Everyday AI Podcast – An AI and ChatGPT Podcast - EP 343: Don’t Make These 5 ChatGPT Prompting Mistakes

Episode Date: August 23, 2024

Win a free year of ChatGPT or other prizes! Find out how.Some people say AI doesn't work and that ChatGPT isn't good. But that's not true. The problem is with the prompts you're using. We want to show you 5 mistakes and help you improve your prompts for better answers from ChatGPT.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 LinkedInTopics Covered in This Episode:1.  Importance of Changing Mindset When Working with ChatGPT2. Approaches to Use Large Language Models Effectively3. Getting Quality Results from ChatGPTTimestamps:[00:00:55] Daily AI news[00:05:12] Is ChatGPT getting lazier?[00:11:15] Prompting is the issue[00:13:55] Mistake #1 - Copy and paste super prompts[00:18:30] Mistake #2 - Looking for outputs vs building skillsets[00:22:18] Mistake #3 - Not using skill-based chats[00:25:24] Mistake #4 - Telling ChatGPT it's an expert in X with X years of experience[00:28:55] Mistake #5 - Using ChatGPT as a shortcut[00:31:30] Final takeawayKeywords:ChatGPT, large language model, mindset change, multi-shot prompting, skill sets, training, expertise, AI news, Everyday AI, generative AI, Accenture, New York Times, AI editorial director, Humana, healthcare, daily newsletter, AI inner circle session, prompting mistakes, live audience input, multi-modality, text-to-text, text-to-photo, text-to-video, video-to-photo, text-to-video, video-to-text, engagement, ChatGPT memory, training, automationSend 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. Some people say,

Oh, AI stinks. Chat GPT is no good. Guess what? Everyone who's saying that, you're wrong. You just have no clue what you're doing. So no, chat GPT does not suck. Your prompts do. All right, I'm excited for today's episode.

Welcome to Everyday AI. My name's Jordan Wilson. I am your host and everyday AI is for you. This is your daily live stream podcast and free daily newsletter helping everyday people like you and me learn and leverage generative AI. We've done a lot of episodes on chat, GPT and other large language models, but we haven't done an episode specifically on prompting. You know, sometimes I forget that even though we do a free prompting course literally twice

a week, it's called Prime Prompt Polish. So if you want access to that, just type in PPP in the comments. I'll send you a link. But we haven't actually talked about the concept of prompting on this show, which is crazy. But I'd hate to break it to all these news organizations that are just covering people complaining on Twitter about chat GPT getting lazier. It's not. We humans using large language models are getting lazier. We are not using them how they should be used, right?

there is a process that you should go through when working with large language models that most people are skipping. So I am going to go over today in today's episode on the five biggest prompting mistakes that people are making as well. But I also want to hear from our audience, right? So if you are a normal podcast listener, maybe you know, you're listening on your commute to work or maybe this is your afternoon time doing the dishes, I love hearing. from people they email me and they say, oh, I listen to the show at this time.

It's always so funny here. But regardless, I want to spend this time hearing from the live audience about your biggest prompting either mistakes or your best prompting tip as we get into today's show about why chat chabit doesn't suck, right? Love, love this from Maybritt. So Maybrit, thanks for joining us as always. She says, this tip hardly fails me. Tell me in a percentage how much you understand of this prompt.

Yeah, it's a great one. I think Maybrit, we've talked about this months ago. It's asking chat GBT for a confidence score is fantastic, right? And the biggest thing with chat GBT or any large language model is you always want to do two things. You want to increase the quality of the output and you want to decrease hallucinations or made up stuff, right? like we talked yesterday, hallucination is dictionary.com's word of the year. So you always want to, when prompting in chat chitp, you want to increase the quality of the output, and you want to decrease the likelihood of having hallucinations.

And May Brits tip is a great one right there. Another one from Tara here. I love, this is something we teach in our course, but the reminder to ask chat chbt to recap everything to ensure it remembers. Yes, you always have to test chat chbt as you are talking to it to make sure it is recalling and retaining the information. That's another great one. All right.

