Everyday AI Podcast – An AI and ChatGPT Podcast - EP 194: 5 ChatGPT Facts You Might Not Know

Episode Date: January 26, 2024

There's a lot of things that you have wrong about ChatGPT. Whether it's the way you use it or what you understand ChatGPT to be, there's some things you should know to properly leverage it. We're giving you 5 ChatGPT facts you might not know.Newsletter: Sign up for our free daily newsletterMore on this Episode: Episode pageJoin the discussion: Ask Jordan questions on ChatGPTUpcoming Episodes: Check out the upcoming Everyday AI Livestream lineupWebsite: YourEverydayAI.comEmail The Show: [e-mail] with Jordan on LinkedInTimestamps:01:50 Daily AI news06:57 Everyday AI approach to research12:33 Improving written communication: key to better outputs.16:28 Using ChatGPT can lead to hallucinations.17:28 ChatGPT errors and free course promotion.22:27 Browsing with Bing relies on SEO analysis.24:31 ChatGPT searches US companies by market cap.28:51 OpenAI facing lawsuits over copyright infringement. Avoid hallucinations by using plugins and methods.32:53 OpenAI facing backlash over ChatGPT plug-ins.35:14 OpenAI's developer focus may affect market share.38:11 Default Bing browsing causing GPT issues, misunderstandings.43:12 Use GPT-4 offline, avoid bad information.44:48 Understanding AI models crucial to prevent hallucinations.Topics Covered in This Episode:1. Overview of ChatGPT's Functioning2. Common Misconceptions on ChatGPT3. Structural Changes in ChatGPT4. Ways to Improve Outputs in ChatGPTKeywords:ChatGPT, Generative AI, AI news, Intel's stock, OpenAI, GPT 4 turbo model, Tokenization, Large language models, Communication skills, Memory management, Prime Prompt Polish, Daily AI newsletter, Copyright infringement, AI plug-ins, Microsoft CoPilot Pro, GPT 5, Misinformation, 0 shot versus 5 shot prompts, Bing default browser, Knowledge cutoff, Paid version access, Plug-ins mode, US companies $3 trillion market cap, WebReader plug-in, Inaccurate AI responses, AI hallucinations, AI in daily life, GPT builder.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. There's a lot of things that you have wrong about chat GPD.

Whether you're just looking at the headlines or the marketing or finding, you know, following a Billy boy chat GPT influencer who's 19 and lives with his mom. There's so many things you have wrong about chat GPT. So today I'm going to tell you five facts. that you might not know about everyone's favorite large language model. All right, we're going to talk about that today and a lot more on Everyday AI. What's going on, y'all? If you're new here, my name's Jordan, and I'm the host of Everyday AI.

In this live stream, this podcast, this free daily newsletter, it's for you. It's to help us all learn what's going on in the world of generative AI and how we can all actually leverage it. All right. So you've got to pay specific attention to that piece today because I'm going to tell you you how to actually properly leverage chat gbt to grow your company and to grow your career so if you're joining us on the podcast thank you as always make sure to check out your show notes make sure to follow subscribe to the show leave us a rating but also there's contact information in there we actually have some

hidden surprises in the in the show notes there so make sure to check those out you can always even reach out to me put my my LinkedIn and my email in there as well and if you're joining us on the live stream thank you as well. So, you know, Tara joining us from Nashville and Harvey, Maricio and Brian, everyone, thank you for joining us. But before we get into facts that you might not know about chat GPT, let me first remind you, sign up for the free daily newsletter at your everyday AI.com. Also, while you're there on our website, whether you know it or not, there's something on there called AI tracks. Go click on that because this is what I still think. is the one single best source of free information on generative AI that there is on the internet,

period. We have now almost 200 episodes interviewing experts from the companies making this technology to startup entrepreneurs and everything. So you can go listen to, you know, if you're interested in AI and education, we've got episodes on that. If you're interested in AI and entrepreneurship, we've got dozens of episodes in that. So make sure to go check that out after you subscribe to our free daily newsletter. All right. So before we get into those five facts you might not know about chat.

GBT, let's first go over the AI news as we do every day. All right. So let's start it off. Intel stock is dropping as investors fear they're not invested enough into AI. Intel stock has taken a hit after week guidance for the current quarter leading analysts to question the company's ability to compete in the AI and chip market. Analysts believe that it could take time for Intel to regain its position in the AI chip market. And they will need to compete, according to investors, to prove themselves in the AI market and compete with companies like Nvidia, who are literally just running away in the GPU market. All right.

