Can AI Replace Me? Evan Ratliff on Letting an AI Clone Live His Life

0:00I think it's upsetting to be tricked in general. and when it comes to friends and family to kind of like start speaking to something someone as if oh hey what's going on and then like even if it happened really fast over the course of 30 seconds of sort of realizing that you feel like an idiot because you you've just been talking to no one and it's just the idea of being fooled in some cases but then in other cases like I had another friend who just said in the moment, just said like, I feel so lonely because fundamentally when you realize it, you realize you're just talking to yourself. Like you're just talking to no one.

0:40Hi, I'm Evan Ratliff. I cover technology, crime, and identity where those three things intersect. I write for Wired Magazine and Bloomberg Businessweek among other magazines. I have a podcast called Shell Game. I also wrote a book called The Mastermind about a criminal Cartel. And today, looking forward to talking about how AI is infiltrating our identity. So Evan, I have had the pleasure of listening to the podcast, The Shell Game, and read a lot of your stuff in general. Would you mind, in your word, kind of like explain what The Shell Game is? And then I will pebble you with questions after that.

1:23Sure. So Shell Game is a six-part podcast. It is, uh, a, an immersive journalistic experiment into how it will feel when more and more of the voices around us are artificial intelligence or otherwise fake. Um, and the kind of premise of the show was that I started messing around with voice cloning and trying to figure out sort of like what it meant for me as someone who does a fair amount of podcasting in addition to my writing and other journalism. And then I had the idea of hooking it up to a chat bot and then hooking that combination up to my phone or a phone line, eventually my own phone line in the form of what everyone now knows is a voice agent.

2:11And then setting that voice agent loose in the world as me, including for work purposes to do interviews and things like that, including dealing with customer service, dealing with scammers. So I set up a specific line to attract scammers and telemarketers. And then I had to answer those calls. And then eventually also my friends and family. So this voice is attached to my mobile number. So I could call someone or it could call someone on my behalf. They would think it was me and they would actually get the voice agent with my voice, but powered usually by ChatGPT. Sometimes I would use another.

2:50another model and so i did that for about five six months and then it's kind of the story of each of the ways i deployed it and how people responded it is very cool if anybody haven't heard it they should go to hear it and also um you did like i guess i think i saw her and npr did like a one episode version of the whole thing together with you yes uh radio lab did an episode that was based out of, it was basically kind of like a merging of two episodes of the show onto one episode of Radiolab. Okay, many, many, many questions. But let's start with, I guess, it's six months ago, is when you did it?

3:32Maybe even a little bit longer? It came out, it started coming out, actually, it's been almost a year. It came out starting in July and it went through August. The last episode's aired in August. And so if you were doing it again, what do you think would be the biggest difference now from when you did it back then the biggest difference now is latency that's the number one difference so latency being the time that it takes to respond for the chatbot to formulate a response and then for that response to get translated into voice and respond on the phone that was the biggest giveaway for the agent when it talked to people was it just took a little too long to respond uh it's still a lot of people did not recognize that they were talking to ai not talking to a human but most of the time that it gave it away that was the main reason and actually i just used it on stage at an event in portland last week and it was i mean it's significantly better than the last time that i used it and i'm using the same services i was using before they've just you know they've just tweaked it to be a little bit better it a little bit better.

4:37It still has some cadence problems. You know, it still has a lot of problems and you can still detect that it is AI, but the technology has advanced as I expected it probably would. Before we dive into what you learned, which I'm really eager to hear about, I'm an idea guy. I spend a lot of time thinking about where do ideas come from. Can you talk for a second about what inspired this whole experiment yeah i mean i had been interested in voice cloning for a while i think anyone who does any kind of audio you know it comes up like there's this company called descript that has a you know a podcast editing software and i worked with one company that was using it and it's you sort of notice in there that they have this thing where they could clone your voice i mean they've had it for a couple of years actually they could clone your voice and they could use that to fill in, you know, instead of doing what we call pickups, where you have to go back and re-record something, they could use that voice to maybe like add a line somewhere.

5:38And the company I was working with did not actually use it for that purpose, but it did get me interested in voice cloning. That's what we do in this podcast. We will actually not use any of your stuff except for what we just recorded. It's just a voice sample. And then we just get you to say whatever we want. It's great. It's great. I mean, that's not that far off. That is, I mean, there have got to be people. I mean, we could talk about this later. Like, my main interest is in, like, crime and scamming. So, like, there have to be people already running a similar type of podcast-related scam.

