Everyday AI Podcast – An AI and ChatGPT Podcast - EP 246: No that's not how ChatGPT works. A guide on who to trust around LLMs

Episode Date: April 9, 2024

Whether you're browsing social media or searching the web, there's so much bad advice out there about ChatGPT. We're cutting down the rumors about what ChatGPT is and how to use it. Here's a beginner's guide on knowing who to trust around Large Language Models. Newsletter: Sign up for our free daily newsletterMore on this Episode: Episode PageJoin the discussion: Ask Jordan questions on ChatGPTRelated Episodes:Ep 217: 7 Steps on How To ACTUALLY Use ChatGPT in 2024Ep 109: LLM Showdown – ChatGPT, Bing Chat, Google Bard, Claude 2 and PerplexityUpcoming Episodes: Check out the upcoming Everyday AI Livestream lineupWebsite: YourEverydayAI.comEmail The Show: [e-mail] with Jordan on LinkedInTimestamps:02:00 Daily AI news05:45 Lots of bad info around ChatGPT08:00 Comparing Copilot and ChatGPT as similar11:45 ChatGPT generates diverse outputs from similar inputs13:40 Comparison of human and ChatGPT writing styles17:08 Stop sharing incorrect GPT technology usage screenshots22:40 trained AI outperformed human.27:35 Ignore exaggerated prompts, focus on authentic communication.32:26 Words analyzed by OpenAI's tokenizer.35:29 Free ChatGPT like landline, GPT4 like smartphone.36:29 Zapier enables custom GPT, data integration, workflow.Topics Covered in This Episode:1. Overview of ChatGPT2. Misconceptions around ChatGPT3. Usage of Generic Business Prompts4. Superiority of Large Language Models5. Risks of Sole Reliance on Tools like ChatGPT with BingKeywords:Chat GPT, ChatGPT Plus, GPT4, generative AI, large language models, NVIDIA GeForce RTX GPU, Free daily newsletter, outdated data, Bing, Jordan Wilson, Nike slogan writer, ChatGPT with Bing, SEO manipulation, Everyday AI podcast, GPU giveaway, US government semiconductor investment, copywriting, AI applications, custom-built ARM processor, Axion, Google Workspace, prime prompt polish chat GPT course, Lewis testimonial, business prompts, misleading information, refining large language model, tokenization of words, refine queue, zero-shot prompting, career growth.Send Everyday AI and Jordan a text message. (We can't reply back unless you leave contact info) Start Here ▶️Not sure where to start when it comes to AI? Start with our Start Here Series. You can listen to the first drop -- Episode 691 -- or get free access to our Inner Cricle community and all episodes: StartHereSeries.com Also, here's a link to the entire series on a Spotify playlist. 

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. No, that's not how chat GPT works.

Whether you're browsing social media, looking something up on YouTube, or maybe you're hearing this from a friend. But there's so much bad advice out there about chat GPT and people talking about what it is and what it isn't. Well, I'm here today to chop down some of the. rumors to cut through the fluff and tell you what chat GPT is and what it isn't as we go through some of the most common things that people get wrong about most people's favorite large language model. All right, what's going on y'all? My name's Jordan Wilson. I'm the host of Everyday AI. We're a daily live stream, podcast, and free daily newsletter helping everyday people. This is for you and me, how we can all learn generative AI and leverage it together to grow our companies and to grow

our careers. So if that sounds like you, and if you want to know a little bit more about chat GBT, whether you are brand new or you're using it every day, I think today's show is going to be for you. Actually, no, I know today's show is going to be for you. All right. So super excited to talk about that. But if you are listening on the podcast, make sure to go to your everyday AI.com. Sign up for that free daily newsletter. We'll be recapping today's show as well as a lot more. You're not going to want to miss today's recap talking about what chat, GPDs. is and what it isn't. All right. So before we get started, let me just tell you, at the end of the show, we're going to be announcing our GPU giveaway from Nvidia. So if you entered, make sure to stick

around to the end of show. We're going to be announcing that live here and in the newsletter as well. All right, before we get started, let's talk about what's going on in the world of AI news. So the U.S. government is investing $6.6 billion with a B, billion. billion in semiconductor chip manufacturing in Arizona. So the U.S. government plans to provide $6.6 billion to Taiwan semiconductor manufacturing company, which is TSM, to build three semiconductor chip fabrication plants in Arizona. So President Joe Biden aims to secure the supply of advanced chips, highlighting the economic and national security vulnerabilities of the U.S. due to decreased chip production capacity. So TSMC.

