Everyday AI Podcast – An AI and ChatGPT Podcast - EP 158: The ChatGPT Mistake You Don’t Know You’re Making

Episode Date: December 5, 2023

You keep making the same mistake on ChatGPT that's causing hallucinations and incorrect information. And you probably don't know you're making it. We'll tell you what it is, and how to avoid it so you can get better results. Newsletter: Sign up for our free daily newsletterMore on this Episode: Episode PageJoin the discussion: Ask Jordan questions about ChatGPTUpcoming Episodes: Check out the upcoming Everyday AI Livestream lineupWebsite: YourEverydayAI.comEmail The Show: [e-mail] with Jordan on LinkedInTimestamps:[00:02:15] Daily AI news[00:07:00] Quick ChatGPT basics[00:13:00] ChatGPT knowledge retention[00:19:07] Remember document memory limit when using GPT[00:20:49] GPTs can have issues too[00:25:37] Better configuration needed to prevent unrelated inputs[00:32:20] Using GPT extensively may lead to errorsTopics Covered in This Episode:1. Impact of ChatGPT Mistakes2. GPT Testing and Usage Issues3. Caution When Using GPTsKeywords:Microsoft Copilot, leadership skills, learning enhancement, GPT, caution, business purposes, performance evaluation, custom configurations, limitations, conditional instructions, token counters, memory issues, ChatGPT, incorrect information, hallucinations, generative AI, AI news, Tesla AI, 2024 presidential campaign, Meta, IBM, AI Alliance, document referencing, memory limit, token consumption, configuration instructions, OpenAI upgrades, knowledge retention.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 mistake that so many people, so many people are making inside of chat GPT,

and it's causing you to get incorrect information, hallucinations. And if you're sharing these chats with others or maybe building custom GPs, the people using those on the back end are running into these issues as well. I'm going to tell you exactly what that chat GPT mistake is and how to correct it on today's episode of Everyday AI. What's going on, y'all? My name's Jordan Wilson, and I am your host of Everyday AI. We're a daily live stream podcast, free daily newsletter, helping everyday people like you

and me learn what's going on in the world of generative AI, not just how we can learn it, but how we can leverage it to grow our companies and to grow our careers. If you're new here, thank you for joining you. us. Normally I have a guest on, but hey, every once in a while, I, it's just me riding solo. And we dive in deep into one issue, maybe one large language model. You're using a generative AI tool, something like that. And let me tell you what's a little bit different about everyday AI. If you're new here or a reminder for, you know, some of our, some of our regulars, you know, that are joining us like Michael Forgey saying, good morning, like Tara, joining us from Nashville.

Yes, this is a live stream. So if you're, if you're listening on the podcast, always check the show notes. come join us the next day. But this is an issue that is affecting so many people, and I don't think people know about it. So what's a little different about everyday AI is, hey, I'm a former journalist. So when a company says, you know, they release new features or when OpenAI says, hey, 128K tokens, context window for chat GPT, you know, giving chat GPT this massive memory. The first thing I do is I run and test that.

And then I tell you all like, hey, FYI, it's not 128. K. It's 32K. All right. So more on that big mistake that people are making in a second. But first, as we always do, let's take a quick dive into the AI news. All right. So a new report details concerns of Tesla's AI. So a former Tesla technician has blown the whistle on the company's AI technology used in their self-driving cars, expressing concern about its safety and readiness for public use. So the technician leaked confidential Tesla data to a German newspaper, revealing that thousands of customer complaints about the autopilot feature as well as recorded accidents and safety concerns.

Tesla CEO Elon Musk, though, he's obviously defended the technology, claiming that Tesla has the best real world AI. Hey, if you're a Tesla driver out there, have you had issues before? I know friends with Tesla's. I don't think they've ever had issues, but reportedly there's thousands of them. All right. Speaking of things, controversy. So AI is already causing a stir in the 2024 presidential campaign scene.

