Episode Transcript
[00:00:00] Foreign.
[00:00:05] Welcome to ChatGPT Curious, a podcast for people who are, well, curious about ChatGPT. I'm, um, your host, Dr. Shantae Cofield, also known as the Maestro, and I created this show to explore what ChatGPT actually is. Really, though, are the files in the computer, how to use it, and what it might mean for how we think, work, create, and move through life. Whether you're skeptical, intrigued, or already experimenting, you're in the right place. All that I ask is that you stay curious. All right, let's get into it.
[00:00:38] Hello, hello, hello, my curious people, and welcome to episode 30 of ChatGPT Curious. I am your grateful host, the Maestro, and today we're talking about how to switch LLMs without starting from scratch and pulling your hair out. So I alluded to this in last week's episode. Let me move this mic here so it's better. I alluded to this in last week's episode, and I'm excited to get into it today. So real quick, though, uh, before we start, we hit 24 ratings and we got a new review from Lauren.
[00:01:10] Cheers to us. Y' all are the best. Podcasting is so unidirectional. I just be talking at this screen. I don't know who I'm talking to. So thank you for making it far less unidirectional. And, uh, cheers to getting 25. Who's going to be the one? Who's going to be the one? Uh, but main topic for today, switching LLMs, large language models. ChatGPT curious. Wow. ChatGPT.
[00:01:35] Claude. Uh, wow, I loosened my words there. Um, Gemini, whatever you may be using, right? How do you switch between these things?
[00:01:43] And, uh, I want to talk about this. Not necessarily because I'm planning on making a switch, but your Maestro does like to be prepared. Also, real talk shit is shitty in the United States, and economic boycotts are absolutely one of the ways that we can strike back. There have already been calls for folks to unsubscribe from ChatGPT. And realistically, right, Realistically, the action item that would have the greatest impact would be to either stop using LLMs altogether or to run local open source LLMs, right? I personally, at this point, I am not willing to give up the informational and educational benefits of LLMs completely, right? Like, if it was like, society is collapsing, I'm like, yeah, get rid of this.
[00:02:30] Go ahead. But in. In general order, from boycotts to work, right, we have to have infrastructure in place so that you don't have to go back to the systems right now. What's happening is people are making call Scott. Scott Galloway, who I actually referenced on one of the episodes when I talked about Will chat GPT get open. Um, Will.
[00:02:48] Will open AI get, uh, Old Navy or Will chat GBT get old Navied. But he's the one that's one of the people that's leading this charge. And I don't know if they realize, like, okay, we divest from it, we unsubscribe, but then people just come back and it all just comes back. And in all likelihood, we unsubscribe stock price drops, which is what he wants, in order to get Trump to do different things and behave like a fucking adult. And what happens then? Rich, rich, rich people buy the dip. And then everybody goes back and the stock price rises and the rich get richer.
[00:03:27] So I understand and I'm. I'm willing to take one for the team. I absolutely am. But this is not like going to work in the long run. There's. You have to think about what the alternatives are, which for some people is not using them at all. But I personally, I clearly, I have this whole podcast, like I. The educational and informational benefits to be had by this, this technology. I am not willing to give them up. Not yet.
[00:03:52] Does that mean I'll never give them up? No, absolutely not. But this is where my, you know, my head is at with things.
[00:03:58] Um, and if I liken this to shopping at Target or Amazon, like, yeah, I stopped. I never shopped at Target though, ever. So that was easy. Which, um, is how some people feel about LLMs. They don't use them at all. So it's like, yeah, get rid of it. Yeah, totally. Um, Amazon. I stopped, I don't know, last year and I don't miss it at all. And, and you know, we, we don't need majority of the things that we get from. From there, but we can get those things, things we do need. I'm thinking about like poop bags, right. For moose. Like, that's the. One of. That's probably been. Probably been the worst thing that we don't have access to is those specific ones, like the, the brand that we get that we got like, only sold through Amazon. So many of these little brands only sell through Amazon. It's so up. Like Amazon is the fogging worst.