investment of $65 billion in Arizona includes the construction of three fabrication plants or fabs, as they're called, and it marks the largest foreign direct investment in the state's history. All right. So the three fabs are expected to create around 6,000 high wage tech jobs and 20,000 indirect jobs contributing to the local economy and the job market. This move reflects the U.S. government's emphasis on on-shoring chip production to reduce supply disruptions, particularly after experiencing bottlenecks during the pandemic.

So this investment will enable the U.S. to domestically produce the most advanced, what they're saying is the most advanced semiconductor chips, enhancing supply chain resilience, and reducing dependency on foreign manufacturers. All right. Next piece of AI news. Google has enhanced its workspace with some new AI capabilities. So let's take a quick look. So Google has introduced new generative AI features to workspace, enabling users to create

collaborative videos for storytelling. So this move aligns Google with competitors like Microsoft 365 and Cisco WebEx, enhancing workplace productivity and collaboration. So the new real-time video collaboration feature in workspace leverages Google's AI research and video expertise for advanced presentation storytelling. Google is continuing to integrate AI capabilities from Gemini into workspace applications like docs, sheets, and Gmail enhancing communication and collaboration. So this new all-in-one video editing and writing production assistant

should be seamlessly integrated with other productivity tools inside of workspace. So if you use workspace like our team does, make sure to go check that out. Last but not least, in AI News, more Google News and more processing, right? So Google has introduced its first custom-built arm processor called Axion, based on arms, the company arm, the manufacturer, based on arms Neuverse 2. So Axion instances offer up to 30% better performance than competitors arm-based instances and up to 50% better performance and 60% better energy efficiency than comparable X86-based instances.

So Google did not provide detailed documentation to support these new claims, but mentioned that technical documentation will be available later this year. Google emphasized that Axion is built on an open foundation, allowing customers to bring their existing arm workloads to Google Cloud without modification. So yeah, a lot of chip in processing news today, y'all, but it's pretty big. I think the world has figured out, including the U.S. government as an example, you know, pairing up with TSM. You know, I think everyone's realized just how far ahead of Nvidia is. So a lot of investments here to increase manufacturing for these AI chips, these GPUs. use that really power all of the generative AI tools that we all use.

All right, let's get into it. Let's talk about reasons. No, that's not how chat TV keeps. I've just been seeing too much nonsense lately, y'all. Like every, I don't know, every month or two, I just see so much bad information. And sometimes it is very smart people sharing bad information. So I wanted to go over some common misconceptions. And essentially, if you are hearing,

these things that we're going to be going over. You know, if you hear people telling you this, right, whether it's someone you follow online, whether it's a coworker, maybe it's an outside consultant that your company is bringing in to help you with your generative AI. If you hear these things, run for the hills, avoid their advice, ask more questions, all right? So these are some telltale signs that someone probably has little to zero clue on what

they're talking about when it comes to chat. GPD. Hey, and let me know. Some of y'all, like Tara, who's joining us live and Brian, have already told me, but how hot should we make it? Y'all, sometimes the more tired I am, the hotter these takes get. So let me know, drop a couple flame emojis. Encourage me to either come with hot takes or to take it nice and easy because, you know, maybe it's, it is call out time. Maybe it is call out time like Woozy Rogers says here on the live stream. So thanks to our live crew for joining us. And as a reminder, if you are joining us on the podcast, check out your show notes. If you want to come join the live stream, so many great AI enthusiasts, leaders who are

implementing generative AI at their companies to connect with here. All right. So let's get it started. Let's get it started, y'all. Ready? So people have chat GPT all wrong. A lot of people talk about chat GPT, like it's just some copywriting tool. That is literally not what it is. That is the furthest thing from the truth. I like to call chat GPD a business operating system, okay, which is a little different. I think that right now, chat GPD is the only business operating system. I think Google Gemini is close to being there, but they're having problems connecting workspace data with enterprise accounts and with paid accounts. So I don't think, you know, Google Gemini is there. Oh, no, I'm wrong. I will say Microsoft co-pilot is there. Sometimes

