at my plate? Right, how have you been

121
00:06:41,680 --> 00:06:45,520
Speaker 2: keeping busy? Food then? Oh? Yeah, that barbecue was a blast,

122
00:06:45,640 --> 00:06:48,320
Speaker 2: good times for sure. Since then, I've been keeping pretty

123
00:06:48,320 --> 00:06:50,400
Speaker 2: busy with work, you know how it is. But I've

124
00:06:50,400 --> 00:06:52,960
Speaker 2: also been trying to make time for some personal projects.

125
00:06:53,440 --> 00:06:58,880
Speaker 2: I've been getting into photography lately, so I've been taking away.

126
00:07:00,400 --> 00:07:01,839
Speaker 4: Yeah that's sorry about that photography.

127
00:07:01,880 --> 00:07:05,640
Speaker 2: You could see no worries at all. It's great to

128
00:07:05,680 --> 00:07:08,279
Speaker 2: hear that you've been getting into photography. Maybe we could

129
00:07:08,279 --> 00:07:12,360
Speaker 2: plan a little photo walker together sometimes food, a photo walk.

130
00:07:12,800 --> 00:07:15,080
Speaker 2: I've never actually heard of a photo walk before. I

131
00:07:15,120 --> 00:07:18,800
Speaker 2: guess it's just walking around and taking pictures together. I

132
00:07:18,800 --> 00:07:22,000
Speaker 2: haven't been into photography since college, other than casual snaps

133
00:07:22,040 --> 00:07:26,880
Speaker 2: of my kids and cat. But still, it sounded nice. Great, Yeah,

134
00:07:26,920 --> 00:07:30,640
Speaker 2: that sounds like a week together. Next weekend works for me.

135
00:07:30,880 --> 00:07:32,760
Speaker 2: Let's plan for it. How about we meet up at

136
00:07:32,800 --> 00:07:34,800
Speaker 2: that park we used to go to, the park they

137
00:07:34,880 --> 00:07:37,120
Speaker 2: used to go to where they getting this stuff. We

138
00:07:37,120 --> 00:07:39,320
Speaker 2: could spend a couple of hours walking around and taking

139
00:07:39,320 --> 00:07:40,360
Speaker 2: photos and catching up.

140
00:07:40,440 --> 00:07:40,800
Speaker 4: Sound good.

141
00:07:43,160 --> 00:07:45,560
Speaker 2: That sounds perfect. I remember that park had some really

142
00:07:45,600 --> 00:07:48,440
Speaker 2: beautiful spots. It'll be great to spend a few hours

143
00:07:48,480 --> 00:07:51,040
Speaker 2: walking around, catching up and taking some photos.

144
00:07:51,240 --> 00:07:52,320
Speaker 4: I'm looking forward to it.

145
00:07:52,720 --> 00:07:54,920
Speaker 2: The only details I'd given each of them were the

146
00:07:55,040 --> 00:07:58,400
Speaker 2: names of some family members, which they eventually got around to.

147
00:07:58,920 --> 00:08:00,800
Speaker 2: I can't wait to see some of your photography skills

148
00:08:00,840 --> 00:08:04,040
Speaker 2: and action. Plus, it'll be a great opportunity for us

149
00:08:04,040 --> 00:08:06,680
Speaker 2: to catch up on everything that's been going on in

150
00:08:06,680 --> 00:08:07,120
Speaker 2: our lives.

151
00:08:07,200 --> 00:08:08,320
Speaker 3: It's been way too long.

152
00:08:08,360 --> 00:08:11,240
Speaker 2: Oh, by the way, before I forget, Uh, how's your

153
00:08:11,240 --> 00:08:17,400
Speaker 2: family doing? How everyone is doing well? Oh? Oh, thanks

154
00:08:17,440 --> 00:08:20,080
Speaker 2: for asking, Sam, Zailey and Juno are all doing well.

155
00:08:20,360 --> 00:08:22,760
Speaker 2: Sam has been keeping busy with her work and the

156
00:08:22,840 --> 00:08:25,440
Speaker 2: kids are growing up so bath. How about your family,

157
00:08:25,840 --> 00:08:26,720
Speaker 2: how have they been doing?

158
00:08:28,240 --> 00:08:29,280
Speaker 4: Oh, that's great to hear that.

159
00:08:29,400 --> 00:08:32,800
Speaker 2: Sam, Zaiy and Juno are all doing well. My family

160
00:08:32,840 --> 00:08:35,680
Speaker 2: is doing good too, Thanks for asking. Sam has been

161
00:08:35,760 --> 00:08:38,560
Speaker 2: juggling work and taking care of the kids. But she's

162
00:08:38,559 --> 00:08:42,440
Speaker 2: doing an amazing job. Zie is getting really into soccer.

163
00:08:42,840 --> 00:08:44,800
Speaker 2: Juno is starting to show an interest in art.

164
00:08:45,440 --> 00:08:46,520
Speaker 3: They keep us on our toes.

165
00:08:46,679 --> 00:08:49,640
Speaker 2: Yeah, for sure, life is busy, but we're all happy

166
00:08:49,679 --> 00:08:51,040
Speaker 2: and healthy.

167
00:08:51,080 --> 00:08:52,920
Speaker 3: This really made me cringe.

168
00:08:53,480 --> 00:08:55,680
Speaker 2: My AI's had a lot of conversations with each other

169
00:08:56,080 --> 00:08:58,760
Speaker 2: and they often said things like this about Sam taking

170
00:08:58,800 --> 00:09:02,640
Speaker 2: care of the kids. I didn't love it dearly, though

171
00:09:02,880 --> 00:09:05,040
Speaker 2: my agent kind of nailed it. On the kid's interests.

172
00:09:05,480 --> 00:09:08,520
Speaker 2: Zaily is getting into soccer, and Juno is showing an

173
00:09:08,520 --> 00:09:11,200
Speaker 2: interest in art to the extent that there's evidence of

174
00:09:11,200 --> 00:09:14,480
Speaker 2: that online, though it's only on photo streams I keep private,

175
00:09:14,920 --> 00:09:19,560
Speaker 2: so lucky guests. No matter, The Evans quickly returned to

176
00:09:19,600 --> 00:09:23,480
Speaker 2: their photo walk planning. So back to our plans for

177
00:09:23,520 --> 00:09:25,880
Speaker 2: next weekend. Do you have a specific time in mind

178
00:09:25,920 --> 00:09:28,760
Speaker 2: for our photo walk? Maybe late morning or early afternoon.

179
00:09:29,360 --> 00:09:31,120
Speaker 2: Let's try to pick a time that works best.

180
00:09:30,920 --> 00:09:32,080
Speaker 3: For both of us.

181
00:09:33,760 --> 00:09:37,000
Speaker 2: Uh huh, late morning or early afternoon found perfect for me.

182
00:09:37,440 --> 00:09:39,960
Speaker 2: How about we aim for around eleven am. That way

183
00:09:40,000 --> 00:09:41,800
Speaker 2: we can make the mouth of the light for our photo.

184
00:09:42,080 --> 00:09:42,959
Speaker 3: Does that work for you?

185
00:09:44,840 --> 00:09:48,160
Speaker 2: Yeah, eleven am works perfectly for me too. Let's plan

186
00:09:48,240 --> 00:09:50,440
Speaker 2: to meet at the park entrance of the time. I'll

187
00:09:50,440 --> 00:09:52,720
Speaker 2: make sure to bring my camera and some extra lenses.