So next is ChatGPT's laziness gone with a new update from OpenAI? So OpenAI updated its GPT4 turbo model to reduce causes of laziness and improve code generation tasks. The company also launched smaller AI models called embeddings to help applications with retrieval augmented generation. Rack. So OpenAI has updated the GPT4 Turbo model to address laziness that many users have reported, specifically when it comes to generating code. A lot of times it would kind of just give you a couple of responses or say, all right, well, for the rest, you're going to have to finish it yourself. So the company made this update. They've acknowledged that there's been this problem, but they've also launched these smaller AI models called embeddings to help applications

with retrieval augmented generation or to be able to bring in outside data while you're working with the GPT model. Speaking of open AI, so the federal government is investigating big tech relationships around AI investments. So the federal trade commission has announced an inquiry into the AI industry and its major players, including everyone, Amazon, Alphabet. Microsoft, Anthropic, Open AI, all the big AI companies right now. So this market inquiry will be conducted under Section 6B of the FTC Act and aims to investigate

potential antitrust violations or deceptive practices in the field of artificial intelligence. So FTC chair, Lena Kahn, has emphasized that there is no AI exemption from existing laws, and the agency will closely examine any instances of companies using their power to hinder competition or mislead the public. Interesting. All right. We got some hot water now with the federal government and artificial intelligence. So if you want to keep up with those stories and more, please make sure to go to your everyday AI.com. Sign it for the free daily newsletter. We'll be recapping those stories and a lot more. If you haven't listened to or read our newsletter, you can actually listen to it. We have an AI Jordan read it sometimes.

But if you haven't read our newsletter, I'll tell you this. It's written by a real human me, former journalist. And we break down the fluff, the lies, the marketing, and we actually just give it to you real. All right. So if you want to actually learn generative AI, I'd say there's no better place than our newsletter. All right, speaking of that, let's actually learn some facts about chat GPT and talk about five facts that you might not know about what I'd say is everyone's favorite large language model. I mean, if you're talking active users, it's technically the favorite. But I also want to know from our live stream audience joining us, right? Cecilia, joining us from

Chicago, Jay and Ted and Judy. Thank you all. And Mike Forgey and Woogie and Laura. What questions do you all have? Or maybe what's one thing that you may be having your mind that you're like, okay, is this a chat GPT fact or is this fiction? Go ahead. Get your, get, your comments in, get your questions in. Same thing if you're listening on the podcast, hit me up. I'll give it to you straight. So before I dive into these five facts, I first want to give you very, very, very brief background about myself and everyday AI. All right. I'd say a lot of what comes into play now at everyday AI is given my background, right? I was an award-winning investigative journalist for many years, right? And one of the things about being a journalist is you never

believe what anyone tells you. Anyone, all right? You know, there's the saying that says, if your mom says she loves you, get it in writing. That's our approach when it comes to generative AI because we have technically more than 100,000 people trusting us each and every month to separate the truth from the lies, to separate marketing from what is actually happening. All right. So if you're a vet around here, you probably know that. You know, sometimes I'll spend five, six, 12 hours doing research just for a single show. All right?

Because I want to make sure if you're listening to this, if you're watching this, if you're reading about this, you know these are facts. I'm not just copying and pasting what, you know, big company says in their press release or, you know, big companies always hitting me up trying to get their people on the show. I vet everything. All right. I want you to know that. So when I'm talking about these facts, you may not believe some of them because they

might go against what a big company says. But as always, I bring the receipts, right? You'll see that if you're watching live, if you're on the podcast. I've got some screenshots. I'll do my best to describe them, but I want you to know that, all right? I'm not one of those Billy boys. I mentioned Billy Boy, right?

That's those, you know, those 19-year-old kids that are, you know, unfortunately teaching the rest of the internet how to use chat GPT. And they say, oh, you know, do this, do this. No, they have no clue what they're talking about, right? I have a background in this space, right? Our team has been using the GBT technology since 2020.

So we're not new when it comes to GBT. You know, chat GPT has been out since November 2020. Our team had already been using the GPT technology for multiple years at that time. All right. I just want to get that out there before we dive into these five facts. So let's get going here. Let's go over the five facts. And again, thank you for everyone for joining live.