I think of co-pilot and chat GBT as one and of the same, right? So I think you have Microsoft copilot that is a business operating system. I think you have chat GBT. I think that Google Gemini is close-ish. And I think that Claude, once Claude kind of enables third-party tools in their chat interface, right, which we talked about on the show that they enabled it for developers recently. But once they enable it on their chat interface, we'll get there. But today I'm just talking about chat GPT because I think that is probably the most, you know, the most capable and the most popular large language model by far. So that's what we're talking about today. All right. So as a reminder, our team has been using the GPT technology since 2020. Right. So yeah,

especially once chat GPT came out in November 2022. And as it's become more popularized, we're just seeing more and more bad takes from people. So here's the first one. Here's the first one. It is call out time. It is call out time. So people who share a chat GPT response and they say, hey, chat GPT is no good. Look at this response.

And then they share a screenshot. That is not how chat GPT works. If you're sharing a low quality output, sorry, not sorry. That just probably means you don't know what you're doing. Let me repeat that again. if anyone shares actually for better or worse, right? Sharing a screenshot from chat GPT means absolutely nothing, right? Because in the instructions before that screenshot, you can tell chat GPT everything it

needs. You can spend the time and get great outputs or you can say nothing, right? And if you just copy and paste something in, you're going to get a bad output. All right. So let me give you an actual example. And this is, I think, I'm going to spend the most time on this one. because people sharing their outputs from chat GPT is terrible. It is a terrible way to share about a generative tool, right?

ChatGPT and large language models are not deterministic. They are generative. So what that means, it's the exact opposite of something like Google, right? If you do a Google search, aside from if it pulls up its new AI search, It's search generative experience. It's SGE, but for the most part, a search online is somewhat deterministic. No matter who puts in the input, everyone, at least at that time and date, is going to get a similar or the exact same output. That is not how generative AI works.

It is generative. You're going to get wildly different outputs, even if you put in the same or similar inputs. And so much of it happens with what happens before you actually prompt a large language model. All right. So here's an example, right? I saw a post, you know, someone posted this on social media. I'm definitely not, you know, naming names, but, you know, I just saw this recently and I thought this could be a good example. All right.

So the example here that we're going to be talking about if you're joining us on the podcast is someone was using chat GPT essentially as a advertising slogan assistant for a Nike ad. So on my screen here, there is an actual end that I'm going to show you. But I'm wondering which tagline or motto is the best. And this is a photo of Caitlin Clark, the record-breaking women's basketball player from Iowa who broke the NCAA scoring record in a very triumphant pose. Right. So she has her hands up in the air kind of looking to the side. Very great photo. So what for our live audience, let me know which one of these is the best taglines, ready?

Or you can play along if you're if you're in the car, wherever you are listening on the podcast. All right, so A, which is the best tagline to go with this Nike ad for Caitlin Clark. All right. A, breaking records and making hoops history. B, you break it, you own it. C. Legends don't chase records. They set them.

All right. Live audience, let me know. All it takes is one little pound of the keyboard. A, B, or C, let's try it one more time. Ready? A, breaking records and. making hoops history. B, you break it, you own it.

C, legends don't chase records. They set them. All right. And I'm going to give you the answer here in like two or three minutes, but one of these was written with a copy and paste prompt and shared online saying, look, how bad chat GPT is, right? That's why I'm going through this process and showing you how chat GPT actually works.

So one of these was done with a. copy and paste zero shot prompt inside chat GPD. One was written presumably by a highly skilled human copywriter that either worked internally at Nike or that Nike hired. And one was written by chat GPT the correct way. All right. So it looks like for the most part, unofficially here, C is is crushing. C is by far ahead of everything else.

Right. So let's keep going here. So this was the, well, I just gave you the answer of the copy and paste version of chat GPT was actually A. So, you know, this person on social media just said, write a headline for an ad featuring Iowa basketball player, Caitlin Kark. She broke the record for most points ever scored by an NCAA woman in basketball. Be clever. And then chat GPT actually gave us that number A there, or option A, which was breaking records and making hoops history. All right.