[00:04:39] So, you know, these things, the few things that we really do need, that we get off of there, we can get from somewhere else. Absolutely. It may take a little longer, sometimes it's a little more expensive, but, like, we can do it it's totally fine. And that's what, you know, very much in the same, same line of thought. You can just switch to a different LLM, right? And you can just use Anthropic or you could use Gemini, which is what I wanna talk about today. But I do think that in that case it is a bit of out of the pan and into the future fire.
[00:05:09] Right? Money has a way of spoiling things.
[00:05:12] Uh, and this is why earlier I said that the actual solution would be to run a local LLM like and use an open source LLM so we're not playing with the big, the big companies anymore. That's the true, the true ultimate solution.
[00:05:27] Uh, and maybe, maybe in some, at some point soon I have so much I'm doing right now, maybe some point soon I look into doing that and figuring that out and if and when I do, I will share it with all of you. Absolutely. But the reason I say out of the, out of the pan, into the proverbial future Fire is that any company that's still somewhat decent, which was, I would say is probably Anthropic, like it always just feels like it's a matter of time and before money makes things worse. Right? So Anthropic, they are arguably, quote unquote, better when it comes to, you, uh, know, all things politics. So my vote would be to go for them that go with them. If you're like, hey, well what should, which one should I use? I also like it the best of the other ones. Like, I don't particularly like Gemini. Rachel, if you're listening to this, I'm sorry, not that you care. Rachel is. She doesn't give a. Um, but like I, I just, it doesn't look the same. I don't like it, it doesn't give me the, the responses I want. Um, Claude is much, is much closer. Um, and we're gonna talk about switching but uh, the next biggest player. And my, my vote would be for Anthropic because also if we think about it, you know, Google donates to Trump, Meta donates to Trump. We're not even gonna discuss Grok. Um, so my vote would be to, to switch to Anthropic. Um, but there's nothing about with money and because we know that, but these companies are always looking to secure funding and that's actually the only real controversy. That controversy that Anthropic has had which was a comment from a leaked memo from their CEO, uh, Adario, I think Adario Amade.
[00:07:01] And uh, he said unfortunately I think no bad person should ever benefit from success, from our success is a pretty difficult principle to run a business on. And this was in regards to a decision to consider accepting um, Gulf State money. And it's like that even. It's like a pretty benign, you know, statement. I really want to, when in, in writing this episode and out in this episode, I didn't want to say anything that could come off as like, you know how fucking people are when they go to report on things and you just like slant things and it's like, oh. And they said like, this is a benign statement in regards to, in uh, comparison to everything else.
[00:07:36] Um, and this was basically, they were talking about accepting money from nations that may have ties to authoritarianism like the United States. Right?
[00:07:49] So that's, they, they walked back, they, they, they walked back a little bit on that stance and that's what that comment was in response to. And that's like literally the only real controversy around anthropic. That's like super front facing. Who knows what else is going on. I'm not saying that anything else is, is going on, but I just think that these, with these big companies, it is out, uh, of the pan into the fire. And so if the call is to divest and, and unsubscribe, it really needs to be from all of them, not from, you know, one of them. But, um, I want to give you the, the tactics for how to do so. Right. So yes, part of the impetus to record this episode is my neuroses like what if and you know, my penchant for preparedness. Another part is the political climate and a desire to, to do the least amount of harm. Um, but actually speaking of the preparedness, I, you folks know that I create the blog, the, and the newsletter for this podcast and for each episode using ChatGPT. Right. I have these super extensive outlines that I speak, uh, from and I record these episodes from. Then I put the outline into ChatGPT and it makes the newsletter and I edit it and tighten it up. ChatGPT seems not be working right now. So actually I ran this week's through um, Claude and I'll probably go with that. Like they put out pretty similar things because I'm giving it such a, like so much of a scaffolding to work from. Um, but that is another reason to use, you know, have the ability to use multiple, uh, LLMs. So let's talk about the tactical stuff, right? How to migrate from one LLM to another without crying. A few things to realize.
[00:09:38] Largely two things to realize to set the Stage with, in regards to expectation management.
[00:09:43] Um, and that is I overarching here. It's going to be different, right? Going to a new LLM, it's going to be different. A new LLM is going to feel different, it's going to look different, it's going to respond differently.
