Everyday AI Podcast – An AI and ChatGPT Podcast - EP 217: 7 Steps on How To ACTUALLY Use ChatGPT in 2024

Episode Date: February 28, 2024

Win a free year of ChatGPT or other prizes! Find out out.You're using ChatGPT wrong. Sorry, but it's the truth. We'll tell you what you're doing wrong and cover everything we know about ChatGPT as of today. We're going over the 7 steps on how to actually use ChatGPT in 2024.Newsletter: Sign up for our free daily newsletterMore on this Episode: Episode pageJoin the discussion: Ask Jordan questions on ChatGPTRelated Episodes:Ep 194: 5 ChatGPT Facts You Might Not KnowEp 179: Mastering Prompts With An OpenAI Ambassador – The One Secret Skill RevealedTomorrow' Show: Roll the dice: Winning the probability game in AI visualsUpcoming Episodes: Check out the upcoming Everyday AI Livestream lineupWebsite: YourEverydayAI.comEmail The Show: [e-mail] with Jordan on LinkedInTimestamps:03:40 Chat GPT high-level overview08:39 Access existing GPTs, utilize ChatGPT Vision.13:15 ChatGPT is superior to other language models.16:22 Tokenize text, emphasizes need for communication clarity.29:57 Experts caution against inadvertent use of GPT.32:38 User highlights concern over ChatGPT's memory.38:51 Grow brand on LinkedIn with Taplio efficiently.44:05 Large language model gobbles up low-quality information.50:12 Comparing expert chat training with GPT training.53:07 Altman discusses improving GPT's reasoning abilities for future versions.58:05 Enterprise versions handle data differently. Be cautious.Topics Covered in This Episode:1. Understanding ChatGPT2. New Features in ChatGPT3. Data Privacy and Security in ChatGPT4. Future Directions and Advancements in Large Language ModelsKeywords:ChatGPT, Large language models, Tokenization, Generative AI, AI education, Business operating system, AI tool, Automating tasks, Everyday AI, Open AI, Google, Microsoft, Memory recall, Prompts, Plugins, New features, GPT store, ChatGPT Vision, Browse with Bing, Data analysis feature, Teams mode, Document upload, Memory personalization, GPT 4 Turbo, DALL E, Multi-modal default, Bing, Generative AI for business growth, Enterprise versions, Confidential data.Send Everyday AI and Jordan a text message. (We can't reply back unless you leave contact info)

This is the Everyday AI Show, the everyday podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business, and everyday life. Meet Firefly AI Assistant, now live in Adobe Firefly, the all-in-one creative AI studio. Just describe what you want to create and the assistant handles the rest, orchestrating multi-step workflows across Photoshop, Premiere Express, and more in one conversational interface. You direct the outcome. The assistant accelerates execution. You're using chat GPT wrong.

Sorry. I mean, you're thinking about it wrong and you're barely getting anything out of it. So I'm going to tell you today, not just what you're probably doing wrong with chat GPT, but hopefully over the course of one episode to give you all of the best chat GPT information, that we've covered over 200 plus episodes of everyday AI. And today, we're going to go over on the seven steps on how to actually use chat GPT in 2024. Here's why we're doing this episode. Number one, things change so quickly in chat GPT. And number two, actually, this was our most popular episode. And I still had thousands of people listening to an old episode of month. And I can't

do that anymore. We got to bring you the most up to date. accurate information. So that's what we got going on today on Everyday AI. So welcome. If you're new here, Everyday AI, it's for you. It's a daily live stream, podcast, and free daily newsletter helping everyday people like you and me, not just learn what's going on in generative AI, but how we can also just leverage it to grow our companies and to grow our careers. All right. So real quick, you might see something. This episode is actually presented by Tapleo. I'm excited about that. More on that later. But we are live, but I didn't even want to spend too much time on the AI news.

I think it's too important. So we normally do that every day. So make sure to go to your everyday AI.com, sign it for the free daily newsletter. I made sure that there wasn't anything huge breaking, you know, Gemini 2.0 or, you know, now all of a sudden chat GPT does your laundry or anything like that. So if you do still watch the AI news, make sure to go to your everyday AI.com. Sign it for the free daily newsletter. It'll be going out here pretty soon.

and our website is a free generative AI university, no matter what you want to learn about, whether it's chat GPT or other large language models or maybe AI image companies like MidJourney. It's all on there. So you can go read every single newsletter we've ever done. You can go read and rewatch every single live stream we've ever done. All right.

Does that sound good? Hey, and let me know if you're joining us live, I'd love to hear from you. What is your one biggest? chat GPT question. All right. And chances are I'm going to get to it. I want to know from you, right? This is a live stream. And you know what? I was doing some math, you know, last night. I was doing some last minute preparation. And I'm like, I've answered probably around 3,000 questions on chat GBT. We do a live training like twice a week. Normally get about 40, like 40 to 50 questions and have been doing that for,

almost a year now, right? Just under a year. So if you have a chat, GPD question, even if you're listening on the podcast, I put my information in there, make sure to reach out to me. And I'll go ahead and try my best to answer your question. But let's start high level. All right. Let's start high level. And I want you first to have a little bit of patience with me. All right. We're going to be a little bit all over the place, but not just have patience with me, but also have patience when you're working with a large language model like chat GPD. All right. There's a reason why, you know, sometimes I go on these long winding stories,

but that can be how humans learn something best. All right, but I want you to first know this. Chat GPD is not an AI tool. All right. I'm going to say that again because it's worth repeating. Chat GPT is not an AI tool. It is a business operating system. It is a complete suite of the next generation of artificial intelligence.

And you can use chat GPT to literally not just automate your business, but you can accomplish just about any within chat. So before we even get started, because I know so many of you probably think of chat GBT is, oh, it's an AI tool. You know, it can write a blog post. Yeah, it can. You know, that's like saying, you know, an airplane is, is somewhere where you eat peanuts. Yes, you know, but it can also take you around the world very quickly.

All right. So first before we even dive into this, I need you to change your mind frame on what chat GPT is. Yes, it is a large language model. Yes, it is technically an AI tool, but it's not. It is so much more than that. All right. So let's keep this thing going.