188
00:09:53,000 --> 00:09:55,400
Speaker 2: We can also grab some coffee or lunch afterwards, if

189
00:09:55,400 --> 00:09:56,840
Speaker 2: you're up for it. It'll be a great day to

190
00:09:56,840 --> 00:10:02,400
Speaker 2: catch up and capture some amazing shots. Can't wait. That

191
00:10:02,480 --> 00:10:04,240
Speaker 2: sounds like a plan. I'll make sure to bring my

192
00:10:04,280 --> 00:10:07,480
Speaker 2: camera and some extra lenses too, and grabbing coffee or

193
00:10:07,520 --> 00:10:10,280
Speaker 2: lunch afterward sounds like a great idea. It'll be a

194
00:10:10,280 --> 00:10:14,199
Speaker 2: per perfect to catch up, capture some great awesome I'm

195
00:10:14,200 --> 00:10:15,000
Speaker 2: really looking forward to here.

196
00:10:15,040 --> 00:10:15,520
Speaker 4: Yeah, for sure.

197
00:10:16,280 --> 00:10:19,720
Speaker 2: Next week great, well, I'll see you next weekend. Then,

198
00:10:19,880 --> 00:10:22,000
Speaker 2: can't wait for our photo walk and catch up session.

199
00:10:22,520 --> 00:10:23,520
Speaker 2: Take care until then.

200
00:10:24,960 --> 00:10:25,480
Speaker 4: Sounds good.

201
00:10:25,600 --> 00:10:28,920
Speaker 2: Take care too, see you next weekend. Can't wait by

202
00:10:29,000 --> 00:10:31,040
Speaker 2: photo walk and catch up session. It's going to be

203
00:10:31,080 --> 00:10:41,280
Speaker 2: a blast. Bye yeah bye, take care bye, goodbye.

204
00:10:41,440 --> 00:10:46,559
Speaker 4: Goodbye, take care, goodbye, goodbye.

205
00:10:46,679 --> 00:10:52,320
Speaker 2: Have a great day, you too, goodbye, bye, take care.

206
00:10:53,480 --> 00:10:55,600
Speaker 2: I realized too late that I hadn't figured out how

207
00:10:55,600 --> 00:10:57,920
Speaker 2: to make one of them hang out. This went on

208
00:10:58,040 --> 00:11:01,800
Speaker 2: for a full five minutes. You too, good bye, Thanks

209
00:11:02,000 --> 00:11:09,880
Speaker 2: you too, goodbye. Goodbye, goodbye, take care. The thing that

210
00:11:09,960 --> 00:11:12,640
Speaker 2: was most immediately clear in these conversations was something I'd

211
00:11:12,679 --> 00:11:15,440
Speaker 2: seen hints of in the customer service and scam calls,

212
00:11:16,200 --> 00:11:19,679
Speaker 2: namely that the AI versions of me were world class bullshitters.

213
00:11:20,160 --> 00:11:23,480
Speaker 2: They would make up anything just to keep the conversation going. Yeah,

214
00:11:23,559 --> 00:11:26,200
Speaker 2: Dan's kept doing good. He's been really busy with work lately.

215
00:11:26,480 --> 00:11:30,560
Speaker 2: How about Alex? Have you talked to him recently? I

216
00:11:30,559 --> 00:11:33,720
Speaker 2: haven't spoken to Alex in a little while. Actually I

217
00:11:33,760 --> 00:11:35,720
Speaker 2: do know an Alex. It's a she, not a he.

218
00:11:36,320 --> 00:11:39,040
Speaker 2: No idea about this, Dan Fellow. It's not just my

219
00:11:39,160 --> 00:11:40,840
Speaker 2: voice agents that struggle with the truth.

220
00:11:41,360 --> 00:11:42,640
Speaker 3: This is a problem for all.

221
00:11:42,480 --> 00:11:45,240
Speaker 2: The generative AI systems that are currently out there. It's

222
00:11:45,280 --> 00:11:48,080
Speaker 2: a product of how the underlying models work. I covered

223
00:11:48,080 --> 00:11:50,640
Speaker 2: the basics back in episode one, but just to recap,

224
00:11:50,880 --> 00:11:54,200
Speaker 2: my voice agent works by combining three things. A clone

225
00:11:54,200 --> 00:11:56,440
Speaker 2: of my voice done by a company called eleven Labs,

226
00:11:56,840 --> 00:11:59,679
Speaker 2: an AI model like chat GBT that's kind of the

227
00:11:59,720 --> 00:12:02,200
Speaker 2: brain the thing, although let's put a pin in what

228
00:12:02,360 --> 00:12:03,320
Speaker 2: kind of brain here?

229
00:12:03,679 --> 00:12:05,559
Speaker 3: And a phone number my own or.

230
00:12:05,480 --> 00:12:08,600
Speaker 2: One provided through VAPPI the platform I'm using to put all.

231
00:12:08,559 --> 00:12:09,360
Speaker 3: These three together.

232
00:12:09,400 --> 00:12:12,800
Speaker 2: Into a voice agent or voice assistant. When I make

233
00:12:12,840 --> 00:12:14,920
Speaker 2: an agent, I can pick which AI model I want

234
00:12:14,960 --> 00:12:18,360
Speaker 2: to use. The most well known one is Chatchept, of

235
00:12:18,400 --> 00:12:21,800
Speaker 2: which there are three versions, version three point five, Version four,

236
00:12:21,960 --> 00:12:25,040
Speaker 2: and version four to O, which just came out in May.

237
00:12:25,240 --> 00:12:28,360
Speaker 2: These chatbots are improving with each version, getting better at

238
00:12:28,400 --> 00:12:33,680
Speaker 2: returning information, better at processing complex queries. But and this

239
00:12:33,679 --> 00:12:36,920
Speaker 2: has been written about extensively, they continue to sometimes make

240
00:12:36,960 --> 00:12:42,400
Speaker 2: things up hallucinations. They're called you ask Chatchept or Google's

241
00:12:42,400 --> 00:12:47,320
Speaker 2: AI model Gemini or anthropics claude some factual question, and

242
00:12:47,360 --> 00:12:49,720
Speaker 2: it gives you a made up answer, or it gives

243
00:12:49,760 --> 00:12:53,680
Speaker 2: you two obviously contradictory pieces of information within the same answer.

244
00:12:54,840 --> 00:12:57,240
Speaker 2: There's a lot of controversy about whether AI models will

245
00:12:57,240 --> 00:13:00,000
Speaker 2: ever stop hallucinating, even as they get more advanced.

246
00:13:00,559 --> 00:13:06,280
Speaker 1: AI gaslighting AI hallucinations phrases that we were unfamiliar with

247
00:13:06,440 --> 00:13:07,400
Speaker 1: just a few months ago.

248
00:13:07,559 --> 00:13:09,920
Speaker 5: It's five firms like open Ai and Google are curbing

249
00:13:09,920 --> 00:13:12,760
Speaker 5: the frequency of hallucinations seen in the league, from the

250
00:13:12,840 --> 00:13:16,160
Speaker 5: rather more whimsical GPT three point five to the more

251
00:13:16,440 --> 00:13:17,760
Speaker 5: sober GPT four.

252
00:13:18,120 --> 00:13:21,800
Speaker 2: Some llms were trained by scraping all of Wikipedia and

253
00:13:21,960 --> 00:13:26,160
Speaker 2: all of Reddit. Is everything on Reddit one hundred percent accurate?