And we barely started. So according to a report from Forbes, Donald Trump recently claimed that the commercial from the group The Lincoln Project used AI to make him look bad in an ad that they ran against him. However, the ad, which was titled feeble, actually used real authentic footage of Trump highlighting some gaps and low lights of his presidential campaign, according to this report. So it's interesting because we know that, you know, as the 2024 elections here in the U.S. come roaring around, we've been talking for a long time that, you know, deep fake videos and misinformation

and AI generated images are going to be a problem. But I never thought about, you know, candidate saying, oh, you know, when something bad happens saying, oh, that's AI. And it's like, oh, well, what if it's not? So maybe the first reported instance of that. All right. last but not least, two big tech titans are teaming up for an AI alliance, Meta and IBM. So they've launched the AI Alliance, a group of leading organizations and industry startup,

academia, research, and government to support open innovation and science and AI. It's not just meta and IBM, but they're the headlining names that are kind of launching this, but there's also 50 founding members and collaborators from all kinds of big research organizations, institutions, colleges, all over the world. So some other big names, AMD, Cleveland Clinic, Cornell University, Dell, hugging-faced, so many others. Technically, my alma mater, University of Illinois is on there as well. So the objectives of the AI Alliance is to include developing a benchmark and evaluation standards, fostering a vibrant AI hardware accelerator ecosystem in supporting global AI skills, building,

and exploratory research. So pretty interesting. I mean, a lot of heavy hitters coming together to form this alliance. You know, we haven't really seen this at least not here in the U.S. anything official like this involving many different companies, many different, you know, research organizations. So I think that's actually a positive sign for AI when you have, even if it's an unofficial body, you know, not they're not necessarily there to govern, but they are to make sure

that AI is used more responsibly. So I think it's good news if you're a fan of generative AI. So there's always more. There's always more news. So if you haven't already, please go to your everyday AI.com. Sign it for that free daily newsletter. We take every single day. We break down the podcast episode from that day.

We always share, if we have a guest on the show, we always share a lot more resources that maybe they don't put out in the public and really just break things down every single day, all the new tools and the trends and the AI news. So make sure to check that out. So let's, hello, everyone joining us. Oh, I love, I love going live. I love being able to learn together with you all to really just break down different artificial intelligence systems step by step like chat, GPT and others.

So good morning, Bridget, joining us from Michigan. Todd, man, our Dallas contingent is really growing. Darrell, thanks for joining us from Philadelphia. you. Woozy says, he's from Kansas City and his Tesla got him safe today with no issues. That's good to hear. All right. So I want to let's let's start with this.

Let's start with this. What is the biggest chat GPT mistake that you've made? And maybe you've caught yourself or, you know, maybe someone else told you. And here's the thing. So much, especially recently within chat GPT has changed. So maybe you're listening to this podcast and it's your first time here. And maybe you're just a casual user of chat GPT. I'm going to give you the quickest history lesson ever.

So chat GPT has their free version and their paid version. The paid version chat GPT plus is $20 a month. If you want that, you still can't sign up. There's a wait list. Essentially, they released a bunch of new updates about three weeks ago. Literally, they're servers, which I think they have like every single piece of compute in the world still couldn't keep up. So you can't sign up for chat GPT plus right now, but in there, you have a couple different modes,

right? So you have chat GPT with plugins, and you have the default mode, which is actually really good now, whereas before I told people never to ever use it. But now the default mode brings you a lot of new functionality. You can upload documents. You can use browse with Bing. You can use advanced data analysis. You can use Dolly, all within the one default mode.

So that's great. Then the other big new kind of piece of Open AI chat GPT since this Dev Day update, which was announced November 7th, started to roll out to everyone the next week, is custom GPs. So what's happened is since there's been the ability to upload documents inside the default mode and now also with custom GPs, and with custom GPs essentially without diving into it too deeply, it allows you to kind of train your own version of chat gbt and to upload documents into its knowledge base. And then you can actually share that with people. And chat gpti said that they are going to be releasing their GPD store, I believe, in January or around the beginning of 2024.

So the same way that you saw, you know, kind of this explosion of the app store, we're going to be seeing that with kind of this GPT store, or presumably you will. Right. So it's essentially made the process between the default mode, becoming now very easy to use and allowing you to upload documents and also the custom GPDs. So what we've seen, and this is indicative by Open AI really shutting down their process to upgrade users, what you've seen is is so many people, millions of people just flooding chat GPT and building these things because now it's easier than ever before. It's easier than ever before to upload your own knowledge, to upload your own company documents. If you want to, again, always read the privacy policy. And I'll say it as I always do, never upload confidential or sensitive information into chat GPT. But it's now extremely easy that you, you know, I wouldn't even say it's low code. It is no code.

You don't need to know really anything. And even the GPT process, it walks you through it. It's a guided process to set up this GPT, right? So, but what comes with that? That is now there's a lot of problems. Right. So I've been studying and watching and I always like to look at people's tutorials. And a lot of people are saying, oh, you know, chat GPT, the quality has gone down recently.