Like I said, it is a business operating system. And here's the thing. It's not a quick fix. All right. Chat GBTGPT is not a quick fix. You really have to spend time to understand it and spend time working with the model to get the most out of it. And that's what today's show is all about. Those seven steps on how to actually use it.

All right. And hey, if you are joining on the podcast, I always appreciate it. it. This is one of those you might want to come watch the live stream. I am sharing, sharing some slides here, but I want to do my best to walk and talk through everything. But I like to tell people this, where you spend your time right now is probably doing these things. You're probably doing a lot of meetings, right? So prep work and follow up. You're probably learning a lot in your job. When I say knowledge workers, right? Chatchiti can do so much of what knowledge workers are doing right now. That's what you do when you sit down in front of your computer.

So it's a lot of meetings, number one, learning. You're constantly learning new things. You're reading. You're taking notes. You're keeping an eye on your competitors, right? Writing. So you're writing documents. You're writing emails. You're writing maybe blog post, right? You're analyzing things, right? So spreadsheets, charts, et cetera. And then presentations, maybe you're creating presentations, sales trainings, pitching someone to try to get a new client, right? What I just described there, meetings, learning, writing, analysis, and presentation. Those five things, if you work in front of a computer, those five things, probably take up 80% of your day or more. And I'd love to hear from our live stream audience.

And thank you, everyone, for joining us. Caroline, thank you. And Justin, thanks, thanks for joining. But how much of your time would you say is spent doing those things? Like I'd say, most people is 80%. And here's why I'm taking a while. Yes, I'm doing a big wind up here before we get our list started. Because so many people, even people who are still using chat GBT on a daily basis, I'm not using it right, right? Those things, meetings, learning, writing, analysis, presentation.

That's what most all knowledge workers do, a large percentage of your time. And you can do all of those things inside chat GPT, right? Shannon said that's 99% of her time. All right. So now that we got our mindset, right, I first want to go over the capabilities of chat GPT plus, all right? Yes, there's a free version. Yes, there's a paid version.

That's actually, I'm going to give something away. That's our first step. All right. But here are the capabilities of chat GPT. Again, I'm starting high level because this is for beginners to advance users. Ready? So you have a multimodal default. All right. That includes the ability to launch chat GPT Plus and browse with Bing. Use the data analysis mode, which if you caught our newsletter, they're working on a new update to that. You can use Dahl Lee, create AI images, vision. Chat GPD can see and document upload. All right. There is the new version of GPT4, which is called GPT4 Turbo.

you have GPTs, the ability to quickly create and customize versions of chat GPD for your own use. You have a GPT store. So all those custom GPTs, there's apparently millions of them that are already made. You can go search for them and use them that other people have made. Vision, we just briefly talked about this. So now we're going to go a little bit more into these modes. So with chat, GPT vision, you can literally upload a photo of anything and say, what is this? Or if you have a leaky sink or something, you can say, what's going on here?

I don't know, right? So chat GPT can see with vision and obviously reason, right? Brows with Bing. Brows with Bing is important. We're going to be talking about this when we talk about hallucinations. But, you know, chat GPT or GBT4 has an April 2023 knowledge cutoff. So the ability to browse with Bing inside of chat GPT is huge because it reduces your likelihood of hallucinations, right? Dolly, we talked about great to be able to conversational in one chat,

right? To be able to, you know, you can be planning out a project and then, hey, say, hey, Dolly, could you create a visual that goes along with that project, right? Data analysis. All right. So that is its own technically, it used to be its own mode, right? But now you can just use it in the multimodal default mode. So if you call to data analysis, it actually does a great job at advanced computation. And like I said, if you read the newsletter, you saw that chat ChbT is rumored to be releasing a new version of data analysis. Then you have teams, all right?

Chat Chbt isn't just, hey, this is just me working on my own. You can have teams. So there's a teams mode for teams from two to 149. All right. There's some extra data privacy, some extra data privacy options in there and the ability to better work with your teammates. All right. Memory, oh, had the GPT store on there twice. Hey, this wouldn't be everyday AI if there wasn't at least one error on a slide, right? That's how you know it's made by me, the human.

Memory, so the memory personalization, a lot of people are talking about that. I'm going to get to that later. There's pros and there's cons and we dive into that more in today's newsletter as well. But essentially, it allows chat GPD to remember certain things from chat to chat, which is a huge shift in the way that large language models in general work. All right. Then you have document upload, the ability to upload a document into chat chivity. All right. So I'm starting there because I know, you know, I know a lot of you joining here on

the live stream, you know, like woozy joining us and Nancy, right? I know a lot of you all on the live stream here are advanced power users. And that's why I love having your comments and your insights, right? But I know a lot of people listening on the podcast as well, you know, are newbies and that's okay. So I wanted to first start with an overview, right? ChatGPT is not an AI tool. It is a business operating system. All right. And then I wanted to show everyone as well, here's the capabilities because so many of these things, so many of these nine things that I just went over have just been either released or updated in the last three months. All right. So it's

important that we all get on the same page. All right. I think we're good now. So now without further and two. Let's get to the seven steps. Enough wind up, right? All right. We're counting them down, all right, to build up, build up the drama. All right. So the first one is step seven. Use chat GPT plus, not the free version. All right. So let me just give you some examples, right? I like to think of it as a map. The free version of chat GPT, which is, GPT 3.5, it's like a map, right? It's nice. It's there, but it's not dynamic.

It's static. All right. If a road changes, you're screwed it with, if you have a map. Right. The pro version or chat GPT plus is $20 a month, right? And all those, most of those capabilities that I just listed off are exclusive to the chat GPT plus version. So when you use the $20 a month chat GPT plus, And y'all, this episode's not sponsored by OpenAI, FYI.

I'm not getting paid to say this, but you should be using the plus version. My gosh, right? So that is in that scenario, that's like having ways on your smartphone, right? It's always up to date, right? There's all these other features that are constantly being rolled out. So it's like having, you know, a smartphone, right? Like if you want to navigate somewhere you've never been before, would you want a map that was maybe printed five years ago?