[00:09:55] If you've been using your current LLM of choice for like two or three years or even a year, right. Or a couple months, don't expect anything, any other LLM to perform in the same way in just two to three sessions, right? It's going to take time and there's going to be that, that break in period, that learning period. Think about making new friends, right? It's like it's not going to be the same as that person you've known for years. You got the, you got the break in time. So as it relates to uh, the actual migration, yes, you can in fact actually export your entire chat history from chat. Um, but I don't think this is actually the way to go. I did it just to see and I was like Jesus, this is not it also. I mean I, I am a heavy user. Um, um, but to do it you go to Settings data controls and there's like a button and this is export or basically like a text but it says export data and then it takes a few minutes and it sends you an email and it's a, ah, I believe it's a zipped folder and in, inside of that folder there's a bunch of files, all of your like, um, if you do any audio things. I do, you know, I talk to Jackie and we have like the talking conversations. That stuff was in there.
[00:11:04] Um, all the things are in there. But it has one file that's exported as a JSON spelled J S O N file that has all your conversations. It's massive. It is a massive, it's so much text in this thing and it's uh, it's too big to upload it anywhere. And honestly it wouldn't yield a good result trying to just, you know, program, quote unquote, program a new LLM. Just being like, here's everything from it before, like it was the context, Windows is too big. Um, and I actually thought about going that route and that's why I'm making this episode. Because I had the JSON file and I was like, oh, I can like you know, create a Python script and then have this be parsed and like doing all this stuff. And I was like one, I just wanted to see how it would do. But two I'm not going to expect the average user to do that. So that to me is not your best bet for migrating. Your best bet is to create what I'm going to call an LLM portable memory doc, or what ChatGPT calls an LLM portable memory dock. A portable portable memory memory document is a short file that captures how you think, how you work and how you want help so that you can reuse it with any chat system instead of starting from scratch and relying on the memory of the, the original, you know, of the, of the LLM being locked into the memory of that specific LLM. So the typical contents include identity and role assumptions, values, principles and non negotiables, domain knowledge or standing definitions, style and communication preferences, ongoing projects, constraints and goals, guardrails and do not do rules. It can have more stuff, but that's just like a, you know, an over, overarching quick list, quick and nerdy list.
[00:12:40] The easiest way my friends to create this, and I think you can guess, is to have ChatGPT or whatever your LLM of choice is, have it interview you and answer the questions. I love doing this. I like sometimes it's fine to answer the questions and I usually I use whisper flow, um, so I can talk.
[00:13:00] But having the LLM, having chat GPT ask and answer the questions, it's the best. And then you can just correct it as needed. All right, so you have it interview you, you have it ask the questions, answer the questions and then you approve, um, or you know, you amend the answer and then it will output the final document. So a sample prompt that you can use, feel free to make your own but just loosely, here's what, what you could say. Please help me create an LLM Portable Memory document. Please interview me. In order to create the document, I would like you to ask the questions one at a time and then provide an answer based on what you know about me and how you think I would answer. I will either approve or revise the answer that you provide.
[00:13:45] Simple. All right. And then when it's done interviewing you, it'll generate a document and you can, you can ask it to generate what's called a markdown file, um, and the like. I don't know what it's called. The little suffix of the, the file name would be dot M D. Or you can ask your chat GBT how to make one on your computer. If you have a text editor, uh, or text edit and it's called Text Edit, you can do it in there, you're going to paste it in and you just turn it into a plain text document. All. Um, um. But I don't want to go through that in this episode. Cause you're gonna be like, what the fuck are you talking about? So just ask ChatGPT how to do that. Um, um. The reason we do this is that markdown files, they are the recommended format for LLMs. They're. They're plain text and it's easiest for LLMs to read. Um, so once it's done making this, this document, you save it somewhere safe and then you can take it with you to your new LLM. Right. The thing to note, like I said earlier, is that all LLMs are different. My ChatGPT does have quite a bit more customization features and customization settings and other LLMs. I've done like entire episodes on it. And then you go to a new LLM and you're like, there's no buttons, there's no toggles, there's no places to like, put this. So that it just applies my preferences to the entire, you know, every chat we have. Um, um. So that, that is worth noting.
