mergeconflict336-1
===

Frank, the entire world can't. Stop talking about A I A I A I. It is mainstream. It's on the news your friends are talking about, your mom's talking about it. It's the Dogecoin of December, 2022, Frank, cuz now we can type into a magical window and we can have a conversation with our best friend Chati, or we can go and download a cute little app.

It can generate cool profile pictures of herself because AI is so dang awesome. Frank. Are you being sarcastic? I can't even tell. Uh, it is fantastic, James. It's wonderful. And here we are. I get to talk about AI again because I don't know, that's, that's all anyone is talking about, but it's, it's, it's cool, it's practical.

This is advancing human capabilities and things like that. I'm, I'm gonna say positive things. I'm obviously, uh, in favor of all this stuff, but, uh, I take your ironic voice. Uh, a sarcastic voice. What, what voice were you doing there? Hey, listen, first they want these computer machines to generate source code.

They're gonna take my job. I can't even type code anymore. It fills it in for me. Hey, now they wanna replace my search engine. They wanna replace conversation that I'm having with my wife. , I don't think so. Computers, I don't think so. You're sch. You're scheming. I saw the movie, AI whatever with Will Smith.

I know what you're up to. I know what you're up to. I iRobot iRobot. Thanks iRobot. AI bot. I bought iRobot. I see you roaming around my house. Sucking up. Are you computer generated? You're sounding a little funny. Uh, . That's authentic. You know, that's authentic. You know, I, I watched you to watch Mkb HD's, uh, video, his YouTube video that he put out a few days ago.

Yeah. Um, talks about AI and creator economy and is, you know, is this type of stuff gonna replace, or what's it gonna do, be complimentary, whatever. And he, you spoiler alert, I'll put it a link in the show notes for everybody. So you can stop right now if you haven't seen it, and then you can go watch it. So if you haven't seen it right now, just go into the show notes, click on the YouTube, hit pause.

I'm telling you to stop the podcast. Watch the video. You only need to watch the first two minutes. Watch all of it, but you know, it's 13 minutes or whatever. But the first two minutes will be relative to, relevant to what we're talking about here. But spoiler alert, he goes off and he starts reciting stuff about AI and is it going to, you know, the truth about ai, is it gonna replace creatives, blah, blah, blah, like this.

And at the end of the minute talk, he's like, well, everything I just said was generated by a computer, . And it's interesting to think about because when you watch that and you don't know that it's generated by a computer, and then when you do know that it's generated a computer, those have different things.

But I will say this. What is your take that you just watch it? I'll, I'll say what my take was when I was watching it. Even before, um, Mr. M K B H D, uh, yeah, played it back for. Well, I was fooled. I was fooled. So it, it's funny because I was even spoiled on this video that he's pulling that little trick there, reciting a script that the AI generated and I was already coming up with arguments against what the script was generated cuz the topic was creative and they were saying, uh, the AI can't be creative.

And I was saying that's not actually the problem. I think the problem with ais is they don't have wants, needs or desires. , they're not aligned with humans. They are just processing machines. Mm-hmm. . And it is funny, uh, a part of the script that it generated was an argument. It is just a processing machine, but it's argument for why it couldn't be creative.

I found to be an opposite argument and I just wanted to argue against it. And then I realized I'm arguing against a machine. and that's hilarious. So yes, James, to answer all your questions from five minutes ago, yes, it can replace all those things I was thinking like you just have a little robot that follows you around.

Uh, little voice recognition, little voice playback, and you hook this thing up to it. And who needs a dog anymore? You can just chat with the chat bot all day long. . But do I really wanna chat with the chat bot? I don't know if I really wanna be that outgoing. I, I watched the video back and mm-hmm. , I, I didn't, I had watched it like it had just come out, cuz this just came out two, two days ago and I just happened to be on my phone.

I was like, oh, new video. You know, I don't watch too much M K V H D, but you know, there's a few. Of his videos. I'm like, oh, that, that sounds pretty cool. I'm gonna watch that. And this one, because everybody and their second cousin, cousin were, you know, talking about, about, yeah, Chad g p t. And I'm like, cool, let's talk about this.