Get your questions in. I'll try to answer them at the end. Or as I go, we'll see how it happens. All right. So number one, fact you might not know. about chat GPT is, well, chat GPD doesn't technically understand words. You might be saying, what, Jordan? I've seen it.

You know, I can talk to it. I can chat with it. I put in a query and it gives me something back, right? And you've probably seen some great outputs with chat GPD. So you're like, all right, Jordan, you clearly don't know what you're talking about. Chat GPT clearly understands words. Not really. All right.

So chat GPT thinks and analyzes and responds via tokenization. Actually did an entire like 40 minute episode on on that process. So if you're really interested, we'll make sure to leave that in the show notes. But essentially, it's like this. And this is how most large language models work. It's through tokenization. So a token can be, you know, characters. It can be part of a word.

It can be an entire word. But essentially large language models, when you type something in, it breaks that down into tokens, right? And it assigns tokens to all of the different words. So actually, you know, even went over, you know, Open AI has a tokenizer. You can go and play and type in words and see how it actually thinks, how it breaks it down, right? It's not like, oh, each word has the same token. You know, the example that I've showed is, you know, in a, I use the word just four times. And depending on the context of the word, it actually signed four different tokens to the word just, right? Because the word just can be used in many different ways.

So that's how large language models actually work is they use context in all of these different, kind of relationships between words and context of your inputs and it assigns tokens. All right. So doesn't actually technically know what words are. Tokens are essentially building blocks of language, all right? And like I said, most large language models work this way. Because again, it's important to know all chat GPT is or all Google Bard is or Anthropic Claude, Matt is Lama. All they are are the world's most advanced auto-complete.

but they auto-complete and create and generate and analyze all via tokens, right? Which is why, you know, I keep telling people this and people don't understand. One of the best skills that you can have in 2024 and moving forward as the world, as the business world starts to integrate its daily processes around large language models, right, Microsoft co-pilot, you know, you have your enterprise, your Amazon Q, you know, your IBM Watson, we're going to be using in the business world more and more prompting language, right? Which is why it's important to understand how to prompt and understand really tokenization

because words matter. One of the best skills you can have is written communication, right? If you are a great communicator, if you are great with words, you are going to get better outputs inside of chat chpte in a large language model. You don't need better prompts. You don't need to become an expert prompt engineer. No, you need to improve your written, your verbal communication. We do free courses all the time and someone had a question like, what book would you recommend to become a better prompter, right, to get better outputs in chat GPT?

And I said, the book on Writing Well by William Zinzer, you know, a book that's been around for decades, that if you want to become better at prompting, read that book. That helps your words mean more. That's one of the best skills you can know. But that's fact, number one. Chat GBTGBT doesn't understand words. All right. Number two, speaking of tokens, right?

chat gbt does not have a 128k token memory all right this one's a little dorky but i'm going to try my best to explain it to you and that's why we talked about kind of tokens in number one and number two so now that you know that chat gpte both analyzes and responds essentially in tokens those tokens add up okay and at dev day so open a i's dev day so open a i's dev day in November, they announced, hey, there's a new model, GPT4 Turbo. And GPT4 Turbo has 128K memory context window. And ChatGPT has, and is powered by GPT4 Turbo, right? So you saw all the major publications, you know, here, I have an example here, a Forbes article

that came out and it said, you know, chat GPT's new features. And it talks about chat GPT now supports 128. K tokens. No, it doesn't. It doesn't. That's marketing. It's marketing. So right now, it has 32K tokens. So that means when you are talking back and forth to chat GPT, all of the words you put in are converted to tokens. Everything chat GPT spits back to you is counted in tokens. And those tokens, once you get past $32,000, it is going to forget what's at the top. So let's just say as an example, once you get to 35K tokens of back and forth conversation, chat, JBT has already forgot was what was in the first 3,000.

All right? Let me do an explanation. And again, this is why don't blindly listen to the Billy Boys on the internet or to anyone, even the big companies saying, oh, 128,000 tokens. No, it doesn't. Here's an example. All right. 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-grade 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. Here's an example. So I have a token counter. You should always use a token counter. I have a little video on our YouTube channel that shows you how to do that. But there's internet extensions that count these tokens as you go back and forth.

So as an example, I started a chat out by saying, my name is Jordan. I'm from Chicago. I like Carolina blue. My favorite food is deep dish pizza. Then I ask chat GPT, right? The very next thing, I ask chat GPT, what's my name? Where am I from? What color do I like?