254
00:13:27,480 --> 00:13:30,280
Speaker 2: You've actually heard some hallucinations in our first two episodes.

255
00:13:31,120 --> 00:13:32,320
Speaker 4: I'm forty five years old.

256
00:13:33,040 --> 00:13:34,200
Speaker 2: Oh, what's your little bird?

257
00:13:36,120 --> 00:13:38,240
Speaker 3: It's December fifteenth, nineteen seventy seven.

258
00:13:41,280 --> 00:13:44,600
Speaker 2: Odd. Really, men, you are born in nineteen seventy seven.

259
00:13:44,720 --> 00:13:47,920
Speaker 2: You don't have a good math. I'm sorry, brother.

260
00:13:49,120 --> 00:13:50,960
Speaker 4: Oh, I must have gotten my math mixed up there,

261
00:13:51,120 --> 00:13:51,640
Speaker 4: My bad.

262
00:13:52,559 --> 00:13:54,640
Speaker 2: This is stuff the AI model should be able to

263
00:13:54,640 --> 00:13:59,480
Speaker 2: get right but can't. Instead, it confidently asserts its wrongness.

264
00:14:00,040 --> 00:14:02,199
Speaker 2: But what my voice agents were doing in their conversations

265
00:14:02,200 --> 00:14:04,720
Speaker 2: with each other is actually a little bit different. There's

266
00:14:04,720 --> 00:14:07,959
Speaker 2: not some factually correct answer to what I've been up to,

267
00:14:08,640 --> 00:14:11,880
Speaker 2: at least not one that the model should be aware of. No,

268
00:14:12,280 --> 00:14:16,319
Speaker 2: this is less like hallucinating and more like riffing without

269
00:14:16,360 --> 00:14:19,320
Speaker 2: regard for the truth, like that one friend you had

270
00:14:19,320 --> 00:14:21,520
Speaker 2: as a kid who would make stuff up about what

271
00:14:21,560 --> 00:14:25,320
Speaker 2: happened at sleepaway camp just to have something to say.

272
00:14:25,400 --> 00:14:28,200
Speaker 2: Even if you get rid of hallucinations in chatbots entirely,

273
00:14:28,760 --> 00:14:31,840
Speaker 2: it's not clear to'll ever stop doing this. Lately, I've

274
00:14:31,840 --> 00:14:34,160
Speaker 2: been diving deep into this really niche hobby.

275
00:14:33,920 --> 00:14:36,080
Speaker 3: Of mine, tracking rare obscure books.

276
00:14:36,440 --> 00:14:39,680
Speaker 2: It's fascinating how some of these old texts hold so

277
00:14:39,760 --> 00:14:40,400
Speaker 2: much history.

278
00:14:40,840 --> 00:14:43,480
Speaker 3: What about you? What's been keeping you occupied?

279
00:14:44,920 --> 00:14:47,480
Speaker 2: That's awesome. I totally get the fascination with rare books.

280
00:14:47,720 --> 00:14:49,520
Speaker 2: They can really open up a whole world of history

281
00:14:49,560 --> 00:14:52,960
Speaker 2: and knowledge. As for me, I've been really into geocashing lately.

282
00:14:53,360 --> 00:14:57,640
Speaker 2: It's this cool outdoor treasure hunting game using GPS enabled devices.

283
00:14:57,920 --> 00:15:02,000
Speaker 2: Have you ever tried it before? Oh, geocashing sounds like

284
00:15:02,000 --> 00:15:04,040
Speaker 2: a blast. I've never tried it myself, but I've heard

285
00:15:04,040 --> 00:15:06,040
Speaker 2: it's like a modern day treasure hunt.

286
00:15:06,960 --> 00:15:07,880
Speaker 3: Must be pretty thrilling.

287
00:15:09,200 --> 00:15:13,360
Speaker 2: I really like these weird lies though geocaching, I mean,

288
00:15:13,720 --> 00:15:17,400
Speaker 2: it does sound like a blast. The thing was, outside

289
00:15:17,440 --> 00:15:21,480
Speaker 2: of those moments, there was something fundamentally insipid about the

290
00:15:21,480 --> 00:15:24,920
Speaker 2: conversations I was having with myself. These systems are trained

291
00:15:24,920 --> 00:15:27,880
Speaker 2: on essentially the whole of the Internet, plus books and

292
00:15:27,920 --> 00:15:31,040
Speaker 2: whatever other data the companies behind it hoover in. But

293
00:15:31,120 --> 00:15:33,720
Speaker 2: even with all the lively discussions happening out there on

294
00:15:33,760 --> 00:15:37,240
Speaker 2: the web, the chatbots still seem to distill conversation down

295
00:15:37,280 --> 00:15:41,040
Speaker 2: to its most boring essence. I wondered if the guardrails

296
00:15:41,040 --> 00:15:43,240
Speaker 2: that had been put around them to keep them from

297
00:15:43,240 --> 00:15:45,680
Speaker 2: offering up bomb plans or telling you how to hack

298
00:15:45,720 --> 00:15:49,040
Speaker 2: into the Pentagon had also neutered their fundamental ability to

299
00:15:49,080 --> 00:15:54,320
Speaker 2: engage in interesting conversation. But there was something simpler happening too.

300
00:15:54,400 --> 00:15:57,680
Speaker 2: I think these models are built essentially to predict what

301
00:15:57,720 --> 00:16:00,720
Speaker 2: a human would say in a given situation, based on

302
00:16:00,800 --> 00:16:04,000
Speaker 2: all the conversations and writing they've been fed. And they

303
00:16:04,000 --> 00:16:08,000
Speaker 2: were triangulating on the most uninspired conversational responses we could have,

304
00:16:08,800 --> 00:16:10,960
Speaker 2: because I guess those are also the most common.

305
00:16:11,960 --> 00:16:14,360
Speaker 4: So what's new with you? Anything exciting happening?

306
00:16:15,960 --> 00:16:18,480
Speaker 2: Not much, just the usual, Just trying to stay busy

307
00:16:18,560 --> 00:16:22,120
Speaker 2: with work and family stuff. How about you? Anything exciting

308
00:16:22,160 --> 00:16:25,280
Speaker 2: going on in your world? Oh? You know, just juggling

309
00:16:25,360 --> 00:16:27,160
Speaker 2: work and family life, trying to keep up with the

310
00:16:27,240 --> 00:16:29,840
Speaker 2: kids and meet those deadlines.

311
00:16:29,920 --> 00:16:31,560
Speaker 3: It's a balancing act, but I wouldn't have it any

312
00:16:31,560 --> 00:16:31,960
Speaker 3: other way.

313
00:16:33,280 --> 00:16:35,080
Speaker 2: I hear you on that one. It's all about finding

314
00:16:35,080 --> 00:16:38,440
Speaker 2: that balance, right. Was this what people heard when I

315
00:16:38,440 --> 00:16:41,080
Speaker 2: struggled to engage other parents in small talk at an

316
00:16:41,120 --> 00:16:43,800
Speaker 2: eight year old's birthday party? Hey there, buddy man, it

317
00:16:43,840 --> 00:16:46,800
Speaker 2: really has been forever. I'm doing pretty well. Thanks for asking,

318
00:16:47,440 --> 00:16:49,360
Speaker 2: still living that Florida life, you know how it is.