And more of a big picture question here from Jay, Jay, thanks. So saying as large language models evolve, will the need to pay as much attempt attention to prompting? Yes, I think so. There will always, I think, be a need for human input. The input type and the input methodology may change, right? Because right now, for the most part, we are, say most people are inputting. with text, right? So within chat GPT, you are typing a text response. Sometimes you may be inputting

a photo or a uploading a PDF with very little text response. I do think obviously the future of large language models is multi-modality, right? So we saw that with Gemini, even though I accidentally roasted them yesterday. But prompts, even how we prompt, is going to change. I think it is going to be much more of our voice. It is going to be video. It is going to to be a combination, multi-modality. I wouldn't even be surprised in the future, right, if we're able to, whether it's through HTML or something else, you know, input all of those things at once, right? To input text, image, photo, or sorry, text, image, video, and audio all in one prompt.

But I think that's where the future is heading. But regardless, the methodology of prompting, I don't think is going to change, but maybe how we input something, Jen. A will. All right. Look at this. People just being a commercial for this, I love it. What's up, Katie?

Katie said for anyone who hasn't joined Jordan's prime prompt polish webinar, get on it. It's been a game changer. That's awesome. Thanks, Katie. And then Tara saying 100% PPP and PPP Pro, yes. Check your emails today. We're going to have an email out about our pro course, which is free as well, right? Yes, I am going to get into this.

You know, sometime this one time someone left a like a one-star. review and says, hey, the host takes too long to get to the point. And you might as well listen on 1.5x. Well, I agree. Might as well listen on 2x. But yes, you've got to get on to the PPP prime prompt polish and the pro. But let's get into these mistakes that you're making because no, to answer that question that I just rhetorically posed five minutes ago about is chat GPT getting lazier according to the reports? No, just people are expecting more out of large language models and they're doing less, right? So let's talk about the five biggest prompting mistakes that people are making that will

show you that chat GPT doesn't suck. Your prompts do, right? And I think we saw this especially maybe like six months ago or so, you know, when chat GPT and other large language models were really starting to pick up steam and, you know, there's this big conversation about will AI take your job? Yes or no. And people would share. And this was actually one of the.

driving factors behind, you know, what was, what was pushing this conversation forward of, will AI take your job or not? People were sharing their prompts and the results from chat GPT and saying, oh, see, AI won't take your job. Like, look at how bad this response is. And I should have responded to those people, but I try to be a nice person of like, no, it's just saying like, you suck at chat GPT or you suck at large language models, right? I like to be nice.

But that's the reality, right? The reality, like, y'all, I've been, and I talk about this in the PPP course, I've been getting paid to write for 20 plus years. Yes, I'm maybe older than the nice camera makes me look. But I've been getting paid to write for 20 plus years. And chat GPT is a better writer than me, period. Will I get that quality of writing with a copy and paste prompt? Absolutely not.

But chat chabit is a better writer than me if you take the time to use chat chbt, and if you correctly, and if you take the time to learn prompting. Because here's the reality, y'all. Prompting is an essential skill set in the future, right? Right now, you may not be prompting every day. But as generative AI comes to our operating system, so I'm looking at things like Microsoft co-pilot coming to the operating system, or with generative AI coming to where you work, right?

Like as an example, maybe you work on AWS, Amazon Web Services, and, you know, Q, their Amazon Q is, you know, being released. Whether you know it or not, sooner rather than later, you will be spending a good chunk of your day prompting. As enterprise companies start to adapt Gen AI, as Gen AI comes to our desktop with Microsoft co-pilot, who knows what Apple is going to be releasing with their Ajax model. But regardless, the same way that we search the internet hourly now, if you are a knowledge worker working at a desk, you will be prompting hourly soon. So you need to understand that prompting is a skill set and you need to learn it. Do not look for shortcuts. All right.

Enough wind up. Let's talk about the five biggest mistakes. Well, you could probably already guess mistake number one is using copy and paste super prompts. The bane of my existence, y'all. 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.adopi.com. Because we all see these, we all see these, right? These things that say 25 insanely useful prompts, right? That'll change your life or blah, blah, blah, right? These follow me everywhere now. Because here's what happens. And here's why you should never use these. All of these people on LinkedIn and Twitter, I think the hot take Tuesday is

carrying over into a Wednesday. They're creating these useless prompt books. right and they want you to either buy them or give an email for them they don't work they're garbage copy and paste prompts do not work is it better than nothing absolutely right but you cannot use a copy and paste prompt and get anything higher than a b than a b minus right if you want actual quality if you want something that is human level if you want something that is human level if you want something that you can actually use without spending hours massaging the content or whatever the output is, you can't use copy and paste prompts, period.

