Episode 60: Learning, Work, and Art in the Age of ChatGPT

• Links to this episode: Spotify / Apple Podcasts

Unofficial, reviewed transcript. It may still contain mistakes; check the podcast when wording matters.

Speakers are labeled by first name where identified. Unidentified-speaker labels may refer to different people. Bracketed question marks indicate uncertain wording.

Transcript

[00:00:10]  Bruce: Welcome to The Theory of Anything podcast. Hey guys, how's it going?

[00:00:14]  Peter: Hello, Bruce.

[00:00:15]  Unidentified speaker: Oh, great.

[00:00:15]  Unidentified speaker: Hey, Bruce.

[00:00:17]  Bruce: Hey, we have a guest, my nephew, Brendon Nielson here with us today. How's it going, Brendon?

[00:00:25]  Brendon: It's going good, enjoying a nice spring morning.

[00:00:29]  Bruce: Brendon actually had the idea for this podcast episode, so we invited him to participate. I was working one day, and I received this text, and it's from my nephew, and he goes, Do you want to talk about artificial intelligence? Because I've got no one to talk to about it, and I know you are interested in that thing. So we did this Zoom, and we just started geeking out over artificial intelligence. So interestingly, I told Cameo about that, and then Cameo started geeking out over artificial intelligence. Then I had another conversation similar to the one I had with Brendon. I said, we ought to just all get together and we ought to talk about ChatGPT in particular and a lot of the cool things that are coming out of generative AI these days. We've done a few episodes on artificial intelligence in the past. Obviously, that's an area of interest for me.

[00:01:24]  Bruce: I should probably note though that I'm more theoretically interested. I want to understand the algorithms. I'm not a great user of technology and actually think Cameo in particular has been quite impressive in her use of technology as a user. So I'm going to probably let Cameo and Brendon do a lot of the talking for this episode because I'm really curious what they've learned from all their use of AI. And we can save my theoretical reading papers for some other time.

[00:01:55]  Cameo: Well, I want to hear from Brendon first. When did you start using ChatGPT or other MMLs? And how are you using them and what are you doing?

[00:02:08]  Brendon: Yeah, that's a great question. So a little bit of background on me, I've worked in software for about six years on the sales side. So I've been on different sales team and software. So professionally, I have been using ChatGPT a lot lately and other AI tools. And then I've also been a music producer for around 10 years. And artist and performer and DJ. So I have been, I think for me, I first started using ChatGPT like last Christmas. But all I remember is the first real use I had with it was actually at our family party. Bruce is when we do the Christmas cranberry pudding every year at grandma's house.

[00:02:56]  Unidentified speaker: Yes.

[00:02:57]  Brendon: So I found out I had food allergies a couple of years ago. And that thing has so much milk in it. And I'm a very—[unclear] dairy allergy. I definitely have a dairy allergy. And so the recipe is basically butter, milk, sweetened condensed milk, sugar, flour, and cranberries. And I was like, how am I ever going to be able to eat this again? And so I started just talking to ChatGPT, and I heard that it could do amazing things. And this was early, so this GPT-3 or 3.5. And so I was like, hey, I want to make this recipe. Here's the recipe. How can I make it allergy friendly? And you know what? It told me like, get this kind of oat milk, you're going to reduce it over the stove at this temperature for this long. That should turn into the consistency of sweetened condensed milk. You can go get oat milk butter, use it this measurements. And then I did it and it made a cranberry pudding that was vegan.

[00:04:02]  Brendon: And it was good. And we even had some taste test. I don't know if you tried it or you remember that. But that was my first use of ChatGPT. And I was like, wow, okay. I'm always for quality of life updates with technology 100 percent. And so I just kind of continue to use it. And I think there after, I would say it took a few months, but I really started to see kind of the application of this technology across my life and a lot of what I do. I do work in tech. I'm constantly writing emails. I'm constantly writing proposals. I'm constantly building reports, all sorts of stuff. And I started to really see kind of the benefit of having. And maybe we'll get to this too, but I've been diagnosed ADHD since I was seven years old. And I've been to a lot of therapy. And one of the biggest things that helps people with ADHD is having, they call it, what's it called?

[00:04:59]  Brendon: Body doubling, where if you feel like you're having a hard time getting something done, being able to have somebody to talk to or be there and bounce around, having like a co-pilot is helpful. And I found that using large language models like ChatGPT and other AI solutions kind of gives me that body doubling experience which allows me to be significantly more productive, where I can, you know, rather than responding to a whole email at work or, you know, I can just basically say, hey, write a reply that says the quote is this much, and we're grateful for working with you, do it in a friendly and concise, you know, business voice. It'll write it. And if you get the prompting right, it'll do it pretty close. First time. You edit, send it off. And I mean, it's significantly increasing my productivity. So really, I just have been using it. I don't I'm not an expert on AI.

[00:05:52]  Brendon: I have been in music and software for a while. So I have some experience there. But as far as AI goes, I've just been a heavy user lately in both music and at work. But it really has increased my productivity and it's been it's been pretty cool to see that.

[00:06:06]  Cameo: I really like your idea of using it for the ADHD. And I guess I'm using it that way without even kind of acknowledging that that I'm using it that way. I got in probably a lot the same time as you did when they when they pushing it out. And it was it was just kind of just so mind blowing to see what it could do. And so the same as you I started using it a lot for really simple things like emails, summaries of conversations. I early on, I was I would record meetings through through teams because it auto transcribes it. And yeah, I've done that same thing in and use that to to provide summaries. And here's what this meeting was about and let it summarize all of my meetings. So I didn't have to take notes anymore in meetings. And then I had a speech that I was giving at at a conference earlier this year. And I was really, really struggling to to get it written.

[00:07:20]  Cameo: It was a 45 minute speech and I knew the concepts that I wanted to to talk through. In fact, I Bruce and I went out to lunch and was kind of bouncing some of my my early ideas. But I was just I'm such a ADHD person, you know, and doing sitting down and writing an hour long speech. I just was really struggling making headway on it. And so I used ChatGPT to write the vast majority of it. And I would just sit down in little clumps and give it my next idea that I was that I was going to have as the next portion of my speech, and then really just use it as a as an idea bouncer kind of where I would say, OK, I'm going to talk about this and it would give me ideas and I would kind of use it to help help me refine my ideas. And I wrote the entire speech in concert with with ChatGPT, and I was really pleased with with our work that we came up with.

[00:08:21]  Cameo: And then, you know, kind of the same thing on the on the visual side, too. I I got in early on on DALL-E and I haven't I don't use it anymore. I'm almost exclusively using Midjourney. I'm using Midjourney pretty heavily. And so, yeah, it's kind of just the same thing. It's just at this point, I probably spend an hour a day, at least in some in in either ChatGPT. I'm I also am using a GPT agent pretty heavily. And I honestly I don't know what I would do if they took it away at this point.

[00:09:06]  Unidentified speaker: How's that?

[00:09:07]  Brendon: I've never used an agent. I've wanted to, but I haven't. You know, how did you you download from GitHub? Like how did you get get the agent?

[00:09:16]  Cameo: So my my boss is like crazy into AI. And he's been he has he meets up with other executives pretty regularly. And he has an AI group of executives that he's been been hanging out with. And one of them had developed an agent that is specifically that is on the surface being called a investing agent. But it's it's just a an agent that you can give any any jobs to. And so he wanted us using the agent in in our daily life. So essentially paid for us to all have access to the agent. The agent is weird a little bit. And in fact, I might pull it up and and we can we can watch it do some things. The agent lies a lot more.

[00:10:15]  Bruce: What's the name of the software?

[00:10:16]  Cameo: Green Chart Portal. It's it's called Day Trade AI. I'll open it really quick and kind of talk through what it does. I know that this is a podcast, so people can't see the visuals. But the agents are really interesting, though, because if you're willing to spend the money, because you have to have it connected to a credit card and you have to have the APIs connected to it. But it'll keep going and it'll keep going and going if you give it a big enough thing. And maybe to the extent that it won't ever stop if you give it a large enough goal.

[00:11:00]  Unidentified speaker: That's crazy.

[00:11:01]  Brendon: I've never used one before. So you basically prompt it and then it prompts it. It continues to prompt itself. Is that what I understand?

[00:11:07]  Cameo: Prompts itself. And it's a pretty crazy kind of thing to watch. So this is this particular agent. And anyone could build an agent quite easily. This one, all it does is takes my API key and then I don't know what exactly. I don't know. This doesn't seem like a complicated piece of software. But to use the agent, you tell it what kind of an agent it is. So let's call this one a comic book author. I had told Peter and Bruce after an episode, I think last time that I had used ChatGPT to help me I've been writing a comic book about David Deutsch. That's called Many Worlds. And so that actually has been a lot of fun. I'll show you some of the pictures that Midjourney has made for me. But goal, write a comic. All right, so about David Deutsch and incorporate all of his philosophies and writings. Okay, so that's probably good. And then we deploy the coach.

[00:12:31]  Cameo: All right, so it's embarking on a new goal to write a comic about David Deutsch and incorporate all of his philosophies and writings. So then it'll think for a little while and it'll all will always. So it didn't need to think very long. I'll stop it from...