6:09But in any case, I was interested in it. But then I was also, I don't know, by the fall of 2024, I was just, I was sick of hearing about AI. Like, I thought, I don't want to cover it. I don't want anything to do with it. I was covering other things. and I had this moment I mean it wasn't like anything prompted this moment but I kind of realized like oh is this just being old like I used to be a person who I embraced every new technology like when Vine started like I got on Vine like I would do because I'm both for my job because I write about tech a lot but also just like I was first who was interested in the future I worked at Wired Magazine you know and I was like actually am I becoming a person who just like refuses to address the new thing.

6:56So I thought, you know what? Voice is going to be the thing. I'm going to clone my voice. I'm going to pro clone my voice and see what it's like. And then once I did that, I started leaving people voicemails, like just calling up and like playing a voicemail, you know, into their voicemail and like fooling them with it a few times. And I was like, well, that's fun. But then I thought, what if I could just call with this thing and hook it up to ChatGPT? and at the time there was like a plugin you could use that could sort of made this possible and that got my sort of old like tinkering uh senses going and i was like oh i can and i spent weeks making this thing so i can make phone calls with it and i started calling my wife with it and some of that is in the show like my very early efforts and i was so proud of myself for building this voice agent and then i discovered that there were like five companies that just they had platforms where you could just do this you just sign up you put in the you just attach it by api to like 11 labs where you clone your voice and they just did it all so the show is actually after the first bit it's all using these platforms which are much much better than the system that i was actually they shut down the plug-in because there was no point in having it anymore what was some of the things that you didn't get to do but you think you could do now well i don't i don't want to give them away because i am doing them now we're working on season two let's go back to be a little bit concrete then so i've called my voice of course also i made the henrik gpt you know like it you can ask a question it looks at all the stuff i've written and then it asked my thing i think we had guy kawasaki on that also have done this and basically things that it's a better version of guy kawasaki because they can actually remember all this stuff that he's ever done and so he's like you should just talk to this um just like a little like any tips on the cloning because I've tried both the professional one with 11 labs and I done like the one where you just upload 30 seconds.

8:51And obviously it's quite like hit and miss a little bit, like how it will clone it. Any kind of pro tips on cloning? Well, the biggest pro tip, and I actually got this when I interviewed the guy who started this company, Vappy, who's one of the platforms that I use to do the calling, is that you got to think about what you want to use it for. So if you actually want to clone, mine is a professional grade clone using, I mean, I have many, many, many, many hours of tape of me speaking into a microphone like this one, a professional microphone, professionally recorded, much of it in a studio. and so it sounds very good but it actually almost sounds too good for phone calls in some ways and so if what you want to do is which not i mean this is kind of a special case but like what i wanted to do well you want to scam your family if you want to call people with it and it sound like you're on the phone the thing to do and i actually did this and did a separate clone which i didn't use in the show because we had already started using one of them and it was like it would be too strange jarring to hear both of them was just like call something and have it record for like a half an hour just through your phone so the sound is actually not that great but the sound of the phone's not that great and so it's really about like the purpose that you want to use it for but i mean the main tip is what i'm sure 11 labs probably says this now it's like the the better the microphone and the quiet environment and the consistency of it because when mine messes up too i can tell that there are some recordings that i gave it that are probably a little more hollow it was probably in a room that wasn't that great and it it will kind of like it i i don't know if it averages or what it does but it like you will hear sometimes like a slight change in kind of like the room tone of the but i think actually that's a super tip i don't think even 11 loves had i did it recently again and uh and i don't think they do like make the tone like what are you going to use it for like even like if you're going to do a professional like a presentation you're going to have one tonality and if you're going to just call your friends and saying hey what's up then obviously you use different words different tonality different kind of like ventilation in your in your voice yeah but now so one thing that's making me think as i listen to y 'all kind of you know geek out on the audio for an everyday person what have you discovered are the applications because i think a lot of people listen they I kind of go similar, you know, we had an artist who's won a pretty prestigious art award, but a lot of people go, hey, I'm not, I'm not making paintings.

11:25I'm not a photographer. And I think a lot of people could be tempted to check out and go, I'm not creating voice recordings myself, right? What have you discovered are some of the practical applications that somebody off the street or, you know, just kind of our typical retail customer, you go, hang on, before you dismiss this, think about using this tech in this way. Yeah, I have those. I will preface this by saying, I mean, I would say like I myself am a skeptic of how this is used. I mean, I'm a skeptic of everything. I'm a journalist. So like my default approach is sort of interest in something, but also like, where is this going to go wrong?