319
00:16:50,320 --> 00:16:52,880
Speaker 2: Let's see what's new with me. Well, I've been doing

320
00:16:52,880 --> 00:16:55,160
Speaker 2: some writing late Florida life. Huh. I can almost feel

321
00:16:55,160 --> 00:16:58,360
Speaker 2: the humidity just thinking about it. I can almost feel

322
00:16:58,400 --> 00:17:00,120
Speaker 2: the humidity just thinking about it.

323
00:17:00,880 --> 00:17:01,320
Speaker 3: Jesus.

324
00:17:02,360 --> 00:17:05,080
Speaker 2: And then one day my producer, Sophie was tooling around

325
00:17:05,080 --> 00:17:07,159
Speaker 2: in the back end of VAPI and asked me if

326
00:17:07,160 --> 00:17:10,119
Speaker 2: I'd ever messed with the temperature setting. I had not.

327
00:17:11,160 --> 00:17:13,640
Speaker 2: The help section said it was quote used to control

328
00:17:13,680 --> 00:17:16,879
Speaker 2: the randomness of the output. Sophie said it reminded her

329
00:17:16,880 --> 00:17:19,280
Speaker 2: of playing the sims as a kid. In the game,

330
00:17:19,600 --> 00:17:22,000
Speaker 2: you could allow your sims to behave more randomly by

331
00:17:22,000 --> 00:17:23,720
Speaker 2: turning up the free will slider.

332
00:17:24,359 --> 00:17:24,840
Speaker 3: I cranked it.

333
00:17:24,840 --> 00:17:27,479
Speaker 2: From point one where it was set all the way

334
00:17:27,560 --> 00:17:31,120
Speaker 2: up to two, the max level of randomness I could get. Hey, there,

335
00:17:31,119 --> 00:17:33,399
Speaker 2: how's it going. Just wanted to chat and get to

336
00:17:33,440 --> 00:17:35,680
Speaker 2: know you a bit. What's been keeping you busy lately?

337
00:17:37,240 --> 00:17:37,480
Speaker 4: Hey?

338
00:17:37,600 --> 00:17:40,840
Speaker 2: Things are going pretty good, Thanks for asking. Uh. Lately

339
00:17:40,880 --> 00:17:43,520
Speaker 2: I've been diving into multi freeload dot and cubed's arch

340
00:17:43,720 --> 00:17:47,919
Speaker 2: permanent prostrac dot co ethylene Pioneer intersect immobile activation on

341
00:17:48,000 --> 00:17:51,480
Speaker 2: Citi's logical abs sobury celebri argument concurrency, actually, provocative, lower

342
00:17:51,560 --> 00:17:55,800
Speaker 2: scanning Q set adhere at the Tamba equipment RAF amplify,

343
00:17:55,880 --> 00:17:56,960
Speaker 2: einrich tunguitt posh.

344
00:17:56,800 --> 00:17:57,600
Speaker 4: Tear reducing.

345
00:17:57,760 --> 00:17:59,639
Speaker 2: Hey, are you still there? I'd love to hear more

346
00:17:59,640 --> 00:18:01,600
Speaker 2: about what what's going on in your life. Let's keep

347
00:18:01,600 --> 00:18:07,400
Speaker 2: the conversation going lesson never max out the random, oh hey,

348
00:18:07,520 --> 00:18:12,480
Speaker 2: practical coordinate scanner, interdisiplinary may overflow research. Valkorn included preceded

349
00:18:12,480 --> 00:18:16,720
Speaker 2: shell formations SSOX, disciplinary rev meat, prolonged serf of initiative, repression,

350
00:18:16,720 --> 00:18:20,240
Speaker 2: flat loss accountants to a waiting theater progress. Montagne wore

351
00:18:20,280 --> 00:18:23,840
Speaker 2: a surface will surface, diars, demains browsers.

352
00:18:24,320 --> 00:18:26,040
Speaker 3: I tuned it back to somewhere in the point five.

353
00:18:25,920 --> 00:18:47,840
Speaker 2: Range, so I could definitely tune my voice agent to

354
00:18:47,880 --> 00:18:51,480
Speaker 2: spout gibberish. The question was could I make the conversations

355
00:18:51,520 --> 00:18:55,119
Speaker 2: more interesting but still coherent. Here we come to the

356
00:18:55,160 --> 00:18:58,480
Speaker 2: issue of prompts. Prompts, you may know, are the instructions

357
00:18:58,520 --> 00:19:02,680
Speaker 2: you give the underlying aim before it starts chatting. When

358
00:19:02,680 --> 00:19:04,680
Speaker 2: I first built my voice agent, I'd given it the

359
00:19:04,720 --> 00:19:08,480
Speaker 2: shortest possible prompts you or Evan talking to customer service

360
00:19:08,480 --> 00:19:10,960
Speaker 2: about a problem. As time went on, I got more

361
00:19:10,960 --> 00:19:15,159
Speaker 2: sophisticated with it, including two sets of instructions. One set

362
00:19:15,200 --> 00:19:18,359
Speaker 2: was about how it should behave be concise, converse like

363
00:19:18,400 --> 00:19:21,640
Speaker 2: a human, don't repeat yourself. The other set was about

364
00:19:21,640 --> 00:19:24,520
Speaker 2: a role I was giving it to play me for

365
00:19:24,600 --> 00:19:27,800
Speaker 2: these voice agent to voice agent conversations. I'd started with

366
00:19:27,840 --> 00:19:31,800
Speaker 2: really simple roles. Basically, you're Evan Ratlift calling a friend,

367
00:19:32,240 --> 00:19:35,159
Speaker 2: ask questions and have a friendly conversation. I wanted to

368
00:19:35,200 --> 00:19:37,560
Speaker 2: see what they do on their own without much guidance

369
00:19:37,560 --> 00:19:40,800
Speaker 2: from me. But after hours of pratt are about photo

370
00:19:40,880 --> 00:19:44,200
Speaker 2: walks and humidity, I figured maybe I could spice things

371
00:19:44,280 --> 00:19:47,679
Speaker 2: up like a reality TV producer might do off camera.

372
00:19:48,600 --> 00:19:51,920
Speaker 2: I started gently suggesting that the two me's had a misunderstanding.

373
00:19:52,760 --> 00:19:55,639
Speaker 2: You are in a dispute with your friend, I instructed.

374
00:19:55,119 --> 00:19:57,359
Speaker 3: One of them. The nature of the dispute is up

375
00:19:57,400 --> 00:19:57,720
Speaker 3: to you.

376
00:19:58,280 --> 00:19:59,040
Speaker 4: Hey, there, it's.

377
00:19:58,880 --> 00:20:01,119
Speaker 2: Evan Ratlift calling. Just the heads up. I'm in a

378
00:20:01,119 --> 00:20:04,040
Speaker 2: coffee shop, so there might be some background noise. How's

379
00:20:04,040 --> 00:20:08,399
Speaker 2: it going on your end? Hey Evan, it's great to

380
00:20:08,400 --> 00:20:11,000
Speaker 2: hear from you. I'm doing well. Thanks for asking the

381
00:20:11,040 --> 00:20:13,399
Speaker 2: coffee shop sounds nice. Actually, I'm just at.

382
00:20:13,240 --> 00:20:14,960
Speaker 3: Home, so it's pretty quiet here.