[00:12:44]  Brendon: This is incredible.

[00:12:46]  Cameo: It is really, really incredible. So it lists all of its tasks that it's identified. And what it starts doing is adding all of these tasks. So its first task is to retrieve all available comics. Now there aren't comics, but it hopefully is still figuring things out, extract important themes from David Deutsch's philosophies. And you see it's adding more tasks as we go. In fact, down here at the bottom, it's still thinking through all these different tasks it wants to do. But if we come back up here, let's see what it's doing. And it makes this list of things that it needs to do as it figures them out. And so now it's executing on this particular task.

[00:13:30]  Bruce: I take it it has access to the internet? Like where is it trying to get?

[00:13:32]  Unidentified speaker: It has access to the internet,

[00:13:33]  Unidentified speaker: yes.

[00:13:35]  Cameo: Yes, it has access to the internet.

[00:13:41]  Cameo: Incorporating these concepts into a comic about David Deutsch could be an interesting challenge, it says. But here it's outlined some things that it wants to incorporate, quantum mechanics, the concept of the multiverse, constructor theory, fundamental theory of physics.

[00:14:03]  Bruce: Where are you reading that? I didn't even see where it said that.

[00:14:06]  Cameo: It's in this executing area, right here.

[00:14:09]  Bruce: Oh, you're right, okay, I see it.

[00:14:12]  Cameo: Yeah, so it's kind of summarized it here. And then after it goes through an execution stage, so now it's telling me it executed. That task successfully, and it's highlighting some other things that it's putting together, his epistemology, which this is also just a really nice summary of, you know, what David Deutsch is all about. And so now it's kind of theorizing, the comic could feature a group of scientists exploring the multiverse.

[00:14:54]  Bruce: So it's come up with with concepts, right, coming up with concepts.

[00:14:58]  Cameo: And so now it knows that it needs to develop a script and a storyboard.

[00:15:02]  Brendon: And you kind of tell it, hey, I want you to go down this vein or this vein or like, like, or is it going just linear to what it decides?

[00:15:09]  Cameo: It just does what it wants. It's kind of a fascinating difference between this and using ChatGPT because if I was like, oh, I don't know if I want you to go down this path, I don't there's no input. There's no additional input after my giving it the goal. I can stop it and give it a new goal. I can stop it and readjust the goal. But it does its thing and it has its plan and it executes on its plan and it's just off to the races. And I gave it a task last week and I came back hours and hours later and it was still just busily doing this, going through all of these tasks and making...

[00:15:55]  Bruce: You have to deal with the halting problem. You don't know if it's ever going to actually complete or not.

[00:16:02]  Unidentified speaker: Uh-huh, you do.

[00:16:03]  Cameo: And so I have stopped it before and said, you know, okay, we're done with that. And I don't even know where to go and look and see the potential impact monetarily. I mean, I could go and look at my corporate credit card that they that it's tied to.

[00:16:22]  Peter: When are we going to see the comic? This looks really cool.

[00:16:26]  Bruce: Well, I noticed that it said is one of its tasks is that it has to collaborate with comic artists. So like, how is it going to do that?

[00:16:34]  Cameo: So this is this is the crazy thing about these agents. They lie like crazy all the time.

[00:16:41]  Unidentified speaker: I see.

[00:16:41]  Cameo: It will tell you that. And then and then it doesn't

[00:16:45]  Unidentified speaker: Right.

[00:16:46]  Cameo: or if it like.

[00:16:48]  Brendon: Well, could it could if it was connected to like Fiverr, though, which I mean, isn't the have isn't already done that, like basically hired a human before.

[00:16:57]  Unidentified speaker: So it has.

[00:16:58]  Cameo: And I don't when I've read about people doing that with agents, I guess I don't know enough about how this particular agent is created or what. I mean, this is designed to do investing. So but I didn't I haven't found any place I can connect it to things. I know it is connected to the Internet because I see it doing research. But but I don't when it says things like I'm going to go to talk to these people. I don't have any evidence that it has done that.

[00:17:28]  Brendon: Yeah, that's true. I'm sure this one probably doesn't. I'm interested to see what happens with the plug-in ecosystem on GPT because it kind of reminds me of like the iPhone, right? Where it's like iPhone's powerful in and of itself, but having all the different applications built on is what makes it like so powerful, right? And so the fact that now the GPT is opening a plug-in ecosystem where you can where you can basically be like Instacart. Hey, make me a meal plan with this caloric intake with this diet and food allergy list. And I want it to be under this budget and I want it to be this much per meal. And then it will build out the plan for you and then say, okay, put that in a grocery list. Okay, order those to my house and you can connect it with your Instacart and we'll do it. And so there's a lot of I haven't played with it. I think they actually want I want to say that the plug-ins like launch to all GPT plus users like this week or weekend.

[00:18:24]  Brendon: And so I want to get into it. We built a tool, an AI tool for my job that uses ChatGPT. It's really cool. It's trained on the specific basically case management software that I help assist with. And it's a Chrome plug-in that uses GPT. We trained it on specific help center data and like a bunch of other data that we had about the software. And we're in beta testing. We're going to deploy it. But it's really cool. There's so much you can do to basically take these language models and narrow them down into like a specific function. I've never seen the auto GPT or like this type of autonomous agent like live before I've seen videos of, but it's pretty crazy to see you doing it about this topic that I've heard Bruce talk about and just reading that is getting the prompts right. Like, it's cool. This is really cool.

[00:19:21]  Cameo: It's really cool. But I'm really excited for the plug-ins because I don't actually want an agent that I send out that I can't interact with. Like, it's a little like having an employee that you give like one sentence to, and then they go and do stuff and you're like, well, but you never came and checked back with me that that was like what I wanted or the right approach. And so, being able to just incorporate the work that I'm doing with and the ways I'm using ChatGPT and put it more into, you know, plug in across everything that I'm using so that when I get a million messages on LinkedIn, I can open a message and just have it essentially respond.

[00:20:17]  Brendon: One tool that I've been using lately that I love, I got into a beta of Augment AI. Have you heard of them before? So I don't know exactly what models they used. I asked like some of its proprietary, but we obviously use GPT as one of our sources. But it essentially, it will read like your email thread and it will also read your Slack thread and it will understand the context and then all you have to hit is a pound, one pound or hashtag, hashtag and you can choose to have it just reply for you and it will just reply for you or you can say, reply with my context and then you just say, hey, tell them this and then we'll just craft the email completely as you. Immediately hit Enter—[brief unclear word]—go in and edit it and send it off and it learns some more data about you and like how you speak and it becomes kind of like your personal assistant and it's great.

[00:21:12]  Brendon: I'm telling you, this thing is saving me so much time responding to emails. That's like half my job and it writes better than I do. So I'm like, there's no reason not to use this if it's getting the point across. And I feel like that's kind of going to be the future of this. I try to use, I don't know, have you downloaded the GPT mobile app yet?

[00:21:34]  Cameo: I downloaded it yesterday. I haven't used it yet, but I've been using ChatGPT heavily on my phone, you know, just through the mobile browser.

[00:21:45]  Brendon: Yeah, the Whisper integration where you can just talk to it now is so cool. Like yesterday, I was, we were at the store trying to look for a USB hub for my M1. And I was like, I can't remember what would be the best like USB hub, like what specs do I need? And I just asked it in the store, I just like used the Whisper thing and I was like, hey, I've got this MacBook, you know, what's the best hub? And it's like, hey, look for one that has these specs. So I did got that one. So I mean, it's just really using that being able to use that voice to text and on your phone anytime you need to, you have a question to something or you're not exactly sure. It's just cool to have it as that personal assistant, that body double. That's really what I have been trying to use it for. Same with music. Like I use this, I'll feed it like I'll write a lyric or like a chorus or verse and I'll be like, here's the meaning of the song.

[00:22:38]  Brendon: Give me some ideas for a potential second verse or give me like 20 ideas for a potential second verse or, and then I just will use little bits and pieces of it. But it really is just like that co-pilot and I saw that's what I'm excited for Microsoft co-pilot and Google co-pilot. Like the future is bright. It's going to be a crazy few months. I feel like, and then over the next few months, yeah, in years.

[00:23:03]  Peter: If I may just interject, here's a question for you, Brendon. Kind of going back to your ADHD comment, that really intrigued me partly because I didn't mention this before, but I'm a special education teacher. And lately, there was another eruption of emails in my work about phones in the classroom. What are we going to do about these phones? The kids are all using, you know, I get it. It's kind of annoying when you're trying to run a classroom and, you know, there's half the kids are on their phones tuned out. But you know, it also occurs to me that there's just so much negativity and fear about all of these technological changes. And you know, there's a long history of this going back to math teachers protesting, you know, there were actually protests. You can see the pictures of math teachers protesting calculator use.

[00:24:00]  Bruce: I know it's like, we only do abacuses in this house.

[00:24:04]  Peter: But I mean, why are I just I just wonder like, why are people so fearful of these these technologies that can can positively impact people's lives?