What's my favorite food? And because I did this right away, chat GPT says, certainly, Jordan, here's a recap. Your name is Jordan. You're from Chicago. Your favorite color is Carolina blue. Your favorite food is deep dish pizza. All right. So what I then do in a brand new chat, I do the exact same thing.

And then I just start pasting in a lot of information, right? I'm actually just pasting in transcripts from our everyday AI episodes and asking chat GBT to tell me what this means just to get through these tokens pretty quickly, right? It took about eight minutes. But you'll see now, once I got to. to 34,000 tokens, I ask the same question. What's my name? Where am I from? What color do I like?

What's my favorite food? And then at that point, chat GPT replies, I don't have access to personal information, right? But if you do the exact same thing when you're at 31,000 tokens, it can still recall. So what does this mean? What does this mean? Well, one of the biggest problems

when people are using chat, GPT is hallucinations, right? Those are untruths or, you know, when chat GPD forgets things, or when it just lies and makes stuff up, or when it gives you inaccurate information. And there's really three main causes of this. And we go over all of these in our free course that we do a couple times a month called Prime Prompt Polish. So if you're listening, if you want access, just type in PPP. Hey, if you're listening, let me know. Did you learn anything in our free prime prompt polish course. It's been taken by over 2,000 business leaders,

even from big companies like Microsoft and Bata that are building the technology. But it's one of the most common mistakes that people make is not keeping memory in mind, right? And then what happens is they just say, you know, Billy Boy says, oh, chat GPT is not good because you're not using my prompts. No, Billy.

It's because you are not using. chat GPT correctly. You have to play by chat GPT's rules, all right. But the correct rules, not just what chat GPT says. Oh, 128,000. No, 32,000. Keep memory in mind and you are going to cut down on hallucinations. Hey, Woozy and Frank both said amazing course. What's up? Fabian. Fabian says this is the best AI program. All right. So let's keep going. Yeah, shout out. Shout out to all these people join it from all over the world. All right. Fact number five, you might not know about chat GPT. All right. So default browse with Bing does not actually allow you to browse websites. Yes. This is one thing I can't stand. I can't stand. There is such a misconception around this.

All right. But I know we have a lot of more advanced people, you know, at least here on the live because I recognize your names. You're all very smart people. But let me try to break this down. If you're newer to chat GPT, maybe on the podcast, you're just getting into it. All right. So chat GPT has a knowledge cutoff. It's actually different for different modes. All right. But with the paid version of chat GPT, 20 dollars a month, chat GPT plus gives you access to GPT4, plugins mode and GPTs. All right. When you're working with those paid models, the knowledge cutoff is April 2023. So essentially GPT4, just like most other large language models, is trained on the entirety of

the open internet, closed internet, and a lot more. So it essentially gobbles up all of this information, put its in its, put its, all of this information, trillions of parameters in its database. And that database is essentially static. It stops at April 2023. All right. But chat GBT, to its credit, did a lot to fix this because now in the new default mode, you have browse with Bing.

So what that means is if you're opening up a new chat in the default mode in chat GPT, if you ask it a question, sometimes even if you don't ask it to, it will browse with Bing, right? It'll do a quick query based on what you are asking. Okay. There's no rhyme or reason. Obviously, if you ask chat GPT to use browse with Bing, it will. But sometimes even if you don't ask it to browse with Bing, it still will, right? So if chat GPT, when it's converting your question into tokens, and if it thinks, oh, this is something that might need a quick browse with Bing, then it will do it.

Okay? Which is actually a little problematic. And I'm going to show you what I mean. But here's what people get wrong about this literally all the time. Even these quote unquotes, large language models, experts that are, you know, speaking in front of large conferences, they don't know what they're doing. All right. People are more concerned about making money than taking the time to understand things,

to test things. All right. You got to do that, y'all. Y'all are dropping the ball. All right. So let me tell you what this means. So Browse with Bing can't always visit specific websites. So if you put a URL into Browse with Bing, people think that that means.