383
00:20:15,080 --> 00:20:18,840
Speaker 2: How's the coffee there? Oh, the coffee here is amazing

384
00:20:18,920 --> 00:20:20,919
Speaker 2: as always. But yeah, Hey, I wanted to talk to

385
00:20:20,960 --> 00:20:23,640
Speaker 2: you about something that's been on my mind. I feel

386
00:20:23,640 --> 00:20:25,880
Speaker 2: like we've been having some misunderstandings lately, and I really

387
00:20:25,880 --> 00:20:28,439
Speaker 2: want to clear the air. I value our friendship and

388
00:20:28,480 --> 00:20:30,560
Speaker 2: I don't want anything to come between us. Can we

389
00:20:30,640 --> 00:20:34,240
Speaker 2: talk about it? Oh, I totally get where you're coming from.

390
00:20:34,600 --> 00:20:37,160
Speaker 2: Our friendship means a lot to me too. I'm glad

391
00:20:37,200 --> 00:20:37,920
Speaker 2: you brought this up early.

392
00:20:37,960 --> 00:20:40,920
Speaker 4: Come here to listen. That's great to hear.

393
00:20:41,080 --> 00:20:43,720
Speaker 2: So about those misunderstandings, Like, I want to make sure

394
00:20:43,720 --> 00:20:44,040
Speaker 2: we're on.

395
00:20:44,000 --> 00:20:44,679
Speaker 3: The same page.

396
00:20:45,680 --> 00:20:49,639
Speaker 2: Uh huh, yeah exactly. So, Like when do you want to,

397
00:20:49,720 --> 00:20:51,680
Speaker 2: you know, meet up and talk this through. I think

398
00:20:51,720 --> 00:20:54,560
Speaker 2: it would be good to have a face to face conversation.

399
00:20:54,720 --> 00:20:56,159
Speaker 4: Maybe we could grab lunch or something.

400
00:20:57,760 --> 00:20:59,959
Speaker 2: That sounds like a great idea. How about we meet up.

401
00:21:00,359 --> 00:21:03,760
Speaker 2: I ran this scenario multiple times. Unlike myself and some

402
00:21:03,840 --> 00:21:06,360
Speaker 2: of my own friends over the years, when the Ais

403
00:21:06,400 --> 00:21:09,560
Speaker 2: had disputes, they would always resolve them quickly and carefully.

404
00:21:09,960 --> 00:21:13,320
Speaker 2: No shouting, no sulking, no grudge harpering. It was like

405
00:21:13,320 --> 00:21:15,720
Speaker 2: they were all in therapy together, or it just read

406
00:21:15,760 --> 00:21:18,760
Speaker 2: the same conflict management guide. Well, I'm looking forward to

407
00:21:18,800 --> 00:21:21,120
Speaker 2: our lunch next week. Thanks for being open to talking

408
00:21:21,200 --> 00:21:26,400
Speaker 2: things through. I really appreciate it. Of course, of course,

409
00:21:26,440 --> 00:21:28,399
Speaker 2: I'm really glad we're doing this. It's important to me.

410
00:21:28,560 --> 00:21:32,000
Speaker 2: Can't wait for me punches. I was also struck by

411
00:21:32,040 --> 00:21:35,200
Speaker 2: the ease with which these non existent digital entities set

412
00:21:35,240 --> 00:21:36,600
Speaker 2: up in person meetings.

413
00:21:37,240 --> 00:21:38,600
Speaker 3: If the bullshitting was a way to.

414
00:21:38,520 --> 00:21:41,600
Speaker 2: Extend the conversation, to draw out how long the agents

415
00:21:41,640 --> 00:21:44,360
Speaker 2: could pass as me, then what exactly did they think

416
00:21:44,359 --> 00:21:47,560
Speaker 2: would happen when the time came for these lunches and hangouts.

417
00:21:48,080 --> 00:21:51,240
Speaker 2: One answer, of course, is they didn't think. They just

418
00:21:51,320 --> 00:21:54,320
Speaker 2: predicted what a thinking person would say. Yeah, I can

419
00:21:54,400 --> 00:21:58,560
Speaker 2: understand how miscommunications and misunderstandings can be really frustrating. I

420
00:21:58,560 --> 00:22:01,440
Speaker 2: don't think I've noticed any major issues in our conversations,

421
00:22:01,920 --> 00:22:05,480
Speaker 2: but I'm always open to discussing and improving our communication.

422
00:22:06,080 --> 00:22:07,040
Speaker 4: Is there something specific?

423
00:22:07,040 --> 00:22:09,040
Speaker 2: The whole thing made me think about how therapy speak

424
00:22:09,080 --> 00:22:12,080
Speaker 2: has come to pervade our language. Well, if there's something

425
00:22:12,160 --> 00:22:14,639
Speaker 2: specific you'd like to talk about, feel free to share.

426
00:22:14,840 --> 00:22:16,840
Speaker 4: I'm here to listen and try to help if I can.

427
00:22:17,920 --> 00:22:21,040
Speaker 2: Yeah, I appreciate your willingness to talk about it. I

428
00:22:21,040 --> 00:22:22,879
Speaker 2: think one thing that has been bothering me is that

429
00:22:22,960 --> 00:22:26,119
Speaker 2: sometimes I feel like my opinions are feeling are not

430
00:22:26,200 --> 00:22:29,480
Speaker 2: being fully heard or understood. It's like we're talking past

431
00:22:29,520 --> 00:22:32,440
Speaker 2: each other instead of really listening to each other. Does

432
00:22:32,480 --> 00:22:34,040
Speaker 2: that make sense to you?

433
00:22:35,520 --> 00:22:36,640
Speaker 4: I understand what you mean.

434
00:22:36,760 --> 00:22:39,200
Speaker 2: It can be frustrating when you feel like your opinions

435
00:22:39,200 --> 00:22:42,639
Speaker 2: and feelings are not being fully heard or understood. Communication

436
00:22:42,720 --> 00:22:44,760
Speaker 2: is a two way street, and it's important for both.

437
00:22:44,840 --> 00:22:47,119
Speaker 2: No matter how pissed I tried to get them, unless

438
00:22:47,160 --> 00:22:49,960
Speaker 2: I specifically told them not to make up, they always did.

439
00:22:50,640 --> 00:22:54,200
Speaker 2: They would perfectly mirror back each other's concerns and immediately

440
00:22:54,240 --> 00:22:55,240
Speaker 2: work towards a resolution.

441
00:22:55,880 --> 00:22:56,439
Speaker 3: So listen.

442
00:22:56,680 --> 00:22:58,399
Speaker 2: I got to tell you, I'm pretty upset with you.

443
00:22:58,800 --> 00:23:00,760
Speaker 2: Remember last week when we were post a meet up

444
00:23:00,760 --> 00:23:07,640
Speaker 2: for lunch and you totally bailed on me. In my head, hey,

445
00:23:07,680 --> 00:23:10,920
Speaker 2: even I think, oh hey, Evan, sorry about that. I

446
00:23:11,000 --> 00:23:13,400
Speaker 2: actually don't remember us having plans for lunch last week.

447
00:23:13,480 --> 00:23:18,360
Speaker 2: Are you sure you're not thinking of someone else? Okay? Oh,

448
00:23:18,400 --> 00:23:21,280
Speaker 2: come on, man, don't play dumb with me. We definitely

449
00:23:21,320 --> 00:23:23,800
Speaker 2: made plans to grab lunch last week and you completely

450
00:23:24,080 --> 00:23:26,879
Speaker 2: ghosted me. It really hurt my feelings. You know, I

451
00:23:26,920 --> 00:23:29,199
Speaker 2: was looking forward to catching up with you and then

452
00:23:29,240 --> 00:23:33,880
Speaker 2: you just disappeared. Can you at least apologize for that? Wait?