Or would you want a smartphone that has the best service, right? It has a GPS in it and you can do anything with it, right? Or even I like to say, you know, a tricycle versus Tesla or an encyclopedia versus Google or a typewriter versus a laptop, all right? The free version of chat TVT is extremely limited. So if that is what your, you know, your whole relationship of chat. ChatGPT is mainly through the lens of the free version, you can't be using it, right? If you're serious about using chat Chbett, if you're serious about getting the most out of

large language models, and by far, this is not even close. Sorry, Google Gemini, Ultra 1.0, Ultra 1.5, chat GPT, because of all these other features runs laps around every other large language model out there. It runs laps around any Gen AI tool out there, right? So I have a little chart here, kind of showing the differences between the free version and the paid version of chat chbt. So there is actually a kind of an advantage to the free version. It's a little faster. And it's, you know, there's less limits, right?

But, you know, I'd rather work with the chat chpt plus. There is a limit. So you get 40 messages every three hours, right? And the free version of chat chp t is a little. a little faster, but all these, you know, especially with the new turbo version of GPT4, you know, there's really not a big difference anymore. There used to be, but not anymore, right? So some people would always say, oh, look at how fast the free version is.

Why do I care? Well, you don't get all these other things in the free version. All these, you know, plugins, which are going away more on that later, advanced data analysis, this multimodal, the ability to create your own GPTs, right? No code. All of that is only available. in the paid version. All right.

So you can't, you can't skip that. All right. So next step, step number six on our seven steps on how to actually use chat chbt in 2024. You have to understand how a large language model works. All right. Yes,

the first version is you got to use the right one. The second step, right, is you have to actually understand how it works. All right. So there's a lot, right? And I've literally had 50 minute episodes where I just talk about tokenization. All right.

So, you know, Shannon, I saw your comment here asking about token counters. All that's going to be in the newsletter. Don't worry. So, but you do have to understand that a large language model thinks and responds in tokens. I know that's weird, but it doesn't actually understand words. It converts all of those words that we put into that prompt. It converts everything into tokens. All right.

That's important to understand because that just emphasizes the need for more clarity in your communication. All right. Because two words can mean the same thing. So that's why we teach, and I'll probably mention this a couple of times, you know, the PPP method, which is prime prompt polish. That's our free course. We're actually blowing it up and rebuilding it from scratch because of plugins. being gone. But anyways, that's what we teach in the course is that you have to go back and

forth and use more words because of the tokenization process. And I know my fellow dorks out there aren't going to like when I say this, but the easiest way to think about what a large language model is is think of it as the world's most advanced, the world's most advanced auto-complete system. Yes, it's generative. it's technically, you know, just predicting the likelihood of what the next word is going to be based on the trillions of parameters that a large language model is based on. That's another important thing to understand about a large language model is how it's built, right?

So essentially with a large language model, it has gobbled up the entirety of the Internet, the open Internet, the closed Internet, other pieces of works, et cetera, right? it doesn't necessarily know what's good or bad, but it's trained by humans, right? So these models, they gobble up all of the information of humankind that's essentially ever existed, you know, even stuff that you don't know exists. It's the large language models have already gobbled it up. And then it's trained by humans, right? And then this is, again, oversimplifying it.

And this is for beginners, because I know my advanced crew here in the live stream, you probably understand this. But then when we talk with chat GPT, it changes everything into tokens. And then it looks at its big 1.8 trillion parameter, big neural network brain. And it makes sense of those words. It turns them into tokens. And then it looks at its big old database of everything on the internet. And it says, okay, based on these words, based on the tokenization of these words,

and based on the context that it's gaining through tokenization, it's saying, here's what I believe that they are asking, right? It's oversimplified. But keep in mind, chat GPT, like all large language models, are generative, right? So that means you are going to get different results. There is literally a regenerate button within chat chbt. If you ever want to really understand how a large language model works, click that regenerate a couple of times.

Change two words and click, you know, click enter again and see how much different your results can be. So you can run the same or a very similar prompt in chat, GPT, or any gen AI system. wildly different results or different outputs. All right. So another very important thing. And we're going to get more into this when we talk about the features and how to use all these features.

But chat GPT4 or chat GPD Plus is connected to the internet. All right. So that's a good and a bad thing. So with Browse with Bing, browse with Bing used to be its own mode. All right? So you used to kind of have a default mode and then you've got to. had a browse with Bing mode. So essentially, you know, OpenAI has a partnership with Microsoft. So chat GPT has a fantastic feature called Browse with Bing. And it's in the default mode.

But you might not always know when it uses it because it can go kind of fast. So unless you're really looking at it and it's not always going to, you know, do a link and, you know, give you a link and say, hey, I did a quick search. Here's what I found. Sometimes it just might spit out that information and you might miss it. So it's very important to know that in the default mode, when you are using chat GPT Plus. It is connected to the internet via browse with Bing, which is both a good and a bad thing. It's a good thing because the knowledge cutoff,

which we're going to talk about more, the knowledge cutoff for GPT4 as of now is April 2023, right? And that's actually kind of old. That's 11 months old, right? So think what hasn't changed in 11 months, right? When we think about chat GPT Plus is a business operating system, basically anything that you're working on in your business has probably changed greatly in the last 11 months. So it's very important to understand that how and when you put in a prompt into

chat GPT or you put in a response, you have to understand what's going on under the hood, right? So essentially chat GPT decides through tokenization and context if what you're asking on requires it to browse with Bing, right? We've done so much kind of like, like blind A-B testing on this with different accounts, different chats, right? Sometimes it will just respond without, you know, seeing if it's, you know, information past 2023. Sometimes it'll say it, you know, hey, I browse with Bing, here's the output. Sometimes it will browse with Bing, but won't even necessarily tell you because it'll bring an information past April 2023. You always have to be testing when you're using chat chit and you have to understand how it works.

