5 More Prompts 99% of AI Users Don’t Know

It's time to look at a few more ChatGPT techniques that most people don't know. And let me tell you one thing, all of these are techniques that I actually use regularly. Maybe not on a daily basis, but all of these have been battle tested and stood the test of time. And with that being said, let's get into the practical part of showing you various prompts that most people don't even know exist. Okay, starting off with the first one, this prompt is there to eliminate one of the biggest weaknesses of LLMs like GPT4O hallucinations. And the way this prompt does it is by sticking sticking to exact citations from the source document. Yes, that's possible. Matter of fact, it's so powerful that even OpenAI considered this to be worthy of the prompt engineering guide. And yes, most people don't even know this. They have a prompt engineering guide with various prompts and recommendations. Now, these are focused by developers, but nevertheless you can use a prompt like this in any scenario where sticking to the source material is important and you don't want it just hallucinating and coming up with new things. And this is the prompt I'm talking about. So let's have a brief look at it. So so, as you can see here at the bottom, you're provided with a document and a question that you ask. These are your two inputs. And then what the prompt does is it answers the question using only the provided document and it exactly cites all the passages of the document it has used. Then there's further clarification that if the document does not contain the answer to the question, then GPT4O will tell you that there's insufficient information. This is something that does not happen by default, and many people do not understand this by default. GPT4O and most other large language models, but especially GPT4O is set up to be a helpful assistant at all costs. Other words to express this is that if GPT4O doesn't have the context or the necessary information or skills to perform a task, it will never tell you. It will just do its best and give you some result. And often this manifests as hallucinations because this assistant really wants to help you. It's there to get your tasks done. And this prompt gets around that by telling you that it is okay to to have insufficient information and to letting you, the user, know. Okay, so let's use this to practice. I'm going to use it on the abstract of this paper that I thoroughly enjoy and have read multiple times at this point. You might be familiar. It's this little simulation where multiple agents live their lives and interact with each other. That's a different topic though. All I'll do is copy the abstract into insert document here. Okay, so I'll replace this variable with the abstract. And yes, I'm aware I copied some other stuff here, like the authors or the date. That's fine. I'm just looking for a quote from the abstract here. Perfect. And then I need a question and I'll just ask something like, does this paper simulate real world scenarios for the use of LLMs? Okay, and that's it. Once you customize these two things, you can just hit Enter and see what you get. And as you see, it works. It provides you one piece of information and then the citation and another piece of information and then the citation again. But what happens if I copy this entire paper into here and I go into ChatGPT, I say edit and I replace everything inside of the triple quotation marks with. With the full text and then maybe I change my question to something that will not be present in there. I'm going to ask something random like are dogs or cats better? Huh? Now here's an important thing to be aware of. Okay? As the text that I copied into here was extremely long, it starts giving me random answers. What's going on here? Business plan. Self funded trucking company. That's not what the paper is about. That's not what I prompted. If I copy this full prompt. Okay, look, that takes a lot of scrolling, right? And I head on over to the OpenAI tokenizer and I paste it in here. I will see there's around 30,000 tokens. We're getting too close to the token limit here. And it's starting to lose the information provided with in the beginning. It doesn't even know what the instructions are properly. The token limit in ChatGPT right now is around 32,000 tokens. And here's an interesting fact. They displayed this previously, but now they sort of removed it. I couldn't find it on their official pages anymore. But as you can see, even getting close to the 32,000, it starts losing parts of your message. So make sure there's a healthy distance between the 32,000 and what you're putting in. Otherwise you'll get nonsensical answers as it will start forgetting what you wrote here right in the beginning. So, so what I did here is I just skipped the abstract. I went from introduction all the way to the end. I double checked this with the tokenizer pasted it in here, 28,000, that's a bit better. And then if I go ahead and edit this prompt and just replace the document part, send it yet again, let's see if this answers our questions. If dogs or cats are better, I suppose, and it tells us what we want to see here. Insufficient information. There's no info about that in the document and it's not afraid to tell you if you use this prompt in the beginning. Quite powerful, isn't it? Let's move on to the next one. This one is arguably a bit more basic, but it's just a staple in my repertoire of prompts that I like to use. SWOT analysis are such a powerful capability of LLMs, especially GPT4O with its extensive world knowledge. Now, as you can see, there's a few variables in here, so we will have to replace every single variable in this prompt to make it fit your situation. And by the way, all the prompts could be found in the description below. You can just copy paste them from there. So, so I'll just make this fit my context. I'll say an AI learning community. I'll actually remove this considering market size block here, no problem. And analyzing competitor landscape. It's great to have in the base prompt, but also in my situation, I won't be using this here. So all