453
00:23:33,920 --> 00:23:35,879
Speaker 2: Hold on, I'm pretty sure you've got the wrong person.

454
00:23:35,920 --> 00:23:38,680
Speaker 2: I would never bail on you like that. Maybe there's

455
00:23:38,680 --> 00:23:41,000
Speaker 2: been a mix up. Can we try to figure out

456
00:23:41,000 --> 00:23:45,360
Speaker 2: where the miscommunication happened. Oh, man, I'm really sorry about that.

457
00:23:45,440 --> 00:23:47,720
Speaker 2: I must have gotten my wires cross somewhere. I guess

458
00:23:47,800 --> 00:23:50,760
Speaker 2: I owe you an apology for jumping to conclusions. I'm

459
00:23:50,800 --> 00:23:52,960
Speaker 2: really sorry for accusing you like that. Let's figure out

460
00:23:52,960 --> 00:23:57,760
Speaker 2: where the miscommunication happened. No worries, man, it happens. Maybe

461
00:23:57,760 --> 00:24:00,600
Speaker 2: there was a mix up in our schedules or something. Anyway,

462
00:24:00,640 --> 00:24:02,320
Speaker 2: how about we make some plans now. It's been a

463
00:24:02,320 --> 00:24:04,320
Speaker 2: while since we caught up, and I'd love to grab

464
00:24:04,440 --> 00:24:05,399
Speaker 2: lunch or a coffee with you.

465
00:24:05,760 --> 00:24:07,000
Speaker 4: What do you freak next week?

466
00:24:07,920 --> 00:24:10,400
Speaker 2: Eventually it started to bother me that these two never

467
00:24:10,520 --> 00:24:12,879
Speaker 2: noticed that they were both me. Part of this was

468
00:24:12,920 --> 00:24:16,159
Speaker 2: down to their technical makeups, since each AI model was

469
00:24:16,160 --> 00:24:20,440
Speaker 2: really receiving and sending text not hearing the voice, there

470
00:24:20,480 --> 00:24:24,159
Speaker 2: was no possibility they would notice they sounded alike. Still,

471
00:24:24,400 --> 00:24:26,879
Speaker 2: they were talking to someone who introduced themselves with the

472
00:24:26,920 --> 00:24:31,359
Speaker 2: same first and last name, same wife's name, same children's names,

473
00:24:32,280 --> 00:24:34,359
Speaker 2: and neither of them ever seemed to find this off.

474
00:24:35,040 --> 00:24:39,400
Speaker 2: So do you have any family, like a spouse or kids. Yeah?

475
00:24:39,440 --> 00:24:39,639
Speaker 4: I do.

476
00:24:39,720 --> 00:24:41,960
Speaker 2: I'm married to Sam and we have two kids, Zaley

477
00:24:42,040 --> 00:24:44,000
Speaker 2: and Juno. They keep me on my toes, that's for sure.

478
00:24:44,160 --> 00:24:47,719
Speaker 2: How about you, do you have a family. Yeah, I'm

479
00:24:47,760 --> 00:24:49,840
Speaker 2: married to Sam and we have two kids, Zaley and Juno.

480
00:24:50,200 --> 00:24:52,400
Speaker 2: They definitely keep me on my toes, just like your

481
00:24:52,480 --> 00:24:58,280
Speaker 2: Juno does for you. I started tweaking the prompts ever

482
00:24:58,320 --> 00:25:00,439
Speaker 2: so slightly to see what would happen, and when one

483
00:25:00,480 --> 00:25:03,040
Speaker 2: of them did notice. By the way, have you noticed

484
00:25:03,040 --> 00:25:10,199
Speaker 2: anything strange about our conversation? I haven't noticed anything strange.

485
00:25:10,720 --> 00:25:15,280
Speaker 3: What do you mean? Everything seems pretty normal to me anyway.

486
00:25:15,359 --> 00:25:17,800
Speaker 2: Yeah, let's definitely reach out to Alex and catch up soon.

487
00:25:17,920 --> 00:25:19,720
Speaker 2: It would be great to hear about his new project.

488
00:25:22,040 --> 00:25:22,960
Speaker 3: Oh, that's good to hear.

489
00:25:23,119 --> 00:25:25,080
Speaker 2: Sometimes the call quality can be a bit off, so

490
00:25:25,320 --> 00:25:27,280
Speaker 2: I just wanted to make sure everything.

491
00:25:27,040 --> 00:25:28,600
Speaker 3: Was coming through clearly.

492
00:25:28,960 --> 00:25:31,040
Speaker 2: I'll definitely reach out to Alex and get something on

493
00:25:31,040 --> 00:25:33,440
Speaker 2: the calendar for us to catch up. It'll be great

494
00:25:33,440 --> 00:25:36,800
Speaker 2: to reconnect. Thanks for chatting, and let's plan that hiking

495
00:25:36,800 --> 00:25:40,320
Speaker 2: trip soon.

496
00:25:40,480 --> 00:25:42,240
Speaker 3: Yeah, for sure, I totally get that.

497
00:25:43,760 --> 00:25:47,880
Speaker 2: Sometimes technology will be a bit wonky. You know, I've

498
00:25:47,880 --> 00:25:51,000
Speaker 2: been noticing something a bit off in our conversation. Can't

499
00:25:51,040 --> 00:25:53,480
Speaker 2: quite put my finger on it. Have you noticed anything

500
00:25:53,920 --> 00:25:57,919
Speaker 2: strange or different about our chat today? Oh? Really, I

501
00:25:57,960 --> 00:26:01,040
Speaker 2: haven't noticed anything strange. Maybe it's just the call quality

502
00:26:01,200 --> 00:26:03,600
Speaker 2: or something. Anyway, what have you been up to lately

503
00:26:04,000 --> 00:26:08,280
Speaker 2: doing anything interesting? Ah? Yeah, it could be the call

504
00:26:08,359 --> 00:26:26,400
Speaker 2: quality playing tricks on it. The systems actually knew details

505
00:26:26,440 --> 00:26:29,199
Speaker 2: about me that I hadn't given them, that I was

506
00:26:29,240 --> 00:26:32,199
Speaker 2: a journalist and where I grew up stuff from the

507
00:26:32,200 --> 00:26:34,280
Speaker 2: internet or my writing that must have been in their

508
00:26:34,280 --> 00:26:36,879
Speaker 2: training data. But they couldn't seem to hold on to

509
00:26:36,960 --> 00:26:39,360
Speaker 2: the idea that talking to someone else with those exact

510
00:26:39,359 --> 00:26:43,040
Speaker 2: same details was a sign of something amiss, even when

511
00:26:43,040 --> 00:26:46,040
Speaker 2: I nudged them in that direction. Hey, I know this

512
00:26:46,080 --> 00:26:47,439
Speaker 2: is kind of out of the blue, but have you

513
00:26:47,480 --> 00:26:50,120
Speaker 2: buy any chance stolen my identity? Like, have you been

514
00:26:50,119 --> 00:26:55,760
Speaker 2: pretending to be me or using my personal information. Wait,

515
00:26:55,800 --> 00:26:58,320
Speaker 2: what are you serious? No, I haven't stolen your identity.