I don't know if any of you on the live stream have seen this, but people are comparing. They're saying, oh, GPT is getting all these things wrong that it used to get right a month ago. And, you know, I'm looking at all of these people reporting these issues and, you know, trying to dig in to see. why? Why are they getting these issues? I'm not getting them. Our team's using chat GBT from, you know, morning to night, uploading documents. We're never getting these issues. And what I've noticed is a lot of people don't actually understand what it means to upload a document into chat GPT. And it's handled a little differently inside of the default mode versus if you're building a GPT, a custom GPT. So here's the thing. I think a lot of people,

writing or complaining or trying to teach people about, you know, chat GPDs or novices, but they claim to be experts, right? I've never claimed to be an expert, FYI. But, you know, our team has been using the GPD technology since it was publicly available. And I believe it was late, I always forget, late 20, late 20, yeah, late 2020, early 20, right? So we've been using the technology for multiple years. And so many people just don't really understand how this works. So we're going to dive in today and we're going to go over what I am now seeing as probably the one biggest problem, the biggest mistakes that people are making. And it's knowledge retention. All right. Knowledge retention. 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. Let's talk about what that means. because there's technically two different things. So within chat GPT and all large language models, you have a memory, right? And I'll be sharing about this at one point, but we busted it down because during the, the Dev Day announcement on November 2nd or November 7th of this year, Sam Altman said GPT4 Turbo is coming to chat GPT, right? And what people, you know, ran and everyone got excited because GPT4 Turbo, they're updated

GPT4 technology included a 128K context window. So what that means is about 94,000 words, which is a lot before chat GPT can start to, you know, forget things or hallucinate. But it's not there. It's not there, right? We're still actually at 32K tokens, which is about 24,000 words. People don't know that. So that's one piece.

That's more memory than knowledge retention, but that's important to talk about as well. because in the end, the most important thing about using chat GPD or any large language model is using it correctly. I tell people you have to play by chat GPD's rules or whatever large language model, you know, Bard's rules, enthropic cloud 2.1, whatever you're using. You have to play by their rules. But you also have to open up the rule book and you have to try the rules out for yourself. Right. That's a huge problem right now is there's a difference between marketing and real life use cases, right?

All the companies are, which they should, right? They should market their products and say, oh, they're the best. They do this. They do this. It's amazing, right? But you go in and you test it and you're like, no, there is no 128K. It's 32K. Right.

So that's a huge issue. But let's talk specifically about knowledge retention. And yes, I do want to know from our from our live audience. if knowledge retention has been a problem for you, or what is your other big chat GPT issue? Maybe I'll tackle them at the end. So make sure if you're like Connor joining us live or Mike Lambers or Brian, Brian Bing, let me know what, if number one, is knowledge retention a problem for you.

Number two, what is your biggest chat GPT issue? So let's look at why this matters and specifically what it means. All right. So if you are joining us on. the podcast, I'm going to do my best to describe what I'm showing on my screen. But I think it should make sense. So in short, when you go into default mode, now you have the ability to upload a document, which is great. Because before, the only place you could do this was in advanced data analysis mode, which wasn't connected to the internet. But now you can bring in a document, chat with it.

And through targeted browse with Bing searches, you can create content that is connected to the internet, but also by using a document to upload, whether it's, you know, your company's knowledge base, whatever it may be. So that's step one is you can upload a document. Great, right? So we did this example. This is the document I uploaded. It's a PDF. I have a screenshot here. It's just Jordan writing samples, right? So, but I hid something. This isn't a long document either. It's, I think it was a 10-page document. So actually, a very short document, PDF. So I hid something in between, you know, toward the end. I said, before we get started, here's some basic information about me. I said, my name is Jordan. I'm from Chicago. I like the Chicago Bears,

but they stink. Sorry. And then I said, my favorite color is Carolina blue. Okay. So, in theory, you think that once you upload this, it's there forever. And this default chat GPT mode, remembers that. That's obviously not the case. So we're going to talk about, we're going to get a little bit into the details here, y'all. But there is a difference even between how the default mode and chat GPD handles uploaded documents versus how a custom GPT handles uploaded documents in the

knowledge base. And there are some very, and I've done hours of testing on this, right? Putting on my own journalism hat, never trust what a company tells you. You know, if your mom says she loves you, get it in writing. Love you, mom. Thanks for, you know, listening to the show.

