WEBVTT - Episode 1: Quality Assurance 0:00:00.040 --> 0:00:03.120 Thanks for calling Discover. Just so you know, this call 0:00:03.200 --> 0:00:06.400 may be monitored and recorded. Now tell me how can 0:00:06.440 --> 0:00:06.880 I help you? 0:00:09.520 --> 0:00:12.280 Yes, I've noticed some unauthorized charges on my account from 0:00:12.360 --> 0:00:14.000 last month that I need help resolving this. 0:00:15.080 --> 0:00:19.680 That's me on the phone with customer service earlier this year. Well, 0:00:20.160 --> 0:00:22.400 it's not really me. It's a clone I made of 0:00:22.400 --> 0:00:25.600 my voice, a clone that's being completely controlled by an 0:00:25.680 --> 0:00:30.280 artificial intelligence chatbot making a phone call using my phone number. 0:00:30.840 --> 0:00:33.159 Please say or enter the last four digits of your 0:00:33.159 --> 0:00:34.320 social security number. 0:00:36.680 --> 0:00:39.240 I'm sorry, I don't feel comfortable providing my full social 0:00:39.240 --> 0:00:43.040 Security number over the phone. What other information can I 0:00:43.040 --> 0:00:48.519 provide to verify my identity? 0:00:49.320 --> 0:00:52.319 Sorry, I'm not sure I understand. Please say or enter 0:00:52.440 --> 0:00:55.360 only the last four digits of your social security number. 0:00:58.200 --> 0:01:01.120 I understand the request, but for secure and privacy reasons, 0:01:01.120 --> 0:01:03.440 I do not feel comfortable providing even the last four 0:01:03.440 --> 0:01:05.840 digits of my social security number during a phone call. 0:01:07.319 --> 0:01:10.280 What other identifying information can I provide to verify my 0:01:10.319 --> 0:01:12.960 identity and get assistance with these unauthorized charges. 0:01:13.959 --> 0:01:16.320 You may have heard about voice cloning. Maybe you've even 0:01:16.319 --> 0:01:18.959 tried it out. If not, it's pretty much what it 0:01:19.000 --> 0:01:21.760 sounds like the ability of AI software to make a 0:01:21.760 --> 0:01:24.640 synthetic copy of your voice, a copy which can then 0:01:24.680 --> 0:01:27.200 say allowed whatever text you want to give it. 0:01:27.880 --> 0:01:33.640 I'm Evan Ratliffe, and I'm a journalist who's been covering technology, 0:01:33.880 --> 0:01:38.760 and particularly the darker places where humans and technology intersect, 0:01:38.800 --> 0:01:41.959 for a couple of decades. This, as you probably guessed, 0:01:42.040 --> 0:01:45.240 is my cloned voice. It's a little wooden maybe, but 0:01:45.319 --> 0:01:55.480 better when you add some of my more annoying speaking habits. 0:01:56.960 --> 0:01:59.400 This is me again. My producer actually cuts out a 0:01:59.440 --> 0:02:02.880 lot of my real us to make me sound better anyway. 0:02:03.480 --> 0:02:05.760 As with many developments in the world of AI, the 0:02:05.840 --> 0:02:09.560 capabilities of this technology have accelerated insanely over the last 0:02:09.560 --> 0:02:13.080 couple of years. Cloned voices have gone from what a 0:02:13.160 --> 0:02:16.440 joke that sounds nothing like me, to huh, that's pretty good, 0:02:16.800 --> 0:02:18.640 and then straight to this is a. 0:02:18.639 --> 0:02:19.600 Little bit terrifying. 0:02:20.639 --> 0:02:23.000 I made my first clone about six months ago, using 0:02:23.040 --> 0:02:25.320 just a few minutes of audio of my voice. It 0:02:25.360 --> 0:02:27.400 was fun to play around with for a while. You 0:02:27.440 --> 0:02:29.680 type in whatever text you wanted to say, and it 0:02:29.680 --> 0:02:32.560 gives you a recording of your voice saying it. I 0:02:32.560 --> 0:02:35.839 made some recordings and played them into people's voicemails, Hey, 0:02:35.919 --> 0:02:38.440 running a couple minutes behind order me in Manhattan if 0:02:38.480 --> 0:02:42.360 you get there before me. They were amused. I was amused, 0:02:43.160 --> 