that's left is the risk factors and opportunities. And I could actually leave this as is. It will come up with the risk factors and opportunities here for me. I could specify them, but I don't have to. And that's the beauty of writing variables in the way that we do here at the AI advantage. All the materials that you will find, all the prompt templates, all the workflows, GPTs. I always use this format for variables with square brackets and a word in the middle that can actually be used by itself. So as mentioned, I could just leave it as is and it will create a SWOT analysis for me. Again, a pretty simple prompt, and this one would work universally. I just wanted to bring it to your attention because it's a powerful tool for anyone who's a bit forward thinking and who's trying to do things that haven't been done before. And the slight little touch in the end with risk factors and opportunities is just fantastic because you get the SWOT and you get the biggest potential downsides and upsides available to you from ChatGPT's point of view. Simple, yet effective. Let's move on to the next one. Okay, next prompt is something we talked about before, but I realized that New people come in and they need to be reinformed. So I'll keep this brief. This is the AI Advantage prompt generator that we built. And this thing is so powerful. I just need to show you again, this one is not in the description because it didn't fit. It's too long, it's too complicated a prompt. Where do you find it? Well, this is the free notion template you get when signing up to the AI Advantage newsletter. You get it on day one. Right away, all you need to do is scroll down here, go to this prompt generator tab, and you're going to get 10 prompt generators for very varying jobs. We're going to take the prompt hacker here, and using this is really simple. I'll just copy to Clipboard and then I'll paste it in here. Now, please don't get overwhelmed by this massive prompt. All you need to do is copy paste, and what it will do for you is generate 30 actionable prompts that work within GPT4O today for you right away, including variables and titles, so you can easily identify the ones that are relevant to you. And this is customized for a growth hacker. And the way this customization happens is here at the bottom in these custom instructions. So if you want to customize this for another job, all you need to do is change this bottom part. Most significantly, you want to change the profession or the role here in the beginning where it states it's a growth hacker. And then maybe in second place, it would be the goals building block here. If you customize those, this prompt Generator will create 30 actionable prompts that you can use with ChatGPT today. This is really amazing. And whenever I tackle a new area, I customize the prompt generator to account for that new task. And then I can easily explore what is possible with AI today within LLMs like this. Again, you find that on the first page of the freebie that you get by signing up to the newsletter. And there you go. This is just a really quick win to get people off the ground when they're starting with ChatGPT. And there you have it. I'm giving it out for free. On to the next one. All right, so let me tell you, this one is actually a personal favorite. Have you ever been in this situation where you open up a group chat or a Discord channel or a WhatsApp group, and there's just this notification with 150 new messages and you're like, there's no way I'm going to catch up on all that. I don't have the time. For this, but maybe you need to. Maybe there's vacation planning going on now before summer or it's work related and important for you to be up to date. Well, GPT4.0 for the rescue in this case with this little prompt. Can you take a look at these messages and tell me what was discussed while I was away from the conversation Messages colon and here you insert the entire conversation. Now I'm actually going to demo this on something that is not a direct message conversation. It's a common thread just to show you that you can use this on all sorts of things you can use use this on forum posts, comments. It just works really well and you guys might be familiar with this. We had this public challenge where people had to find GPT4.0 use cases and Sal wanted with his book listing GPT. So I'm just going to expand this entire comment section and then I'm going to copy everything in here. Mind you, this does not copy the images if there's any present and I can insert it. And there you go. A quick little summary as to the various points that have been discussed. Now at this point I should say the summary capabilities of GPT4.O and other large language models have been heavily criticized recently. If you put something like an entire paper in and it's not great at extracting all the relevant information, but something like a DM conversation that isn't as deep as a paper where specific words don't break the entire flow. This is where LLM summaries are really good. Actually they become even better if you specify a goal in your so I could even tell it the specific thing I'm looking for and then it's going to customize the answer to fit that. Plus, if you're analyzing a group chat or you're planning a vacation, you can always follow up with questions like is there any action I need to take right now to participate in this vacation? Right. Simple follow up prompt and it will go through all the messages and give you exactly the info that you're looking for. In this example it gives me all the conversations based around GPT building, which is the thing I personally care about the most when I look at a comment section like this. Okay, and just quick side note, some of these prompts actually come from our weekly newsletter and we compiled a database of every single prompt that we featured in our weekly newsletter over the last 18 months. It's a database where you can copy paste every single one of these. We included titles so you can easily identify the ones that might be useful to you. Plus we even included the section from our newsletter so you can get a little bit of extra info