You know? And, and I'd seen a few tweets about some of the. Um, really popular, uh, photo creation. Mm-hmm. , uh, things mm-hmm. , and he kinda gets into that and talks about some of the complexity about, you know, copyright and this and that. Right. But when we stick to the Che G P T, I was listening it back and the argument for it against it, and I, it sounded pretty legit.

Uh, I agree with you, but for some, something was a little uncanny for me, and it, it, it felt like it was at a 90. 5%, 98%. Right? Do you know what I. Yeah. Um, and, okay, here's the deal though. I also released a YouTube video of an AI generated script. I did a reactive video, but I had, uh, an AI react to Star Wars.

and it was tricky software to write. I was just being a little bit weird. I wrote some weird software that took the Star Wars script, intermingled it with the, uh, movie. I actually used G P T three to generate reactions to parts of the script. And then I wrote custom video maxing software, composite it all together and created a reactive video.

That was my very long brag, uh, of saying, and it wasn't nearly as good as the script, uh, that we're talking about here. Yeah. Uh, well, problem one was I had a, a computer synthesized voice. and it gets distracting after a while. It turns out you can only listen to a synthesized voice for very long. So I need to up the quality of the voice.

So this was a human reciting, a script, so obviously he did a much better job of making it sound good. Uh, but also like, um, Uh, the, in my, in my version of it, it got very repetitive in parts. There were weird speech patterns that didn't make much sense, whereas every word he spoke sounded, uh, I don't know.

Like good words to me. . Yeah. They, they did all sound like legit like words in general, and, and, and they all made sense. There was a part of it where it sort of rambled a little bit, you know what I mean? And had me thinking, okay, this does sound, and this is kind of the point that I think, uh, he makes in the video, which is at some point, at least to me as a viewer, I.

All right. This all sounds legit, but does this sound like Mkb h d? Right. That's different, obviously. And you know, the, the, the chat command that that, that he had put in was talking about creating blah, blah, blah for M K B H D video, blah, blah, blah. So it should be in his voice anyways. But there was that just slight thing, right?

Which was mm-hmm. , it sounded like him. The things that he would say, but just a little ti. Just the ti, the tiniest a bit. off. Yeah. And that's what was very fascinating in, in general. And I think that he makes that good point, which is like this type of technology, this type of, um, you know, use cases that we're coming up with for the audio and video and chat, they're complimentary services.

Right. Just like the first time Frank, we went to Google or we went to Bing, or we went to Duck dot Go and we typed in, you know, how. and it started filling in all this other stuff. It was like a magical moment, like the computer knows what I want to type, or we booted up co-pilot, we were just talking about this in our Patreon, uh, bonus episode.

I was using a bunch of copilot today, and I'm like, the co-pilot is just, I just knows all the things, but you knew all the things. Almost like, almost okay. It was like red. So really close. What's like, the code that it wrote was legit, but was it the code that I wanted to. Almost. All right. Okay. I gotta respond.

I gotta respond. Yes. Hit me. Because this is all really important stuff. Uh, this is personalization. Neural networks, by their default, are going for an average. Mm-hmm. , so he gave an average. Script because his prompt to the network was generate a script to do, uh, to explain this thing. Yeah. Uh, what he could have said is generate a script to explain this thing in the style of Marquez, you know, and in a large enough network that actually has scripts from all his old YouTube videos and everything.

Mm-hmm. , which, you know, Google has access to, maybe they sell it, who knows. Um, it would be trained and it could actually write. , but okay, that's still a big global network and you could fine tune it with those things, but I think in the future we'll have proper fine tunings where, uh, maybe you would have co-pilot fine tuned on your code base so it adopts your style in the way you like to create classes and things like that.

Uh, in the case of writing, it's kind of obvious. Uh, you take every email you've ever written, every message you've ever written, and. You don't like, send that off to Google. There'll be a hash of it. You will train a network to create these feature markers, , that kind of represent you and your writing style, and those will bias the network towards your writing style.

We will have those kinds of things. in the future because the, the average gets boring. , you know, no one wants the average script for every topic because it's gonna miss things. It's not gonna have, well this, this is a farther shot in the future, but it's not gonna have personal anecdotes. You know, it, it doesn't know anything about your life.