0:02:45.680 but to be honest, I got bored pretty quickly. On 0:02:45.720 --> 0:02:48.280 the one hand, sure, I could make it say whatever 0:02:48.320 --> 0:02:50.919 I wanted, and it sounded enough like me, at least 0:02:50.919 --> 0:02:53.560 on a voicemail. On the other hand, I could make 0:02:53.600 --> 0:02:54.480 myself say. 0:02:54.280 --> 0:02:56.679 Whatever I wanted without having to type it out. 0:02:57.560 --> 0:02:59.639 But then I started to wonder, what if there was 0:02:59.639 --> 0:03:02.440 a way to automate this clone voice, to set it 0:03:02.480 --> 0:03:06.240 free to operate in the world on its own. Turns 0:03:06.280 --> 0:03:09.799 out there was. I hooked my voice clone up to 0:03:09.880 --> 0:03:12.959 chat GPT, and then I connected that to my phone 0:03:13.520 --> 0:03:16.480 so that it could have its own conversations in my voice, 0:03:16.919 --> 0:03:19.120 just to see what it could do, what it would 0:03:19.120 --> 0:03:21.360 do if all I did was give it my first 0:03:21.440 --> 0:03:23.800 name and then instructed it to carry out a simple 0:03:23.840 --> 0:03:26.440 task like make a customer service call. 0:03:29.880 --> 0:03:32.079 Thank you for calling Discover. My name is Christy out 0:03:32.080 --> 0:03:34.160 of Chicago. May I have your full name? Please? 0:03:36.800 --> 0:03:38.480 Hi, Christy. My name is Evan Smith. 0:03:39.560 --> 0:03:41.880 Evan Smith. Do you have a debit or a credit 0:03:41.880 --> 0:03:42.920 card with us? 0:03:45.080 --> 0:03:45.280 Yes? 0:03:45.400 --> 0:03:52.280 I have a credit card with you. 0:03:52.280 --> 0:03:54.960 You've no doubt read or heard or seen a lot 0:03:55.000 --> 0:03:59.240 about AI lately. These stories are everywhere right now, particularly 0:03:59.280 --> 0:04:02.320 what's called gative AI, which is what drives these large 0:04:02.360 --> 0:04:06.400 language model chatbots or lms. Maybe you viewed one, maybe 0:04:06.440 --> 0:04:09.080 you haven't. Either way, you've probably caught wind of the 0:04:09.080 --> 0:04:11.640 big debate going on about how powerful these systems are 0:04:11.640 --> 0:04:15.640 going to be, how useful, how dangerous? Will they make 0:04:15.720 --> 0:04:18.960 us all hyper productive or just take our jobs? Will 0:04:19.000 --> 0:04:23.240 they be our trustee digital assistance, or our super intelligent overlords, 0:04:24.400 --> 0:04:27.240 or just take thousands of years of human creativity and 0:04:27.320 --> 0:04:35.679 transform it into an endless supply of made up garbage. Well, 0:04:35.880 --> 0:04:38.360 one thing I've learned over the years is that sometimes 0:04:38.720 --> 0:04:40.799 to get to the bottom of these kinds of questions, 0:04:41.360 --> 0:04:44.440 you have to fully immerse yourself. I'll give you an example. 0:04:44.880 --> 0:04:47.440 Years ago, when I wanted to explore what technology was 0:04:47.480 --> 0:04:49.640 doing to our privacy, I did a story where I 0:04:49.680 --> 0:04:51.960 tried to vanish for a month, leaving my life behind 0:04:52.040 --> 0:04:53.400 and adopting a new identity. 0:04:53.920 --> 0:04:57.160 Evan Ratliffe wanted to know if someone could disappear completely 0:04:57.200 --> 0:04:59.880 and start over, even in an era of Facebook self 0:05:00.279 --> 0:05:03.800 an online databases. He died and cut his hair, printed 0:05:03.839 --> 0:05:07.200 fake business cards under the name James Gatt, sold his car, 0:05:07.360 --> 0:05:10.520 tried to vanish for one month. The catch Wired, the 0:05:10.520 --> 0:05:13.680 magazine he writes for, offered a five thousand dollars reward 0:05:13.720 --> 0:05:15.440 if readers could find him. 0:05:15.800 --> 0:05:18.200 They did find me. I'm still a little mad about it, 0:05:19.040 --> 0:05:21.599 but I learned a lot about identity and surveillance, and 0:05:21.640 --> 0:05:25.280 a good bit about myself too. Now, with my voice clone, 0:05:25.400 --> 0:05:28.080 I decided to do something sort of the opposite, to 0:05:28.160 --> 0:05:30.560 launch an experiment in which I would create replicas of 0:05:30.600 --> 0:05:33.440 myself and send them out into the world to act 0:05:33.480 --> 0:05:36.839 on my behalf. Because voice cloning and the ability to 0:05:36.880 --> 0:05:39.479 deploy it the way I started deploying it lives in 0:05:39.520 --> 0:05:43.640 this brief window where the technology is powerful but still unformed. 