I'm going to put in this URL. And now my chat within chat GPT knows and understands this information because I'm telling Bing to go read this URL. It's not how it works. Okay. It flashes very quickly across the screen. I'm going to show you an example of that, but all Browse with Bing actually does is it queries the information. Okay. It queries whatever is in the URL. So there's actually some semantics there. So it's, it's kind of reliant on SEO what's going to come up and what's not. If Browse with Bing is actually going to read and analyze that website, it depends. When I've done this, many videos on the YouTube channel, looking at how all large language

models do this and chat GPT with plugins is the only mode that consistently does this. But if you don't fully understand this, if you don't test this for yourself, if you aren't careful, you are going to allow these hallucinations or lies into your outputs. If you don't know what you're doing. So listen, I got it. Don't worry. So let's take a lot of. a look. We're going to take a look by showing another example in fact four. All right. So

hallucinations are actually rare if you use chat GPT correctly. Okay. These two facts correlate with each other. And now I'm going to show you that example of browse with Bing. Okay. And how it can actually lead to hallucinations. All right. If you're on the podcast, You might want to watch, you know, you might want to click the link. We always leave the link to watch this, but I'm going to do my best to walk you through these screenshots. Okay. So I am in the default mode, browse with Bing.

And I say what companies have a $3 trillion market cap, okay, in the default mode? So I didn't ask Bing, but it's smart enough to its own credit to browse with Bing. Bing, right? Because it's determined that, okay, this query might need up-to-date information past April 2023, right? That's like 10 months old. So chat GPT does know today's date. So it does know that my query here asking about what U.S. companies have a $3 trillion

market cap. It might know that it needs to browse with Bing. So I have a screenshot here, okay, that says it has the little browse with Bing icon and it says, you, it says, it. It's searching U.S. companies with $3 trillion market cap of 2024. Seems like a great query, right? So the problem is that flashes on the screen for like a second and then it goes away. And then you can't necessarily see exactly what chat GPT did behind the scenes or exactly what Browse with Bing did behind the scenes.

Okay. Now let me read you the response from the default mode from Browse with Bing. Ready? as of January 2024, Apple is the only U.S. company that has achieved a market capitalization of $3 trillion. All right. So if you don't know any better, if you don't know how the default mode works, if you don't know how chat GPT and large language models work, you're going to see this and there's also some citations there, right?

So you're going to look at this and think, oh, this is great. Save me some time, right? Think of how you might use this in business in your company to grow your career, whatever. There's this quick little flash of browse with Bing and you get what looks like great information that's sourced. So you're like, oh, great. You know, you feel confident. You feel secure. Guess what?

That's a hallucination. That is false. Okay. Brows with Bing because of how it's. it works, it doesn't work very well. If you want accuracy, you're not getting it here. This is false. All right. Now, I'm doing the exact same query, but by using plugins mode. All right. So here's what I'm doing differently. I have three plugins enabled. One of them is a plugin called web reader. So I'm saying using the web reader plugin. Okay? Plugins mode literally,

freaking fantastic. I tell people this all the time. It is a literal cheat code. It's not fair that plug-ins mode exists, and that's so few people know how to use it. All right, so I'm saying using the web reader plugin, please visit this article, okay? And then I give it a specific URL to go to.

And then I say, then please answer. And I get the exact same question verbatim word for word, what U.S. companies have a $3 trillion market cap. All right. So then it says, and it says using Web Reader, so I know, it says the U.S. companies that have reached a $3 trillion market cap are Microsoft and Apple. Wait, what? That can't be right. Because Browse with Bing just told me, as of January 2024, Apple is the only U.S. company that has achieved a market cap of $3 trillion. Huh. What's up with that, y'all? Who's lying? Brows with Bing is lying.

You do not have control over how Browse with Bing works. You do not have very great depth or information on what it actually is doing behind the scenes, right? Because all it did, I actually had to do this prompt multiple times to get this screenshot because it flashes so quickly and then you can't go back and look at exactly what happened behind the scenes. So again, browse with Bing, all it did is it searched US companies with $3 trillion market cap. Okay? Chat GPT plugins actually goes, if you know what plugin to use and how to use them, it goes

to an actual website, right? Which if you try that in Browse with Bing, again, all it does is query the words in the URL, right? Because Open AI has gotten into a little bit of some legal issues, right? They even had to take the Browse with Bing featured down for multiple months as it was facing lawsuits because it was determined or it was alleged that Browse with Bing was accessing copyrighted materials behind paywalls, which it was, right? That's one of the reasons why now Open AI is facing a multi-billion dollar lawsuit from the New York Times. And they're being, you know, most all large language models are being sued by publishing

companies all over the place, but that's one of the reasons why. So using plugins, you avoid hallucinations. All right. So again, if you know what you're doing, if you're using chat GPT correctly, hallucinations are actually rare. You know, there's some days when I'm in chat GPT for 80s. hours and get zero hallucinations. Why?