So you can keep voting. You can keep voting. So we now know at least that what one of them was. But let's talk about how a prompt like that, right? And again, I'm not just calling this person out. It was just a fun and timely example. A prompt like that is always going to give you bad results. Okay.

And so many people, even very smart people online are just sharing copy, like sharing copy and paste prompts. They're sharing screenshots of, hey, look at Chad GPT. It's no good. Well, if you don't use something the right way, of course it's not going to be good. If I open Google search and just pound my head on the keyboard and hit enter, I'm not going to get good results. Okay? That is the equivalent, y'all. Large language models are not one good input, one good output.

That is not how they work. A 1.8 trillion parameter large language model, which is what GPT4, which is chat GPT's latest first, that's what it is. If you think you can just go in there, open a new chat, and type a short prompt and get something great, you're wrong. That is not how generative AI. That is not how large language models work. You have to teach them.

You have to coach them. You have to give them examples, right? We're going to talk about that a little more. But zero shot prompting. A zero shot prompt is essentially when you just happy and pay something in there without examples, without going back and forth. Okay. So we have, and hey, has anyone out there taking our PPP course, our free prime prompt polish?

Let me know if you have, you know, maybe drop a line in there. But we actually did a brand new version. We blew the thing up. We've actually taught more than 3,000 business leaders from huge companies and small startups how to properly prompt. Prompt Engineering 101. And we actually redid the whole thing. So we have a new process as part of priming called Refine Q.

So yeah, if you want to access, just let me know. I'll send you. We're doing another one here in about four hours, right? So you can even join us live then. But zero shot prompting like that is always, almost always going to give you bad results. So everyone, please, can you stop sharing screenshots of your results? Okay. For better or worse, that doesn't help anyone.

If I'm being honest, one of the reasons I started. everyday AI was, yes, our team's been using the GPD technology for three and a half years, and we saw that mostly everyone's using it incorrectly. But it's also very dangerous. It's very dangerous for your company if you're listening to someone who has no clue what they're talking about. And you don't know. That's why I'm telling you today, no, that is not how chat GPD works. All right. So here is how, here is the correct way. All right. And if you're listening to on the podcast, I'm sharing something with our live streamed audience. live stream audience. I did this last night. This is prime prompt polish. So this is our

recommended way to prompt engineer any large language model, including our new Refined Q method of priming. Okay. This is an example. This is a lot. All right. So this is 6,000 tokens or about 4,500 words to build a Nike focused sluble. Logan writer. All right. And hey, before anyone, I don't want to hear any like nonsense that are like, oh, Jordan, like you wouldn't know anything. Hey, guess what? I used to work very closely with Nike and Jordan Brand. I worked with professional athletes all the time. I worked with Nike's global head of marketing. I helped ghost write tweets for professional athletes. I know Nike, right? And I know what it takes to build copywriting, right? I've been in, you know, Martec and comms for 20 years.

So this is the process that you should be going through, right? If we want to share screenshots, let's share this. Because essentially, I'm showing my entire process. I did this last night. I took this. I built an expert chat, which is what we teach people to do through prime, prompt polish, PPP, an expert chat on writing Nike taglines, right? And one of those three came from the result of this. 4,500 words of back and forth iterative conversation with a large language model. That is how you use a large language model. You don't just go in there, copy and pay something, or go in there and say, hey, here's

some information about Caitlin Clark, write me something, you know, from that. That means nothing. That means nothing. People don't understand. You have to refine a large language model if you want it to perform well for a specific purpose. That's why I'm sharing. Hey, if we want to start sharing screenshots, let's start sharing this. Let's start sharing this, the process, the hard work, using it correctly, right?

And yeah, if you're on the podcast, essentially, even our live stream audience can't really see anything because I zoomed out all the way, right? And this looks like a like a 30 page book, but zoomed out because that's how you do it. That is how you use a large language model correctly and get the maximum output. live stream audience. Are you feeling this? Right? Hey, Cecilia said she loved PPP, so she needs the updated version.

It's kind of like what, yes, Ben, it's kind of like what Ben said. And Ben, I'm getting out old man Wilson here for you. You know, garbage in, garbage out. Yes, that is the same thing. If you put in, even if you put in a copy and paste prompt, you find something great on the internet. And you find a super long prompt. That's garbage in. You're going to get garbage out.