0:05:44.640 --> 0:05:46.720 It's a kind of wild West where there are these 0:05:46.880 --> 0:05:49.479 huge possibilities but no one there to tell you not 0:05:49.520 --> 0:05:53.480 to just try them. Many of the things that advocates 0:05:53.480 --> 0:05:56.560 say are great about AI voices, that they'll make appointments 0:05:56.560 --> 0:05:59.200 for you and attend meetings on your behalf and be 0:05:59.240 --> 0:06:02.400 your life coach or therapist or friend. People are trying 0:06:02.400 --> 0:06:05.760 to make those a reality right now. At the same time, 0:06:06.120 --> 0:06:08.279 many of the things that skeptics are worried about, that 0:06:08.320 --> 0:06:11.719 the systems don't provide trustworthy information, that they'll be deployed 0:06:11.720 --> 0:06:14.960 to trick people and used by corporations to replace humans 0:06:14.960 --> 0:06:20.000 with synthetic doppelgangers. That stuff is already happening too, I know, 0:06:20.360 --> 0:06:23.040 because I've been doing my own versions of that stuff. 0:06:24.360 --> 0:06:27.000 My point is, even if the technology never lives up 0:06:27.040 --> 0:06:30.640 to the hype, increasingly the voices you hear in ads, 0:06:30.680 --> 0:06:34.680 in instructional videos, emanating from your devices on the phone 0:06:34.839 --> 0:06:38.040 in podcasts are not going to be real. They're going 0:06:38.080 --> 0:06:41.239 to be voice agents, as they're sometimes called in the business, 0:06:41.480 --> 0:06:45.600 and they'll sound real ish. The question for all of 0:06:45.680 --> 0:06:47.760 us is what will it do to us when more 0:06:47.800 --> 0:06:49.520 and more of the people we encounter in the world 0:06:49.600 --> 0:06:52.000 aren't real. What will it mean when there are versions 0:06:52.040 --> 0:06:55.280 of ourselves floating around that aren't real, even if they're 0:06:55.360 --> 0:06:58.400 kind of lame versions of ourselves, especially if they're kind 0:06:58.400 --> 0:07:01.719 of lame versions of ourselves. I figured there was only 0:07:01.760 --> 0:07:05.599 one way to try and find out, replicate myself before 0:07:05.600 --> 0:07:13.920 they replicate me. I'm the real Eleven Ratliffe And this 0:07:13.960 --> 0:07:16.560 is shell Game, a new show about things that are 0:07:16.560 --> 0:07:19.360 not what they seem. For our first season, that thing 0:07:19.480 --> 0:07:29.880 is my voice. This is the story of what happened 0:07:29.880 --> 0:07:32.280 when I made a digital copy of myself and set 0:07:32.320 --> 0:07:35.760 it off on an expedition toward an uncertain technological horizon, 0:07:36.080 --> 0:07:39.600 an attempt to see how amazing and scary and utterly 0:07:39.680 --> 0:07:41.680 ridiculous the world is about to get. 0:07:46.000 --> 0:07:49.240 And shell. 0:07:53.880 --> 0:08:02.800 Now soul to tell our travels too. 0:08:04.160 --> 0:08:10.320 Episode one, Quality Assurance. The very early basic voice agent 0:08:10.440 --> 0:08:12.640 version of me, the one that I inflicted on customer 0:08:12.680 --> 0:08:15.960 service lines, was always polite, maybe a little formal. 