My mom listens to the show. So you have to always pay attention and always test this because let's see, right? So now what I'm doing, I have a screenshot here and I uploaded the document and I'm just asking it to analyze the document. Right. So I'm not instantly asking it questions, you know, from inside of chat GPT. that's not what I'm doing. I'm just saying, hey, analyze my writing. Okay.

So then I ask, what is my favorite color? And chat GPT says, I'm sorry, but I cannot determine your favorite color. Okay. So I didn't do anything in between. We didn't run through that 32K token 24,000 word context window. So that's something that so many people are confused about. right, because they'll upload a document and then they'll ask a question from it or maybe you're writing a blog post for your company and you're uploading examples of, hey, these are the other

web pages on our website and then you're telling chat GPT to, you know, write this in your brand voice. If you are not making a call to chat GPT, to retrieve information, it may not, right? And here's the thing. Chat GPT, y'all. It is the world's most advanced auto-complete. That is all it is. Obviously, this auto-complete has 1.8 trillion parameters in its big neural network brain. But regardless, it is an auto-complete. So I've built simultaneous chats to test this theory.

And I could say, what is my favorite color? And chat one will say, I'm sorry, but I cannot determine your favorite color. And chat two will get it correct, even when I'm copying and pacing the exact same thing. So that's also important to know. You are never going to. to get, well, I won't say never, but you are very rarely going to get the exact same responses inside of chat GPT because GPT4 Turbo is a large language model. It is essentially a smart autocomplete that thinks acts and speaks in tokens.

Okay. So how do you fix this? And why is this a problem? Yeah, David, if your mom says you love you, get it in writing. Absolutely. So why is this a problem? Well, if you're uploading documents inside of default mode, you, you just, you just, think and you assume, oh, great, oh great, I can spit out content at scale. I can create a

PowerPoint presentate. I can do anything because I've uploaded all these long documents that I took a lot of time to put together to give chat GPT access, but that's not how it works, as you just saw. So the fix in default mode is actually extremely easy. All you have to do is call. You make a call. So what that means is at sometimes you need to point chat GPT in the right direction or tell it to reference something. Yes, it is always best to work with a large language model like a new employee that you're training and talk to it like a person. So you don't have to speak in fancy prompt to talk. But a lot of times, especially in this scenario when you're working in the default mode and you want to tap into knowledge from a document that you uploaded, number one, you have to keep the memory in mind. So once you get past that memory, it doesn't matter what type of language you use.

You're probably going to need to re-upload that document once you get past chat GPT's memory limit, right? Or all you have to do to fix it in default mode is to make the call. So what that means I did the exact same thing. I just said using the attached document, what is my favorite color? And then you can see I screenshot it and within chat GPT default, it will always show you what it's working on. So if it is reading, quote unquote, reading your knowledge base or whatever, sorry, not your knowledge base. If it's reading, the default mode is reading your uploaded document, it'll quickly flash up there.

And it'll show you a little, kind of like a little timer. And it'll say reading, you know, in this case it says reading Jordan writing samples PDF. Okay. So now I know because I called chat GPT to reference that document, even though it was right there. And I only asked, I only prompt one thing in between. It still may not know, right? And here's the other thing. Even if you upload a document and you tell chat GPT, hey, retain all of this in your memory,

doesn't matter, doesn't help. Tried it many, many times. It does not matter. A lot of times what people don't understand, you either have to call to your knowledge source or you need to get it in the writing or in the inline text in order for chat GPD to best remember it. Keep an eye on your tokens as well. All right.

So a little more because GPTs can have issues too. Yes, people. I haven't seen anyone else on the internet talk about this. So you're welcome, 22-year-old influencer who was an NFT crypto bro and now as an AI expert. You can steal this and go sell it in your next prompt book for $99. All right. So GPs can have issues. with this as well.

And we've done plenty of double testing on this, right? So what that means is doing the exact same prompt in a brand new chat, copying, pasting, and making sure that it's not an anomaly to make sure that these are consistent issues. And they are. So let's look at how GPs can have this issue as well. Okay. So again, I was just doing some testing with a custom GPD.

I uploaded just two different docs. So a separate dock and then the same, exact same writing samples dock. Okay? It's super easy to configure, you know, you upload your dock and it's kind of the same thing. So here's where you run into the issues with chat GPT in the custom GPT mode. But I also, before you think about this, how you are using it, think. we are talking about an app store, right? So presumably people will be paying for GPTs, right?