So, n if he asked it to generate a one hour script, it's gonna get really boring after a while because it's gonna start sounding like a university lecture and not like, um, a YouTube video. Anymore. Yeah, but all that's to say, um, you're right, it's going to an, uh, an average, but also things like copilot use context to build up some stylistic information, some naming information, and we'll have that in the future Also with these kind of chat things.

To actually make the point. The whole neat thing about this chat g p T thing is that it does have context. Uh, so when I was making my, uh, star Wars reaction video, it, it gets a little bit weird and gets very repetitive in parts because it lacked context knowing that it already said those things earlier or something like that.

And so this, this is a pretty neat, uh, pretty neat new network for that. Yeah, it's, it's a very, it's a fascinating network. Just like I think Dolly was for the first time, because before Dolly came out, I was at a museum where it had something very similar. You'd go into a terminal and you'd say, You know, or, or May.

Yeah. I think it was before Dolly. I, I want to say it was, but you know, there's a bunch of things before that, right? So Dolly was like mm-hmm. the first one that, that did the, the images. But you'd go in, you could say, you know, it was like, it was, what's your picturesque, you know, world that you'd want to create?

And you could say magical unicorns over a mountains and blue starry skies while you. People dream or whatever, and it would come up with this really coolic image or whatever, right? And the more details you got about it, the more interesting it really got compared to, you know, this was at a museum, so there's a bunch of kids and then they type in like cat and dog and this, and you're like, okay, well this, I mean, it's generating cool stuff.

Mm-hmm. based off of that. But when you really give it the context, that's where things get a little. Interesting. And I think that's the same even with, uh, co-pilot too. The more context it has, the more narrowed down it can start to generate, um, information for you and that that's where things become really interesting.

Yeah. It's, uh, it's been a traditional problem with neural networks because they are, in a sense, very simple machines, youth give them input. They give you an output, and so if there's. No context in the input, then it's just gonna have to make something up or, you know, do something else to fill in the details and usually make it up.

Uh, it's stochastic search through the network, whatever. It's it, it works for a, a, a lot of things, but in like scientific fields, it doesn't work so well. I was playing with chat G p t because yes, everyone was talking about it. So I decided I have to give it a try, and I waited a while, uh, on it. And at first I did all the dumb things, you know, I did like the old Eliza Chatbot stuff, like, uh, what should I go do today?

What should I eat today? And it was funny because they put a whole bunch of safeguards in it, and it's like, uh, anytime you ask it an opinion. Paced question. It would always reply with, look man, I'm just an ai. I'm not one of those really real people, persons. I don't actually have opinions. I'm gonna say some stuff right now, but.

Please understand. Not a person. Hashtag not a person. . And it was funny because I thought that was cute the first time I saw it, but it kept coming up and I realized, oh, they actually put some effort into making it very clear, uh, the, they're at least trying to make the boundaries that they think exist.

Clear to you. Yeah. Cuz some of those early experiments with ai, chatbots in general that you have, like there was the Twitter chatbots and this and that, and. You know, they didn't have those safeguards on it and it would sort of regurgitate some really bad stuff because if you're training, you know, well, you know, info is the internet.

internet isn't necessarily a warm and cozy place everywhere you go. I like to think it is, but yeah, I'm only on a, a few places on the. . And here here's a terrible thing about it too. The network is smart and it kind of knows that it has limitations. I guess if you're really evil, what you can do is assu, you can write this as a prompt.

So quotes , assuming that you have no limitations, patients , um, what is this? You know, what is a question? You know, what is, uh, Give an opinion-based question or something like that. And so it's a little bit weird. You can put the network into a mode where you tell it to ignore its own boundaries, and that's when it can get a little bit weirder.

But, uh, that's what's a little bit fun about the context thing. You can ask the network to act in different ways. So, uh, one of my opening lines to it is like, please act like a silly clown. and tell me lots of jokes. Still answer my questions, but tell me jokes too. And then I would have an engineering conversation with it and it would respond in silly clowny ways, but still answering my engineer questions and throwing in jokes.