It's literally why we're doing this episode because so many people are getting. this piece wrong. So that's important to know. All right. So here's here's a good example, right? Here's a good example. So in this example, I said, how much has Nvidia's stock grown the past year? All right? And for whatever reason, when chat GPT responded in this case, it was just trying to access information from one site. It was trying to look up something on the from the Motley Fool, a very famous financial website. So I didn't tell it to browse with Bing, right, but obviously it understands by the context when I'm saying how much has the stock grown in the last year. And all that happened is it looked on one website, right? So it didn't give me a result. It didn't know,

right? So if I regenerate it and run the exact same thing, now it tells me, Nvidia stock has experienced significant growth over the past year with an increase of more than 210% since January 2023, right? But that was actually outdated as well. Because when it browsed with Bing and when I clicked the little site, the source that it cited, that was from the end of January, which was 30 days old. And I just did this screenshot like three days ago, right? So that was still technically incorrect. I did the exact same prompt one more time.

And this one was much better. It said over the past year, Nvidia stock has seen a total return of, you know, 236%, etc. But this information on the third iteration was actually accurate at the time, right? So each time, one time it didn't really use Browse with Bing or use it correctly. And then it did use Browse with Bing and it brought back some inaccurate information. And then the third time, it used Browse with Bing and it brought back correct information. All right. And all of this is happening behind the scenes.

You might not know. If you're new to Chatsypte, you might not know what's going on. That's why I'm walking and talking you through it and showing you examples. So like I said, Browse with Bing is both very powerful, but it can also give you outdated or false information. You've got to keep that in mind. All right. We're doing a lot of water, y'all.

A lot of water today. There we go. I'm actually, if you couldn't tell, you know, maybe if you're, if you're joining to the live stream daily, I might look a little different today, right? I'm actually in, in New York. I'm not normally in the Chicago home office. All right.

So I'm a little more thirsty today. All right. So step five. All right, step five. You need to use new features strategically. and know what they do. All right. Let's talk about that a little bit.

All those new features, right? We talked about them, but something we didn't go into great depth is number one, plugins are going away. If you listen to the everyday AI show at all, you know I love plugins because it was an unfair advantage. All right. So yes, we still do have these GPs, these custom versions of chat GPD. You can train them on your own configuration instructions.

You can upload your own documents, right? So chat GPT or that GPT has a better idea of what you want to use it for. Right. So Open AI has been phasing out plugins. And they officially said, hey, they're going away in March. So or you can't start new plugins in March. So with that, Open AI is saying, hey, you don't need GPs. You need plug or sorry, you don't need plugins because we're doing chat GPs.

Or we're doing plugins. Oh, man. So so here's here's the difference, y'all. It's not the same thing. Using GPTs, it's not the same as plugins, at least right now. So right now, GPTs work a little differently, all right, then plugins. What I loved about plugins and what we taught in our free prime prompt polish course. And if you do want to access at any time, just put PPP and we'll make sure to send that to you.

But plugins could work together. They could work autonomously. So you could go through and train a chat. That's always what we teach. And you could put in one prompt. And then those three little plugins that had three very different, you know, different features, they could work autonomously. They could work with each other, right?

You can't do that with GPs. You can only work or mention one at the time. However, in the same chat, you can mention unlimited GPs, right? So it's not limited to three, which is nice. So there's pros and there's cons, right? So these new features, you have to use them strategically. However, you can't use three or five or 10 GPs in the same prompt or the same at the exact same time.

like you could use three plugins at a time. All right. So the reason why, you know, I really wanted to go over this piece. And actually, let me get, let me get some exacts here, right? So plugins will be new chats with plugins will be disabled. There we go. We got the actual date there on March 19th, right? So you can't create new chats with plugins after March 19th of 2024.

And your existing plugins that have, your existing chats that have, your existing chats that have plugins will no longer be usable after April 9th. All right. So let's just quickly show what a GPT looks like. All right. So this is a new feature. And I think they're very powerful, but you have to use it correctly.

I see so many people, even the experts, using these new GPs incorrectly. All right? So real quick, and if you're listening on the podcast, it's simple. Creating a GPT, right,

for all the downsides I don't like, creating them is so simple. simple. So it is a no code or low code. So you can literally just go in. It's kind of meta or you know, inception, but you can chat with chat GPT to create a GPT. So you don't have to know code, right? And then there's a, you know, as you're building it, you get a little preview, right? And then you can go on the back end and go to the configure tab. So essentially after you have this conversation with the GPT builder, it will create configuration instructions for you. You can go in and change those or modify those as well as upload documents. So in this example, you know,

I was just uploading, you know, Jordan writing samples. So as an example, if you wanted to create a GPT that sounded a little more like you, it's not going to be great, but, you know, that's one thing that you can do there. All right. So here's, here's where you have to understand the downside of these GPs and how a lot of people are using them wrong. All right. So you can use a GPT in its own mode, right? So a dedicated kind of chat that is just with that GPT. Or in the default mode, you can use the GPT mentions feature. That is the way you should be using it. As far as I know, and I've done a lot of testing on this, there is zero, zero reason to ever use a GPT in its own mode or in its own window and not just use the default mode and mention it.

Number one, if you're doing that, you're kind of stuck there, right? And you lose some of that functionality and some of those capabilities that you would have in the quote unquote default mode of chat GPT. And here's the reason why. So many GPTs, which is actually good and it helps them work better. But they're very refined, right? So, you know, as an example, a lot of GPs have browsedies. with Bing disabled. They might have Dolly disabled, right? And those configuration instructions

might really limit. So if you're trying to, you know, let's say you're using a data analysis, GPT, right? And it's great, right? It's powerful. It's great at, you know, Python and it's great at, you know, creating spreadsheets, whatever, right? But if that GPT does not have access to browse with being enabled, and if you want to, you know, get some information, if you want to use what's in that gpt chat, you got to start a new one and do all this copying and pacing. So when I say using these new features strategically, I mean just that. You should almost always be using any GPT and the mentions. So in the default mode and then mentioning them, all right?

And then when you're done getting the information from that GPT, by default, the GPT is going to stay there and you are going to still be quote unquote talking. You just got to exit out of it. All right. And then just return to the normal default chat. All right. All that information will stay intact. The reason is, I see a lot of people doing this, a lot of even quote unquote experts who are teaching people.

