Bonus Episode: Finding Your Voice Addendum - Alex Gets Meta in the Woods

Show Notes

Summary

Alex gets all meta and talks about his latest episode, ChatGPT and the challenges and delights of podcasting, all recorded (badly) while he walks through some woodland!

Transcript

The following transcript as transcribed my OpenAI Whisper and is presented completely unedited and therefore might contain some errors of even total nonsense. I take no responsibility for anything herein!

Welcome to the Art Against Mental Illness podcast. My name is Alex Loveless and this is my podcast about the healing powers of art for artists, art lovers, the art curious and anyone with an interest in mental health and well-being. So I guess you’re wondering why I’m back so soon. Well here’s the thing, I’ve taken to preparing for these episodes by going out for walks and recording my thoughts via my headphone microphone into my phone, which I later then transcribe and then edit down into a proper episode. I tend to think better when I’m walking. After I wrote my last episode on finding your voice, I went out for a walk as I often do and was thinking about that episode and also some of the things I’ve been doing around large language models, Gen AI, chat GPT, things like that and decided to just record my thoughts there and then. And I was going to format that into a follow-up episode and then I thought actually maybe you should just hear the original recording as it came out as a bit of a view into my world and how I think about these things in real time. And so yeah that’s what you’ve got here. The sound quality is terrible because it was recorded by my headphone mic while I was walking around some local woods here in central Scotland in Octorada. So you’ll hear me sort of sniffing, crunching leaves. I think there’s some rattling sound towards the end which is me fiddling around with my medication, my little medication box I carry around with me that carries my ADHD meds. And it’s a little bit meta and self-referential but I think you’ll find it quite interesting. So I’m just, this is completely unedited. I’m just going to cut and paste it in there and you can listen to it or not but otherwise I’ll be back soon with a more formal episode. Enjoy. And addendum to my piece on finding your voice. I don’t use chat GPT or any other language model to write these episodes for me. I do increasingly use record my thoughts in my own voice and then use a language model to transcribe that which I didn’t edit it. Most of the ones I’ve created so far, the ones I’ve scripted rather than the monologues, are, you know, I’ve just sat down in front of my laptop and with an idea of what I wanted to say, written some on my phone and just gone at it and redrafted a few times and then performed it. But it takes quite a while and it had occurred to me as someone who regularly works with language models and has as part of my day job, forms natural language processing and various other data science related tasks that relate to understanding of the written language and creation of the written language via computers. It had occurred to me that I could remove some of the labor required to create these episodes by feeding my thoughts into a model and asking it to write, write them for me and I can just edit it or whatever. And I have tried this and if you just put in my thoughts, you know, please write a podcast episode script for, you know, the covers about this topic, the covers, the following points in the following order. You just get chat GPT, whichever models do that for you and it always does a pretty decent job of it and you might think I could just use that, but every time I read back, I really cringe at the choice of language and to be more specific, the voice it uses, which isn’t mine and it tends to err on the side of some corporate American or perhaps American media voice that is fundamentally not how I talk or how I write or how I want to talk or write and I think would sound very weird coming from me and I think you’d probably figure out very quickly that I probably didn’t write or I’ve got someone else or something else to write for me and that’s certainly not what I want to put across. If I’m going to say it in my actual physical voice, I would prefer to say it in the way I would usually say things now. That’s not to say that the words that I write down and then perform are exactly how I would say it. I’m not quite that eloquent in real time, but certainly they’re the way I want words to come out of my head and if I slow myself down enough and clear enough head, the words sometimes do come out like I wrote them down. So broadly speaking it is how I speak. So obviously that’s just with the vanilla GPT-4 or whatever model and either other models and B, you can give the model instructions to tune its output and I’ve tried this too. For example, I’ve given very specific instructions on how to use language, UK English into a semi-informal style and given a bunch of pointers and then giving it actual examples of how I write things. So for example, transcripts of previous episodes and it’s like well here’s the points I want you to write it like that and it gets it closer, but it’s always saying things in a weird way that I don’t like and tends to miss some of the nuance that I would put into it. So I don’t tend to think it’s more work then to rework that than it is just to write it from scratch, but it also ignores another factor of how I write these in that I don’t know what I’m going to write when I start. I know the point I want to make or perhaps multiple points and perhaps have an idea of structure and especially if I’ve dictated it, i.e. I’ve done a guy out for a walk and done a monologue about my thoughts about a particular subject and then transcribed that. So I might have a whole body of text to edit into a more palatable form, but even then I don’t really know how it’s going to come out, how much of what I said is going to be left in there, how much of it was even coherent or was really the point I wanted to make. And so if you take my last episode on voice as a case in point, I went out for a walk, I recorded into my headphones which are attached to my Bluetooth attached to my phone and I recorded my thoughts and had that transcribed which did a decent job. I then just went at that text directly without passing it through a language model and I probably changed 70 or 80 percent of the words. The structure is roughly the same and then I sort of stumbled across some bits and I’m like yeah this is good but I kind of need to bring that alive