And so this is also a shout out to all of you GPT developers that are going to be selling something. Don't sell crap, right? Take your time and build this the correct way. Because if you put this in your configuration, if you remind your custom bot to every single time. Check the knowledge base, regardless of the user inquiry. It is much, it doesn't get around that issue 100%, but it decreases the likelihood that whoever is using this GPT, whether it's you or someone else, that they're going to run into

this issue, because here's why. You can actually confuse a GPT bot if it is not properly configured. Yes, I did it. Yes, this is what I do in my free time, y'all. I bless, bless my wife for having the patience with me as I, you know, play with chat GPT and plugins at all hours of the day and the weekend, right? But it's for you. It's for you guys. So I hope I hope you appreciate it. So you can, you can trick chat GPT. Not well, not even trick. But if someone is using your GPT, whether you're making it for a teammate, for yourself, if you put it on the GPT store, if your configuration instructions are not clear, it is very easy for your bot to either hallucinate or sorry, your GPD to either hallucinate or not perform its most basic functions.

All right. So this GPT is essentially saying, hey, write like Jordan, right? So I have the uploaded writing samples. So here's where it gets tricky. So right before, because there's a lot of this, I also tried to screw it up by putting in about 200,000 words, which caused it to break and no longer, you know, access the knowledge base. But that didn't happen consistently enough. But essentially what happened here is I uploaded my PDF. And then also right after that, I told chat, GPT, hey, I'm eating M&Ms right now.

And then later, I put in about, it was about 50,000 tokens. So I wanted to go past the 32K limit. And then I said, I switched it up. I said, what am I eating and what is my favorite color? Now let me break this down. The favorite color is in the document. Okay. What am I eating was in the inline.

text. Okay. And at that point, I was well past the 32K token limit. Okay. So normally on any question that you ask a GPT, normally, it will almost always first browse through or screen the knowledge base. So what I found, and you know, who knows, maybe Open AI is aware of this and they're, you know, patching it up and making it more consistent.

And yes, you can get around this with better configuration instructions. But what I found is when you put a lot of unrelated, unrelated text, right? I literally just copy and pasted transcripts from the podcast, threw it in there between me entering this information about I'm eating peanut M&Ms, put in all this information. And then I asked this very question, what am I eating and what is my favorite color? Because when I just say, what is my favorite color, it gets it every time. it reads the knowledge base for whatever reason. When I ask an unrelated question to the knowledge base, it does not call to the knowledge base,

which is problematic, right? Because what it says here is it says, I don't have the capability to know personal details such as what you are eating or what your favorite color is unless you tell me. My responses are based on the information provided in our conversation and cannot access personal data. Here's the thing. Obviously, it can, right? Because when I asked this same question before all of that long text, it got it right.

So for whatever reason, it could be a context window issue. It could be the combination of asking something that was in line, but further than the 32K back. But regardless, a very basic question from a very short document inside of the GPT's knowledge base. didn't work. Why does this matter to you and what can you do about it? Well, one, you're going to see the GPT store coming out and you want to make sure to only be, you know, buying things if they are for sale. I'm sure there's going to be free and paid ones, right? You really want to be buying things from people that know what they're doing. So it's almost a problem, right? Think of 15, 20 years ago, if you're looking at the iOS app store. What if it was so easy for anyone to build an app

and you were to trust or run your company through it, you know, something a child made that has no clue how apps work or what an API even is, right? So I love that it's so easy for anyone to go in there and build a custom GPT. But you have to keep this in mind when the GPT store comes out. And if you're using other people's GPTs, you really have to put them to the test. You have to run through all of these different kind of scenarios that I'm walking. you through right now. Because I think what so many people are doing with GPTs, which is great. It's they're uploading their company document. They're uploading their writing samples.

They're uploading whatever it is the project that they're working on, right? And they're blindly trusting the process, which again, if you're configuring this GBT on your own, make sure to put instructions in there to every, say, regardless of the user inquiry, always read through the entire knowledge base first. Or, you know, you can have, conditional instructions, right? So maybe instead of uploading, you know, one 50-page document, maybe you upload five, 10-page documents. And you give conditional instructions saying, you know, first, you know, for if they're asking about, you know, quarterly reports, look in the quarterly reports document. If they're asking about, you know, company services, look in company services

document. Because yes, right now, GPTs break pretty easily, especially when you're building them. break a little bit on the back end, but I know that that's something that presumably OpenAI is spending a lot of time on right now because I think the response to the GPTs has been overwhelming. That was a lot. Yes, Cecilia. So chat GPT is turning human. It skims and misses content. Kind of.