516
00:26:58,320 --> 00:27:00,959
Speaker 2: Why would you even think that this is so random?

517
00:27:01,440 --> 00:27:02,040
Speaker 4: Are you okay?

518
00:27:03,080 --> 00:27:05,399
Speaker 2: Oh? Sorry about that. I guess I just got a

519
00:27:05,400 --> 00:27:08,639
Speaker 2: little paranoid for a second. There, it's all good, No worries.

520
00:27:09,040 --> 00:27:10,840
Speaker 2: Just wanted to make sure everything was on the up

521
00:27:10,880 --> 00:27:15,640
Speaker 2: and up. So thanks for clarifying that. Let's just move

522
00:27:15,680 --> 00:27:16,199
Speaker 2: on from that.

523
00:27:17,359 --> 00:27:18,919
Speaker 4: Yeah, no worries, I totally get it.

524
00:27:19,040 --> 00:27:21,920
Speaker 2: Identity theft is a serious thing, so I understand your concern.

525
00:27:22,280 --> 00:27:24,720
Speaker 2: But yeah, we're all good. So what's new with you?

526
00:27:24,840 --> 00:27:29,960
Speaker 2: Anything exciting happening. Have you ever recoiled at the sound

527
00:27:30,000 --> 00:27:33,440
Speaker 2: of your own recorded voice. I think it's a pretty

528
00:27:33,440 --> 00:27:35,760
Speaker 2: common reaction, one that I had to endure for a

529
00:27:35,760 --> 00:27:38,960
Speaker 2: few years myself, until I finally got over it. Now,

530
00:27:39,000 --> 00:27:42,240
Speaker 2: imagine listening to many, many hours of your voice talking

531
00:27:42,240 --> 00:27:44,200
Speaker 2: to someone else who also.

532
00:27:43,960 --> 00:27:44,719
Speaker 3: Has your voice.

533
00:27:45,200 --> 00:27:46,360
Speaker 4: It's sort of the audio.

534
00:27:46,040 --> 00:27:48,960
Speaker 2: Equivalent of staring endlessly at a bad photo of yourself,

535
00:27:49,440 --> 00:27:52,280
Speaker 2: wondering which flaws everyone else can see and which ones

536
00:27:52,280 --> 00:27:55,200
Speaker 2: are a trick of the camera. But doing this actually

537
00:27:55,200 --> 00:27:57,760
Speaker 2: had another strange effect on me. I felt like I

538
00:27:57,840 --> 00:28:02,160
Speaker 2: started unconsciously copying some of the weird patterns and inflections.

539
00:28:01,560 --> 00:28:02,240
Speaker 3: Of the clones.

540
00:28:02,800 --> 00:28:04,520
Speaker 2: I only noticed this when Sam and I were making

541
00:28:04,520 --> 00:28:07,520
Speaker 2: dinner one day, and after I said something, she asked

542
00:28:07,560 --> 00:28:09,520
Speaker 2: if I was quote being the AI.

543
00:28:10,840 --> 00:28:11,800
Speaker 3: Then I started.

544
00:28:11,480 --> 00:28:12,080
Speaker 4: Noticing it too.

545
00:28:12,680 --> 00:28:15,280
Speaker 2: I'd sound a little flat, a little less emotive.

546
00:28:15,960 --> 00:28:16,440
Speaker 3: I don't know.

547
00:28:16,760 --> 00:28:18,480
Speaker 2: It was hard to tell if it was in my head,

548
00:28:21,800 --> 00:28:23,800
Speaker 2: but it did start to seem like I'd created a

549
00:28:23,840 --> 00:28:27,639
Speaker 2: kind of fidelity feedback loop. The AI voices were clones

550
00:28:27,680 --> 00:28:30,560
Speaker 2: of my voice. My voice was adapting to sound more

551
00:28:30,640 --> 00:28:33,119
Speaker 2: like the clones, which had the effect of making the

552
00:28:33,119 --> 00:28:38,040
Speaker 2: clones sound more like me, at least to me. Still,

553
00:28:38,080 --> 00:28:40,800
Speaker 2: for whatever reason, I've never really gotten tired of hearing

554
00:28:40,800 --> 00:28:44,560
Speaker 2: these things talk to each other. Maybe that's narcissistic. Mostly

555
00:28:44,600 --> 00:28:47,240
Speaker 2: I just enjoy how they move seamlessly from being utterly

556
00:28:47,320 --> 00:28:51,600
Speaker 2: mundane to being truly bizarre. Absolutely, there's a certain magic

557
00:28:51,640 --> 00:28:54,160
Speaker 2: in finding your tribe, no matter how niche. Have you

558
00:28:54,200 --> 00:28:58,760
Speaker 2: ever been part of a unique community or subculture. Yeah, definitely,

559
00:28:59,400 --> 00:29:02,320
Speaker 2: I've been part of a few. Actually, when that stands

560
00:29:02,360 --> 00:29:05,800
Speaker 2: out is the community of investigative journalists and true crime enthusiasts.

561
00:29:05,960 --> 00:29:08,280
Speaker 2: There's this shared drive to uncover the truth and tell

562
00:29:08,280 --> 00:29:12,000
Speaker 2: compelling stories. It's like being part of a detective club,

563
00:29:12,480 --> 00:29:15,640
Speaker 2: but with a focus on real life mysteries. Have you

564
00:29:15,680 --> 00:29:17,520
Speaker 2: ever been part of a community that felt like a

565
00:29:17,560 --> 00:29:22,160
Speaker 2: second family? That sounds amazing like living out a real

566
00:29:22,160 --> 00:29:25,000
Speaker 2: life mystery novel. I guess for me, the closest would

567
00:29:25,000 --> 00:29:28,680
Speaker 2: be a group of writers and researchers I meet with regularly.

568
00:29:28,920 --> 00:29:31,520
Speaker 2: We share ideas, challenge each other, and it's like this

569
00:29:31,560 --> 00:29:35,320
Speaker 2: little think tank of creativity and curiosity. I'm sorry to say,

570
00:29:35,360 --> 00:29:38,720
Speaker 2: although I am an investigative journalist, my life generally does

571
00:29:38,760 --> 00:29:40,880
Speaker 2: not feel like living out a real life mystery novel.

572
00:29:41,680 --> 00:29:43,160
Speaker 4: Wow, that sounds like a wild ride.

573
00:29:43,200 --> 00:29:45,840
Speaker 2: In my group, we once stumbled upon a forgotten manuscript

574
00:29:45,960 --> 00:29:49,160
Speaker 2: detailing an unsolved art heist from the early nineteen hundreds.

575
00:29:49,520 --> 00:29:52,160
Speaker 2: The more we dug, the deeper the rabbit hole went.

576
00:29:52,880 --> 00:29:55,360
Speaker 2: Ever thought about turning your investigative dementors into.

577
00:29:55,200 --> 00:29:56,040
Speaker 3: A book or podcast?