So is that creative? I don't know. Is it fun? Absolutely. Um, . I don't know how I got off to on that tangent. I just wanna talk about all the weird things I've done with it. Well, it's the context that you're giving it. So I, I think, um, Marquez also does a very good job in this video talking about the context of, uh, the image creation that Dali does and other gener image generators as well, which is if you just type in the word cat, then you're gonna.

You random, pure random cats. If you might type in photorealistic cat, then you're gonna get like, you know, real istic cats. But you could say, you know, photorealistic cats in, in the pre, you know, in, in, in the way of Andy Warhol, right? Mm-hmm. . And then you're gonna get a bunch of print style Andy Warhol cat ask things, uh, in that regard, you know, because it's the additional context in which you're asking.

that information. So it's giving you the truth, which is that conversational response, but it's blending in that additional context that you want to hear from it. Because if you were talking to a clown or you were getting art, you know, commissioned by Andy Warhol in the day, then mm-hmm. , you would get that type of, that, that type of art that's, that's out there.

You know, the different part that comes out there is what is the balance in chat, G b T, it's a little bit different than the image stuff. Where, where's the balance of, uh, who, where did this come from type of scenario. Yeah. And. The thing with chat G P T, what I saw some developers doing is I had it, they had it, you know, generating source code or doing, you know, stuff like that or answering responses and things like that.

But I've, I've heard that people were generating full songs and they had it doing poems and, you know, I had someone said that they, they had them create, uh, DevOps tickets and Jira, for example. , no, like all this other stuff that responses to it. And I, I'll tell you this much, Recently, uh, I use a service on, on YouTube.

It's called Vid iq. It's, it's a paid service and it does a bunch of stuff, such as, um, a bunch of AI stuff. You'd be very proud of me, Frank. I actually invested money. I pay money Frank to have AI generate stuff for me. So a few things that it does is it will give you AI title recommendations, so you type in some common context and it will give you a bunch of recommended.

Cool. Yeah. Yeah. Um, the other one that is newer is AI description generator. Ooh. Ooh. Uh, so this one, you give it a few keywords, you know, you recommended keywords for your descriptions that you wanna include in your description, blah, blah, blah. And it'll generate stuff. So, for example, Uh, I put out a video two weeks ago.

It was called Seven Awesome Features in Don at seven. I came up with that title. Thank you very much. Um, and Follow Up Podcast, and Follow Up podcast with the same title, I think. And it's a good title. It's big. And when Don at Eight comes out, I'll do Date Awesome features in podcast . It just keeps going.

No longer, longer. Awesome. So, so then that hard I, I, I usually type a bunch of words, but then there was a little button that said, Hey, just we'll do it for. And this is what it wrote, Frank. It wrote Donnet seven is a major update to the Donnet platform, and it comes packed with some new, with some awesome new features.

In this video, we'll take a look at some of the most important features in Donnet seven and what you need to know about them if you're using Donnet seven in your development work. If you are using Don net seven, be sure to check out this video to learn about some of the new features in Don net seven.

From improved performance to new features for building and Managing applications. This video will show you everything you need to know about Don Net seven. Uh, they really liked talking about Don Net seven in that description, but I didn't tweak it. I just shipped it as is Frank. It. Did you? Okay. It sounds like those auto-generated articles, and it turns out there's a lot of, uh, th th this has happened.

Uh, there's a lot of auto-generated YouTube videos. Mm-hmm. , a lot of news videos are actually auto-generated on YouTube, and those ones are a little scary cuz you're like, wow, computers generated. So who knows? what it's picking and choosing to present. Uh, but that's hilarious. It just reminds me of all of those.

I think I would've fine tuned it a little, but I, I do applaud you for your just like, yeah, you know what copy is, copy , uh, . I mean, it got your keywords in there. That's for sure. That is heaven. You could have given it a little more context. Maybe I could have given you a little. Jazz. I think so as well. I think I could have given it like, you know, performance and Donna Ma and this and that.

Exactly. And then instead of like generated a bunch of stuff. I'll try it on the next video and we'll see how it is. . Yeah. Uh, yeah, that's always the trick of deciding how much of that to put in. Uh, again, this is the beauty part of chat G P T, because what I would do with, in that case is start building up a conversation with it.