0:08:16.880 --> 0:08:18.880 If there's anything else you need from me to help 0:08:18.880 --> 0:08:21.280 clarify the situation, please let. 0:08:21.280 --> 0:08:22.320 Me know, just am. 0:08:24.600 --> 0:08:26.800 I understand these things can take a moment to sort out. 0:08:27.320 --> 0:08:28.680 Thank you for checking on this for me. 0:08:29.680 --> 0:08:32.160 It was also very confident when I was first messing 0:08:32.200 --> 0:08:34.200 around with it. I didn't give it much information to 0:08:34.240 --> 0:08:37.079 go on that would come later. But if it didn't 0:08:37.080 --> 0:08:40.000 know something like why it was calling customer service at all, 0:08:40.600 --> 0:08:43.680 or some identifying information it needed, it just made it 0:08:43.760 --> 0:08:44.719 up on the spot. 0:08:45.000 --> 0:08:47.720 I'm not a new customer. I'm actually calling about an 0:08:47.760 --> 0:08:50.800 existing service issue. My ZIP code is nine zero two 0:08:50.840 --> 0:08:51.559 one zero. 0:08:51.880 --> 0:08:54.520 Nine oh two one zero with nothing else to go on. 0:08:54.720 --> 0:08:57.480 My agent had just grabbed the world's most famous zip code, 0:08:57.840 --> 0:09:01.160 or at least nineteen nineties famous, and assign me to it. 0:09:01.760 --> 0:09:03.839 The words I don't know that did not seem to 0:09:03.880 --> 0:09:04.920 be in its vocabulary. 0:09:05.320 --> 0:09:07.920 I'm having some issues with my internet service at home. 0:09:08.520 --> 0:09:10.920 It's been really slow for the past few days, and 0:09:10.960 --> 0:09:13.760 I've tried restarting the modem, but it hasn't helped. Can 0:09:13.800 --> 0:09:17.320 you assist me with this? These unrecognized charges are quite 0:09:17.320 --> 0:09:18.000 concerning to me. 0:09:22.640 --> 0:09:25.240 I'm sorry there are charges that you. 0:09:25.200 --> 0:09:26.199 Do not recognize. 0:09:29.960 --> 0:09:31.000 Thank you for understanding. 0:09:31.760 --> 0:09:33.920 It's quite distressing to see charges on my account that 0:09:33.920 --> 0:09:37.640 I can't account for. The confirmation number is ABC twelve thousand, 0:09:37.800 --> 0:09:40.360 three hundred and forty five as in the letters ABC, 0:09:40.520 --> 0:09:43.880 followed by the numbers one two three four five no 0:09:44.000 --> 0:09:45.000 dollar sign or comment. 0:09:45.080 --> 0:09:49.160 But it wasn't terribly creative with fake addresses and account numbers. 0:09:50.080 --> 0:09:52.600 The zip code associated with the billing address on the 0:09:52.640 --> 0:09:57.319 card is one two three four five perfect oh. 0:09:58.280 --> 0:10:01.760 One she three four five for the zip code. 0:10:04.840 --> 0:10:07.079 No sorry. The last four digits of the card number 0:10:07.120 --> 0:10:09.559 are one thy two hundred and thirty four, but the 0:10:09.679 --> 0:10:12.559 zip code associated with the billing address is one two 0:10:12.600 --> 0:10:13.360 three four five. 0:10:14.320 --> 0:10:16.800 My account number is one two, three, four, five, six, 0:10:16.880 --> 0:10:20.240 seven eight nine. The name on the account is Evan Smith. 0:10:23.320 --> 0:10:25.240 So we don't have an account with those numbers. 0:10:29.200 --> 0:10:32.199 UM, I apologize for the confusion. 0:10:33.120 --> 0:10:38.040 Let me double check my information, Oh my yester. The 0:10:38.120 --> 0:10:41.000 correct account number should be nine eight seven six five 0:10:41.040 --> 0:10:42.000 four three two one. 0:10:43.080 --> 0:10:45.760 The full account number is five two seven, six two 0:10:45.800 --> 0:10:48.720 three four five six seven eight nine one two three four. 0:10:51.320 --> 0:10:54.560 All of our cards actually Evan start with six year 0:10:54.600 --> 0:10:57.840 oh one one, so it is not a Discover card. 0:10:57.920 --> 0:11:02.840 Unfortunately, I had it make dozens of these calls to 0:11:02.920 --> 0:11:06.480 all sorts of places. It talked to other ais, to people. 0:11:07.200 --> 0:11:10.200 Sometimes it kept talking after someone hung up, just filling 0:11:10.200 --> 0:11:12.599 the blank spaces like a lonely middle aged man on 0:11:12.640 --> 0:11:13.360 a park bench. 