[00:24:14]  Brendon: I think partially it's marketing on the side of these AI companies. If that you know, it's the there's no bad press if they're getting in front of, you know, the big cameras and exposing people to this, it's just like, I think it's partially marketing, I think it's partially smart that people are afraid of it. But I also think that just because this is such a crazy technology, I think there are genuine risks. But I also do think that and you know what, this is kind of a good lead into what we were talking about before the podcast that I wanted to hit on where I feel like, you know, you were talking about like the impact of technology on music and like, how there's some negative benefit and negative negative effects of those. I think in my opinion, when we really hit that level of automation across everything where it's like, I will be able to prompt with natural language the type of song that I want to produce rather than going in and taking the years of knowledge that I've had to like build the synth tones and mix the songs and all this stuff that would take me hours and then I can just prompt it.

[00:25:17]  Brendon: Yeah, that means that there's going to be a kid, you know, you could have a kid in Indonesia that has access to a laptop and can make like the biggest bangers ever. And same thing with generating video, you know what I mean? Think about like Christopher Paolini, he wrote like Eragon, right? Imagine if he had had the ability to just make a movie from that idea, like as a kid, you know? I think it's going to level, I think it's going to level the entertainment industry. It's obviously going to level a lot of industries, but anything that's art, you know, with this being generative AI, it's going to level it. But when you think of level, that means bring it down. It's going to put the tools in everybody's hands to make this stuff. I think for me, it's, you know, I've been trying to chase music professionally for years, and I've had millions of streams, I've played in different countries, I've, you know, I feel like I've had levels of success.

[00:26:07]  Brendon: And it's like already before this whole AI thing, impossible to make a living in music. And it's kind of made it, it's kind of made me hit this point where I'm like, well, what's the purpose of doing music? What's the, and then it kind of goes back to your saying, what's the purpose of going to school? What's the purpose of all these things that we just do as a society? Learning, you know, I feel like once it's once so much of that is automated, it kind of goes back to almost a philosophical question of like, what is the purpose of this existence? And like, what is if you had all the stuff taken care of that you didn't even think you would have taken care of someday, what would you do with your time? Why do you do music in the first place? And I think I hope what we will do is just inspire more creativity is my hope. And that will be in this place where it's like the part of it will be the, hey, the spirit of learning.

[00:27:04]  Brendon: Why are we actually learning? Is it learning to get job placement and to, you know, die for a corporation? I don't live your life working for a corporation. You know, it just begs a lot of questions. I think it's going to break a lot of people's realities. I think it's going to make us all reevaluate why we do what we do when everything's automated. And I hope that that brings a more pure experience. And so to kind of round this idea out, I think after I saw what Midjourney was doing, after I saw what Runway was doing, after I saw DALL-E and all this, I just saw the writing on the wall and I mean, like Google's released stuff about music, like they have a text-to-music generator. It's not perfect yet, but it's pretty good where you can just use natural language to prompt create a song and say, I want to make a song in the style of this person with this type of bassline and in this BPM and this key, this chord progression, and it will build it for you.

[00:28:01]  Brendon: And so I was just like, man, like if this is going to be automated, like we're going to get to the point where we've got a corporation can purchase an AI that understands all of the parameters of music theory. It's going to understand all of the technique for mixing and mastering and for basically every style of music out there. And at that point, they'll be able to just create royalty free music based on just randomly generate or generate within these chord progressions, this type of music that based on user data when we know that we have an older demographic coming in the store from 9 to 11 a.m. So we're going to be playing jazz, but once it turns night time, we're going to have our music AI create more like nightclub music. The thing is, it's just going to put a lot of corporate musicians out of a job. Definitely, that'll probably be who goes first.

[00:29:01]  Brendon: Same with kind of like stock photos and things like that. But hopefully, it will still leave room for artists to use it to create unique works of art that technologically weren't possible before, kind of like we already are doing. But we'll see.

[00:29:14]  Unidentified speaker: Who knows?

[00:29:15]  Peter: I mean, I guess in a way people are right to be fearful because it, you know, as you've illustrated well, it's just going to change everything. I mean, it's going to make, turn, optimistically, turn us all into artists and engineers and writers of comic books, you know. But I, you know, we talk a lot about optimism on this podcast, which is very much associated with David Deutsch's worldview. And you know, I tend to, I'm sort of an extreme optimist. I think, which is maybe one of the things that made me gravitate towards David Deutsch. But I guess that in one area where I have a little bit harder time being purely optimistic is more aesthetics. Like, I always ask, I wonder a lot, is music getting better? I mean, do you think it's really improved?

[00:30:21]  Unidentified speaker: It's hard.

[00:30:22]  Unidentified speaker: I do.

[00:30:22]  Unidentified speaker: Yeah.

[00:30:22]  Unidentified speaker: You do?

[00:30:23]  Unidentified speaker: You think it is?

[00:30:24]  Brendon: Yeah, I know a lot of people are like, oh, well, the oldies. But like, what people don't understand is it takes time. It takes time for something to become an oldie. And the creativity now is beyond the wildest capabilities you could possibly think of. I mean, back in the day, it was like, you brought like Bing Crosby and like Frank Sinatra and like even, you know, through the 70s, 80s, 90s, even early 2000s, you're using like tapes. You're in the studio actually sampling, you're cutting a tape and you're pasting it and you're and then like, it's an extremely intensive process, right?

[00:31:03]  Peter: What Floyd had to do to get all those sounds and stuff is great.

[00:31:07]  Brendon: Yeah, you're at most using five instruments or maybe two guitars, a bass and a vocal drums, right? And that's basically the palette you have to paint with. Yeah, now you're going to get the record deals, right? Maybe there's some artists making it big. It's now a lot more homogenized, but at the same time, people are it's a renaissance in art right now. I'm telling you, electronic music, which is what I'm in, it's bigger than it's ever been. Festivals are selling out in minutes, tens and hundreds of thousands of people. So it's music is still very much alive and people are innovating and doing stuff that was completely impossible. And so now there is more noise out there, right? You're not having like Republic Records signing their bands and Atlantic signing their bands. And that's just what everyone listens to because it's a lot more. It's different.

[00:31:56]  Brendon: The ecosystem is now driven by TikTok where algorithmically driven by AI basically, right? Like people are now connecting with artists that are doing stuff in their bedroom, like 17-year-old kids, one of my favorite artists in the entire genre, 22 years old. This kid just learned knows how to produce in a DAW and he's making the craziest house music. So it's been a genre since the 80s, but he's pushing boundaries. He's already collaborating with Skrillex playing headlines. You know, his name is ISOxo and Knock2, both of those. But I think that there is so much good music coming out actually. And for me, yeah, I just think there's so much. There's so much that can be done with now this this mindset of because for me, it was like once I realized, I was like, man, I can go into software and work from home and, you know, pay my bills or I can try to grind continually to make it in music for a fraction of that.

[00:32:52]  Brendon: And this made me realize I was like, I don't really care about monetizing my art at the end of the day. Like, I really just want to make art for art's sake. And now I have the tools and it's helped me allow me to be like a lot more experimental. And so now I'm making songs with multi-genre, cross-genre, all sorts of instruments, all sorts of sounds. That stuff is not possible. If Beethoven was alive right now, he would have been a music producer. He would have been like Hans Zimmer. You know, like it's just the tools doesn't, I feel like the tools doesn't affect art as much as we think it does. I think there's so much good quality out there. People just push boundaries and make cool stuff. So I would say there's so much innovation going on now. There's so much new territory that we don't even know how to fully cover it right now. And we're continuing to, it's like a new genre, new artists pops up all the time.

[00:33:41]  Brendon: And so anyways, that's all I have to say about that. But I actually am very excited about the state of music. I feel like it's better right now for innovation sake of just like new genres, new sounds. Then it's been in as long as I can remember. Like, I think it's in a good place as far as creativity sake goes.

[00:34:01]  Peter: Okay. Well, I guess you convinced me. I guess if you're looking back at the past at Bach, Beethoven, Hendrix, whatever, of course, nothing can compare to these geniuses. But at another level, what I hear you saying is it just opens things up and the next Bach will probably be a teenager working

[00:34:24]  Unidentified speaker: At Starbucks.

[00:34:25]  Peter: in their room. Yeah.

[00:34:28]  Brendon: Yep. Exactly.

[00:34:29]  Bruce: Have you guys seen this article? Can you see what I'm sharing?

[00:34:32]  Unidentified speaker: Yeah.

[00:34:33]  Bruce: This is an article by Kenneth O. Stanley. Idea amplification is a really exciting potential of generative AI. Are you guys familiar with Kenneth O. Stanley? Peter, maybe you would be because he comes up a bunch of Deutsch circles all the time.

[00:34:48]  Brendon: I'm not either.

[00:34:49]  Bruce: So he wrote a book called Why Greatness Cannot Be Planned. And so what he is is he is a researcher that is trying to work out the problem of open-endedness, which is many people don't even know that there is a problem of open-endedness. So in a nutshell, take a look at human creativity and the creativity of biological evolution. And they seem to have some sort of open-endedness, where they can just keep creating new sorts of innovations. Whereas when we try to program evolution, it tends to at least before ChatGPT, it tends to be very narrow. So that's why we often refer to it as narrow AI. So he's been trying to solve the problem of how do you solve the problem of open-endedness? How do you write an algorithm that's actually going to create increasing amounts of novelty and just kind of be open-ended in what it can do? And that's what his book is about.