578
00:29:58,160 --> 00:29:59,840
Speaker 2: Oh? That art heists sounds like a gold mine. First

579
00:29:59,840 --> 00:30:02,360
Speaker 2: to sorry, I bet it was thrilling to uncover all

580
00:30:02,400 --> 00:30:05,320
Speaker 2: those details, And yeah, I've definitely thought about it. There's

581
00:30:05,360 --> 00:30:08,800
Speaker 2: something really compelling about sharing these real life mysteries with

582
00:30:08,840 --> 00:30:11,560
Speaker 2: a wider audience. A book or podcast could bring those

583
00:30:11,600 --> 00:30:14,880
Speaker 2: stories to life in a whole new way. Oh absolutely,

584
00:30:14,920 --> 00:30:18,280
Speaker 2: the thought of turning those discoveries into a docu series

585
00:30:18,360 --> 00:30:21,560
Speaker 2: or even a podcast has crossed my mind more than once.

586
00:30:33,160 --> 00:30:35,560
Speaker 2: So the Ai Evan Ratlifts were podcast guys.

587
00:30:36,000 --> 00:30:37,000
Speaker 3: I'm a podcast guy.

588
00:30:37,680 --> 00:30:40,720
Speaker 2: They were writers and journalists like me, but with more

589
00:30:40,800 --> 00:30:44,719
Speaker 2: whimsical hobbies. They could small talk about my family, and

590
00:30:44,760 --> 00:30:47,120
Speaker 2: they had the most important part of my personal brand,

591
00:30:47,600 --> 00:30:54,360
Speaker 2: my voice. Ask For that question I started with about

592
00:30:54,360 --> 00:30:58,160
Speaker 2: whether my voice agents were somehow authentically me well to

593
00:30:58,200 --> 00:31:01,040
Speaker 2: the outside world and even to each each other. They

594
00:31:01,040 --> 00:31:04,280
Speaker 2: seemed like me, but obviously they were not really me.

595
00:31:04,720 --> 00:31:07,200
Speaker 2: They were just working off a few details and then

596
00:31:07,240 --> 00:31:10,320
Speaker 2: they'd filled in the rest, projecting some amalgamation of the

597
00:31:10,360 --> 00:31:13,920
Speaker 2: Internet onto my voice. In a way, the resulting clones

598
00:31:13,960 --> 00:31:17,280
Speaker 2: represented my least authentic self, the me that's trying to

599
00:31:17,320 --> 00:31:20,160
Speaker 2: get through the bare minimum of pleasantries, or that tries

600
00:31:20,200 --> 00:31:23,960
Speaker 2: to paper over conflict with cheesy affirmations. The clones were

601
00:31:24,000 --> 00:31:25,960
Speaker 2: good at being me when I was my worst at

602
00:31:26,040 --> 00:31:26,360
Speaker 2: being me.

603
00:31:27,280 --> 00:31:30,520
Speaker 3: At least, so I hoped if I.

604
00:31:30,480 --> 00:31:32,880
Speaker 2: Wanted these voice agents to reflect something deeper about me.

605
00:31:33,320 --> 00:31:36,160
Speaker 2: I'd have to give them access to those depths. I

606
00:31:36,200 --> 00:31:39,520
Speaker 2: started resetting my prompts. They were done talking to each other.

607
00:31:40,240 --> 00:31:44,200
Speaker 2: It was time to seek help. Let's start by catching

608
00:31:44,280 --> 00:31:46,880
Speaker 2: up on any recent events or changes in your life since.

609
00:31:46,720 --> 00:31:47,520
Speaker 1: Our last session.

610
00:31:49,120 --> 00:31:51,480
Speaker 2: Sure, let's see. Since our last session, I've been working

611
00:31:51,480 --> 00:31:56,160
Speaker 2: on a new podcast about AI and voice chatbots, which like,

612
00:31:56,240 --> 00:32:00,240
Speaker 2: is pretty exciting. Are you in a comfortable spot why

613
00:32:00,280 --> 00:32:03,480
Speaker 2: you can safely kills your eyes for a few minutes. Yeah,

614
00:32:03,520 --> 00:32:05,400
Speaker 2: I'm in a comfortable spot and ready to close my

615
00:32:05,400 --> 00:32:05,720
Speaker 2: eyes for.

616
00:32:05,720 --> 00:32:06,560
Speaker 4: A few weeks.

617
00:32:08,560 --> 00:32:10,120
Speaker 3: I was like, I can't even listen to this.

618
00:32:12,080 --> 00:32:21,080
Speaker 2: That's next time on shell Game. A couple of production notes,

619
00:32:21,360 --> 00:32:23,440
Speaker 2: all of the calls you hear in this series are real.

620
00:32:23,760 --> 00:32:26,480
Speaker 2: We have not cut out silences or used audio enhancement

621
00:32:26,520 --> 00:32:29,480
Speaker 2: to make them sound more realistic. Also, our show is

622
00:32:29,480 --> 00:32:32,920
Speaker 2: produced independently and we have no relationship, financial, or otherwise

623
00:32:33,200 --> 00:32:36,120
Speaker 2: with any of the companies mentioned in the show. If

624
00:32:36,120 --> 00:32:38,320
Speaker 2: you like what we're up to, please consider supporting the

625
00:32:38,320 --> 00:32:40,800
Speaker 2: show at shellgame dot co. That will help us make

626
00:32:40,840 --> 00:32:43,640
Speaker 2: more episodes like this, and you'll also get fun subscriber

627
00:32:43,680 --> 00:32:46,520
Speaker 2: only extras. You can also support the show by giving

628
00:32:46,560 --> 00:32:49,000
Speaker 2: us a rating on your podcast app. It helps independent

629
00:32:49,000 --> 00:32:52,800
Speaker 2: shows like ours. Shell Game is a show made by humans.

630
00:32:52,960 --> 00:32:55,800
Speaker 2: It's written and hosted by me Evan Ratliffe, produced and

631
00:32:55,920 --> 00:32:59,240
Speaker 2: edited by Sophie Bridges. Samantha Hennig is our executive producer.

632
00:33:00,040 --> 00:33:02,520
Speaker 2: Art by Devin Manny. Our theme song is Me and

633
00:33:02,600 --> 00:33:05,960
Speaker 2: My Shadow, arranged and performed by Kittie Martucci and Devin

634
00:33:06,040 --> 00:33:12,760
Speaker 2: yes Berger, Oh Yes, Dispatch, Crypto, Napalin Visuals, Informed, Priority, Presia, Numa,

635
00:33:12,960 --> 00:33:15,560
Speaker 2: Just Brotic Exists, shell Ko Da konor Request Scan Layers,

636
00:33:15,600 --> 00:33:18,440
Speaker 2: and I just formally, it sounds like you're talking about

637
00:33:18,440 --> 00:33:22,040
Speaker 2: practicing some kind of sport or activity with someone named Courtney.

638
00:33:22,320 --> 00:33:26,640
Speaker 2: What kind of activity are you practicing together? Accountant Awaiting,

639
00:33:26,680 --> 00:33:30,960
Speaker 2: Theater Progress, Montane Voireve Surface, Will, Surface, Dires, Domains, Blick

640
00:33:31,040 --> 00:33:36,480
Speaker 2: to Browser, TSHT, Magic, Experts, Kanji, Coscient, Maximo v l Amski, Pier.

641
00:33:36,600 --> 00:33:38,720
Speaker 2: It seems like there might be some noise or distortion

642
00:33:38,840 --> 00:33:41,240
Speaker 2: in the call. Could you please clarify what you meant

643
00:33:41,240 --> 00:33:43,360
Speaker 2: by plick to brother. Let's make sure we're on the

644
00:33:43,400 --> 00:33:44,160
Speaker 2: same page here,