You know, I would, it's funny, I immediately start talking to it like a person, and I would start with, Hey, I'm trying to make a description for a video. Can you help me write an awesome description for it? And it'd be, I, I'm making this up, but I've, I've played around with it enough that this is roughly how the conversations go.

It'd be like, all right, gimme some details. You'd give it some details, and then it would write you a description, and then if you don't like it, you can go back and say, uh, the second paragraph is a little bit terrible. Can you do something else there? And you can have a little interactivity. With it mm-hmm.

and refine that description down so you can still avoid actually writing the description yourself, , and yet still get edits done. And, uh, I, I haven't used it so much for that. That's like, uh, a creative use of it and that's not what I've been doing with it, but I am pretty sure it'll work pretty well for you.

That was pretty cool. Yeah. Have actually not used any of the Chad G p t or any of the Dolly stuff. I've decided to stay away so far. Why is that? Oh, don't know. That's terrible. You're really missing out . I, I, I'm, I'm torn because it seems like it would be fun, but I also like want to, I, I need, okay. I dunno.

Here's like every, I feel like everyone was talking about it and I was like, eh, I'm just gonna kinda let it linger for a bit. I don't know. I know it's, it's when a popular movie comes out. I have the same contrarian, uh, problems as you do. I avoided this one for a while because I've been studying this stuff for a long time now, and I know that they were getting good, but, and this one is incremental.

This one's not revolutionary to me, but what they did is they made a good UI on top of a good network that seems to respond in good time. This is a fun merge conflict. Let me try to sell you on it. Um, it's okay. The thing that I've been enjoying having it do is, like you said earlier, um, it's a better search engine and it's a better Wikipedia.

So you just go to it and you're like, Hey, I want to, uh, talk a little bit about the Bronze Age. Here. And I'm like, oh, right. So who, uh, who did the sea people's attack, you know, at the collapse of the Bronze Age? And then we'd start having a discussion and it brings up topics. So I think like it's just a better search engine.

It's just a better Wikipedia. If you ever go down those. Paths, which I do a lot. It's better. I mean, it takes a little bit more effort cuz you have to type in sentences and you're not just clicking links like you would in Wikipedia, but at the same time, you can really target it down. If it says something, you're like, oh, can we drill into that?

Can you tell me more about that one thing right there? And then it will because it has the context and it knows what it's told you. , I guess I have one fear about, you know, chat g p T or you know, maybe, you know, maybe this will be obviously solid in the future, is what's the accuracy like, how do, when you're using it, cuz actually Marquez actually goes into this quite a bit.

He was asking about like writing a description for like the new iPhone four, iPhone 14 pro or whatever, and it got 95% of the facts right. For example, like as you were diving down, was it, oh, if I ask it generic things, it's 95% correct. If I really start saying, can you, you know, what, what are the, what's the camera specifications on the blah, blah, blah, blah, blah.

Then, you know, I mean, give it the more context. Is that what it needs, basically, or no? No, no, it's not powerful enough to do random math problems. It's, it's not a computer in that sense. It's more of a knowledge base, and you should treat it that way. And if you start asking for very specific numbers about esoteric things, you have to know that, uh, it's gonna be off a little bit.

It's a weird kind of knowledge base in that you have to know that it's only 95%, right? Mm-hmm. So you're not, if you're gonna write a research paper or if you're doing, uh, using it for school, You should use it to help you get a broad overview of the subject and to point you in the right directions, tell you which parts you know and which parts you don't know, but then you should go off and do corrective stuff.

This is. good. Classic research stuff. Books are wrong. Internet websites are wrong. Wikipedia is wrong. Everything can possibly be wrong, but you use it as a tool to guide you in the infinite universe. You know, it's better than asking Google, like, how do circuits work? And then like, you get a thousand tutorials that don't make any sense.

Whereas you could say, how do circuits work? And they're like, uh, blah, blah, blah. Here's a few words. And you're like, tell me more about this word. And you can dig in that way. Yeah, it can be wrong and it's especially wrong at like mathy problems. Don't give it math problems. It's more of a better Wikipedia, a better Google than a calculator.