a bit more. And that only happens sort of reading through it, changing this, but that doesn’t really say what I want it to say. I understood the words that I said and they make sense. I don’t think it’s going to translate well. And so if you look there’s a whole block in there where I put in some examples of some pretty radical voices. And that came out after. That was one of the last things I did. I was like I thought I need something in there that sort of makes this feel a little bit more human. And I sort of started thinking about how to illustrate some of the points that I was going to make. And I started coming up with some stuff. I mentioned Stephen Hawking, Yoko Ono, Lamont Young and others, Napalm Death. And they sort of just started coming out. I went and did a little bit of research to verify what I remembered of these artists and their work. Found some new stuff along the way. Learned some really interesting things. Went on a minor detour where I actually really bothered to find out what the situation with Jerusalem, the West Bank Wall, Gaza Strip and Israel and this Palestinian people. Because I had a fairly high level knowledge of that. But I sort of really didn’t understand how that situation, well how that state came about in the way that it is. And so while looking, verifying my understanding of Banks’s work on the West Bank Wall, I learned a bunch about Israel and Palestine. And I’m not going to make comment on it. But I’m now more educated than I was before. Hasn’t massively changed my view. But that’s for another day, another podcast maybe. But the point is that not only did all this come about as I was producing the piece, I learned a bunch. And the piece was a lot better for it. And I just simply wouldn’t have conjured that up by sitting around thinking about the structure of the piece and then farming that work out to a language model. Not because I’m being lazy by doing it. I mean if I can make a tool work for me, I’m going to do it. I use Python to do things that would be boring, laborious and slow or not even possible. Whereas I try and do it myself. So I don’t shy away from using tools to help me do what I want. But only if they’re actually helping me achieve the same thing I would have done otherwise but faster. And in this case, not only would I have probably not put these extra embellishments in, I would unlikely to have gone and researched and discovered some new wonderful things in the art world and understand more about the situation in the Middle East. And I just would have missed that. And I wouldn’t. That’s good. I’m really pleased that happened. And therefore, you know, maybe I could get a model to talk in my voice. But would it make the choices that I made? Would it, if I said to it, insert some examples in here, would it chose the examples that I chose? Because they’re very much part of me, my brain, the things that I know, the knowledge that I have. And GPT doesn’t know everything. It knows probably all the facts that I know, but it doesn’t know which facts I know. It doesn’t know which I’m likely to choose. And maybe there’s some point in the future where we’ll have our own personal agent that reads and hears everything we do and does know. You know, what we would do or what we would think. And that’s theoretically possible now. But at what cost? And I think it’s a big realisation of mine where I’m sort of toying with the idea and feeling slightly icky about using a language model for any of my work. And let’s be clear about this, where I’m doing my day job and what I need to do is write corporate, slightly American sounding English for a client or for my website or whatever. If that’s what I need, I’m damn well using the model to do it. I almost certainly will have a model that more closely suits how I would say things. But it’s for me, in a sense, for a lot of that sort of stuff, I sort of want my voice out the way. Certainly, well, depending on the type of communication, but absolutely no words. But using it for this stuff, it feels anyway, a lot more personal. And I let my voice roam a lot more freely. I let my opinions and thoughts and philosophies flow. And and that’s that the process is the whole process. So cutting, scooping bits of it out for the sake of convenience has totally changed the process. And so would change the output. And and I think for the worse, not the better. So I’m not going to use language model to produce my scripts for these. I will continue to use them to transcribe. I might use it, use them in the process of doing any research I need to do, answer any questions I have about the world or whatever. And that’s fine. But the words that get written, the structure, the tone of voice, the word choice, the phrase choice, the artist that appear, I want them, it’s important for me, them to come from me. And when I speak these words out, I’m allowed to have conviction. And that when I say something, it’s because I meant it, rather than I read something about it and wanted it to sound like I came up with it or I meant it, which is really not the case. I’m not claiming I’m some sort of philosophical genius art wizard. But, you know, I talk about what I want to talk about when I want to talk about it. I’ve got my spin on things, which may or may not be controversial. I have no idea. But these are the things at this point in time that I believe and that I think and that are motivating me and driving me, giving me enthusiasm and joy. And therefore, that’s what I talk about. And that’s me. That’s my voice. That’s my choice. And I hope that any of you, if anyone’s listening, please let me know if you are. This is a value to you in some way. If you learn something or it helps you make you feel something or helps you progress or well, maybe you think it’s terrible and you just love to hate me, it’s all good. But this is me. This podcast isn’t like a magazine, you know, where lots of different people contribute, although it’s not out of the question of bring some others into the mix at some point in the future. But really, this is about me and my thoughts on creativity based on my experiences and my understanding of the world. And I don’t know if this is interesting or valuable, but I enjoy doing it. So I’m going to keep on doing it. And it’s a good way for me to capture my thoughts on these things. Anyway, it’s a matter of prosperity. It’s like a diary. So feel free to listen to not listen, but do know that if you’re listening to me and I’m saying it, and I wrote it, I meant it, my ideas, my thoughts, minimal from robots, computers, language models.