So stop falling for all these tricks. These are just literally, this is online marketing 101 from the early 2000. Stop giving your email and buying prompt books or using these prompts, right? It's fine if you're using it for an idea starter. But prompting is a skill. If you're just trying to use copy and paste prompts, all you are doing is screwing your future self over. Right?

Because when you are having to prompt on an hourly basis, yeah, it's nice. Have a prompt library, right? Like if your job is repetitive, there's probably some type of prompts that you will be able to use consistently. But it's always better to create an expert chat. It works better. We're going to get into that in a second.

All right. So that's mistake number one. And also, as you're going, as we're going along here, if you have questions, please let me know, drop them in the chat. I got to make sure to include the awesome everyday audience. Sometimes when I get in a rant like Ben here, what's going on, Ben? So Ben's saying, initially, most of us were not good using Google to search. Did we get better asking Google or did Google get better at answering?

Likely, it'll be the same with Gen AI. That's a great observation, Ben. I think the difference is, for the most part, Google never changed the rules of searching, right? Even going back to the first days of Google, you could always do Boolean search. I guess the only major change in the last, you know, 20 plus years is being able to search with an image. But for the most part, searches functionality has not changed. Whereas large language models functionality is changing rapidly, right? It's no longer text to text.

It's text to photo. It's text to video. It's video to photo. Video to text. Video to photo. Right. So the functions of large language models are changing in a much different way than search. All right.

So mistake number two, looking for outputs versus building. skill sets. So so many times when people are prompting, they're looking at a large language model the wrong way. They're saying, oh, I don't want to do X right now. So let me just jump into chat GPT. Nothing wrong with that. But again, you're going to get C plus B minus results. Okay. And if you want that human level quality, you have to change your mindset when working inside of chat GPD or any other large language model. Right? I just gave a speech at the AI summit in New York City,

which I think is one of the biggest AI summits in the world. And I talked obviously about chat GPT. And I have so many people come up afterwards and say, I've been using chat GPD and other large language models just to get an output, right? And that in the long run is such a time waster, right? Because you're just going to click new chat. You're going to go in there and you're looking for, you know, a new policy for your HR department

or you're looking for a response for an email or whatever it is, right? And is that going to save you sometime? Yes. But not as much as you think because copy and the copy and paste zero shot prompting is not going to get you anything of quality. So you're still going to have to probably spend some time on the back end, quote, quote fixing it, right? If you change your mindset and start to build skill sets inside of chat GPT,

which is what we teach in the prime prompt polish course, that's when you can actually get usable, consistently high quality outputs from chat GPT because you build a skill set that way. Yeah, more on that later. More on that later. Yeah, because we've been trained, kind of like what Ben was talking about. Yeah, we've been using.

search engines for a quarter century. We've always been trained one input, one output, right? And that is not how a large language model works. It's not. Can it give you one output with one input? Yes. Right. But that's the whole concept of, you know, without getting too technical.

But we kind of talked about it yesterday on the, uh, the Google Gemini. You know, when they were comparing, you know, five shot prompts to 30, two-shot chain of thought, right? Without getting too technical, you always get better results, right? If we're talking about zero shot, five-shot, whatever. A shot is essentially a back-and-forth or an example. And obviously, every single benchmark ever shows, you get much better results on all of these benchmarks when you're five-shot versus zero shot, 32-shot versus five-shot, right?

So think you should be prompting the same way. You shouldn't be zero-shodding, which is just putting in a prompt and looking for a response without giving examples, without going back and forth. You need to be building a skill set going back and forth, not just one input, one output. That is a search engine, y'all. That is not a large language model. Stop using it like this. It's all the fault of these people on social media that have these pods, they have these

engagement groups and they say, all right, I'm going to post my prompt book today. Make sure all 50 of you say this is the best thing ever. And then all of you people out there think it's the best thing ever. And then you try these copy and paste prompts. They absolutely stink. And then eventually you come to the PPP course and you say, oh, I know how to use a large language model now. Just skip it.