This is how you do it. This is how you properly prompt a large language model, specifically if you are looking for a tailored output. You got to put the work in, y'all. 6,000 tokens, 4,500 words of back and forth conversation with a 1.8 trillion parameter large language model will finally get you to something good, a good output. Right? And I promise, I promise, you know, by far, right.

I'm going to go ahead and do a quick, a quick count here. So it looks like we have about eight, nine, ten, eleven, twelve, thirteen, 14, 15. It looks like we had about 15 people vote for C. It looks like we had about two, one, no, yeah, one person vote for A and one or two people vote for B. All right. So guess what?

The copy and paste prompt finished last. the one written by a human, which was you break it, you own it. That finished second. And by far, the one that finished first with about three to four X higher than the human copywriter was the one that I did through chat GPT. Again, I think I had an unfair advantage, right? I've worked with Nike, so I understand I've created slogans with Nike. Mikey, you know, so maybe I had a little bit of an unfair advantage. But that just goes to show you a

couple of things. Number one, zero shotting a large language model or just giving it no information, clicking new chat and saying, hey, you know, Caitlin Clark broke this record, write a cool tagline for, you know, a Nike ad. If you do that, that's a zero shot prompt. You're not working with a large language model. You're going to get garbage. Garbage in, garbage out. The human copyrighter, I'm actually surprised because I kind of like the human one. You break it, you own it, right? She broke the NCAA scoring record. But, hey, at least our live stream audience said, by far the best one was the one that properly went through a large language model, an iterative process back and forth teaching and training a large language model.

Now, guess what? If I wanted to go in there, I could spit out new Nike motto, Nike slogan after one another, one another, once it's trained because now it's repeatable and I can use it. That is learning and using generative AI the right way. Is anyone surprised that the chat GPT version outperformed the human copywriter by that much? I am personally. I thought, if I'm being honest, I thought it was going to be head, you know, kind of neck and neck between the Nike, the human written ad copy. You break it. You own it versus the kind of the prime prompt polished version, which is legends don't chase records.

They set them. I thought it was going to be neck and neck, but apparently not. All right, let's get this. Let's keep this thing going. The second thing, if you ever hear this, if you ever see this, avoid, avoid this person. Avoid, avoid at all cost. All right. Anyone that says chat GPT is never out of date because of Bing, right?

That's not how chat GPT works, y'all. All right. So if you don't know, now if you are using the paid version of chat GPT Plus, which is GPT4, GPD 4 Turbo, you get essentially live access to Bing, right? And so people think, oh, I'm not going to get hallucinations because if I type something into chat GPT, right, it's training data ends at April 2023, which is now a year old. So people think, hey, if something's out of date, that's okay. I'm not going to get lies or ambiguity out of chat GPT because it's going to use Browse with Bing. it's kind of true, but also very false.

All right, let's talk about how this actually works. Here's the thing. Browse with Bing can actually read URLs. I see so many people sharing their prompts with with screenshots. If you upload a URL into chat GPT and say, hey, Bing, go look up this information, right, and help me write this better, right? All Bing does is it looks at the words in the URL and it queries them. So you can't actually point Bing to a specific web page.

Sometimes it'll work and sometimes it'll go to the correct page. I've done this live on the show before. I've done five minute tutorials on this and YouTube, y'all. You have to pay attention. You have to know how a large language model works. Because Bing can actually, you can't direct Bing to a specific URL. It'll work sometimes. But it's just an SEO.

It's an SEO game. And unfortunately, SEO is finicky. You can game it. Guess what? You know, if you want up-to-date information, let's say you, you know, use Bing and you say, hey, Bing, I want, you know, a marketing plan for everyday AI, you know, use Brows with Bing to find the most advanced tactics for multi-channel marketing in 2024. So what it's going to do is it's going to Bing. It's going to query multi-channel marketing tactics 2024. Okay. So are we following? That's what it's going to do. guess what SEOs are great at. People who do SEO, they're great at gaming the system because there's a good chance,

and I've done SEO y'all for 15, 20 years. Someone probably had an article that from 2020 or 2021, and all they do is they update. They update the title. They update the URL, but everything else remains the same. And they game the system. So it might say top multi, multi channel marketing strategies for 2024. Guess what? it might be from 2020 and maybe someone's game the system.