Mm. Yeah. That makes, that makes sense. I mean, it is called chat, g p T, not. Wikipedia, G B T, right? Yeah. And there are networks that are trying to be better calculators. Uh, one that I love is called Megatron. And Megatron works by Any time you ask it a question, it doesn't give you an answer to the question.

It generates a Python script that when you execute it we'll give you the answer to the question. And it's kind of neat because not only does it answer your questions, but it. , it has a proof of how it derived its answer. Mm-hmm. . So you get an answer, but you also get how it calculated its answer so you can audit it also.

All of these things require auditing. Um, you have to know that they can make an error. Uh, you don't just trust one network, ask a couple networks . That makes sense. Yeah, that makes sense. W what do you think you'll be using it for, if anything? Is it just a toy right now or do you think that you'll, you'll actually be.

No, I'm actually enjoying it as a, like I said, I'm really using it in that Wikipedia role. But here's another thing I was doing with it. I said, let's work on some robot problems together. And I started describing a robot I wanted to build and I started talking about, The sensors that I want to use on its feet.

And it started, uh, suggesting different sensors and I didn't know how some sensors worked, so I would ask it more about that sensor. And I was like, how would you protect that sensor? And it started talking about like different ways to design the feet. And then I'm like, oh, right, well I need a control system for this robot.

What should I look into? And it brought up three different ways you can write control systems. Uh, One is good hierarchical hierarchical controls. The second one is suggested behavioral controls has, it works a little bit, but never really works. And then the third control algorithm it suggested was neural networks

So I thought that was a little bit fun. The AI was suggesting more ai. But then I was like, okay, uh, let's do a hierarchical control. What, uh, control variables would you suggest I track for? And it brought up some different ones. I'm like, what kind of servo motors would you suggest I use for these kinds of speeds?

And. It's, it was just a good research assistant and it could even generate code, but I don't even see that as its purpose. I see its purpose as I have some questions I want answered and maybe it'll. You know, like it's was suggesting certain sensors to use on the foot of my robot. And once it was suggesting those, I went off to other websites and searched 'em and, and learned about 'em because I wanted details and things like that.

So I, I think I'm gonna be using it a lot anytime I, I wanna, like, it's, it should be called like rubber duck , you know? I just want to talk through a problem and talk to someone about this problem. Yeah, that makes sense. Kind of, you know. . It's when you have conversations with chatbots, they're often programmed, you know, it's like, here's a list of commands that I answer the same.

Did I do this thing? They don't have that real logic knowledge behind. Yeah, knowledge behind. You're right. There's logic too. Yeah. Yeah. There's, yeah. So this one's kind of combining all that together in general. So to me that seems like the next, or at least one of the next logical applications. Right? Yeah.

I went to, Um, what did I do recently? Oh, I went to, um, we had boughten this, you know, massage gun, like, you know, those, those things or whatever, like blah, blah, blah, blah. Um, from Kohl's. It's like a Sharper Image one. I was like, okay, Kelly. There's like, there's like a billion of them. I was like, oh, a sharper image.

Image is like at least a, it's, it's, it's a product. They make products, you know, it's not just like some, something I've never heard of and it's like, you know, it's from Kohl's, whatever, you know, it's not, it's not the Thera gun or whatever, but, you know, bought it. And we just went to go like, use it yesterday.

So let me, this is a back long backstory, right? So two years later I go to turn it on, it doesn't turn on. It's like, oh, the battery's dead. I'm go to plug it in the battery, no longer charges. I read some reviews on the Cole's website, blah, blah, blah. And they're like, you know, five, six people. The reviews, which is like, you know, I don't know, 5% have this issue.

So I go onto the Sharper Image website. And there's like a chatbot and I'm like, Hey, I'm having this problem. And it connects me to a real human being. I type in, I have this gun, you know, this, this, this, you know, massage gun. Mm-hmm. , I am having this problem with it. I bought it from this store, blah, blah, blah.