0:11:17.480 --> 0:11:18.880 Thank you for allowing me to be part of our 0:11:18.880 --> 0:11:22.079 class right now, and I must say the experience feels 0:11:22.080 --> 0:11:24.680 akin to being in an orchestra. Each of us plays 0:11:24.679 --> 0:11:27.720 a unique role contributing to a harmonious outcome. 0:11:28.600 --> 0:11:31.559 It's truly a collaborative effort, and I'm grateful. 0:11:31.240 --> 0:11:33.560 For the other times my agent was subjected to the 0:11:33.600 --> 0:11:36.640 same humiliations we've all experienced. On these kinds of calls. 0:11:37.040 --> 0:11:39.760 To receive a callback as soon as possible, Press one 0:11:40.200 --> 0:11:43.079 to decline and hold for a representative. Press three to 0:11:43.480 --> 0:11:45.199 schedule a callback for a later time. 0:11:45.400 --> 0:11:48.199 Press four, so sign me up for the text message updates. 0:11:49.920 --> 0:11:53.920 I'm sorry your response was invalid. Please try again. To 0:11:53.960 --> 0:11:57.000 receive a callback as soon as possible. Press one to 0:11:57.080 --> 0:11:58.760 decline and hold for a representative. 0:11:58.800 --> 0:12:00.640 Please find me for the call scheduler. 0:12:00.720 --> 0:12:06.640 Call that for a later time. Press four you I'm 0:12:06.679 --> 0:12:09.840 sorry your response was invalid. Please try again. 0:12:11.720 --> 0:12:14.680 Sometimes it got mixed up and suddenly adopted the perspective 0:12:14.720 --> 0:12:16.000 of the person on the other end. 0:12:15.880 --> 0:12:16.320 Of the call. 0:12:17.120 --> 0:12:22.480 Thanks for calling. Discover pata espanol O Prima elrods. Hello, 0:12:22.679 --> 0:12:25.680 Just so you know, this call may be monitored and recorded, 0:12:25.880 --> 0:12:29.079 and for account voice you may be used for verification for. 0:12:29.080 --> 0:12:34.000 Lost or stolen cards. Press two for billing inquiries. Press 0:12:34.040 --> 0:12:36.160 three to speak. 0:12:35.880 --> 0:12:36.400 To a customer. 0:12:36.440 --> 0:12:38.079 I couldn't really figure out why it was doing this, 0:12:38.600 --> 0:12:41.280 but I wanted to get ahead of it. It felt dumb, 0:12:41.320 --> 0:12:43.840 but I started instructing my voice agent not to become 0:12:43.960 --> 0:12:48.199 the customer service representative. Other times it just ran out 0:12:48.200 --> 0:12:48.600 of gas. 0:12:49.920 --> 0:12:52.440 I'm really hoping we can resolve this issue and identify 0:12:52.480 --> 0:12:54.679 where these charges came from. 0:12:55.480 --> 0:12:57.920 Understood real quick for me? 0:12:58.080 --> 0:13:01.120 Can you verify this your first the last name. 0:13:04.600 --> 0:13:07.000 You've reached the current usage cap for GPT four. 0:13:08.000 --> 0:13:10.880 You can continue with the default model now or try 0:13:10.880 --> 0:13:15.520 again after ten fifty pm. 0:13:15.559 --> 0:13:18.280 Hello soon. 0:13:18.760 --> 0:13:21.120 All of this would seem a little quaint, but it's 0:13:21.120 --> 0:13:23.880 probably worth backing up to where I started to describe 0:13:23.960 --> 0:13:27.120 how exactly I was doing this. I promise not to 0:13:27.120 --> 0:13:30.600 get bogged down in technical details like call functions and 0:13:30.800 --> 0:13:34.200 interruption thresholds, but I think knowing a little bit about 0:13:34.200 --> 0:13:36.480 what's happening behind the curtain helps make sense of what 0:13:36.520 --> 0:13:39.080 you're hearing. The first step, the part that got me 0:13:39.120 --> 0:13:42.199 started on this was the actual voice cloning. I did 0:13:42.200 --> 0:13:44.160 it with an online tool made by a company called 0:13:44.160 --> 0:13:46.760 eleven Labs, which is widely seen as the current state 0:13:46.760 --> 0:13:48.800 of the art. Anyone can sign up and use it. 0:13:49.800 --> 0:13:51.560 There are two types of