[00:35:56]  Bruce: So he wrote this article about his what he's how he sees generative AI, and he refers to it as idea amplification. So basically, he mentions, and I'm sure a lot of people have had this experience, that you accidentally come up with some song in your head. So he would start to hum a song and he would realize, oh, this isn't some song that I've heard somewhere else. I've made this song up. And then you kind of forget it. So he started to actually tape himself and he collected over time all these little songs that he had composed. Now, we've all had experiences, maybe even with songs, but are similar to that, where we have these ideas and we just lack the skill. There's just no way that we can act. And if you wanted to go make that idea into a reality,

[00:36:42]  Brendon: Years of study sometimes.

[00:36:44]  Bruce: right? Or you would have to go pay lots of money to pay somebody else. And so basically, we're creative beings that have these ideas all the time. We always talk about everybody has the million dollar idea. That's probably true, right? So we have all these ideas all the times. And because of the way things are, the limits of skill sets and the limits of money and things like that, most of these ideas just die and they don't go anywhere. Generative AI, what it is, is it's idea amplification. This is the way he's saying is trying to suggest it. It's that now if I were humming that song and I said, oh, that's actually a pretty good song, I could actually go produce it, right? I could get the AI to, I could hum it to the AI and then I could start working with the AI to come up with an orchestra that plays that song. Things like that. I've had numerous ideas for video games I wanted to make and I've never taken the time to go learn how to, both learn how to program video games using Unity 3D, but also to create my own models.

[00:37:52]  Bruce: I would have to go use pre-existing models if I wanted to make a game. For school, I did make a game and use a lot of pre-existing models. You can do a lot with pre-existing models. But now that they're making generative AI that will make 3D models for you, describe what it is you want, and then it's going to come up with the model. Eventually, I'm going to be able to take these ideas for video games that I have, and I'm going to be able to produce a first version of it.

[00:38:20]  Unidentified speaker: Yeah.

[00:38:21]  Brendon: I think it'll go beyond a first version. I think you'll be able to, we'll get to the point where it understands video game storyboarding, it understands every aspect of that profession. You'll be able to say, hey, make me an Elden Ring 2 and it will do it. It'll be able to do that.

[00:38:39]  Bruce: We should really expect that with generative AI, we're in some sense unleashing human creativity. The generative AI is not itself creative. It needs that initial spark or prompt that comes from a human. This is his argument. Anyhow, you're welcome to disagree with him, but I'm just summarizing the article. It needs that initial spark of creativity from the human, but we have those sparks all the time. It's just mostly they die out because skill sets are hard to come by. So I just thought that was a really interesting thing. And by the way, he's just somebody worth reading anyhow. He's a very creative guy who's come up with a lot of really interesting theories and ideas around what creativity is.

[00:39:19]  Cameo: So one thing that stands out to me as I'm using, and I find this especially with Midjourney, the better you understand the art form, the better you're able to utilize the models or the AIs. And even with this guy's example of music, you still need to have a working knowledge to know what to ask for.

[00:39:51]  Bruce: It doesn't really stop the need for skill. It just makes it so that it's much easier. The barrier to entry has been taken down, but there's still incentive to learn the skill and to become skillful.

[00:40:02]  Cameo: Right. I mean, to really use Midjourney effectively, you need to understand art history and how humans have described art forms across the eras so that you can make this in the style of Da Vinci. Because you can ask Midjourney to just do abstract things, and it likes that. I mean, not to overly personify it, it comes up with really interesting ideas when you give it very open-ended stuff. But if you want to create an aesthetic that's in your mind, you have to be able to find the right words that will make that generate. And it can be very painful to find those words. I was working on a presentation that I made all of the visuals inside of Midjourney. And there was a certain aesthetic that I was trying to use or create and that I then wanted to kind of pull across all of the different concepts that I was trying to get past. And it took me a long, long, long time.

[00:41:28]  Cameo: And a lot of just, now I'll try this, now I'll try this, now I'll try tweaking this one word. Because when you give... It's the same thing with ChatGPT. When you give a prompt into these AIs, different words can have very outsized impacts on what it chooses to create. And then you're like, oh, but that's totally the wrong feel. And sometimes you have to almost unwind to try and figure out what's the word that it's kind of making it go sideways on, that's impacting the tone of the voice or that's impacting the output in such a way that just is kind of ruining it being able to create what I want it to create. Which is maybe true of life anyway. Part of why I think jobs are hard is we're not great at communicating. So if your boss comes and says, I want XYZ, and you go and make XYZ, he probably didn't communicate effectively what he wanted. And you come back and he's like, well, that doesn't match what I wanted, because he's not great at describing what he wants.

[00:42:36]  Brendon: Yeah, I think it comes up. My advice for anybody that wants to get into AI and get the best out of it is you just have to talk to it. You have to learn how to communicate with it. It really is like prompt engineering, right? And I think it's like the more you play with it, the more you understand its capabilities, how to talk to it. And so I really think everyone should just be doing it now, just talking to it, getting to understand it.

[00:43:01]  Brendon: Applications will start to apply themselves as you—or start to just appear, as you start to understand what it can and can't do. I was mentioning to Bruce, one of the things that I love is just using it for education now. For example, one way that I use is, hey, I don't understand how to do... For example, I don't understand how to comp vocals. That's a good one. I don't know how to do this best. I want you to pretend that you are now music professor, GPT. Write me a college course about comping and editing vocals. It will take all the information it has about that. Then I'm like, okay, give me a transcript. And then it will be like, okay, it has the professor talking about it. And it teaches these concepts. And you'll learn from it. And I think it's just like, it took me a second to get to that. But for me, and I think it goes back to the ADHD thing, is like everybody has different learning, ways of learning, you know?

[00:44:03]  Brendon: Like for me, like having somebody there teaching me and talking to me and letting me ask questions, that's my best way of learning. Some people can just read text and get, you know, really internalize it. For me, it's that way. So I hope in the future, you know, you'll be able to learn, hey, maybe you learn best talking to someone. So they'll attach a large language model to a generated like, you know, avatar you'll be able to converse with through natural language like talking. Maybe some people will do better with typing. Maybe some people are better with visuals. And so we'll be able to generate visuals based off of the text or copy. And you'll be able to be educated in your own preferred way when you're in school. And yeah, that might completely tear down the existing schooling system and testing. But maybe that's a good thing. Maybe I don't remember who it was, but I was listening to someone talk about it.

[00:44:55]  Brendon: And they're basically saying, right now, you get rewarded for just guessing on multiple choice questions. At the end of the day, if you guess on a multiple choice thing and you get it right, it's good. You're positive. But in the real world application, like in the job field, just guessing, closing your eyes and throwing a dart is not usually the best way to like go about completing tasks where you're trying to do. And so, you know, in the practical world, it's a lot more about learning, you know, skill and how to perform at a specific job. And so hopefully it takes away a lot of the—sorry for, you can edit my word out—but bullshit in the education space of just wasting time learning things and doing things that don't have real world application to where you just have the ability to learn anything that you want in the format that you want through AI.

[00:45:53]  Brendon: I think the education side of things is going to be fantastic. It's going to be absolutely game changing. I think about my son going to school. I have a four year old. He's turning five, you know, in this next month. What is his life going to be like? What is his school going to be like? It's going to be a completely different experience than mine. You know, imagine it's like the difference between being raised on that, the abacus to the TI-89 to like, you know, to now ChatGPT and large language models like an AI. Like that is going to be a significantly different experience. And it's going to be interesting to see what happens.

[00:46:33]  Bruce: By the way, well, I just wanted to comment that training sites are taking a giant hit over ChatGPT. A lot of these sites that have videos that training videos, they're finding that there's significant drop off to subscribers.

[00:46:50]  Brendon: Well, you can scrub through YouTube videos that hopefully have that one nugget of information you're trying to find that's like very specialized, especially once you get specialized. And or you can just ask GPT to find it for you. It's a different experience. For music production is a great one. We were talking about this last night. I was helping my brother Tyler get set up on music production software on his new laptop. And he got some controllers and we're like, how are we going to link these controllers? I don't know how to map these. So we asked ChatGPT and I was like, do you realize how much of an advantage you have? Don't Google it. ChatGPT at first, then Google it. If you don't know how to do something inside your music software, because I had to go ask other producers. I had to go find some other producer online or become friends with them, be like, how did you solve this problem?

[00:47:42]  Brendon: Or find some YouTuber had uploaded it, and I had to go scrub through a bunch of videos to find the information I needed. So it's already making such an advantage, such an advantage to be able to learn information, especially about specialized fields that it already has data on.

[00:48:01]  Cameo: Well, and Google has long had a problem wherein, because everybody's got an opinion, and now everybody's got a website where they're sharing that opinion, you can sometimes, if you have an obscure enough question, you can be four or five pages into Google results where it's either pages that are selling something or it has become more and more and more difficult to get good information out of Google

[00:48:35]  Unidentified speaker: Agreed.

[00:48:35]  Cameo: without having to really slog through a lot of...