12:02And so what I was trying to get people to think about was not using it themselves. It was that you are going to encounter this in the world more and more and you need to be ready for it. So with that preface, I do think that even people who are very skeptical of it don't think about some of the uses. The very obvious one is there are people who have lost their voice for medical reasons in all variety of ways who can use these types of voice agents to absolutely transform their lives. So like, that's the baseline. There are people who don't have a voice who can use this to have a voice. So we should think about that.

12:39I think then there's the question of kind of deploying it for practical purposes. And, you know, there's a, there's a wide range of things and feelings about this. So for instance, after the show came out, I had people ask me, well, what about using it? Could I set one up to like speak to my, uh, you know, mother or father, grandfather, grandmother who has like, is in memory care and basically like can't remember when I'm, I've contacted them, but it's nice for someone to check in with them every day. And I feel like they should have this communication and they could communicate with this AI. There's even now, I think there's a company that maybe does this.

13:20now some people are just absolutely disgusted by this idea you know they think that it's actually inhumane and horrific that you would sick a chatbot on your mom or dad or grandmother and other people i think are like well if you've ever dealt with this situation and how difficult it is like i'm open to it like i'm open to anything that makes them feel better and so So I, again, like I tend to be skeptical of it and think like, well, I wouldn't do that. But, you know, people are going to have a wide range of views of what's appropriate and what's ethical. Similarly to another idea that a lot of people are into, and I do get into this in the show, is sort of creating a voice agent of a loved one or of yourself so that when you are dead, other people can talk to it or you can talk to a loved one who is now gone.

14:17So feeding it a lot of information, giving it the voice. I don't want to spoil the show, but like, ultimately I conclude like, that's not for me. That's not something I would want to do, even though I have a member of my family that actually would actively want to do that in some ways. So I think those are the different uses. And then there's this sort of like everyday stuff that AI companies sell. And that's where I'm the most skeptical of it. like, I don't know, creating, creating agents. I think it's telling that whenever they talk about creating agents, the example they almost always use is getting restaurant reservations.

14:52And like the number of people that even in America get restaurant reservations, much less the number, and then any of the people who get restaurant reservations and then can't use any of the available services, like, oh, it's too hard to use resi or open table. Like I do get restaurant reservations. And so they're like, you can create an agent that will call 200 restaurants and get the best reservations. Like that is a solution looking for a problem. Well, it's hard. I would say, I would say it's hard to know what good niche versus bad niches, you know, right. So Uber, you know, started as a black car service.

15:29You go, actually not a lot of people want a black car. Well, when you realize that it worked, which is to say, it's really difficult to know kind of which edges of the possibility space are relevant or irrelevant. I, I love the example of somebody who lost their voice. You know, we had Josh too, who runs Meta's Orion project. So he's working on, uh, on that product. And he said that one of the lead users that they're getting a lot of inspiration and getting a lot of great feedback are folks who've lost their sight, right? Folks who've lost their voice. So I think there's kind of this augmentation of a lost human ability that, that makes a lot of sense to me all the other stuff i don't know if it's a good niche or a bad niche so you don't have you don't have anything that way that you kept in operation where it just was so much more convenient to deploy the agent that you were like you know what i'm just gonna keep having them do that no i think and i think that that's a little bit particular to voice like i know you know some of these sort of like power users of ai who like automate the writing of their email and all this stuff you know but i think when it's voice how many phone calls do people like make these days like most people don't make that many phone calls and so most of the phone calls that you make the amount of time you spend writing the prompt currently the amount of time that you would spend writing the prompt exceeds the time on the phone call right like i've used it for some during the show i was trying to see like well i need to change a flight you know with Delta and I'm like, you know, it says you're going to be on hold for two hours or something like that's something better.

17:00I used it at the DMV, but you have to prompt it so perfectly. Like I had a car registration problem. I sent it to talk to the DMV, like the real DMV. It, it did get to a human user. Like it navigated the phone tree and it got to a human. And then it just made an appointment for me to go to the DMV, which is like, I didn't, that's the thing I was trying to avoid. That's the last thing I wanted. But it did it. And it's, it was kind of cool to listen to it, do it. And the person at the other end of the line didn't know. Okay. You, you, you saying the, the person on the other line didn't know, let's talk for it.