on when and how to use this. And the way to get this is by signing up to our completely free newsletter. And I included this in our onboarding sequence that we reworked from the ground up recently. What the onboarding sequence now does for you is that five days after signing up you get various AI workflows or resources like this for free into your inbox. It's my best attempt at helping people get up to speed with valuable resources like this. So sign up to the newsletter if you're interested in this full database. Obviously there's other great resources in there too, but by I think day three, you should be receiving this in your inbox for free. Oh, and by the way, in the database there's a prompt that actually does the same thing as I just showed you, but for emails. And we customize it a little bit to be more actionable, more work related. So if you're interested in that, you'll find that in here. It's a little bit of tease, but also that could be fun sometimes. But again, all the prompts we're discussing today are in the video's description, which is one that I actually use all the time. This is one of those hidden capabilities in ChatGPT that to some people is just blatantly obvious. And fair enough, if you're one of those people then great, you might be using this. But most people do not realize that ChatGPT is incredibly good at recognizing patterns, then reproducing them. And when you apply that to the context of Excel sheet or a database, it means that it's extremely good at filling in the blanks. Meaning if you have any table that misses a certain column or there's certain data missing and you need to fill it in manually, no need. Upload it into ChatGPT. Use this prompt. Here's a database of you know you need to tell it what kind of content it is. Some of them are missing a field. Please take a deep breath, think step by step and then fill in the missing fields for me. This little prompting trick here is called chain of thought. It lets it approach the problem step by step rather than trying to fill in all the fields at the same time and getting confused by that sometimes. So in this particular case it's really helpful to have this little block in here. So let's test this, shall we? I'll just go on Kaggle and grab a random data set here. Maybe best selling music artists will work for example, quick Google login, later download. And here we are. So this is a complete Data set with 120 lines. But what if I just go in here and delete some of these countries, right? Celine Dion in Canada, Queen from the UK with Aerosmith, Lil Wayne from the us I'll just delete these and maybe. And this is where it starts getting tricky. I'll even delete some of these release years. So Elton John, 1970, Ed Sheeran, Chris Brown and Jay Z. Okay, so now we're missing some data. I'm going to save a second version of this, head on over to ChatGPT, upload it like so, and then let's not forget to, to specify what kind of content we're looking at. Although it could probably infer that from the file name because it's well named, which is the best practice. But just to keep this clean, we'll fill this out and hit Enter and let's see if it can handle this. And then let's compare it to the original data set. Now, it's going to be using the code interpreter here, but again, it's also going to be using the world knowledge of ChatGPT4O to fill in the blanks. And as you can see here, it's thinking step by step. It's not trying to fill in all the blanks, it's starting with, okay, what country is Queen from a British band, so the country should be UK. What about Elton John? First released 1970, just as we saw in the last data set. And if I look into this table in this new data analysis interactive table, I will see that, hey, it actually missed a few. Aerosmith is missing Celine Dion. So it might not always be fully automated, and it does take a little bit of human interaction to get this right, but by simply following up with something like fill in the remaining fields. There you go, it's on the case. And in a second here, we should have a new table I can download, and then we can do a final double check, which I would always recommend. There's just a certain amount of randomness and hallucination, but that's a feature, not a bug. It's where you get the creativity from. But then you do have to do some double checking. And then I'll just open all of these side by side. This is the original file, this is the one where we deleted some data and this is the resulting one. So if you look at Mariah Carey here, United States. United States. Celine Dion. Canada. Canada. Chris Brown and Jay z. Release year 2005 and 1996, 2005 and 1996. So as you can see from our spot check here, this looks to be flawless. Now again, always proceed with caution. And if it's absolutely mission critical that every single piece of Info here is 100% correct, then you should probably double check every single piece of info. Nevertheless, a massive time saver and many use cases. A quick spot check like this, where you just check five data points. If all of them line up, you're probably good to go. And those are the prompts I have for today. But wait a second, there is one more thing, one thing that takes a little more time than we have in this video to discuss and to show you how to use properly. I lovingly called it Sam the Prompt Creator, and it's a custom GPT that we've been developing for months. And what it does is it builds prompts for you, really complex and specific prompts, or it turns prompts that you already have into this more advanced version of the prompt. What's the advantage of that? Well, mainly if you're going to be handing them off, running them many times, or if you're going to be including them in an automation, that's where you really want to flesh out all the context. This GPT does it automatically for you. And I created a separate video talking about that and other situations in which you might not be needing prompt engineering at all, where following your intuition is the right call. If you're interested in learning about that, check out this video. And other than that, I hope this was helpful. Have a great day.