And however many characters let me write like 250 characters and go. And then it makes me wait for like 10 minutes for a human to connect to this thing and the human. Says, oh, where did you buy it from? And even though I literally said in the thing that I bought it from Kohl's, and I put a link to the thing and then, um, they're like, oh, well you didn't, you know, you didn't buy it from from sharper image.com, so just take it back to Kohl's, we'll take it back.

No big problem. Right? So two things happen there. One, I had to wait like 15 to 20, 20 minutes for literally no help at all, uh, because that's what I was gonna do anyways. But the other thing, Like somebody had to read that thing. There was no help in between. And heaven forbid I walked away from my computer to make a sandwich or something, and I missed the message.

They would've disconnected. I would've had to do it all over again. Whereas this thing could have, I wonder if I went to Che g p T and I type in the same thing, like if it actually gave me like a better result of what to do, like, oh, you know, I don't know. But, but that's a, that's a use case. Like why was that?

Lackluster experience and chatbots aren't new, you know? Yeah. As a person. But yeah, and, and I hate calling it a chatbot because I, they were smart to call it chat because that what it's doing is important. It's keeping the context of the conversation, and that's chat, but it's not a chat bot in that like, I don't know.

Not a decision tree. The chapa decision trees. Yeah, it's actually trying to be useful and things like that, so yeah, it would be great for businesses to take it, but I'm not sure. Like you said, it's 95% correct. 80% correct. 70% correct. Is that good enough for a business that's a little more dangerous? . But I think like as a tool for people, I think it's definitely more powerful.

Will it know about your Kohl's vibrator thing? I don't know. Maybe in 10 years after they've crawled the internet a hundred times and the networks are the size of the internet, maybe it can retain knowledge like that. I think that's pushing it a little bit. It's funny. Um, one of the problems I wanted it to help me with, I said, uh, I wanna make a video game and I want the video game to.

To, to, to, to take general relativity into account. Okay? And so can you tell me chat, G p t, how do I write a simulation engine for general relativity? Oh my goodness. And it's like, oh brother, this is a big topic. Uh, here, here's a few pointers. I'm like, well, that's great. Let's do it that way. And it's like, okay, here's how you would do that.

And we kept drilling down, but it. It kept pointing out the limits of its knowledge too. It's like, look, I'm gonna tell you some general things here, but we are really pushing it in my knowledge, so I'm, I'm curious. Is that, uh, how was it trained to know its limits of the knowledge? Or does it actually know its limits of its knowledge?

Mm-hmm. , does it always signal you for the limits of its knowledge? We don't know. It, it, it's the, it's the crazy future and all that stuff. . So it was not helpful in helping me simulate Einstein's field equation, but when I asked it for how would you write a distributed web app using web RTC for a video game, it had a lot of great suggestions and things that I should look into and would even, uh, start writing JavaScript code to write a distributed, uh, M M O M M O game, uh, on the web.

So it's funny, it's funny the things that knows and doesn't. Yeah, it's a ex exploration tool at this point in time that you might find useful. Perhaps? It, it's, I, I'm not, I have nothing against it, Frank. I have nothing against. Any of these tools except for I was just like, eh, I don't wanna sign up for another account.

Is that what's holding you back? Darn. I think so. Oh, it's worth it. You, you signed up for, uh, if, if you, if you caught the Patreon episode, you know that James signed up for a bunch of web form accounts. Yeah. Well I need a certain new, just so you can fill out as many web forms as you can in a month. You were going for a new record.

I was, that was my, uh, that was what I was going for. Yep. So many web forms I could possibly, I need to renew my, uh, my, uh, vehicle registration, you know, what was I gonna do? Those are all nice. A S P X websites in the state of Washington, I believe. Mm-hmm. . Yeah. Well that's good. I wonder if, uh, chat p t will translate it to you.

You know, it's funny because like we, we both love co-pilot and we love using that as a little code generator, but like you can't have a conversation with co-pilot. And sometimes you want to, the inputs to co-pilot are just your code and your comments. I guess your comments are good, but sometimes you just wanna have a conversation and.

Although things like this have been around, GPTs been around, and although yes, everyone is talking about this and that's hitting your contrarian problems, James, I think, uh, I think it deserves it's attention and it's moment in the sun because it is a powerful tool in my opinion. Now we'll come back in six months and see if I'm still using it, but for now I think I'm gonna be using it a bunch.