[00:15:05] Right? Um, but your best bet is still to, to take that, um, that document with you and then ask the LLM that you're migrating to ask it might ask it how and where you can upload your MD file so that it will be referenced and that the LLM can get to know you faster. Right.
[00:15:23] For Claude, I uploaded. I uploaded it as a skill. A Claude skill. Um, I couldn't find a place to put it in Gemini. Ah. Which seemingly I was like going back and forth with it. It seems like it just like learns you over time and like, that's just what it does.
[00:15:37] Um, again, Gemini wasn't my favorite. I think that. Nano Banana.
[00:15:41] Great. You're looking for images. Amazing. You're looking for Google Notebook lm. That shit is literally in fucking insane. That episode. I did go back, listen to it if you haven't like, that Google Notebook LLM is insane. But for just its regular LLM, I'm not so much of a fan.
[00:15:58] The other aspect of changing LLMs is any projects or workflows that you may already have. So I have a ton. I am what I would call and what I'd see other people call a power project user.
[00:16:08] Uh, I use this. I love, I love projects and I have a bunch of them and each of them has their own instructions and their own associated files.
[00:16:15] Uh, and the nice thing with this is that Claude has projects as well and Gemini has gems. So you can't just like drag and drop them. But this now just becomes a matter of copying any instructions that you have for those specific projects and then re uploading any supporting files to the new LLM. Right. It's mildly tedious, but, but you know, it's pretty straightforward and it's probably worth making like a master document that has these things in it and so it makes it easier to copy and move over.
[00:16:44] Super simple though.
[00:16:47] In general, speaking of files, files are going to be your best friend and uploading any kind of writing samples to the new LLM will be helpful in having it Learn your voice Worth noting, in episode 22 I went over how to have chat create what's called a voice anchor.
[00:17:02] Right? Having a voice anchor is another way to help your, you know, your new LLM quote unquote, learn you. I will link that in the show notes at episode episode 22. But for those who don't know you didn't listen to the episode or you just don't forgot, a voice anchor is a short of a short set of rules, patterns and preferences that help an AI tool write in your actual voice. Instead of some generic or AI sounding version, it captures things like how your thoughts naturally move, how you open and close ideas, the tone that you default to, what you never want the AI to do, and the internal logic that makes writing makes your writing feel like you. All right. It's not a style guide, it's not a list of favorite words. It is a baseline map of how you think on the page.
[00:17:43] Um, very valuable to have. Definitely check out that episode if you haven't already and spend some time and you can make one with, with chat and to do in order to make one, you're going to have to upload some of your like your, your best writing. So it's nice to have access to those as well because you can upload those to the new LLM and just give it examples. Right.
[00:18:03] Like I said earlier, I'm gonna say keep saying this, but all the LLMs are different and so there is absolutely value in just using the new LLM and letting you get to know you and get to learn you in that way. Right.
[00:18:17] And it to do so in the way that it learns about its users. Right. Which is likely different from how other LLMs do it. So it may pick up on certain, you know, different things and um, that's just something that's gonna happen over time. If you use these different LLMs, you, you see it, you feel it, you're just like this is like talking to a different person.
[00:18:33] So there is value in talking to that different person, that new person for, you know, a longer amount of time. And then it will get to know you in the way that it gets to know, you know, quote, unquote, know people.
[00:18:44] A nice thing that you can try that you could do here is to perform any rep, any repetitive tasks that you use ChatGPT for. Um, so like, you know, in this case, creating my curious, um, companion, right? Things like that I can. Or any workflows that you have, systems prompts, things like that.
[00:19:00] Try it on the new one, right? It's a really nice test to see how this thing is, how, how similar this is. All right, so one, you note the difference and then you're going to provide feedback so that you get it to output what you want. These things do learn.
[00:19:14] So you, you have your typical. And you can do it on, on both, right? You could have both LLMs perform the same task. I could have them both create a curious companion newsletter for me.
[00:19:24] And then if Chat does like a way better job than uh, Claude, I can feed the output that Chat gave me and be like, yo, this is what I want. What do I need to do or change or say differently in my instructions so that you match this output, right? These things learn this, it's simple.