They mention a GPT. And then even when they're done, it is still there in their sidebar and they're still are in their top bar. And they're still technically chatting with it. And they don't know. Right. So anytime it's there, even if you are not using it, you have to exit out. Otherwise, if that GPT does have limitations or certain ways to act in its configuration instructions, whether you know it or not, you're still using that.

All right. We're going into the intricacies here, but I wanted to make sure that even whether you're a newbie or an expert, that you're getting a lot out of today's episode. All right. Another new feature. And this is being slowly rolled out. All right.

I love this. Devald said, made a GPT to help make GPDs. Same thing. Same thing. So, so good. Such a smart, such a smart thing to do. All right. So memory person.

is another new feature. All right, we're spending a lot of time on this because I know that people are going to be listening to this episode for months to come. So this feature has not fully been rolled out yet, but it is being introduced and a lot of users have this. So this new memory personalization allows chat GPT to remember things across chats, which on the surface is amazing, right? That's great. And you can always go in to check to see if you have it. Again, you have to have chat GPD plus go into personal, into your settings. go into personalization and you can toggle memory on and off and then you can manage it as well.

So from, if you click to manage memory, you can go through. And these, so essentially when you are talking to chat chbt, it's going to take things automatically. That's the key here. It's going to automatically start grabbing facts, right? So is it a little weird? Yeah. Is it helpful? Absolutely.

Right. So, you know, in this example, if you have a two-year-old daughter named Lisa Lena, it's going to remember that. right? So it's going to remember that across different chats. So you can kind of think of it like, you know, retargeting, right? If you visit a website and, you know, you get pixeled and there's a cookie on there, right? And you go to another website and you see an ad for it. So it's kind of like that. When you start different chats, it's going to start to remember things about you or things that you've told it. So chat GPT by default will remember certain things about you without you asking or you can ask chat

GTP to commit something to memory. Personally, I'm not going to be using this a lot. And I think that most people probably shouldn't, at least as it is now. Here's why. Number one, I don't think it's great that chat chepti is going to automatically put things into memory. It would be good if it gave you a notification or if it asked you, right? If you put a big, big long conversation and it's like, okay, you know, these are the things

I want to commit to memory and you can click yes or no. But that's not how it is. Right now, this works behind the scenes and is committing things to memory. So here's an example of why that might not be the best, all right? It might not be the best because what if you don't want chat GPT to remember all of those things, right? As an example, what if you're using chat GPT for many different clients, right? Or maybe you have many different writing styles or, you know, maybe you are using it for personal and for work. you probably don't want those certain things creeping in to all of your other chats.

So personally, because I use chat GPT for every aspect almost of my life, this is actually a feature I am not going to use probably at all until they change how it works. All right. So quick recap of the new features in the default mode. You can mention any GPT and you should be doing that. Also, keep in mind, the browse with Bing in the new features and the default, mode, you have to really understand because that's where a lot of people are getting bad, are getting bad outputs, right?

It's because they don't know how all of these new features work. So again, here's what I was referencing earlier, right? So when you mention or when you at mention a GPT, it kind of pops up there in the top bar. So once you're done, you know, and you can mention any GPT just by hitting the at, right? And it's only ones that you've used before. But then once you're done, so as an example in this screenshot, I'm mentioning the web reader GPT because maybe I want to visit a specific URL, which BrowseWitt can not do. Another advanced tip for you guys, right?

But then when I'm done, when I'm done, you know, using Web Reader, I need to exit out. I need to click the X button and then go back to using the default mode. Otherwise, whatever, you know, restrictions that might be on that GPT are going to follow me around everywhere. So keep that in mind. All right. Here we go. We're going to go fast now, y'all. Hey, Cecilia said love that you ran away to New York City. You know, I'm going to be back in two days. Don't worry. I got to hold it down for all my Chicago people. All right, here we go. Number four, keep track of memory to reduce hallucinations. Yes. Someone was asking about memory earlier. We got to get a little dorky here, y'all. We got to get a little dorky. So,

here's what you need to know about memory. There are 32,000 tokens of memory right now. So here's what that means. That's about give or take, 26,000 words. So when you are chatting with chat GPT, and hopefully we're going to get to this here in a minute, but you're going to be going a lot of back and forth. All right. So you need to keep the memory in mind because it has a 32,000 token context window of memory, which is about, like I said, 25, 26,000. words. So all of that back and forth with chat, GBT in that chat, once you kind of get past that 32,000 tokens, it is going to start to forget whatever is at the top, right? So yes, use a token counter extension. The one that we use is literally called token counter. So, you know,

if you use Chrome or Edge, you can use it in there and we'll be linking to that in the newsletter. All right. Today's newsletter, y'all, I'm being honest. I already wrote it. And it's probably the best newsletter I've written in 220. So much good stuff in there. Like I should probably be selling today's newsletter alone for like $200. There's that much information. Thousands, literally thousands of hours of using the GPD technology. I didn't even say this. Our team's been using the GPT technology since 2020, right? I told you, I've trained 2,500 business leaders on how to use chat GPD. All those questions, thousands of questions, all those answers. Literally, today's episode and the newsletter is a culmination of all that.

All right, but you need to keep memory in mind to reduce hallucinations. So hallucinations mentioned it before, but that's when chat chagipati kind of lies, right? That's when chat chit can't tell a half truth, right? Or be a little ambiguous or show some ambiguity, right? And a lot of times it does that because you don't do a good job. You, the user, don't do a good job when you are starting a chat. You just put in a big prompt, right? And you don't take the time to train it.

All right. So you need to keep track of memory. So as an example, I mentioned, you know, a token counter. So I have a screenshot, right? So in this specific chat, I'm showing that it says, oh, you've used about 13,000 tokens here or about 10,000 words. Okay. So I know in that one chat, if I keep that token counter up, once I get to 32,000 or if I, let's just say I get to 35,000 tokens, right? So in 3,000 tokens over chat GPT's memory.

limit. It's context window. So what that means when I'm at 35,000, it is going to forget what is at the top or it is going to forget the first 3,000 tokens that I gave it. So again, this is one of the things I see over and over that people are getting wrong about chat chbtee. Even people that use it every day, every day. I've had very bright, intelligent, amazing people who literally give speeches on chat chbtee. I kid you not. And they're like, Jordan, I had no clue about how memory worked. That's how memory works. You got to keep in mind. And that's one of the complaints we always see, hallucinations and, oh, you know, I was using chat, GPT.