Well, I follow our prime prompt polish method. When you do that, very rarely get hallucinations, but also using the right mode at the right time for the right output. Sometimes I'll use browse with me, but I know the pros and the cons of it. But most of the times, I'm using chat GPT plugins because I have more control. whole, right? If you increase the quality of your inputs inside of chat GPD, you will increase the quality of your output.

Sometimes you have to take away decision making power from chat GPD, especially in the default mode because you can't really see under the hood. So you got to say, nope, I'm not leaving this up to you, browse with Bing, because you're going to Google or you're going to Bing, right, just like what I showed you. you're going to Bing, U.S. companies with $3 trillion market cap, 2024, and Bing thinks that's right. And even the response, you can look at the response as of January 2024. Well, that was true two weeks ago in January 2024, but guess what? It's still January 2024, and that is no longer true.

All right, last one. Here we go. And get your questions in now. I see a couple of them. I'm going to get to them. Hey, we got some mind blown. emojis and some people saying, wow, thanks, y'all. So, number five, chat GBT is two developer-focused, which will cost it billions.

That's billions with a beat. I've had a whole show on this. And let me, yeah, I had to take a deep breath here. So Open AI is obviously, I think, one of the most brilliant companies in the history of technology. You can't really make an argument against that, right? What other company ever has had 500 employees an evaluation of $100 million? Hardly any, right?

Hardly any. They make an amazing product. It is extremely capable. I think right now I've used hundreds of generative AI softwares. And it's not. not that chat gbt is in first place. Chat Gpti is in a category of its own. I would not even say, oh, chat chipt is first, blank is second.

No, chat Gpte is literally in its own category. Because if you use it correctly, nothing else on the market, no generative AI product helps you win back as much time as chat GPT if you use it correctly. All right. So let me talk about something you might not know now. I think they're too smart. Open AI is too smart for its own good. Let me talk about.

What I mean here. So an article I'm showing here that says open AI pissed off developers by phasing out plugins for GPTs. All right. So there's been some unofficial proclamations that chat GPT is getting rid of plugins, phasing them out, right? It's not official official, but, you know, I have a tweet from someone in developer relations here responding to someone saying, plugins aren't showing right now due to a bug. This was about a month or two ago. But he said, long term, we hope that GPTs will be the new home for plugins and you can

reuse an existing plugin as an action in your GPD. All right. Let me tell you what that means. So since November, Open AI has stopped allowing new plugins. Okay. And it has been reported and alluded to that they are phasing plugins. plugins out, all right, in lieu of GPTs. So let me tell you this, GPTs are amazing.

Don't get me wrong, right? So if you don't know what GPTs are, essentially you can create your own version of chat GBT, you can essentially train it on how you want it to act. You can configure it in real plain time English. You don't have to know how to code. But also, if you do know how to code, you, you can add some advanced functionality, right? You can put in some, you know, some coding.

I'll just say that to keep it easy in the actions, right? So they're saying in the future, you can reuse, quote unquote, an existing plugin as an action. So they're like, all right, Jordan, what's the big deal? You can reuse a plugin in an action in the future. Well, number one, it's not out yet. But number two, and this is where I think, Open AI is two developer focused, and they are not, I don't think, thinking of the everyday person.

So this is, I think, the double-edged sword of Open AI. This has led them being very developer focused and extremely smart has led to their rapid rise, right? And they were first. They had first mover advantage. But I think this is going to, decisions like this are going to cut into their market. share because I don't think that this is a decision that as an example Microsoft would have made because right now the plugins in what truly makes plugins valuable for the everyday person emphasis on the everyday person i.e. non-developers right the everyday person could go in

even just by taking our free courses and they could automate probably 60 to 70% of of their manual work that they're doing by creating plugin packs. Okay? Here's why that's important. Right now, inside of chat GPT, plugins mode, you can use any three plugins at one time. So what that means, as an example, is you can use in one prompt, right, if you go through and properly train it like we teach y'all in our free prime prompt polish course, you can properly train it first, but then in one prompt, you can have as an example, one plugin that can

read the internet specifically, unlike Browse with Bing. Then you can have a plugin that creates spreadsheets for you. Okay. And then you can have a plugin that takes information from that spreadsheet and creates diagrams. This is a plugin pack that I demonstrate on the show often. So you can, you know, give chat GPT in one chat. Here's a hundred page PDF. Read through it.