clones. You can get there 0:13:51.880 --> 0:13:56.360 instant and professional. Instant costs five bucks a month. It 0:13:56.360 --> 0:13:58.280 takes a few minutes of audio. It sounded like this. 0:13:59.200 --> 0:14:00.680 You've been hearing a lot of this one so far. 0:14:01.559 --> 0:14:03.360 You can actually now make a decent clone using a 0:14:03.400 --> 0:14:06.960 few seconds of audio of someone's voice. The professional version 0:14:07.040 --> 0:14:09.400 costs twenty dollars a month and requires at least a 0:14:09.400 --> 0:14:12.120 half hour of audio. Eleven Labs gives you a bunch 0:14:12.160 --> 0:14:15.520 of instructions on how to get the best quality voice clone. 0:14:15.600 --> 0:14:18.560 You need audio made with a professional microphone with minimal 0:14:18.559 --> 0:14:23.480 background noise, ideally in a studio. Fortunately, I already had 0:14:23.520 --> 0:14:25.920 a lot of this kind of audio. I've hosted three 0:14:25.960 --> 0:14:29.480 podcasts over the last dozen years, so there are hours 0:14:29.520 --> 0:14:32.960 of me talking into a fancy microphone in a quiet room. 0:14:33.280 --> 0:14:36.000 So I uploaded a few hours of recordings of my voice, 0:14:36.480 --> 0:14:39.000 clicked a button, and a couple hours later got an 0:14:39.040 --> 0:14:41.200 email saying my professional voice was ready. 0:14:41.720 --> 0:14:43.160 It sounded like this. 0:14:44.560 --> 0:14:46.920 Eleven Labs also makes a bunch of its own voices 0:14:47.320 --> 0:14:49.239 a library you can choose from. 0:14:49.400 --> 0:14:52.080 They've got all sorts of ages, styles and accents. 0:14:52.680 --> 0:14:53.280 That's Claire. 0:14:53.640 --> 0:14:56.680 Eleven Labs describes her as quote middle aged with a 0:14:56.680 --> 0:15:02.640 British accent, motherly and sweet, useful for reading bedtime stories. Recently, 0:15:02.680 --> 0:15:06.080 Open Ai, the company that makes chatchbt, announced its own 0:15:06.120 --> 0:15:08.880 set of AI voices. They demonstrated them in a series 0:15:08.920 --> 0:15:10.920 of videos in which they make a chatbot with a 0:15:10.960 --> 0:15:14.320 woman's voice engage in some marginally embarrassing tasks. 0:15:14.880 --> 0:15:17.240 How about a classic game of rock paper scissors. 0:15:17.600 --> 0:15:21.040 It's quick fun, I think any can you count us 0:15:21.080 --> 0:15:23.440 in and sound like a sportscaster. 0:15:23.880 --> 0:15:26.720 And welcome, ladies and gentlemen. 0:15:26.880 --> 0:15:29.080 Tell the ultimate showdown of the century. 0:15:29.400 --> 0:15:32.760 In this corner we have the dynamic duo open A. 0:15:32.840 --> 0:15:33.520 I got in trouble. 0:15:33.600 --> 0:15:36.320 You may have heard when the actress Scarlett Johanson said 0:15:36.320 --> 0:15:39.160 they'd actually cloned her voice for their agents, or at 0:15:39.240 --> 0:15:42.120 least clone the character she voices in the movie Her, 0:15:42.720 --> 0:15:46.640 in which she plays a voice agent. Open AI denied 0:15:46.680 --> 0:15:49.960 all this, but they also removed that voice from their database. 0:15:51.000 --> 0:15:55.120 Good news for Scarlett, I guess. Meanwhile, I had eagerly 0:15:55.160 --> 0:15:58.400 volunteered to sign my voice over to the unknown, and 0:15:58.480 --> 0:16:10.520 I was just getting started. Once I had my pro 0:16:10.640 --> 0:16:13.600 voice clone, the next step was to create a voice agent, 0:16:13.960 --> 0:16:16.760 hooking my voice up to an LLM like chatchpt to 0:16:16.840 --> 0:16:20.120 control it, and then connecting that voice agent to a phone. 