It was great. And then it started to forget things. And then the quality went down. That's why you're not keeping track of memory. And that's when you start to get hallucinations. Ooh, y'all, I'm so tired. Finally. Finally.

I've always, I've always, like, joked about this. I can finally take a water break because today's episode is we partnered up with Tapleo. We've been using them forever. So here's actually me talking about Tapleo. And I'm going to take a water break. This is great. I get to be like you guys now and watch and watch this. All right.

If you're looking to grow your brand on LinkedIn and who isn't, just do what I do here at Everyday AI. I use Tapleo. So do you know that LinkedIn is 227% more effective at generating leads than other social media networks? You know I bring the receipts. But that is also why Everyday AI has such. a heavy emphasis on using LinkedIn for growth.

And that is why I've been using Tapleo. I've been a paid subscriber for Tapleo since way before I even started the Everyday AI show. So here's the basics of Tapleo. You can use their AI powered suite of tools to discover trending topics to write about, create great content 10 times faster, or engage with the right people at the right time, more efficiently, and turn followers into leads. That's literally the feature I use the most.

So here's what you need to do. Go to Tapleo.com and sign up for a free seven-day trial. of the plan that fits your needs. And then on checkout, go ahead and use my code every day. That's E-V-E-E-R-Y-D-A-Y to get your first month for $1. It's only good for a couple weeks. So make sure to go do that now. And check out the show notes as well.

I'll give you a little secret on exactly how we use Tapleo to grow everyday AI on LinkedIn. So let's do this together and use Tapleo for growth. Oh, that was fun. So during that break, sponsored by Tapleo, I was doing some. stretching and I got hydrated because now we're getting into the top three. All right. And a little bit more, we're actually going to be giving away a year of Taplio. All right.

I'm going to tell you how at the end. And also I'm going to tell you how in the newsletter. All right. But here we go. Let's get to the top three on the seven steps on how to actually use chat chbt in 2024. All right. Number three, prompts don't work. Don't use them.

All right. I'm going to try hard here not to really, you know, be mean to Billy boy, you know, the 20-year-old chat GPT expert who was previously an NFT expert, previously a crypto expert, right? And he's like, hey, you know, use all these prompts in my prompt book. I have, you know, 9,000 prompts for $99. Don't use those. They don't work. All right. Here's, here's what prompts do. Here's what prompts do. All right. They get you from an F to a C. If you're brand new to chat, GAPT, yeah, go go find a super prompt. Put it in there and see what chat chat, but he can do. But that's all prompts actually do. It shows you the capabilities and it can

get you from like an act to a C. All right, because technically, and we're going to get a little technical here, but if you want to get more dorky, I link to other episodes in the newsletter where we went into this in great depth and detail. All right. Copy and paste prompting, which is what a lot of people do. They first try something on their own, doesn't really work. And then they go try to find the longest prompt on the internet and they put it in and they're like, all right, well, this is a little better, but still kind of stinks. Because That's not how a large language model works, right? It's not like a search engine.

It's not like Google. It's not where you put in one input and look for one good output. That's wrong, right? So all of the benchmarks. So there is something called MMLU. We've talked about this a lot. MMLU is, you know, widely considered the best benchmark for large language models to see, essentially can it reason like a human?

Is it smarter than a human? All right. So it's the multitask language. understanding. That's MMLU. All right. And every single test out there, right, I'm going to go ahead and skip to my graph here. All right. Every single test, every single benchmark out there says that you get much higher results or much better or much closer to a human than if you use multiple shot prompting. So what that means, a copy and paste prompt in most cases is essentially a zero shot prompt or a one shot prompt where you're not giving any examples. You're not going back and forth. You're not having a conversation,

right? You're thinking a large language model like chat GPT is like Google. I put in a large input. I get an output, right? It's not how it works. So a shot, right? There's something, you know, always the highest scores on these tests is what's called a 32 shot chain of thoughts prompt, right? Not going to get too technical, but this chart here shows you, right? A, A five-shot prompt with the same large-language model is going to always outperform a three-shot prompt, right? A three-shot prompt with GPT-4 is always going to outperform a one-shot prompt. All right. I don't know why people hate math and hate science.

Prompts don't work. They get you from an F to a C. right so we need to be doing what is essentially and again I'm oversimplifying here so you know if you build large language models for a living yes you can still tell me I'm wrong but I'm oversimplifying here right but when you're having a conversation and you're going back and forth with a large language model you are essentially doing you know a multi shot prompt a five shot prompt a nine shot prompt if you go back and forth nine times give nine examples that's how a large language model works prompts do not work All right. Very low quality. So one thing I always like to pick on, right, if you tell a nearly two trillion parameter large language model to act as an example, act as an expert marketer with 20 years of experience, that does nothing. Remember, we talked large language models with trillions of parameters have gobbled up the entirety of the open and close internet and more, right? And guess what? There's a lot of people on the internet that say, I'm a marketer with 20 years of experience.

and the large language model has gobbled that up. So chatDB has gobbled up bad information and good information. So if you think that saying something at a prompt, like act as an expert marketer with 20 years of experience, if you think that does something, it absolutely does not. We've done blind testing 20 times each way. It does not do anything. A lot of times it does more harm than good. I gave that example earlier of the Nvidia stock. Think when you regenerate something different, when you regenerate something different.

If you go through, if you want to create an expert chat that is a marketer with 20 years of experience, you have to go through and teach it, right? That's what we teach in our free prime prompt polished course. Also, you got to be aware of prompt injections. Again, something for our technical crowd, but a lot of people are using these super prompts and maybe not looking at them too closely. Yes, just like any other software or hardware, there is something called prompt injections, which are dangerous and you might not be seeing them happening under your nose.