And Bing and Google don't know any better. They don't know any better. All right. So you can't rely on Bing in that facet as well. Also, in this instance, right, browse with Bing may bring back outdated info from before the training data even, right? So people think, oh, if it's after April 20203, I need to use Bing. You can't trust it.

You can't trust a blind query because you. You don't know what's on the other end. Oh, Mike felt personally called out by that one. But, hey, Mike's an SEO specialist. He knows that's how that works, right? It's not like an SEO, like it's not like a dirty secret, but you can't necessarily trust a single source from a blind browse with Bing call. All right.

So if someone says, oh, use Browse with Bing and Chad Chip, and you're fine. You'll never have out of data information. It'll always be accurate. Not true. Not the case. All right. next. I'm going to try to not spend too long on this one. Yeah, it's getting, it's getting hot in here, Mike. It's getting hot in here. All right. So if someone says, try these 20 prompts to 10x year workflow or,

hey, here's, you know, seven prompts that'll instantly double your business, right? Or here's 30 prompts for, you know, marketing your startup, right? That's not how chat GPT works. We kind of already touching this with the Caitlin Clark example, but if you just start a new chat and you find some, you know, 19 year old who lives in the basement with his mom, who's, who's now an AI influencer, right? And they say use these 25, right? I got to go back. How many, how many, how many, how many flames did we get? We got a lot of three. Okay, we got a two. All right. So it looks like people kind of want the heat. All right. I'm going to have to call this out. I agree with Brian. This is so annoying. Those prompts don't work.

People sharing those prompts. Share them because they know they can trick you, right? They want your email and then they're going to sell you some crap in their newsletter. All right. And they say, hey, you know, these 50 prompts, we spent so long making this prompt playbook to grow your business. Those prompts don't work. They are garbage, garbage in, garbage out. If you go into chat, GPT, and this is literally what we teach and we show step by step in our free prime prompt,

polish course, all right, in our PPP course. We take the highest rated, you know, prompt for a certain topic. We break it down, we dissect it, and we talk about everything that goes wrong, and then we teach you for free how to do it the right way. Y'all, copy and paste prompts is not how chat GPP works. Chat GPD is not deterministic. So if anyone's telling you, yes, these prompts will work for your business. That means they don't know what they're talking about.

They have no clue. Let me give you a little hint here, right? Let me give you a hint. Adobe just introduced an entirely new way to create, bringing the power and precision of its creative suite into one conversational experience. Meet Firefly AI Assistant, now live in the Adobe Firefly app, the All-in-One Creative AI Studio. Powered by Adobe's Creative Agent, Firefly AI Assistant lets you start with your vision, just describe

what you want, and shape the outcome as it takes form with the Assistant. The Assistant orchestrates multi-step workflows, drawing on 60-plus, pro, and you know, and upgrade tools across Adobe Creative Cloud apps, including Photoshop, Illustrator, Premiere, Lightroom Express, and more to help bring your ideas to life. You can also get started with creative skills, a growing library of pre-built workflows for common creative tasks, like batch editing photos, creating mood boards, portrait retouching, and creating social variations. Every step the assistant takes is visible so you can refine, redirect, or take over at any time. You stay in the driver's seat as the creative director.

Adobe Firefly AI assistant now in public beta. See it today at firefly.adobie.com. If someone reaches out to you and they're like, oh, yes, I'm a prompting expert, ask them questions. You're like, what are your thoughts on, you know, 32 shot chain of thought prompting versus zero shot? Hey, explain to me the difference between a few shot and chain of thought. right like say explain your your prompt engineering methodology ask them right maybe maybe your

company's hiring a you know an outside experts to come in and help your team with prompting ask them hey what are your thoughts on MMLU benchmarking okay y'all I spent my weekend reading research papers about new prompting methodologies that's what I do You can't trust people. They're just tried to game you. They're trying to make you share their crap. They're trying to make you sign up for their newsletter to sell you crap. It is crap.

Sorry. Sorry for my PG-13 language, right? Copy-upase prompts don't work. That's not how generative AI works. That's not how a trillion parameter large language model works. It is not deterministic. Those prompts aren't going to get you anywhere. If you think prompts are going to save your business,

if you think they're going to grow your startup, if you think they're going to help your company simple copy and paste prompts. They're not. That is not how chat GPT works. All right. Let's keep this one going. We got one, I think we got one more here, one or two more.