Just come straight to the course or just stop using copy and paste prompts. All right. Number three, not using skill-based chats, right? So those two kind of go hand in hand. But let me give an example. And this is literally taking a page out of our free prime prompt polish training. So again, if you want to access, just say PPP. I'll tell you how to access.

But you need to treat a new chat inside of chat GPT, like an employee that you are trying to train. Okay. So even think of, you know, these long super mega prompts, right? They're like, you know, five pages long, you know, because someone out there is selling those. Stop buying them. But that is the equivalent of if you have a brand new employee and you throw a giant 200-page training manual on their desk and then instantly say, now go do your job.

Right. I use this analogy all the time, but it's the best way to illustrate how you should be prompting, if you do that, that employee is going to fail. And you are going to fail in your role overseeing that employee. So you need to have that same mindset that you are overseeing an expert chat inside of chat GPT. Because when you start a new chat inside of chat, it both knows nothing and it knows everything at the same time.

All right. So you need to go through, and this is the priming, right, without giving away our entire course here. but this is the priming phase when we talk about prime prompt polish. You need to go through an onboarding phase, just like you would with a new employee, going back and forth, back and forth, an onboarding, training, reinforcement learning, feedback, testing, knowledge sharing. All right.

And that is where you come up with this expert chats. This is another thing that we teach. But this is just an easy way, kind of like a marketing and messaging thing. but every single chat that you start inside chat TPT should be an expert at one very specific skill set. Okay. You don't just start a chat and say, all right, this is my Wednesday chat and I'm going to go in there for anything. Or you don't just start a new chat for every single prompt.

You build a skill set and use that skill set, that one very specific skill set whenever you need it. You scroll down and you say, all right, here's my recent. and analysis skill set that's based on this element of my role. Here is my creative copywriting skill set for email marketing for my company, right? Every single chat is a skill set and you need to train it accordingly. All right. Mistake number four. Yeah, like Tara says, onboarding.

LinkedIn user, sorry, I can't see your name. Examples. I don't want this to accidentally turn into. well, an hour long, an hour long show. But yeah, we give examples in our free prime prompt polish course. All right. So mistake number four saying you're an expert in blank with blank years of experience. This is hilarious.

This is hilarious. Guess what? That does absolutely nothing. That does absolutely nothing, right? A large language model has the entirety of the open internet, the closed internet, literary works, movies, right? It has one, you know, GPT4 has 1.8 trillion parameters. So saying or telling chat GPT that it is an expert with blank years of experience, that does about nothing, right? Go, go test this on your own.

Do do that version and then start a new chat and go. go through prime prompt polish and you'll see every single time by training a skill set, you will wipe the floor with saying you're an expert in this with blank years of experience. Here's why. And I use this example in the course, right? Let's say copywriting to say you're an expert level copywriter with 20 years of experience. Okay. Remember how I just said large language models are trained on the entirety and the history of the

open internet and the closed internet? There's a lot of people putting out information online that say, hey, I'm a copy. with 20 years of experience, with 30 years of experience. Here's all of the tips you need. Guess what? A lot of people out there putting information on the, on the internet, claiming to be experts, stink. Stink. So if you are that open ended with a prompt inside of chat GPT without giving examples, right?

So like saying, hey, these are the three copywriters. You should be emulating. These are the three copywriting styles you should be using. These are the three copywriting conversions or the three copywriting outputs that we need out of our copy. In giving examples of all of those, if you just say you're an expert with blank years of experience, that's why you're not getting anything usable out of chat GPT or you're frustrated or you're like, oh, hey, I won't take our jobs. Look how bad the output is.