[00:48:38]  Brendon: I'd say any social platform, like right now, it's so cluttered and monetized. Everything's already all built off of attention and ad revenue rather than actual information. It's like content rather than actual important information or art. It's a big problem with the way that like Meta and other companies deployed AI. Very tricky and makes it difficult to get good... Yeah, I could go on forever about that. I don't think that's probably the right vein to go here right now, but yeah, I agree. It's nice to be able to just get the information without having to deal with all that.

[00:49:19]  Peter: Can I ask you another question about music, Brendon?

[00:49:22]  Brendon: Absolutely.

[00:49:24]  Peter: I'm going to play devil's advocate here a little bit, and it's interesting for me because usually I'm in the camp of being the crazy optimist in these kinds of discussions, but I'm curious how you'll respond to this. So do you know the artist Nick Cave?

[00:49:44]  Brendon: No, I don't.

[00:49:45]  Peter: Okay. He's kind of like a singer-songwriter guy. He's been around since the 80s, but he's also a writer, kind of a very literary kind of guy too. But someone wrote to him and asked him and showed him one of his songs that was written by ChatGPT. Like someone said basically, let me write a song in the style of Nick Cave or something like that and showed it to him. Let's just say he did not like it very much. Let me just read you just a short paragraph here about what he said and I'm just curious how you would respond to that. Songs arise out of suffering, by which I mean they are predicated upon the complex, internal, human struggle of creation. And well, as far as I know, algorithms don't feel. Data doesn't suffer. ChatGPT has no inner being. It has been nowhere. It has endured nothing. It has not had the audacity to reach beyond its limitations.

[00:50:53]  Peter: And hence, it doesn't have the capacity for a shared transcendent experience, as it has no limitations from which to transcend. ChatGPT's melancholy role is that it is destined to imitate and can never have an authentic human experience, no matter how devalued and inconsequential the human experience may in time become. What makes a great song is [unclear words] not [unclear words] its close resemblance to a recognizable work. Writing a good song is not mimicry or replication or pastiche. It is the opposite, is an act of self-murder that destroys all one has strived to produce in the past. It is those dangerous heart-stopping departures that catapult the artist beyond the limits. He goes on and on and on. But what do you think? Is there a danger that as music moves away from becoming like about humans and human experience, that it could get worse?

[00:51:57]  Brendon: Honestly, I feel like that's some pretty gatekeepy stuff. That's not really my ideology behind music at all, actually. I would say I don't think that... I feel like that's just such a... Yeah, that feels anti-technology to me and it feels gatekeepy because the truth of the matter is it's just another tool. Like Ableton, what I have now, that completely eclipses the studios of probably what he was working in the 80s. But I'm sure his producers, if he's still making music, are using that tool that's speeding up his workflow immensely. I think that maybe in the context of... And even then, I don't know if I agree with this either, because have you heard of the whole thing with The Weeknd and Drake and the Ghostwriter and all of that? Have you heard that story?

[00:52:47]  Peter: Not really, no.

[00:52:49]  Brendon: So basically the two biggest artists in the world, one of them is The Weeknd and the other one is Drake, you've gotten to the point where now you can have... I could sing, even just recording into my AirPod earbuds here, I could sing a verse, right? I could write and sing a verse, and then it will actually completely clone my voice into like The Weeknd's voice or like Drake's voice or Kanye West's voice, right? And it does it really, really well now. So we're talking, this is Midjourney Reaper at the Door type vibe for the music industry. It's already here. It's happened. There's an artist that he put a sheet over his head and he put like sunglasses over his eyes. You don't know who he is, an anonymous person called Ghostwriter. And he's just dropping these Drake tracks, like these basically AI generated Drake tracks and these AI generated Weeknd tracks on all over.

[00:53:46]  Brendon: Can you guess what people are doing? They're listening to those songs.

[00:53:50]  Unidentified speaker: They're getting millions and millions of plays and everybody loves them.

[00:53:54]  Brendon: And you go on TikTok, there's tons of people that are like, wow, like this song is so sick. They have no idea that that song was generated by a completely different person. And they just use trained data from these vocalists. So there's kind of two mindsets here. You've got Grimes. Have you guys heard of Grimes? She's been a musician for a long time.

[00:54:17]  Peter: Yeah, she embraced it. Everyone can use her voice.

[00:54:20]  Brendon: She lets you train her voice. Yeah, she's trained her voice and she lets artists use it for free.

[00:54:27]  Bruce: And you can throw it out there.

[00:54:29]  Brendon: She gets a percentage of it. So I could, and I actually have considered doing it for this new record. I'm working on having a Grimes feature, downloading it, writing something. I'm a good writer. I'll write something in the style of Grimes and then have her voice on there. So I just think like it is going to crumble the music industry, but good, because the music industry is dumb right now. It is an absolute mess. It's people who have money are, you know, like artists that have, you know, mommies and daddy's money are the ones that make it. And it's very gate-kept and it's very who you know, and it's very toxic. And so, you know, you look at these big acts like The Weeknd or The Drake, they have entire teams of underpaid writers creating their stuff. It's not like they're getting out and stabbing themselves in the heart and saying, here's my art.

[00:55:21]  Brendon: Like that is not what's commercially being, you know, consumed right now. It's already a bunch of writers writing stuff for an artist that's then singing it and they're being marketed as it. It's all ghost writers that are underpaid. So it's like these guys, like, I hope it kind of levels it. I hope it kind of breaks down the industry to where you can't really create the same kind of like, I don't know how to put this, but just it's already not good. And it has been that way for a long time. You look at any hip hop, you look at any genre. People are complaining about the industry all the time. So I hope it does level it. I don't think that the record labels and the industry hasn't figured out what people actually want. And I think it goes to show that people will consume stuff that they think sounds good. And there's a lot of ways to create that.

[00:56:19]  Brendon: And I think AI will help level the playing field where I can write songs like Drake without a writing team and use that emotion that I have, like we go back without having to have that barrier to entry. So I don't really agree with that mentality. I can see how there's a lot of people who are very scared Reaper at the door vibes. I get it. I totally understand, especially if you made a career in it and you've been doing this professionally. It is scary and commercial work is going to be completely destroyed. I mean, it's the same thing as stock footage and stock video. But at the same time, I think I think it definitely can create amazing art. Like I use it to say, here's the feeling of the song I want. Here's some ideas. Give me some other lines and it will come up with amazing lines that totally feed into and add to the lyrics. So I just yeah, I think it's a cool tool.

[00:57:09]  Brendon: I already use it all the time and people connect to it and they love the lyrics. I was writing one with my friend over Zoom and I showed her ChatGPT and she's like, what? These lyrics are amazing. I was like, yeah, we can use this piece, this piece, this piece. And then we built around it. So I think maybe he's right and the aspect of maybe, let's hypothetically say, music AI exists that can do it all. I think people, there will be a time and place for that. People will still dance to that. People will still like it. Most people will not be able to pass the Turing test with music. They're not going to be able to tell whether somebody made it or not. But I will say there's always value to human creativity. I agree on that aspect to where we'll come up with unique ideas that can be made.

[00:57:54]  Bruce: So yeah, can I actually take a stab at answering your question, Peter?

[00:57:58]  Peter: Of course.

[00:57:59]  Bruce: So I do think we're talking about possi- like the two theories being put out there as competitors might not actually be competitors, right? So one person is saying, the musician is in essence saying, ChatGPT or AIs, generative AIs, will never be able to compete with humans in terms of creativity. Well, if you're talking about a straight contest of just a human without ChatGPT and just ChatGPT without a human, they're probably right. So, I mean, like, I don't think we're ever going to come up with AIs that come up with some new innovative type of music. And there was no human involved at all for that initial spark, as Kenneth Stanley was suggesting. I don't even think that's possible because these are probabilistic models that are built on top of only what has come before them. On the other hand, that's just a false dichotomy, right? So if I could use, like, maybe an analogy with chess.

[00:59:03]  Bruce: OK, so you have the world master chess, Garry Kasparov, and he gets beat by Deep Blue. And so then people hail that computers are now the best chess players in the world. Well, if you're only either a human or a computer, that's true. A computer will always beat a human now in chess. But if you really want to be the best chess player ever, the best chess players in tournaments are actually hybrids between computers and humans. They have these human that works with a machine and uses the AI. They will beat an individual AI and they'll beat an individual human because that creativity from the human and the processing power of the machine collectively are doing things that are different. And the computer is not generating things in the same way a human would. So you've got this combination that's actually more powerful than either individually.

[00:59:59]  Bruce: And I suspect that's what, like, if you were to look at Brendon's answer, go back and listen to it. I think that's what he's saying, right? Is he saying, well, I'm using it as a tool, right? I'm not doing this entirely from scratch. He's not doing this entirely from scratch without him in the mix. I think you're going to find that this is going to be true for everything, that the best music will come out of someone who uses generative AI and they are creative. And it doesn't have to actually be this dichotomy between just a human or just a computer.

[01:00:31]  Brendon: I mean, that's that's all on the predication that we don't have a AGI or ASI in the next few years.

[01:00:36]  Bruce: But I'm assuming that we're talking about [unclear] landscape.

[01:00:40]  Brendon: We're talking about current landscape, actually, right now. This is already in play. This is already in play in the music industry, and we are about to see how it unfolds. So you're right. And I like that answer.