17:34Cause, cause you started and we went down a different rabbit trail, but you started by saying, I wanted to know what do we need to be aware of as, as this becomes a reality. And that kind of phrase, the person on the other end didn't know. If you think about the user who we should be designing for potentially as the person who's on the receiving end, what have you learned about living in a world where this is now a possibility? What should folks be aware of? I would say for the time being, the big issue is that there's no requirement on anyone's part to disclose that AI is being used. You know, some places choose to do that on their own.

18:09There's obviously no law about that. There's a little bit of FCC stuff around robocalling where it's like illegal to use AI for robocalling currently. So the question is like, what happens when someone encounters AI, but they're not expecting to encounter AI? And in my experience, people find that quite upsetting for the most part. Like they find that either irritating or almost like existential, you know? Existential is more when you're talking to like someone you know. In my case, I was calling up friends and family who did not know they were going to encounter AI and then suddenly realized it.

18:43and some of them found that like pretty profoundly disturbing that they thought they were going to be talking to me for a minute they were thinking they were talking to me and then realized oh no this is something else and you think the latency was the biggest thing that gave it away or what do you think is the latency for for them it's more just like little subtle you know differences in the way i talk like it's a little bit faster sometimes like actually in its speech like it has the latency and it's time to respond but then it would speak a little faster and it was like a little more enthusiastic than me that was a common problem like people described it like evan on cocaine was like the how they my friends that was like their shorthand for it uh that it would just i'm not necessarily like an enthusiast in that way um and then the other thing was like sense of humor and uh it it has i mean i'm not saying i have the greatest sense of humor it has a shit sense of humor.

19:42Like it's absolutely just, it's jokes, not jokes that not only that I would make, but like you just wouldn't make with friends. They're the jokes skimmed from, you know, training data and the guardrails make it just sort of like, it's just lame. So like. So given those deficiencies, because to me that those seem like pretty, um, glaring deficiencies, why, despite those deficiencies was the, you said it's quite upsetting to people. So what is it that's upsetting? Um, I mean, I think it's upsetting to be tricked in general. Um, and when it comes to friends and family to kind of like start speaking to something, someone as if, oh, Hey, what's going on?

20:30And then like, like, even if it happened really fast over the course of 30 seconds of sort of realizing that like you feel like an idiot because you you've just been talking to no one and it's just the idea of being fooled in some cases but then in other cases like i had another friend who just said in the moment just said like i feel so lonely because fundamentally when you realize it you realize you're just talking to yourself like you're just talking to no one uh So, and then I had a more extreme example, which was a friend who didn't realize it because the way that the conversation went, he thought it was sort of me being sarcastic.

21:10And then he thought it was me being strange. And he thought I'd had like a, like a mental breakdown. And so he was actually very upset in the moment because he was trying to figure out what he should do about it. Did you then break cover? I wasn't on the call. Like for 95 % of these calls, I wasn't there. Oh, you didn't even send listen in? No, no, no. I just sent it away to do its thing and it records all the calls. And so then I would pick up the recording later. There were a couple of exceptions to that. But yeah, for the most part, I then would listen to the recording and then call my friend and say like, hey, I'm okay.

21:48And you think when we have all these, obviously audio is pretty developed now, even more so when you did it. They've started to create all these different video clones too. We hear stories of people who had transferred money because their CFO called them a Zoom call and stuff like that. What's your view on how far are we away from this definitely being used for scamming purposes, but then secondly, for use for kind of like other use? We're so far into it being used for scamming purposes already. It is the best. it might be the best technology past just like the internet and the telephone like the best technology for scamming that has ever been created i mean there are already thousands of thousands of what they call the grandparent scam which is you know they can clone a voice and then they use the voice for you know a few seconds to be like oh i'm in trouble they call a relative oh i'm in trouble you're gonna hear now you're gonna i'm gonna give you my lawyer we need money and then like supposedly handing the phone off that scam is now routine not to mention that like the normal scam architecture they often now use ai to sort of like weed through the marks and i know this because i have this dedicated scam line that gets like now it gets like 40 or 50 telemarketing scam calls a day and so they'll just use that to like because scamming is a volume game So like they'll use it to go through the marks.

23:14And then if they get someone who's actually engaged, then they pass them off to a human to like close the deal. Basically. Is there something about that? You said you still have a live kind of scam receiver. I remember Mark Rober did this thing with grandparents. I don't know if you saw it, but he actually, he ended up working with the FBI to shut down a call center in India that was scamming grandparents. And he was able to verify it and prove it because of his kind of YouTube channel. Um, is there, do you have any altruistic or otherwise kind of mission oriented goal with keeping this scam line alive?