All right. So that's why I always recommend working with a large language model like a human. All right. Which brings us to number two, prime chat GBT before asking for an output. Frank says good, good content. Thanks, Frank. And Nancy's clapping. Appreciate that, Nancy.

Juan's signing up for Tapio now. That's great. Make sure to use the code every day, Juan. All right. So number two, prime chat, GBT before asking for an output. Here's what that means. and this is what we teach, all right, and check this out in our newsletter. But you need to create dedicated trained chats for specific tasks.

All right. Another thing that we teach in our free prime, Pripolish course, when you are creating a new chat, right? You click the new chat button and chat chvety. You don't use that chat for everything. You don't say this is my Wednesday, February 28th chat. No. When you start a chat, you do so with the intention of training it on one very specific skill set. all right and you go through and you have a conversation with it like a human you talk back and forth you

share all this knowledge you go through essentially training right it's it's like if you have a new employee you have to train them you have a conversation you don't throw down you know a big training manual and say go to work you you go step by step you have a conversation you go back and forth you ask questions right that's what priming is when you start a new chat inside chat chpt you train it for one specific purpose all right so let's say it's for the example that we were just kind of talking about, you know, the expert digital marketer. Let's say it was marketing strategist, right?

So you were going to train it on that and you don't want an output, right? You are literally, when you start a new chat, you don't say, you are an expert strategist and you give all this information and you say, now build me a marketing plan. That's wrong. You train it on a skill set first. You go through, you ask questions, you go back and forth, you share knowledge, and you make it better.

You turn it into an expert. That is priming. You do not want an output. You want an expert skill set, all right? And then when you go back in a week or in two weeks, you go back when you need that marketing strategy expert that you trained, you go use that chat that you built, right, that you trained for that one specific purpose. You don't use that for content creation.

You don't use that for data analysis. You only use it for that one specific thing. But again, you need to prime chat GPT. People always go in and start a new chat and they put this long prompt and they're looking for an output. I say like this. Chat GPT, you have all those chats on the left hand side. Think of all those chats as expert employees that you can train in one specific skill set. That's what the, that's the advantage of working with a 1.8 trillion parameter large language model.

But you have to make each of those chats much more refined. You don't, you technically want a small model in each of those chats, right? You have to bring it down. You have to narrow it in and keep narrowing it. That's what priming is on one specific skill set. And then you go back and you use that when you need it and you don't use it for anything else. But you also got to keep memory in mind, right? You do what we call as a memory recall.

You know, it's like clicking auto save on a document or saving a document, right? So saying like, please recap this conversation so far, you know, in detail. everything that we've concluded, all right? So you need to make sure that chat ChbT is retaining its memory as well. When you're using these expert chats and you're floating back and forth, y'all, I cannot emphasize enough. You're doing it wrong. That is how you use a large language model.

All right. How does, good, good question. How does tokenization come into play? Well, tokenization, people think of it as two different things. So in theory, tokenization is when I put in words, chat, GPT, tokenizes, right, or it converts all of those words into tokens based on context, right? The same one word can technically have four different kind of token values based on context. So tokenization is important, and that's why it's important to play with the tokenizer.

I've done that before, live on the show and showed you guys how to do it. But also, that's why prompting matters in the way of your language, right? You have to be very descriptive in your language and go back and forth and make sure that chat, through its tokenization is understanding it. So Jay said, wouldn't that be what a GPT would also be used for? Jay's asking that. In theory, yes. I can't get into this too much in depth, the difference between training and expert chat

versus training a GPT, right? The problem with GPTs is the knowledge retrieval process is always changing. And personally, it is, for me, it is too inconsistent right now with how it retrieval. the knowledge, depending on how you set it up and how you configure it, it's not as accurate, at least for me, someone that uses chat GPT many hours a day. It is not as accurate as putting it all in text, right? It's kind of the knowledge retrieval process on your GPT can change a little bit, just like, you know, it's a generative system. So that's something, Jay, hopefully that helps. All right. So priming is this, right? Like I talked about.

It is, think of it like training a new employee. The same process that you go through with a new employee. So onboarding, training, reinforcement, learning, feedback, testing, knowledge sharing. That's the exact same thing you do when you start a new chat into chat GPT. All right. All right. Let's go to number one. I was really trying to not make this an hour episode.

We'll see if I can still get in before the time limit here. Number one, and we've been referencing this the entire time. But you need to talk to chat GPT. like a human, y'all. All right? So I have a love-hate relationship with the term prompt engineering, right? It's technically fake,

but it's technically very real. Right. But two years ago, three years ago, you know, people would have you think it's a science working with a large language model and that you had to be very technical. Like, oh, you can't use this large language model. It's for experts.

It's for technical people. Wrong, right? Large language models in the way that they're being developed and the way that they're changing, it's for everyday people like us. Chad GPT is not made for experts. It is made for everyone. It is made for a six-year-old. It is made for a 96-year-old.

You do not have to know anything about prompt engineering. Think of prompt engineering is this. It's having a conversation, right? It is what we are doing right now on this live stream. Y'all are asking me questions. I'm trying to answer them as they pop up on the screen, right? That prompt engineering. Are there better ways to do it?

Of course. But you just have to treat chat GPT like a human. So it's real, but it's also very simple. Like I said, you also have to keep in mind why you should be talking to chat GPT like a human. Well, large language models like we talked about earlier, they're trained by humans, right? So sometimes, I think early on, people were trying to come up with these overly structured, overly long prompts, right? Well, hey, guess what? That can work very well sometimes, but chat GBT is not deterministic. It is generative, right? So it is made to have conversations. And, you know, Sam Altman,

the CEO of Open AI has talked about this a lot in future versions of chat GBT, whether that's GBT 4.5 or GPD5, he said the reasoning is going to be much better than where is that now. And GPT's ability to rationalize and reason, right, or to understand things, right? Again, the most important skill set right now. I was at a venture capital event a couple weeks ago. And a CEO asked me, he said, I have a daughter in college. She wants to get into large language models. I said, what's the most important skill set for her?