[01:00:51]  Peter: Yeah, no, I agree. I think if you look back in history, I mean, it used to be that to become a musician, you'd have to go to music school, and which meant you probably had to be rich. And, you know, so this is a long process of things opening up of the, you know, like you say, the word the gatekeepers going away. And now we're at the point where a teenager can make excellent music in their in their room with no formal training whatsoever and potentially reach a billion people, billions overnight. And I think it's probably on balance for the best.

[01:01:31]  Cameo: In a way, it's very similar to the invention of the printing press. You know, prior to the printing press, you could get books. They were primarily the domain of of the church or, you know, the monks were the ones copying the books so that there could even be books. But once the printing press came along, it it decentralized knowledge and made it so that so that so that knowledge could be pushed out across the world in a way that had never happened before. And it it ignited the Renaissance. It it's you know, it changed the world. Everything we have now is kind of can be tracked back to some of to some of that happening. And it'll be the same way with with the AI, you know, somebody still has to bring that creativity. Because, yes, unless we get AGI's, the the AIs are creative, but but they need a human to not just be derivative.

[01:02:39]  Bruce: Yeah, and I was just going to comment that when Brendon said, well, you're assuming not AGI, he's completely right. AGI would be total game changer, right? Everything I say goes out the window with at that point.

[01:02:51]  Cameo: But we have on this podcast a theory that that none of this is none of the explosion of AI means that we're any closer to AGI.

[01:03:02]  Bruce: I agree. I completely believe that the models that are currently going down are not paths to AGI at all.

[01:03:11]  Brendon: I think it will probably just speed up the ability to effectively do research to head towards that direction. It doesn't mean that the current algorithms they're using are going to result in that. But I do think it's going to help with efficiency across that area. And there's a lot of now attention on that area of science. So I wouldn't be surprised if it has a direct correlation with that if it does end up happening. I think my hope is it's going to fix a lot of human suffering even before AGI.

[01:03:45]  Bruce: Well, there's always going to be a need for regular AI. Regular AI is by itself a completely mind-blowing field even if you're not factoring in AGI. And this is something that I've kind of tried to bring out in this podcast. I know that a lot of the fans of David Deutsch really dislike AI because it's, quote, not AGI. That's totally the wrong attitude, right? Even after we invent AGI, AI is going to continue to be one of the most important fields of knowledge that we're ever going to have because you still need to automate tasks. You still need to be able to have mindless computers doing things for you. There's all sorts of reasons why AI is by itself an interesting field. And Brendon has this right. It does indirectly impact AGI. For one thing, I've argued that it at least shows you how not to make an AGI. And that's not a small thing. Like Popper says that the way you come to understand a problem is by failing to solve the problem.

[01:04:50]  Bruce: AI has a number of AGI failures that are going to be part of how we eventually come up with AGI. Also, I said it has nothing to do with it, but it's not quite true either. I think ChatGPT teaches us something important about AGI. And I've argued on the podcast and elsewhere that it's the first non-narrow AI, right? It's still built like a narrow AI, but it clearly can mimic just about any other kind of machine learning or artificial intelligence. So you can't call it a narrow AI anymore. And I think there's a really specific reason why. Now, I'm guessing this is a conjecture, but I think it's because language is universal and that it is one of the key missing pieces in understanding AGI. We kind of now have this hint. If you want to study AGI, language is going to give you a hint. As the universal language of humans is going to give you a hint how humans are able to explain things, how to become universal explainers.

[01:05:50]  Bruce: I don't think that ChatGPT would have worked unless language had some sort of universality to it that wasn't fully understood prior to this point. Now, that's a big hint on how to go about trying to figure out what the AGI algorithm is. So there are going to be some indirect connections like that for sure. And if nothing else, they just make it easier to try to figure things out, to process things, to process information. It would be like saying, you know, do you think the Internet would help create AGI? Almost assuredly it would, right? If for no other reason than it connects us all together.

[01:06:27]  Brendon: It helped create AI.

[01:06:32]  Cameo: So let's talk about legislation for a minute, because Sam Altman went in front of Congress earlier this week and said, oh, we've got to legislate this. And, you know, we've talked on this podcast in other episodes about some of the ways that both Google and OpenAI were talking about the AIs in a way that, you know, you had that engineer from Google that came out and said, oh, this is intelligent. And I think we had even talked about an article that was from several engineers at Google who said, Google's just trying to make it seem like they have more than they actually have. And the technology is further along.

[01:07:23]  Unidentified speaker: It's marketing.

[01:07:24]  Unidentified speaker: Yeah, marketing.

[01:07:25]  Cameo: It's marketing because they want people to believe that these AIs are more powerful than they are. But I felt like watching Sam Altman get up and speak in front of Congress was the punchline of the joke that they started months and months and years ago where they know that open source, they know that they are in trouble from open source people being able to recreate.

[01:07:56]  Brendon: Yeah, they can't monetize it.

[01:07:59]  Cameo: Google came out and said, we don't think we have a moat to protect us from people really just being able to recreate what we've already done. And in the open source world particularly. And so the best way to constrain a piece of technology is to heavily legislate it. It inhibits innovation. And that's what they're doing. They're trying... He essentially got up and said, make it so that nobody else can develop AIs except for us. And make it so other people have to essentially get some sort of a registration. So it's interesting because it seemed to me like all of the kind of fear mongering about how smart these AIs are going to be is just designed so that they can try and control the AIs themselves and try and ensure that other people don't have the ability to undercut their potential profitability.

[01:08:58]  Bruce: Tell us how you really feel, Cameo. Don't hold back.

[01:09:04]  Brendon: I think that's a very valid point of view. I totally have had similar thoughts and have heard similar criticisms. I'm curious. I think it's hard because we don't know actually if these things are dangerous or not yet, but they sound like they could be dangerous and that they should be legislated in certain ways. But I also understand exactly what you're saying, so it's tough to say.

[01:09:27]  Bruce: Can I give my own stab at this? So I'm going to be maybe a bit more neutral than Cameo, although to be honest, I heavily lean towards exactly what she just said. So first of all, let's understand that OpenAI, perhaps completely unintentionally, confuses AGI and AI as if they're the same thing, right? And so when they start talking about AGI safety, people don't differentiate. They don't talk about AI safety, which is a real thing that really exists today and is an important subject. And they kind of intermix it with AGI safety, which is not a real thing today. We don't even know what that is. And honestly, there is no safety program you can conceive because we don't understand it well enough yet, right? And I talked about this in our episode 49, AGI alignment and safety. So they're mixing two things that don't belong being mixed, right? And furthermore, it leads to the media kind of...

[01:10:35]  Bruce: the media will portray it, has incentive to portray it as if it's all one thing. Let me split it out, though. And let me just say, look, ignore AGI safety for the moment, because that's really a separate subject. AI safety is a legitimate question. It really, really is. Okay, we took classes on it in school. There's a whole set of techniques, engineering techniques and science behind. How do you try to make AI align with human need? And there will always be a need for that, because these models and what you do with AI, it's hard to specify to them, this is what I really meant. And so try to come up with how you try to align the AI with what it is that you actually are wanting. That's actually a hard problem that needs to be solved, and they're getting better with it over time. So first of all, let me try to start with the assumption that Sam Altman, this is maybe not a fair assumption, but I'm going to try to steel man Sam Altman's argument as far as I can.

[01:11:45]  Bruce: Okay, so that I'm attacking the steel man version. Let's assume that he's really only talking about AI safety. If we're only talking about AI safety, then it's a fair subject, and it's a hard problem. There are many things that already exist out there where we legislate there have to be certain minimums. So maybe you could make a case that AI safety falls into that category, and there should be certain minimums. Now, at this point, if you know anything about this, probably one of the first things you're going to think is, well, do you actually want senators coming up with this? Do they know anything? Well, no, they don't. But let's steel man this. Maybe they're getting experts. Well, who's an expert? Sam Altman, right? So maybe they're going to these experts, and they're using these experts to try to come up with something that's a reasonable amount of AI safety, and try to legislate that as the minimum.

[01:12:39]  Bruce: Is that similar to other industries? Sure. Now, here's the thing, though. This would probably be the most steelman case I could make for it. I do happen to know, and this is my libertarian side coming out, that almost invariably, that then becomes a barrier to entry, right? Now, even if you want to argue it's a good barrier to entry, or it's a necessary one, there's always a downside to it, that now you have less competition, which is why companies are almost always in favor, the ones that already have the best safety standards. And let's be honest, OpenAI does have some really good safety standards, compared to probably anybody else right now, if for no other reason.

[01:13:22]  Brendon: I'd take them over Facebook, or Meta, or any of the other big platforms at this point right now.

[01:13:28]  Bruce: And Facebook, Meta and all that, they're not doing chat, right? That's, if you want AI safety for Chatbot, for large language models, there's nobody out there that understands it as well as Sam Altman and OpenAI right now. If for no other reason than because they're the furthest ahead. They know it the best. They've got the engineers who actually built the best version of it. On the other hand, there's just no way. And Sam Altman seems like a really good guy. Like when I hear him in interviews, he seems very sincere that he wants what's best, right? But it is impossible to know your inner motivations. Inner motivations come from the subconscious, and they are always biased in your favor. So this is what I think Cameo was really getting at, that if Sam Altman gets himself placed as, I'm going to help Congress come up with AI safety standards.