23:51Not for me. I mean, there's all these sort of scam baiters on YouTube, uh, and there's a little bit about them in the show. And even since the show came out, actually some phone companies like O2 in UK, and I've seen a couple of others, they have implemented these bots where they, the bots answer the phone and pretend to be a grandma and keep the scammers on the phone. But I actually, I just think the whole premise of sort of like wasting scammers time, there's scamming is an industrial strength business now. It's, I mean, these are massive, massive endeavors. And so the idea of like even keeping them on for 20 minutes and then, I mean, the FBI doesn't have trouble now like finding scams like there's millions of scams the problem is like the jurisdictional issues of trying to chase like they have to be so big to even spend the time for law enforcement to go try to track them down that it's it's really it's actually like an incredibly difficult problem and so no i mean i'll i'll say up front like i'm just interested in what the scammers are doing and looking for stories for myself i don't think that my scam line it does anything to like Do you then have, we have safe words now in the family.

25:08Is that something you recommend too? Yes, absolutely. I mean, I think everyone should have that now. Everyone should have that. I mean, it doesn't have to be, doesn't necessarily have to be a safe word. It's as simple as like, if you get an emergency call from me, just also text me. Like, that's all. Like, that solves it. You know, it's like a problem that if you know about it, you can avoid it for the most part. It's the same with what you described before, which is like more advanced forms of business email compromise where people are, you know, companies get contacted and like someone pretends to be the CEO and gets them to like transfer money.

25:44And it's like, if you know about the scam, you could avoid it. That's the main defense against it. So, so sorry, just to be clear, because now, you know, as, as I'm recognizing my heart rate is elevated, thinking about the industrial string scamming, I want a clear kind of checklist. How do we protect ourselves? you're saying safe word as individual yeah yeah you should have some way of identifying yourself to your family that like only you would know and someone outside wouldn't easily find out so if you get a call from your daughter or sister whatever and they say i'm in trouble and i need money from you right away for a variety of reasons like you can say like oh say the word uh and they could say the word and if it's them then obviously you need to go into action mode but what they're trying to do is get you to go into action mode get your adrenaline going and not even think of making that call making that text i don't i mean i find it to be quite common now that my parents get a text where it says hey this is henrik i just got a new mobile number you know heads up and then they kind of tried to take it from there like this happened at least a few times uh and i think both my parents have also been going through you know scams and some of them at one point successfully right and so i i mean i i i have 100 with you like i am i just assume that it happens and it happens over and over again i think you know at least i can help my parents uh kind of not not jump in the the trap twice yeah it's hard it's hard though because people you know everyone's living their lives more online but like the basic principles of if something incredibly strange happens double check it and also like if you buy something and it's way too good to be true in terms of price like it probably is to go back to the beginning of the conversation if you don't mind because there's something you said that i thought was really interesting that i just find personally curious and i would love to know more about it.

27:49You mentioned when we asked where this idea came from, that in the fall of 24, you were sick of hearing about AI or fall 23, whatever it is, sick of hearing about AI. And then you ask yourself this question, is this just me being old? And I wonder about the temptation to opt out in this moment, you know, for folks who have experience because there's always technologies that pass us by, right? And, you know, I think of Bryce Shalamel, who's the head of AI at Moderna. He was on our show. Amazing episode. If you haven't had a chance to listen to Evan, you should, and every listener should. But one thing Bryce said is they've made stickers at Moderna that say, don't be Fred, use AI.

28:30And that is a reference to their old CEO's boss, who when email came out, had his assistant print his emails, and then he wrote his response. And then his name was Fred. He would hand the written response to his assistant, who would then type their response via email. And it's kind of an allusion to the fact that we always have a tendency to check out from technological changes. And when you said, is this just me being old? I'm curious if you have any thoughts on that human tendency broadly or specifically, how do we know whether we should be checking out or leaning in? What have you learned about that?

29:09I think it's very difficult, especially when you become a middle-aged or older person, because it's very difficult to separate this sort of natural tendency to think that things were better when you were younger or the way you came up or the way your career went from something being genuinely negative in society. And, I mean, there are many examples of this over history, but the one that I use in the show is I found this, there's this article from the new york times which is like my favorite article that i've ever uh read from 1925 or maybe it's 1924 um which is an article about like a guy in new york city who has a shop who's one of the last people to like adopt a phone in his shop like he he has refused to have a phone in his shop he's like the incessant ringing and it's very it's a beautifully written story and it's sort of about this idea of well it's true that when the phone came like it fundamentally changed people's relations with each other and people certain people were very upset about that and people resisted it because they were like i don't want people to be able to contact me all the time like i what about the quiet of the city and all these sorts of things and now we look back back at that and we're like the phone like come on but you see it you know know, it's happened with all of these different technologies, but then it's also a trap because some technologies are dangerous.