All right. So another one, paste in your writing style. And then chat GPT will write just like you. That's what? That's not how chat. So here's the thing, y'all. So many people think that this is how chat chatt chatt chpt works, right? Hey, give it some examples of your writing style, of your brand voice, of your company voice.

You know, just paste it in there, give it an example. Then click enter and all of a sudden it's going to write like you. It's going to write like your company, right? You're not going to need that copywriter. False, false, false. Okay? What most people don't know who aren't dorks like me and read, you know, hundreds of pages of research papers on advanced prompting methodologies and scientific research papers on prompt engineering, most people don't understand. Large language models don't understand words.

They don't. They tokenize, right? They assign values, right? So I'm just showing an example for our live stream audience here. You know, I'm just going to read two sentences here. So it's talking about the word just and how the word just has different meanings. Right. So as an example, it'll say, it'll be there just as soon as I finish this task.

And then another usage of the word just. It's important to do what is just and fair in all situations, right? So essentially, the word just can mean a lot of different things. And this is a screenshot from OpenAI tokenizer, which I encourage everyone. If you want to learn how large language models work, go use Open AI's tokenizer. It's free in their playground, right, if you have an account. But what we're saying here is the word just. Open AI has assigned just in this, I didn't mean to do that, just in this use case,

just in this small example, four different meanings or four different contexts, you know, meetings to the word just because the word just can mean many different things, right? So if you think you can just copy and paste a bunch of examples of your writing into chat, GBT and get a brand voice or get this. Y'all, can anyone, and I'm sorry if this hits home and I'm sorry if this is you and I'm sorry if I'm calling you out, that's what we're here for. We're here to give people the truth. I'm tired of BS, y'all.

It's important that we all understand how generative AI works. I've been saying this for a long time. 24, things are going to get bad. Things are going to get bad. AI layoffs, things are going to get bad. I want you to succeed. You have to understand that's not how chat, GPT works. You can't just copy and paste a bunch of stuff and say,

They do it like this because of the tokenization process. Chatypc doesn't actually understand words. So if you can't give it enough examples to understand the word just as an example, it's not going to work. You're still going to get nonsense out. But why? Why are you still reposting all this nonsense, y'all? Why are you still supporting people that share bad information?

Stop doing it. Learn to do it the right way. Here's an example. I actually gave a talk on this. at one of the largest AI summits in New York City. This is what you have to do if you want to get a large language model to write like you, right? Hey, our podcast audience, you can't see this. But again, we have a lot of screenshots and very tiny text, right?

You have to turn unstructured data, which is content writing, right, into structure data. You have to create new rules for chat GPT so it can work within their rules, right? Again, we teach companies this. If you want to do it the right way, it's possible. but you don't just copy and paste a bunch of examples. You literally have to turn the art of copywriting into a science of rules for chat, GBT, to follow. Right?

We call it our pattern recognition framework. It's not done the other way. If anyone's sharing that, they don't know what they're talking about. Sorry, not sorry. All right. And then last, last but not least, last but not least. If someone says this, run. Do not, do not follow their advice.

Do not listen to them. If someone says, the free version of chat GPT is good enough. Nope, that's not how chat GPD works, y'all. The best analogy I can give, and I give this one all the time, the free version of chat GPD is like a landline telephone. Chat GPT Plus with GPT4 is like the newest smartphone. It's a huge difference. There's very little similarity between a landline and a smartphone.

With a smartphone, you can run your entire business. You can become productive in your personal and professional career. There's very few things you can't do with a smartphone. Well, it's kind of hard for me. I got fat fingers, y'all. But you probably get what I'm saying. The paid version of chat GPT Plus because of its outside connections that you don't get on a free plan, you don't get that in the free version.