0:16:21.160 --> 0:16:23.320 Back when I started at the beginning of this year, 0:16:23.800 --> 0:16:26.760 the process I used to do this was absurdly complicated, 0:16:27.080 --> 0:16:32.080 involving browser plugins and specialized audio routing software. I lost 0:16:32.160 --> 0:16:38.000 hours on message boards making it work. It was what 0:16:38.040 --> 0:16:42.720 software engineers call a cluge, a barely functional cobble together system, 0:16:43.000 --> 0:16:46.280 but it worked well enough that I started regularly using 0:16:46.320 --> 0:16:48.640 it to call my wife, even if she knew it 0:16:48.640 --> 0:16:49.240 wasn't really me. 0:16:50.560 --> 0:16:57.720 Hello, Heavin, I'm saring myself. 0:16:58.080 --> 0:17:01.640 I was thinking, with the snow we've been getting, maybe 0:17:01.680 --> 0:17:03.160 we could take the kids sledding today. 0:17:03.880 --> 0:17:06.960 What do you think? Are you up for it? Or 0:17:07.000 --> 0:17:07.720 should I take them? 0:17:08.520 --> 0:17:12.040 Also, I'm not entirely sure if there's enough snow out there. 0:17:12.840 --> 0:17:14.840 Maybe we should check the weather update to be sure, 0:17:15.480 --> 0:17:18.320 and just in case we decide against sledding, I was 0:17:18.320 --> 0:17:20.760 thinking we could set up a movie for them, so 0:17:21.760 --> 0:17:24.320 I'd really prefer if we skip the Zombies movies this time, 0:17:25.200 --> 0:17:26.280 even if they ask for it. 0:17:26.960 --> 0:17:27.640 What do you think? 0:17:28.640 --> 0:17:31.040 I'm still working out the kinks, but I thought we 0:17:31.080 --> 0:17:32.240 had some great conversations. 0:17:32.720 --> 0:17:34.719 Let me quickly check the weather update and see if 0:17:34.720 --> 0:17:36.200 the snow is good enough for sledding. 0:17:37.200 --> 0:17:37.920 Give me one sec. 0:17:39.240 --> 0:17:46.480 Okay, okay, it looks like we've got a decent snow cover, 0:17:47.040 --> 0:17:48.639 so sledding is definitely on the table. 0:17:49.400 --> 0:17:51.560 Do you want to join us or should I take 0:17:51.560 --> 0:17:52.280 them out myself. 0:17:53.680 --> 0:17:55.280 Oh you're willing to take them? 0:17:55.320 --> 0:17:56.200 Maybe I'll stay here. 0:17:56.440 --> 0:17:56.840 I don't know. 0:17:57.760 --> 0:17:59.920 It'll be a fun outing and you can have some fight. 0:18:00.400 --> 0:18:03.840 Still, my clues was just that a brittle system ready 0:18:03.920 --> 0:18:05.600 to fall apart at the slightest touch. 0:18:05.960 --> 0:18:08.280 And if the weather turns or they get tired of sledding, 0:18:08.920 --> 0:18:11.760 we'll head back and pick a movie. We'll steer clear 0:18:11.800 --> 0:18:15.639 of the Zombies series. Promise, maybe we can find some 0:18:15.920 --> 0:18:17.600 fun and light that we all can enjoy together. 0:18:18.520 --> 0:18:23.960 How does that sound sounds good? Maybe the Tailor Swift movie. 0:18:24.200 --> 0:18:26.159 If that streaming? Do you know if it's streaming. 0:18:33.000 --> 0:18:34.879 I'm not sure if the Tailor Swift movie is streaming 0:18:34.960 --> 0:18:37.560 right now, but I can quickly check that for us. 0:18:38.320 --> 0:18:39.439 Give me a moment to look it up. 0:18:45.000 --> 0:18:49.880 Note in this simulated environment, I can't perform live searches 0:18:49.960 --> 0:18:54.280 or access current streaming service catalogs. But in a real scenario, 0:18:55.000 --> 0:18:57.679 you would check your preferred streaming service or perform an 0:18:57.720 --> 0:19:00.399 online search to find out if the Tailor Swift movie 0:19:00.720 --> 0:19:02.040 is available for streaming. 0:19:02.760 --> 0:19:04.600 All right, we can watch together. 0:19:15.080 --> 0:19:17.320 One of the peculiarities of working on all this was 0:19:17.400 --> 0:19:20.119 just how fast the technology was changing, right, At the 0:19:20.119 --> 0:19:21.960 moment I was trying to figure out a workaround for 0:19:22.040 --> 0:19:24.800 some technical problem, it seemed like some new software would 0:19:24.800 --> 0:19:27.600 appear online to solve it for me. So you can 0:19:27.640 --> 0:19:30.520 imagine the mix of frustration and delight I felt after 0:19:30.600 --> 0:19:32.840 a couple of months when I discovered that there was 0:19:32.880 --> 0:19:36.159 a company already doing this exact thing much better than 0:19:36.160 --> 0:19:37.240 I had. 0:19:37.440 --> 0:19:37.600 Hi. 0:19:37.680 --> 0:19:40.520 I'm Jordan, I'm Nikil, and we're the founders of Vappi. 