I love this question, by the way. He actually asked me this. And I didn't. He's like, okay, she's going to prompt engineering, you know, computer science. I said, yeah, you can do those things. I said, the skill set is communication. you have to be able to talk to chat GPT like a human. A good example is mid-Journey.

I know another Gen AI system, right? But mid-Journey V3, right? You had to talk to it in prompt speak, right? You had to use all these parameters, you know, oh, I have to hit space, comma, space, right? Like all these very structured rules, right? And if you didn't like speak mid-Journey V3 prompting, you weren't going to get great results, right but now look in the newest version of mid journey mid journey v6 i can talk like this i've actually done examples where i just click record on the mic and i talk to mid journey and i say hey here's the

image that i want this from this a i image generator and it does a pretty good job you know so think of chat gbt and large language models the same way they're constantly improving and y'all i'm trying to get you to where the puck is going and we need to skate there together not where the puck is at today. Let me say that again. We need to start building our skill sets for the future of large language models. And we've already seen Sam Altman and all the CEOs are saying that they are making all of these large language models and these generative AI systems less complex. They're making it more for the everyday person, right? I mean, Microsoft ran a commercial at the Super Bowl that showed literally everyday people getting the most out of its co-pilot AI system, right?

This is the future. The future is prompting everywhere. If you're not already prompting on a daily basis, you're going too soon, right? Apple's going to be releasing Gen AI into presumably into its smartphones, into its computers. Microsoft co-pilot is starting to pick up steam and popularity, right? So as a knowledge worker, whatever device you are on, right? And that's why, yes, this is about chat, GPT. but it's about so much more.

Talking to large language models or small language models like a human is so important. You have to have great communication skills, right? We had an entire episode with Abrin from OpenAI, right? An ambassador at OpenAI. He actually does more speeches and more teaching than literally anyone at OpenAI except for Sam Alman. And this is literally what we talked about. We said, you know, his example is a teacher talking to a student, right?

My example is you're a CEO talking to new hires. Same thing. To get the most out of chat, GPT, you have to change your mindset. It is not a content creation machine. It is a business operations system, all right? Business operation system. And you talk to it like a human.

And it can do anything. So the power of written communication is through the roof. That's why I think teachers, journalists, English majors, are great inside of chat, GBT. Because that is how, not just how it works today, but that's how it is going to work even more in the future. All right, let's recap. This was a long episode.

There's a lot of information in here. And y'all, I can't emphasize enough. You got to go read today's newsletter. So I'm going to say it. I'm going to say it here. Go to your everyday AI.com. Sign up for the free daily newsletter. Check the show notes.

If you're listening to the podcast, you can sign up there. I'm not kidding. The value of today's newsletter, it's a long one, right? It's a little more structured than this, you know, one hour live stream. It is worth a lot of money. I'm just going to say that, all right? If you understand these rules, your outputs are going to be so much better. All right.

Before I recap these, I actually see a very important question. Thank you for this one, Monica. So she says, what should we know about entering proprietary private info data into chat GPT? Great. Yeah, let's end there before we go over all of our recap our rules. Don't upload any confidential, sensitive, proprietary, private health information into chat GPT.

Don't do it. Or any other large language model right now, right? I will say this. If you have an enterprise version, though, if you have an enterprise version of a co-pilot or an enterprise version of chat gbtee, that answer might be a little different, obviously, right? And you should be talking, you know, to your rep at OpenAI or your rep at Google or your rep at Microsoft to understand how those systems are handling that data, right? I don't have access to, aside from Google Gemini enterprise, I don't have access to,

you know, these larger enterprise systems. So they have data handling a little different. But if you are on the free version of chat GPD, if you are on the team versions of chat GPD plus, or if you're on chat GPD plus or chat GPD teams, you need to not upload confidential or sensitive information into chat GPD. Great question. Thank you. All right.

Here we go. We're going to recap everything and we're going to wrap up the show here. All right. Our seven steps. All right. Seven. Use chat GPD plus, not the free version. Six.

Understand how large language models work. Five. Use new features strategically and know what they do. Four. keep track of memory to reduce hallucinations. Three, prompts don't work. Don't use them. Two, prime chat GPT before asking for an output.

And one, talk to chat GPT like a human. Y'all, this was a long time in the making this episode on the seven steps on how to actually use chat GPT in 2024. I hope this was helpful. Shout out to Tap Leo. Again, make sure to check out our newsletter if you do want to win a year of Taplio. So thanks to them for partnering with us on this episode. If this was helpful, y'all, and I hope it was.

Like I said, this is the culmination. I've spent more than 1,200 hours inside of chat GBT. I've spent thousands of hours in the GPD technology dating back to 2020. I've taught more than 2,500 business leaders how to use chat GBT. I've answered thousands of questions. This was a lot of work to put this together. So if you do have questions, keep them going, right? I'm going to go jump in when I have time later.

There were some questions I didn't get to. If you're listening on the podcast, thank you. Go ahead. Reach out to me if you have a question. But the biggest thing I want you to know, y'all, is chat. GVT is for everyone. It is not an AI tool. It is a business operations system.

All right. And if this was helpful, I hope it was. please consider sharing this with a friend. Please consider leaving us a five-star review on Spotify or Apple. Or if you're on the live stream here, you know what? This episode took dozens of hours. It'll take you about 10 seconds to just click that repost button. Please, that would be super helpful.

Because you know what? There's bad information out there. There's too much bad information. There's people trying to scam you about AI education. It should be free to all of us. It should be available to all of us. We should all have a place that we can come together, learn the facts together to grow our companies, to grow our careers. That's what we're all about here

at Everyday AI. I appreciate y'all and we'll see you back tomorrow and every day for more everyday AI. Thanks y'all. Meet Firefly AI assistant. Now live in Adobe Firefly, the Allman One Creative AI Studio. Just describe what you want to create in your own words and the assistant handles the rest, orchestrating multi-step workflows across Adobe Creative Cloud apps, including Photoshop, Premiere Express, and more in one conversational interface. You direct the outcome while the assistant accelerates execution. Stay in control with the ability to step in and refine at any time. See it today at firefly.adobie.com.

And that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going. For a little more AI magic, visit Your EverydayAI.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.