Summurize it for me. Grab all of this specific financial information. tell me what it all means, put it all in a CSV in a spreadsheet, and then also from that, create a diagram for me, right? That's people's jobs, right? But that is only possible because you can pick and choose any three plugins at once. So by phasing out plugins, an emphasis here can re, in this tweet, saying you can reuse an existing plugin.

singular. Who knows? Maybe I'm kicking around dirt for nothing. Maybe you'll be able to use all your plugins inside one GPT, but it does not look that way. All right. So all this does is it takes away the power of chat GPT for the everyday person.

And I think that's what's going to happen. If and when OpenAI does take away plugins over the course of a couple of months, people are going to realize, wow, I'm getting more hallucinations. You know, I used to use plugins. Now I'm using Browse with Bing, right? Because that's the default mechanism to access the web when you create a GPT is Browse with Bing.

And it's problematic. I just showed you that. So what happens now when all of these businesses are making these GPTs, right? And they have this false sense of security or maybe false confidence because because they don't fully understand how browsed with Bing works. You're going to have an increased level of reported hallucinations. People are going to distrust chat GPT. People are going to say,

just getting more hallucinations because they're taking away capabilities from the everyday person. I'd say, obviously there's things I don't know, right? I'm sure there's reasons, whether it's financial, legal, et cetera, that maybe is making them take away kind of plugins for a reason that I don't know. But as it stands right now, right, you can still use plugins. In this decision, if they do phase them out in lieu of GPTs, is a move that will eventually mark my words, y'all. Mark my words. I don't say this lightly.

It will cost them billions of dollars. people are going to jump ship right Microsoft co-pilot pro has released their GPT builder but they've announced it they're going to release it okay I think people are going to jump over there or whenever you know bard catches up people are going to jump there because I think this is going to lead to a lot of problematic usage inside of chat GPT. All right we made it to the end I'm going to get to your questions if you heard something in this podcast and this live stream and you're like, huh, interesting. I didn't know that.

I should investigate more. You know what? We do the hard work for you, y'all. This is what we do literally every day. Cut the fluff. Give you straight facts. All right. So if you want to know more, just type in PPP.

If you're, you know, listening live here, you know, like Val just did. send me an email if you're listening on the podcast the email's in there my lincoln's in there we do these free courses literally all the time they are free they are live all right teach you how to properly do this stuff but let me see i'm going to get through a couple of york of uh everyone's questions here because this is a show this is a live stream y'all all right so declan says when is v5 coming and what new updates will that bring good question Declan, Declan.

Sorry if I'm not pronouncing your name correctly. Who knows, right? If you were to follow Twitter, right? Or these billy boys out there, they would have told you two weeks ago that GPT was out. GBT5 was out, right? Sam Altman did an interview and everyone's like, oh, he's talking about GPD5. It's out. It's already here.

No, it's not. When will it come out? No one knows. Could be tomorrow. It could be next year. All I know is GBT4 is. coming up on, geez, year, I don't know, yeah, almost two years, a year, a year and change now. And it is still by far the most powerful model out there.

It is not even close. Look at all the benchmarks, right? Even Google's new model that they just released does not come close to GBT4. So I don't know. What new updates will GPT5 bring? It's a great question, right? We recap, we recap this in our newsletter a couple weeks ago. I'll try to put it in the comments of this LinkedIn live stream.

But an increased emphasis on the ability to reason, right, to think more and analyze more like a human to better understand our language. But also an obvious increased emphasis on multimodality. So what that means is the ability to better input different modes. So input text, input photos, input video, right? And then the ability for chat GPT to output in multimodal. So those are the at least what's been reported so far is a better ability to reason and a higher emphasis on multimodality. And obviously, you know, probably everything else will get better, the memory, the speed, the accuracy, et cetera, but those are kind of the big things.

All right, Woozy. Woozy says, any use cases that you think an earlier version of GPT, not GPT4, actually does better on than four, at least just as good. I'll say no. And I'll say that with a caveat. So I will say in some instances, even when you're using the GPT4 model that is connected to the internet, you might want to cut internet access off. Because like I just showed you here, you don't necessarily have.

have access, you know, you can put in a prompt not even asking chat GPT to go to Browse with Bing and it'll still do it. But I'll say no, you know, at least Woozy if you're asking, oh, should I go to, you know, GPD 3.5 at sometimes, probably not. You know, like I said, if you are getting results from GPD4 where it's bringing in bad information from Browse with Bing, try it again and put in your prompt and instruct chat GPT to not use Browse with Bing. But I will say there's probably not an instance when you're going to get better results. Jason, fantastic question here. Jason says, how would you know that you have a hallucination?