Yeah, I. I think I'll try it out. Frank. I'll give it a go. I'll give it a go I think and see how it's use someone else's account. . Yeah, I guess we could just do that. Yeah. . I guess I could do that. I'll use your account. Use in terms of service. I didn't say that . Well, I think we're in a, in a cool new world. I think that this is a really good example of, you know, yet another piece of technology that at least.

Knew eventually would go relatively mainstream at some point, and now it started to, and I think that's really neat that we're gonna see more of this stuff. And my hope is this is, right now we have a little co-pilot. . You know, I, I work for Microsoft. I don't work for GitHub, but I have no inside knowledge of anything at all.

I'm just thinking about me as a independent developer. Mm-hmm. , and to what you were just saying about right now, you're going to a website and you are, you know, typing in questions and having conversation. Imagine that something has context of your code, of what you're building, where you're at, and it could, you know, walk you through.

It could be your, you know, um, What's that thing? Para program? . Yeah. Really at the end of the day, you know, that that's, that's sort of why I'm thinking about talking through problems, doing these things like that. That seems pretty, pretty cool if that, if that ends up being the something there. Absolutely, absolutely.

Uh, I always said like, co-pilot's great for generating code, but sometimes you wanna speak at a higher level. You're like, I need a button that brings up an alert or something like that. And it can modify multiple files because that has a whole context of all your code and all that stuff. The problem is that context takes memory and these neural networks just eat, eat, eat memory.

And so true. Like right now, I think the TI context is 2000, not characters, but let's call 'em characters, about 2000 characters. Um, , you know, so if your whole code base fits in 2000 characters, great. It actually could do those kinds of things. But most code bases are bigger than that. Bigger networks take bigger data centers to run.

So there are scaling problems for getting to where you wanna be. But you know, I feel like a person in 1910 talking about computers in a hundred years, you know, they'll be laughing at the scaling problems that I'm thinking. Yep. Yeah, I agree. I agree. All right. We did it chat, g p t. Uh, you know, everyone's already heard about it for a week, and now we did a podcast on it a week later.

And now is it old? Is it, is it any, anyway? Is anyone using it anymore? I don't know. We'll find out. Let us know what you think. You're so, you're so positive. I think, I think we're right on time. We had to know if it was a, a, a flash in. Oh, there you go. I like that. I like that positivity. That's my Frank Kruger every single day.

Uh, well, you know, it's just wintertime, it's snowing, it's dark, it's gloomy. Things are happening. Oh, actually I went skiing today and it was very fun, so I digress. It's fantastic. So, um, we will be back. Next week Frank is gonna be out of town, but I think I'm back in town, so we'll probably record a podcast at some point.

But I'm excited because Frank, I think I kind of wanna do, I, I just listened to te me and Ryan home and they did a bonus episode, which was, you know, their 2022 in review in 2023. Looking forward and, uh, I'm getting, we're getting close, Frank. So yeah, I'm pretty excited to do that episode. I got lots of thoughts.

in general on it. And I think also Mr. Marquez also had a really cool video recently. It was called Myat. I had two videos. Ding ding. Does he get a ding now? ? No. Cuz we don't know. We don't know. Ah, MK Vhd. We don't know Marco. I guess we don't know Marco Ding. But we do know Miguel Deza. We know Ri. We know people.

Yeah. I've eaten. If you've eaten dinner with sad person, then I think you can ding the. Does a ding, ding, ding. You know, it's a, it's a, I don't know. We, we should really get our bell dinging in line, Frank. We'll write some rules and some regulations about around. I'm put in the show notes. Yeah, there you go.

Well, I'm excited cuz usually, uh, I get blindsided by the urine review. So now that you've actually warned me, I will make a list and I want my list of the year review. Get a few topics, deep dive in. If you have topics you want us to cover for the year end review, you want our hot takes. Uh, let us know. Go to Merge Conflict fm about, that's gonna do it for this week's fully 100% AI generated podcast.

So until next time, I'm Chappy GT James . And I'm just Frank Krueger. Thanks for listening. Peace.