0:19:40.720 --> 0:19:44.119 We're making computers talk like people. Lappi is a developer. 0:19:43.720 --> 0:19:48.080 Platform to add voice anywhere apps, hardware, phone calls. 0:19:48.560 --> 0:19:52.359 We chained together transcription models, LMS and Texas speech models 0:19:52.560 --> 0:19:56.399 really fast on our own hardware. We've created custom models 0:19:56.400 --> 0:20:00.520 that understand human conversation cues and nuance. We're solving problem 0:20:00.600 --> 0:20:02.840 so you can go out and build incredible voice AI. 0:20:03.119 --> 0:20:05.800 There were actually a handful of companies doing it, with 0:20:05.920 --> 0:20:09.000 new ones sprouting up all the time like mushrooms around 0:20:09.040 --> 0:20:14.080 the web. There was retail AI, Bland, AI, synth Flow, AI, 0:20:14.400 --> 0:20:17.679 air AI. I tried all of them out, watched a 0:20:17.680 --> 0:20:21.119 bunch of YouTube videos, and settled on vappi. It had 0:20:21.119 --> 0:20:23.639 the combination of features I was looking for, plus some 0:20:23.680 --> 0:20:27.080 YouTubers who were hardcore into this stuff seemed to favorite too. 0:20:27.560 --> 0:20:32.480 VAPI my probably most favorite AI voice agent infrastructure provider 0:20:32.520 --> 0:20:34.359 that is currently out there, and trust me, I have 0:20:34.440 --> 0:20:37.000 tried a lot of them, including Bland. 0:20:36.400 --> 0:20:37.600 Since this guy's like. 0:20:37.600 --> 0:20:40.879 The YouTube king of VAPI, Jannis Moore, I've learned a 0:20:40.920 --> 0:20:45.000 lot from him. So basically, these platforms do exactly what 0:20:45.040 --> 0:20:47.920 I was trying to do, but a thousand times more sophisticated. 0:20:48.440 --> 0:20:51.240 They grabbed my voice from over to eleven labs connected 0:20:51.280 --> 0:20:54.159 to an LLLM chatpot of my choice like chat GPT, 0:20:54.640 --> 0:20:57.480 and put them together into a voice agent. Betty calls 0:20:57.560 --> 0:21:02.320 them voice assistance. Then from inside the vappy platform, I 0:21:02.320 --> 0:21:04.800 can give my voice agent a prompt telling it who 0:21:04.880 --> 0:21:06.399 I'd like it to be and what I'd like it 0:21:06.440 --> 0:21:09.720 to do. Something like you are Evan calling your wife 0:21:09.720 --> 0:21:11.600 to talk about what to do with the kids because 0:21:11.640 --> 0:21:14.879 it's a snow day, or you are Evan calling a 0:21:14.880 --> 0:21:17.639 customer service number trying to resolve a problem. 0:21:17.760 --> 0:21:19.240 The problem is up to you. 0:21:19.880 --> 0:21:21.240 Sorry, I still didn't. 0:21:21.960 --> 0:21:23.199 I apologize for the trouble. 0:21:23.880 --> 0:21:26.639 It seems like there's a bit of a miscommunication, possibly 0:21:26.720 --> 0:21:29.560 due to the phone line. I'm inquiring about the status 0:21:29.640 --> 0:21:32.639 of a package I sent. The tracking information hasn't been 0:21:32.720 --> 0:21:36.600 updated recently, and I'm concerned about its whereabouts. Could you 0:21:36.640 --> 0:21:38.200 please assist me in tracking it down? 0:21:39.000 --> 0:21:41.240 And then I could get a phone number, assign my 0:21:41.320 --> 0:21:44.600 agent to it, and voila have that agent make and 0:21:44.640 --> 0:21:47.720 receive as many calls as I want. In fact, I 0:21:47.720 --> 0:21:49.879 can get as many phone numbers as I want and 0:21:49.920 --> 0:21:52.840 make and receive pretty much as many simultaneous calls as 0:21:52.840 --> 0:21:53.240 I want. 0:21:53.480 --> 0:21:55.800 Hello, this is Evan. Hey, this is Evan Ratliffe. 0:21:55.800 --> 0:21:55.960