Well, you're not doing it correctly. That's number four. All right. Cecilia, thanks for the question. Cecilia asking, how does the best? building of the skill set of your chat affect the usage of tokens. Yes. Great question. It eats your tokens up, right? So no, chat GPT does not have a hundred twenty eight thousand token memory. Sam Altman on November 7th said that GPT4 Turbo was coming to chat GPT. GPT4 Turbo has

128,000 tokens. I just tested this last week. I haven't tested it yet this week. So unless it's changed, but I doubt it has. But right now, that's a great point, Cecilia, because when you are building a skill set, what you are probably doing is you are sending a lot of text back and forth. So all of that text back and forth essentially eats up chat GPT's memory. Right now, chat GPT plus, if you're on the paid version, has a memory of 32,000 tokens, which I believe is about 25,000-ish words, give or take, right? So at a certain point, chat GPT will start to forget things. So yes, as you are building that skill set, in theory, you are starting to eat away at chat GPD's memory. But that's why in our free prime, prime, proud polished course, we teach you ways around that.

But great question, Cecilia. All right. So like I said, number four is saying you're an expert in blank with blank years of experience. Garbage doesn't help. All right. And number five. And here we go. Looking at chat GPT as a shortcut.

All right. Here's the thing. And I'm kind of twisting this one on its head. Chad GPT is the ultimate shortcut, right? To grow your company, to grow your career. But you have to build the road. You have to build that shortcut. Okay?

You can't just copy and paste something in or do a halfway job of prompting. Okay. What I like to tell people, again, think of it the exact same as an employee. New employee comes in who knows nothing and the ability to know everything at the same time, right? That's the thing. If you train chat GPT the right way, if you use it the right way, it is better than humans at most knowledge-based tests, right? We've seen that with these recent benchmarks, you know, GPT-4,

allegedly Google Gemini Ultra, large language models are getting better than humans at knowledge-based tests, but only if you use it the right way. Okay. So it is an employee that is smarter than the smartest human who sits down at your desk and everyone just wants to throw a giant 200-page training manual at it and say, go, because it's a shortcut. It's smart. It can work for me, right? Yes. But you have to put the work in, right?

Prompting isn't just about trying to get the best quality something as quickly as possible. It is training a replacement for one of your skill sets. Using chat GPT correctly is replicating one of your skill sets. It is automating one of the things that you do every day. It is systemizing your manual work on autopilot. But you have to put in the work. You have to do it correctly. You have to build the shortcut.

And then once you do, then chat GPT is a cheat code. But most of y'all just want to put in copy and paste prompts. So no. So to wrap things up, no, chat GPT does not suck. Your prompts do. Stop reposting those people on social media. Stop buying into using other copy and paste prompts. Like I said, if you're brand new and if you just want to see the capabilities, it's an okay start.

And yes, Cs get degrees as the saying here in the U.S. goes. But all those copy and paste prompts are going to do, the best case scenarios, get you a C. If you want to get to the point where chat GPT can work at the same level as you, where it does become a true shortcut, where you can start to truly automate your work, where you can start to, you know, 5x, 10x your outputs and get usable content in return. You have to go through the proper channels. You have to build it up. And that's where our free prime prompt polish course comes in, right?

People already said it here. I didn't say it, right? Tara said it, 100% people. Katie said it. Other people said, you know, we've had thousands. We've literally had now thousands of people take our PPP course from entrepreneurs, solopreneurs to Fortune 100 business leaders. Yeah, we've literally had people from top 20 companies in the world take our free prime

prompt polish course and give five star reviews. All right. It's free. There's no upsell at the end. This is why I do this, right? I firmly and truly believe that prompting isn't something that should be smoke in marketing mirrors online. It is an essential skill set.

And it can be hard to learn the basics, right? Because in the end, someone's always trying to sell you some crap. Or, yeah, you can kind of learn how to prompt from a big company. But in the end, they're really just giving you, you know, what's going to work best for their systems or they're pushing you into their system. So with everyday AI, with our free prime, prompt polish course, You get an unbiased, free way to learn an essential skill set. And we do it twice a week, right?

We do it Tuesdays and Thursdays. So if you want access to that, just let me know. Thank you all for joining us. I hope you learned that no, chat GPD doesn't. Chat GPT doesn't suck. Your prompts do. Please, this was helpful. Share this, share this post, share this podcast with a first.

friend, leave us a rating if you can, but more than anything, if you haven't. My gosh, go to your everyday AI.com. Sign up for that free daily newsletter and we'll see you back tomorrow and every day with 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. Adobe.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.