[00:19:44] Last thing here, uh, with that, you know, Talking to the LLMs, the LLM Convo as a segue, tell the new LLM that you're new and you're wondering about the fastest way to get it up to speed. So it feels as similar as possible to using your original LLM, right? Let it know, like, I've been using ChatGPT for three years. I, uh, you know, want to use you. I'm m coming over here.
[00:20:06] How do I get up to speed? Fat, how do I get you up to speed as fast as possible? What am I going to do? Sometimes, you know, you want to interview me, you can ask about that MD M file that the MD M file, the markdown file, and where you can upload that, right? As per always folks, when in doubt, just ask the robot. Simple as that. All right, let's head into the final section and then we will wrap this thing up. So how I use Chat GPT this week? Each episode I include a section where I briefly discuss how I use ChatGPT that week. This week we were in the computer and I use it to help me code what is called an OG card. Uh, OG stands for open graph. And it's simply a shareable social card that you can use to Promote things on the Interweb. So substack generates these for individual blog posts. I don't have a substack, but I have blog. And I was like, hey, I want to have this. My friend K, he had one in. In his stories. And I was like, that, this looks cool. I would like to make that.
[00:21:04] Um, I wanted to be able to make that from. For my blog, my chat GPT Curious blog. So I had chat GPT helped me code one up.
[00:21:11] Right now all I have to do is I just input the URL. It's like, I had it. Had it. Um, I should say that I actually had it help me code an OG card generator. Not just an OG card. It had me. I had it. I'm like reading this and I'm like, that's wrong. I had it. Help me code an OG card generator so I can make these cards.
[00:21:35] Uh, so now I all have to do is go to my little generator. It's a URL, so it's a web page, and then on it there's an input field and a button, and I input the URL of my blog post, and it generates that OG card. And the card has the title of the episode on it and has the featured image underneath it. And it looks dope. Yes, I could do this in Canva, but it's faster. To do this, all I have to do is input the URL. If you ever have to go on camera, then one, you have to go into Canva, and then I have to type in the title, which means I have to remember the title. And sometimes I don't remember the title. And then I have to go and pull the featured, uh, image. Right. And the featured image is either at this point in Google Drive or it's, like, already uploaded to the website. That's just fucking annoying.
[00:22:17] This is literally so much faster. Just take the URL of the blog post and boom, it generates it. So it's faster. And honestly, it's just really cool to be able to generate my own tools and rather to create my own tools. Um, little vibe coding.
[00:22:32] So, yeah, I would call that AI Assisted coding. Because I didn't just say, hey, make this for me. I asked it, uh, for the prompts that I would need. I asked it for the code largely, like, I understand the framework that's needed. And using this, this, um, uh, interface, we'll call it. It's an ide, but it's. It's called Cursor. It's like an AI generated. Excuse me, AI assisted, um, Integrated Design Environment.
[00:23:00] Uh, Excuse me, Integrated development environment.
[00:23:04] Um, but there's like a lot of jargon in there to say that I also was in the computer making the files. It was the thing that I had chatgpt do was actually make the code, but I made all the files and, uh, working in the. The terminal and things like that. So I think it's just really, really cool to be able to do this. And so that's why I did it. Solved my own problems. Right? But that my friends looking at the time, that is all for today. Hopefully you found this episode helpful. If you did consider leaving a little rating or a review, uh, let's get to 30. And by 30, I mean 25. You get to 25 first. Uh, if you do switch LLMs, let me know. Remember, I am not here to force anyone to use anything. If you decide to stop using LLMs altogether, I want to know. I want to know the things. All right? I am not here to force anyone to do anything. I just want to hang with my curious people and, and hopefully create some informed and knowledgeable consumers. I do not forget, folks, I have a companion newsletter, the Curious Companion, that drops every Thursday. That's basically the podcast episode in text format. So if you prefer to read or you just want a written record of the things, join the newsletter fam. You can head to chatgpt curious.com forward/newsletter or check out the link in the show notes.
[00:24:20] I think that is officially all that I got.
[00:24:26] As always, endlessly, endlessly. What's that? One more time. Endlessly appreciative for every single one of you. Until we chat again next Thursday, stay curious.