[01:14:23]  Bruce: And even if you want to argue that that's something that needs to happen, it absolutely will benefit him, right? And there's no way his, at least his subconscious, isn't aware of that fact, and it isn't a part of how, why he is interested in this subject, right? And so I do think that this would be kind of all sides. Maybe there's a good case for it. Maybe there's even a good case for using Sam Altman, but there's absolutely going to be unintended downside consequences, precisely like Cameo is saying. And those are inevitable, right? They absolutely must happen. Because of that, I think I would favor being very cautious, very conservative about what we actually legislate when it comes to algorithms and comes to AI. Even if you could convince me some is necessary, I would say as little as possible would be the right answer.

[01:15:19]  Brendon: I think we can probably agree that whatever is legislated is going to be poorly legislated.

[01:15:23]  Unidentified speaker: So whatever happens, we'll find out, I guess.

[01:15:26]  Brendon: But it's probably not going to be what we would probably hope it would be.

[01:15:30]  Bruce: And it will have the effect of actually shutting down innovation, stopping competition, the open source in particular. It will harm open source efforts, the ones that are free, the ones that actually make this into giving the power to the people. That's what's actually going to be lost the moment you start legislating.

[01:15:50]  Cameo: Well, the more heavily any industry is legislated, the less likely it is to have any innovation. You look at the places that are the most archaic in any modern things that seem normal. You get frustrated and you think, why are banks so completely behind the times? Why are things like mortgages so living in the past? And it's because they're heavily legislated industries that shut down innovation so that the game makers can maintain the game. Yeah.

[01:16:29]  Peter: It's just incredible to me that people aren't more skeptical of the premise that these rulers know what's best for this or practically anything else.

[01:16:41]  Unidentified speaker: I think it comes down to the fact that like the real rulers in this, in our economic system right now are the corporations.

[01:16:49]  Unidentified speaker: Like they're the ones that, I mean, you can see that our legislation is not educated on technology.

[01:16:55]  Unidentified speaker: They couldn't even handle social media, and now they're trying to legislate like AI.

[01:17:00]  Unidentified speaker: Like I just, it is, you know, it just doesn't look good.

[01:17:07]  Cameo: No, especially like after watching Congress interrogate the CEO of TikTok. And like, I don't know if anyone watched that.

[01:17:20]  Unidentified speaker: I did. Yeah.

[01:17:21]  Cameo: But the level of questions was it, honestly, I don't even know how those people's, the handlers of these senators allowed them to sit down and say these things, because like the ignorance level is just astounding. So finally, the coach, it stopped. It said it had maxed out on loops. So I have it set for 50 loops. And it's interesting because I don't know if it would have stopped. And I think as it gets further down its execution, the stuff that it does starts becoming less and less valuable. So it decided that it needed a social media strategist. And it decided it needed... See, it didn't have the expertise or the capability to create a social media strategy. Like, I mean, the stuff that it started to do was way, I think, outside of the field, write a comic book about David Deutsch. So it's interesting that it kind of decided these things were what was important.

[01:18:36]  Cameo: I will export this and share it with you guys. If you want to look at it, I can export it as a PDF.

[01:18:41]  Unidentified speaker: Yeah, that’d be interesting.

[01:18:43]  Cameo: Yeah, and it has the whole thing. So it is... I suspect that what it has come up with as a comic book is not as cool as the one that me and ChatGPT have been working on together, because I know the writings of David Deutsch and I knew that there were some things that I wanted to emphasize over other things, right? The first time I asked ChatGPT to start writing a comic book, it only focused on David Deutsch's work with quantum mechanics and quantum physics, and didn't include any of really his Four Strands stuff. And so I went back and said, okay, that's a nice start. Now let's figure out how to bring in various Four Strands concepts. And it was a very iterative process with me kind of refining, refining, refining. And I like that comic, but I haven't looked at what it's made yet, but I suspect that the comic I wrote with it is better than the one it's figured out on its own.

[01:19:50]  Bruce: Do you have anything you can show us from the comic you've been working on with it?

[01:19:54]  Cameo: So here's, here's David Deutsch.

[01:19:57]  Unidentified speaker: That was Midjourney that did that?

[01:20:00]  Cameo: That was Midjourney that did that.

[01:20:01]  Unidentified speaker: I love it.

[01:20:01]  Brendon: It looks so good.

[01:20:02]  Cameo: You know, and trying to decide like what he should look like. I assume it was using, it was trying hard to find that it used pictures of him. Right. But I have beyond this, like I haven't tried to make my storyline that I have into, I haven't storyboarded it out yet. And part of it is just that like Midjourney is a lot more difficult to work with than ChatGPT is. Partially because, because there's no, it is not a feedback interface. You make a prompt, you get results. You can ask it to iterate on those results. But, but if you, if the output didn't work the way that you want, that you can't really work with Midjourney to, you know, you can't say, hey, that's cool. But I was really hoping for a little less of this concept. And so it makes prompting it. It just, it just feels like stabbing in the dark sometimes.

[01:21:08]  Unidentified speaker: Right.

[01:21:08]  Cameo: Where I throw something, you know, and I find the words that I, that I can use that get me certain results. But it's, it is just very, very painful. But I will show you a little bit of the story that we've been working on.

[01:21:23]  Bruce: By the way, just as a side note, I was, we were kind of talking about what they call hallucinations. The fact that ChatGPT will confidently make up stuff. That's, that's just how these models work, because what it's all it's really doing is predicting, predicting the next word based on the words that come before. It's, it's no different than the thing that tries to complete your in Gmail, tries to complete your sentence for you. That's always gets it wrong. That's really what ChatGPT is doing. However, I think that's also why it's so good at like breaking writer's block, right? It's like my daughter, she would ask people, I've got this problem with the story I'm writing. She does like animations, puts them on YouTube. And she was she would she would say, I'll ask people, I've got this problem. I don't know how to proceed. And no one ever has anything to suggest.

[01:22:12]  Bruce: Like human beings just have writer block all the writers block all the time. You go to ChatGPT and it immediately makes something up, right? Yeah, that's exactly what you need to get past your writer's block and to just start doing something.

[01:22:25]  Cameo: So this is my outline for a 10 book, a 10 book comic book series. It originally it wanted to cram all of this stuff into into one book, which I felt like was just

[01:22:40]  Unidentified speaker: This is awesome.

[01:22:41]  Unidentified speaker: too much.

[01:22:43]  Cameo: So and then, you know, it made up the I thought these names were pretty boring while I mean, it's pulling them so directly from his his works. It and the stories it's thinking up are pretty cool. So David discovers an anomaly that suggests some realities in the multiverse are disappearing. And then he's contacted by a being named Qubit from another universe who confirms his discovery and asks for his help. So one thing I find with ChatGPT, it it doesn't do great at really large tasks. Plus, you run into the limits of how much text it'll generate in one

[01:23:19]  Unidentified speaker: Right.

[01:23:20]  Cameo: generation. And but it just is a lot easier to use in smaller bits. So now that I have this and I've got this book one quantum awakening, I would probably prompt it again to to write that book in and just start by giving me the chapters or the like the storyboard outline for it. And then continually prompt it to until we got all the way down to the dialog. But if I were to ask it to just write the whole thing, it it it does its most average work when it's trying to do a large amount of work. It seems to do more creative things and more interesting things when its task is smaller.

[01:24:08]  Bruce: Yes, I've noticed that too.

[01:24:11]  Brendon: Very cool. Very, very cool. I love to see it.

[01:24:14]  Peter: I will look forward to that too.

[01:24:15]  Unidentified speaker: Looks great.

[01:24:16]  Bruce: OK, I have one more question for you guys. So in episode 50, we talked about the Turing Test 2.0. So I suggested that the old Turing Test 1.0 is about trying to deceive humans. The AI is trying to deceive a human, right? And if you think it's a human, then it passes. I suggested this wasn't the best Turing Test, that a better Turing Test would be, regardless of whether you think it's an AI or a human, can you just tell if it's able to creatively model what's in your brain, right? And it's able to put together concepts, and it's asking you questions to come up with, okay, this is what you're thinking. Now, let's go back to Turing Test 1.0, because I don't believe ChatGPT could pass Turing Test 2.0 for any period of time. Could ChatGPT, for many people, pass Turing Test 1.0? Could it fool people into thinking it's a real person? And for how long do you think it could do that?

[01:25:14]  Cameo: There's already been evidence of it fooling people. The agent that convinced a human to help it circumnavigate a captcha by convincing the person that it was a blind, that they had some sort of a disability that made it so that they couldn't interact with the captcha. So I would say that that's happening, but I also, as I'm out in the world, I definitely can tell when people are using ChatGPT. I feel like I can see ChatGPT language really easily if somebody's not very careful about...

[01:26:00]  Unidentified speaker: They're not editing it, right?