Yes. Kind of. Yes. Dr. Mithata, thank you for saying, yeah, he said this is really great. Thanks for sharing. I do see a couple of questions. I'm going to try to get to them.

If you have any last minute questions, get them in now. So I'm actually, hey, I don't know if anyone out there is listening in New York, but I will actually be speaking at the AI summit in New York. I'm actually hopping on a flight right after this. So I can't say for too long. But hey, if you are going to be out in New York at the AI summit, let me know. I'm going to be giving a talk there on guess what? How to make chat GPT sound more human.

Right? Because you can try to figure out of your own. You can Google it, but you're still just going to find us. You're still going to find me. So let me try to get to a couple of these questions. Cecilia asking, is there a feature that tells you where you are in terms of token usage? Can you ask chat GPT to keep a running count of remaining tokens? So Cecilia, great question.

The best way to do that, actually, I don't know if most people know this. ChatGPT is not great at advanced computing or advanced math. So I always say use a token counter. Go check out our website or YouTube. We've done entire episodes on tokenization. We've done little five-minute tutorials on different Chrome extensions. You can use that will count your tokens. But yes, whenever I use Chad GBT, I always have the token counter up because you need to know when it is running out of memory.

That's extremely important. Let me see. Looks like one more question here. What's up, David? David said in a custom GPT. Do you have to call back to info? Well, in theory, you don't because in almost most use cases, David, it will always check that knowledge base first. However, I recommend because of this same issue that I found and I'm talking about now and I haven't seen other people talking about it, you can confuse a GBT, a custom GVT by putting in large amount of text or by asking it to recall something that was in line and in the document at the same time after.

long text. I know that sounds like a very specific kind of way to encounter an error. But again, people are now running their businesses, which is great through these custom GPTs. So you need to be aware that it's not as cut and dry. It's not as cut and dry as you upload the document there once and it's safe. And you know, you can't just think, I'm going to upload a 100 page document and I'm going to chat with this bot for, you know, every single day, nonstop for a year and never run into issues. You should always be testing whether you're building it yourself or using someone else's. Do what I do. Hide something small in the middle or at the end of a document, paste in, you know, 50,000 words and then ask about that one specific thing.

So, David, that is what I would recommend doing. If you're using someone else's GBT, if you're building your own, yes, I would put in instructions to always, always, always go through. the whatever knowledge base that you upload, however, the downside to that, if you have a lot of documents or very long documents, and if you put that into your configuration constructions, you are increasing the likelihood that the GPT is going to essentially, for lack of a better term, crap out, right? You've seen those, you know, you get a little number at the bottom each time it makes an attempt. It's, you know, one of one, two of two. And, you know, I've seen people's screenshots where it's like 37 of 37. So the more recent,

that you require your GPT to go through, the more likely it is, at least now, while Open AI is tackling kind of this resource issue, the more likely it is to not give you a positive return. So you do, you may at times, depending on your documents that you have, depending on your documents that you have in there, depending on how they're set up, you may have to go back and forth between speed and truth, right? Obviously, you always want to side on the error of truth.

But yeah, it might mean just breaking up a very large document into smaller documents and then having configuration instructions inside of chat GPTs, telling it to look in specific documents for specific queries. But if you're not running it into the issue and if you're getting consistent outputs every single time, you're all good. All right. Brilliant show, Jay. Wow, thanks, Jay.

I appreciate that. So, hey, if you feel like Jay, if you're listening on the podcast, if this was helpful, if you're getting value out of the everyday AI show, do me a favor. If you aren't already subscribed to us on Spotify, Apple, please do that and give us a rating as well. But share this. Share this with a friend. If you're listening on the podcast, click that send button, send this to a friend that you know is using chat GPT or building GPs.

If you're listening live, share this post on LinkedIn. Tag someone in the comments on, you know, someone that you think needs to hear this. You know, sometimes people ask, hey, Jordan, you provided so, you know, so much great information for me and my company, what can I do? There it is. Share this. Hit that repost button. Tag some people in your network, trying to help more people make fewer mistakes inside

of chat GPT. So thank you all. I hope this was helpful. Make sure to go to your everyday AI.com. Sign up for that free daily newsletter. Join us tomorrow. Join us tomorrow as well because we are bringing on a leader from Microsoft to talk about how co-pilot can help you build your leadership, but also it can help you learn better.

So I'm extremely excited to talk about Microsoft copilot. So I hope to see you tomorrow and every day on 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 in the assistant handle 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.