30:40And I think social media is a good example of like, oh, ha ha, the joke was on us. Like we, perhaps our guard was too, was too far down. Yeah, exactly. We adopted it so easily and we let ourselves be manipulated so thoroughly, uh, by the companies that were creating it. And they, you know, if you look now, like they weren't really hiding their incentives. Like they were just growth, growth, growth, growth, growth. And the fastest way to do that was to create incentives for us to yell at each other. And so all of which is to say, I try to like all, I think all you can do is kind of reflect on that and think about those questions and like have those questions in your mind.

31:22And I think in this case, I do know a lot of people that are rejecting AI, you know, writers, journalists who, who are just, they're very angry about it. And rightfully so, like it is built on stolen copyrighted material, like whatever. They'll never make it to court. Probably they'll settle. But like, I think in our hearts, we all know what exactly happened there. But the question is like, what are we going to do about it? And what I want is for, because the adoption is happening so quickly for people to try to take on some of these questions themselves and think about them because no one is going to do it for us.

31:59And I think one of the best examples is AI therapy. Like AI therapy is happening, including voice. Chat is like already a huge thing. Voice talk therapy is happening. You could talk to an AI voice therapist. I had my AI talk to an AI voice therapist during the show. But like I went to look at the number of controlled studies about AI voice based behavioral therapy and its effects. And there was one. There was one study. And like millions of people are going to be using this very soon. So it's just getting ahead of where we are. One of the things that I've been thinking a lot about is because I was proud of building the social web.

32:42And obviously we didn't think much about taking pictures of each other's food and putting on an app would kind of destroy kind of like a generation of mental health. Right. It wasn't really something that we were planning at the time. And so there's all these other consequences. And one of the things that seemed clear to me now with AI is that all this loneliness that kind of like has slowly kind of crept in and got worse during COVID could potentially be even more kind of bad now that we have all these agreeable, very charming kind of synthetic people that we can chat with. And at one point we might cut more and more real people out because it's nicer to talk to the agents that say nice things back to us all the time.

33:22And then we might kind of wake up one day and be incredibly lonely. So, I mean, that's definitely seemed to be kind of one thought. But on that note, one of the things that you spent a lot of time was talking to your agent, or at least kind of like thinking about how your agent would be a representative of you. Was there something in how AI works that you kind of got surprised about, or you learned was an important kind of discovery in taking as much of you as possible and then synthesize it down to a voice button that then should represent you. I mean, yeah, I learned a lot. Like, I feel like it was a valuable exercise.

34:05I mean, I learned some things. I got more of a personal sense of, like, the way AI works. And I think as much as you read about it, and, you know, you guys know it very well, but, like, you know, a lot of people use AI and they don't really think about what it is or why it can do what it can do, you know, that it's a predictive engine, that it's built on this training data, that it's sort of like trying to synthesize what a human would be expected to say in this particular moment when you put it into conversation. But if you listen to a lot of those conversations, and particularly if you try to have one represent you in conversation, like it becomes very clear.

34:44It's like using my voice, but speaking in this manner that sounds like the average person it doesn't sound like me like the content of the conversation so it was interesting to kind of like really feel almost like how the models are working um but then the flip side of it is like in this particular type of situation where you're trying to create an agent of yourself it's again it's like a little bit of a trap because like it gets better the more information you give it like and do you and this is a super geeky question using these rack models that you did did you figure out what would be the best kind of repository of content to give it because i would imagine just taking your wired articles wouldn't kind of like make it sound like you in voice so was it more transcripts of kind of conversations or or how did you kind of like hone the tone uh you couldn't i felt like the the the knowledge base that you could give it didn't really change the tone so i think to if you wanted to do that you'd have to sort of like train up a mini model um but i didn't that's not what i did i basically like basically using the standard models but then pasting on this knowledge base of information that it could access which is like it's post training so it's not really in there it's all i believe i may not have the technical aspects exactly right but like i believe it's like almost like giving it a gigantic prompt like if you could put all this stuff in the prompt so i would do things like write a dossier you know, I was sending it to therapy.