GPTs, poor plugins, they're gone now, right? But GPTs, browse with Bing, code interpreter, dolly, right? All these things that turn chat GPD from a copywriting assistant to a business operating system, right? Being able to work with Zapier or bringing, you know, being able to build your own GPTs and kind of mini-rag, right, with your data, being able to upload your data and build a custom GPT, right? All of these things you can't do in the free version or people that say, oh, well, you know, I just get the GPT4 via Microsoft co-pilot. Okay, that's better than using the free version, but you don't have all those outside tools and being able to bring in your outside workflow. So if someone is telling you, oh, hey, your company, right, if you're hiring someone, if you're following someone on social media, if they say, hey, the free

version is good enough. Hey, here's some free alternatives to chat GPT. They're selling you snake oil. They don't know what they're talking about. That's like if someone were to really come up to you and say, hey, your company, I see your company is using computers. Have you thought about this much more affordable version? It's called paper, a pen, a landline. You should try it. Now, shit, here's my guide on how to save money in your business by using paper, pen, and a landline. Repost this and I'll send you the guy, right? It's not how chat GPT works. The free version, sorry, garbage, not good.

You shouldn't use it. Paid version can change your business. You really want to grow. It can change your career. You can literally run an entire small business inside of chat GPT if you know what you're doing. If you do it, if you do it the correct way, if you understand prompt engineer, If you understand how to safely bring your data into chat, GPD, working with automation, I mean, having Zapier in there, being able to connect your own data.

It is a business operating system. The free version is not worth your time. Was that too hot, y'all? Oh, sometimes I get done and I'm like, did I just offend someone? Right. Hey, did I just offend? Brian, Brian, Brian said this one, the last one is laughable, right? hopefully I didn't offend you, right?

But there's so much bad information out there, y'all. I had to set the record straight. Because people who have no clue what they're talking about are trying to tell you how to run your business. They're trying to tell you how to use chat GPT to grow your career. Stop listening to them. They're selling you snake oil. I'd rather you figure it out now versus a year from now when it might be too late.

Your company might already get passed up, all right? If you're trying to implement generative AI, large language, models into your company, which you should. If you haven't already, you've got an expiration date. Sorry. You have to understand large language models. You have to do the work. Yes, generative AI can be a shortcut, but you've got to build it first.

You've got to build the bridge, and that's hard work. We're lazy as humans sometimes. We want those copy and pay shortcuts. We want the crap guide from the 19-year-old kid, the 19-year-old Billy Boy in his basement. We just want it to work. It's not going to work. You got to put the work in, but then generative AI will work for you. All right.

Let's do this. Huh? Let's do this. Let's see if we can properly, let's see if we can properly give away the GPU, right? So I should have had the GPU ready to show me. It's literally behind me here in my home office. So if you entered into our contest, all you had to do, it's over now. All you had to do was sign up.

You got free access to the Nvidia GTC conference. And then you were entered. So we had, I don't know, like 150 people. Let's see if we can do this. So I entered every single person's name onto this list here. Don't worry. I'm not giving away your personal information. I kind of redacted everything.

So we only have first name and last initial. All right. Hopefully y'all can see this here. Let's see if we can give away this GPU. It's random. That's why I'm doing this. Hopefully it works. If not, I got a backup one.

All right, ready. Click to spin. Here we go. All right, we have, right, Justin L. All right, I'm going to be sending you some information, Justin, about this GPU. And also, keep in mind. Oh, wait, there it is.

Look at this. It appeared out of nowhere. Look at this. We're going to be sending you this, this invidia, GeoForce, RTFs, right? So you can also run chat with RTX once you install that. All right. Also, we're going to be having, we'll have more giveaways. We'll be sending those to you.

We also have some DLI credits to give away. Those will take a little longer because there's like 10 or 11. So make sure to check your email for that. All right. I hope this was helpful, y'all. So make sure if you haven't already, repost this. If this show is helpful, please consider giving us a rating on Spotify, Apple, wherever you're listening to your podcast. Please share this with a friend, right?

Because here's what? Here's what you may not know. Someone in your life, a friend, a brother, a coworker, your sister's neighbors, babysitters, dog walkers, best friends, boyfriend. Someone out there is getting bad advice when it comes to using generative AI. When it comes to leveraging large language models to grow your company, grow your career, share this with them, tag them in the post, repost this on LinkedIn. that would help. It would also help you to go to your everyday AI.com. Sign up for that free daily newsletter. We're going to be breaking today's episode down, going over more AI news and a lot more.

Thank you all for tuning in. 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. Stand 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.