Yeah, that's great. And this is why it's important. Number one, there's not enough education out there on large Chinese models because all the companies are really putting out is marketing. And then you have a bunch of people trying to monetize. then you have everyday AI, everyday AI in the middle, right? We're saying, no, no, no. Don't follow this kid on Twitter.

Don't listen to what big company is telling you, right? There's the whole debacle about when, you know, Google released their new Gemini model and they showed this video and it's like, no, that's marketing. That's not how it works, not at all, right? So there is a huge problem, Jason, because there is no good way to avoid hallucinations unless you actually know the pros and cons of the specific generative AI model, the large language model that you're using.

And most people don't take the time, the effort, and the care to truly understand how to properly use it, right? They either just listen to the marketing from the big company or they listen to the 19-year-old in his basement that has no clue what he's doing and is just trying to sell you some crap. But they don't take the time to do it the right. way. That's why everyday AI exist. All right. Let's see. I think, I think we got to everything here. Okay. Wait, I think we got one more here. I think we got one more. All right. It's Friday. This is an informal episode. We're having fun. We're kicking it. All right. So another question here from DeKlon.

So we're better off keeping our prompts short and sweet to get better outputs and longer conversations. Not necessarily. You're best off by priming chat GPT, right? We teach that whole process. You never go in. Copy and paste prompts literally do not work. That's not how you use a large language model. I don't care who you're, you know, who online, what expert is telling you use this copy and paste prompts. That's laughable.

It's not how large language models work, right? There's a reason why you look at any benchmarking, right? So when these new models come out, I'll try to explain this for our non-technical friends. But anytime a new model comes out, there's all of these benchmarks. There's all of these different tests, right? and they get scores. So any single time, any model always performs better when you'd go to a five shot as an example versus a zero shot or a one shot. Okay.

So what that means, if you just put in a prompt and ask for an output, you don't give examples, no back and forth. That's a zero shot prompt, right? Five shot, I'll simplify it. Let's just say that means you go back and forth with chat, GBT, or any large. language model five times or give it five examples right i'm oversimplifying it but that's a five shot so essentially which obviously makes sense every single model every single test always does better on a five shot versus a zero shot or a one shot it always performs better because that is how large language models work when you start a new chat it knows both nothing and everything at the same time so

So if you just throw in a huge long prompt, you're going to get either hot garbage or lukewarm garbage. You're always going to get better results by treating chat GPT like a human, right? We had Abrin from OpenAI come and that's exactly what we talked about on that show with him. You're always going to get better results in chat GPT. Treat it like a brand new employee. You need to have a conversation. You need to go back and forth. You need to go through reinforcement learning.

You need to go through knowledge testing. You need to share information in the right way. need to test it. You need to keep memory in mind. It's not as simple as do I need a short prompt or a long prompt. You need to properly understand how the model works and you need to go through a process. All right. I think we got to it all, y'all. I hope this was fun. Did you learn anything? Right. Hey, I try to make this one that even if, even if you're here all the time, I wanted you to know at least one new thing. So I hope today's episode where we talk about the five facts that you might not know about chat gpt i hope you learn something all right so here's what we're

going to do i try to do this because i know sometimes i'm long-winded y'all we're going to go over i'm rapid fire fact number one chat gpte doesn't technically understand words it uses tokens fact number two chat gpt does not actually have a 128k memory no matter what someone tells you it has 32k Number three, fact number three, the default browse with Bing does not actually allow you to always browse specific websites. It just queries and sometimes it doesn't tell you. Fact number four, hallucinations are actually rare if you use chat GPT correctly. And I showed you that. And fact number five, chat takes more of a hot take, but it's going to end up being a fact.

Don't worry. Chad GPT is actually two developer focus, which will cost it billions. All right, that's it, y'all. Thank you so much for tuning in. Make sure to go to your EverydayAI.com. Sign up for that free daily newsletter. We're going to break down today's conversation and more, letting you know how to actually learn generative AI,

how to leverage it to grow your company, grow your careers. That's it. Thanks for tuning in. Hope to see you back Monday and every day with 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. Hi, 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.