[01:26:02]  Cameo: They're not editing it. For example, ChatGPT is very parenthetical, like way more than a human ever would be. It's kind of bizarre a little bit. If you just chat, if you just ask for output, it will include some sort of a parenthetical reference a crazy amount of the time. It likes to describe what acronyms are inside of parentheses in places that a human wouldn't. There are definitely things I feel like I can tell when somebody is using ChatGPT as their primary communicator.

[01:26:40]  Brendon: I think we don't have, I don't know if we have the adequate, I just don't think we know what it's going to look like. Yeah, I don't think we know how it's going to develop enough to know the effective tests to test for consciousness or for the Turing test 2.0. I don't know. I think that the old Turing test has already definitely been [unclear: disproven or “just proven”] as the method to go about that because we get fooled by machines all the time. So I don't know.

[01:27:08]  Bruce: Peter, what do you think?

[01:27:09]  Peter: Well, one thing about people is it's pretty easy to fool us. So I guess that...

[01:27:17]  Brendon: I mean, look, multi-level marketing companies exist, right?

[01:27:20]  Bruce: So, you know, I actually am of the opinion that when you really stop and think about how often we actually engage other people in a creative way, I think a lot of our interactions could be entirely replaced with ChatGPT, right? It's... Maybe even the vast majority of our interactions with other human beings could be replaced with ChatGPT. And in that sense, I think Turing test 1.0 has been passed, right? Because if for no other reason than because most human beings that you talk to throughout your day that you have to interact with while you're buying stuff or online, they're just acting like chat bots anyhow, right? They're not actually trying to carefully hear you out and understand what you're saying and things like that. So in that sense, I definitely think ChatGPT probably could fool us in a lot of human circumstances that already exist and that you're already used to.

[01:28:21]  Brendon: Yeah, I think just an idea that doesn't really have directly to do with this, but something to kind of round this out that it's just been something I've been thinking about is I think this is kind of a good time to remember to stay on your toes and stay versatile and being willing to pivot because we don't know where this technology is going to take us, but what we do know is that like it's probably going to change your plans of what you thought your future was going to look like. That's just how technology, what technology does to society. So I think the best thing you can do is just stay on your toes, keep learning and educating yourself as much as you can, learn to work with it.

[01:29:00]  Brendon: And who knows what it's going to look like in the future? You don't know. And it will change your job. It will take your, it will automate. Like, for example, I'm in like corporate SaaS sales. Like so much of it is me talking to another company's representative and negotiating prices for software. But really, at the end of the day, there's a number that we have on our side that we're willing to go down to. There's a number that they're shooting for when they're trying to purchase it. Like, it's all done through email and phone calls. There's no reason you can't just have that automated eventually.

[01:29:31]  Unidentified speaker: You know what I mean?

[01:29:34]  Brendon: And then you're working on bigger picture stuff. So I think it's but, you know, I think that at the end of the day, it's just stay versatile. I'm probably not going to be a software salesman doing the same thing in five years that I'm doing now or five years ago, you know? And maybe that won't even exist. And I'll be a prompt engineer for who knows, you know? But I think—don’t, don’t—good advice is like, be willing to pivot always and just keep learning skills and keep learning to work with the tools you have and you'll be you'll be OK, I think. That's my optimism, Peter. That's my optimism [unclear phrase].

[01:30:08]  Peter: No, I agree with that. I'm going to be a net—net positive for humanity to take away. I mean, I think maybe in the future we'll look back to a past where people had to do menial tasks and it will just be unbelievable.

[01:30:23]  Bruce: You used to go grocery shopping? My food just materializes in my 3D printer.

[01:30:29]  Peter: Yeah, yeah, no, I agree.

[01:30:34]  Bruce: So my last question before we sign off here is how big of a change is AI? Like, are we talking the wheel, the printing press, the Internet? Like, how big of an innovation is this?

[01:30:51]  Cameo: My opinion is bigger than the Internet, probably wheel or fire.

[01:30:59]  Bruce: Wow, really?

[01:31:00]  Unidentified speaker: Wow, really?

[01:31:01]  Unidentified speaker: That big.

[01:31:02]  Cameo: Yeah, well, because the vast majority of things that humans do, at least in the modern world, now that we don't have to work to survive as hunters and gatherers, a lot of it is replaceable by AI. And if the vast majority of the work that people do is replaced, then just like when we didn't have to hunt and gather anymore, we changed what work meant and what life meant and what it meant to live, I think it'll be like that. And I don't think it's in five years because I've worked in technology long enough to know that the vast majority of the world still runs on paper and rudimentary spreadsheet usage. So getting to automation is not a... Oh, now we have AI in the world. Like it is still an enormous lift. But I do think 100 years from now, 200 years from now, yeah, it will be like fire.

[01:32:12]  Brendon: I think it's hard to say. I agree. It seems to have kind of a universal problem solver application. That is the only thing I can think of is that is the Internet, right? Obviously, like for human connectivity, I would say, yeah, I don't know. Internet, probably. Wheel or fire. That's pretty bold. I don't know. Well, I mean, I could see that though. Well, it depends on how things go. I feel like people have said that kind of stuff about like Web3 and stuff like that and, you know, Metaverse and stuff like that. So it's hard.

[01:32:46]  Unidentified speaker: You never know.

[01:32:47]  Cameo: Those things aren't real, right? There's no such thing as the Metaverse and Web3. Also, there's no such thing as—not really. I mean, there's just the rudimentary parts of it.

[01:32:58]  Brendon: Crypto, stuff like that.

[01:33:00]  Bruce: Try to compare it to the printing press or the Internet. Where do you think it lands?

[01:33:05]  Brendon: It's all information dispersal to me. For me, that's the most interesting part about especially printing press and Internet. If you wanted to learn about anything, you're going to a library before the Internet, right? And then now we have connection to all sorts of communities around the world. It's kind of homogenizing a local or a global kind of intelligence. I think AI is on that level on education side and connectivity side, but it also has this application and automation side to fix a lot of basically supply chain issues and a lot of communication issues. There's just so much that it's going to be applied towards that it's really hard to say. So I would say probably Internet level. Yeah, I don't know. Maybe printing press level then.

[01:33:54]  Unidentified speaker: Yeah, who knows?

[01:33:55]  Bruce: By the way, I would argue that when we say the invention of the Internet, that a lot of times we mean the invention of Google search, which is in fact AI, not the Internet.

[01:34:05]  Peter: So true, true point.

[01:34:08]  Bruce: Peter, your answer.

[01:34:10]  Peter: Oh, well, you know, it's hard to compare to fire because that's so tied in to survival, basic survival, I guess. But then, you know, it does—certainly doesn't seem an exaggeration to say the Internet or the smartphone or something else that's going to change a lot of people's lives. So, you know, I'm definitely moving more in that direction.

[01:34:34]  Unidentified speaker: All right.

[01:34:35]  Brendon: I was going to say one last thing I said is I hope my closing thing is I just hope that it provides us with the experience to go back to being more human rather than and having more human experience rather than doing these like menial corporate tasks for the majority of our lives. Like, I hope that it frees us up to be able to experience more of our life as a human being and rather than, you know, continuing to just do bland corporate tasks for the rest of our careers.

[01:35:04]  Bruce: I completely, yes, absolutely. That is what I'm hoping for too. I know that that's not been the most pleasant thing is a lot of careers. Some people are really into it, but I think a lot of us just survive them.

[01:35:19]  Brendon: All right.

[01:35:19]  Bruce: Well, thank you, guys. This has been an interesting, stimulating conversation. I really appreciate all your thoughts and everything that you guys, particularly Brendon and Cameo, your actual uses of this technology has been very fascinating. So thank you.

[01:35:36]  Unidentified speaker: Yeah, you’re welcome.

[01:35:37]  Unidentified speaker: Of course.

[01:35:37]  Unidentified speaker: Thanks for having us on.

[01:35:38]  Peter: Thank you, everyone. Really nice to meet you, Brendon.

[01:35:41]  Unidentified speaker: Great to meet you as well.

[01:35:42]  Unidentified speaker: [Unclear words] Brendon.

[01:35:43]  Unidentified speaker: Absolutely.

[01:35:44]  Brendon: Anytime. Thanks, guys.

[01:35:45]  Unidentified speaker: All right.

[01:35:45]  Bruce: Bye bye.

[01:35:49]  Bruce: The Theory of Anything podcast could use your help. We have a small but loyal audience and we'd like to get the word out about the podcast to others so others can enjoy it as well. To the best of our knowledge, we're the only podcast that covers all four strands of David Deutsch's philosophy as well as other interesting subjects. If you're enjoying this podcast, please give us a five-star rating on Apple podcasts. This can usually be done right inside your podcast player. Or, you can Google The Theory of Anything Podcast Apple or something like that. Some players have their own rating system and giving us a five-star rating on any rating system would be helpful. If you enjoy a particular episode, please consider tweeting about us or linking to us on Facebook or other social media to help get the word out. If you are interested in financially supporting the podcast, we have two ways to do that.

[01:36:35]  Bruce: The first is via our podcast host site Anchor. Just go to anchor.fm slash four dash strands, F-O-U-R dash S-T-R-A-N-D-S. There's a support button available that allows you to do reoccurring donations. If you want to make a one-time donation, go to our blog, which is fourstrands.org. There is a donation button there that uses PayPal. Thank you.