If you have no experience with ChatGPT but want to use the GPT-4 to its fullest, watch this video. In the next half an hour, I will go from a complete zero to pro. I've already made many videos about ChatGPT, but here I've combined everything into one video to teach you how to correctly write prompts, generate images, create custom GPTs, and much, much more.
So, open up your ChatGPT and repeat after me: I figured I should start with the basics and slowly go to more and more advanced things. The first thing we must do is create an account. Just go to chatgpt.
com, click sign up, write your email address and password—boom, we're in! Now, if you don't pay for the subscription, you will still get access to the GPT-4, the latest and fastest model, but you will only get a limited number of responses. So, to get the full advantage, I recommend getting a $20-a-month subscription.
Then you'll have no cap on messages and get all the new features on day one. Not sponsored by OpenAI, by the way! But if you don't have the subscription, after you reach the limit of free responses, ChatGPT will switch to the GPT-3.
5 model. It's smarter and faster than GPT-3, but not as capable as GPT-4 or GPT-3. 5.
You'll definitely feel this change in the quality of responses, so the subscription is the way to go. Here we are on the main page. On the left, you'll see the list of your GPTs and all chats.
Right now, I have no chats, but if we start one, it will appear here. A chat can be renamed, but I rarely do that; the AI is doing a pretty good job here. Slightly to the left, there is a drop-down menu that lets you pick which model you want to use—GPT-4, the GPT-3.
5 model. GPT-4 is the oldest one, and it lags in terms of smartness. There is also an option to create a temporary chat, which means anything you ask for or input won't be used for training the model, but I'd still avoid uploading any sensitive information.
If you click on your profile in the top right corner, a whole new menu will pop up. Here, you can check your ChatGPT subscription, the custom GPTs you create, or customize the ChatGPT with custom instructions. We'll get to that in a minute, don't worry.
Then there are some suggestions from ChatGPT, but they're mostly useless and can be modified. All the action is happening at the bottom in the prompt field. Prompting for ChatGPT is very important.
Most people never pay attention to it and end up with terrible text. Remember: garbage in, garbage out. The better your prompt, the better the response.
Sure, most people don't need a master's in prompt engineering, but even if you do a few minor adjustments, you will significantly improve the results. For each prompt, there are three things you need to include: formatting, style, and restrictions. The clearer your instructions will be, the better.
There is no one format for prompting; you can just describe what you want. So, let's say I want a pasta recipe. I take my simple prompt and add all these three things: explain how to cook pasta as if I were a total noob.
There's the style—ChatGPT will write it in a very simple manner. Now, you can directly ask for an academic style or for a scientific one; pick anyone. Format it as a step-by-step guide with a full list of ingredients and cooking times, as well as a calorie counter.
Now we've defined the formatting for the response. As far as formatting goes, go crazy—ask for tables, bullet points, lists, and so on! Lastly, restrictions: don't use complicated recipes; I want the basics.
I recommend adding some restrictions to minimize the chance of hallucinations or going off course. Now, ChatGPT knows exactly what I want from it, and the actual response is much better. I also like to provide ChatGPT with some additional context.
Try to give it more information that's actually relevant to the task. For example, "I am a student in my last years of college with a GPA of 3. 5.
I have an assignment to write an essay about the French Revolution and its impact on the world. Can you give me 5 to 10 bullet points? " Any context gives ChatGPT more information to work with, and don't worry; it won't interfere with your style choice or formatting.
And of course, you can always ask ChatGPT to elaborate on something from its response. You can ask questions, and it will answer. A good way to learn these basics would be through prompt tools.
By the way, we’re starting to list our favorite tools on our new website, aimmaster. me. There are quite a bit of AI tools here already, the ones we've tested ourselves, like this Brisk, human-like AI generation.
Slowly but steadily, we'll add more tools, but only the ones we are sure about. One of the best tricks you can use is roleplay—asking ChatGPT to act like someone is super effective when you need to use specific language or access certain data. For example, imagine you are a chemistry teacher with a Nobel Prize nomination, and I'm your high school student trying to understand ionic bonds.
If you got the Easter egg here, leave a comment! This response is super technical and scientific, so I'll just edit the prompt to this one: "You are a high school student, and I'm your 10-year-old brother. " Now, if we compare the responses, we'll see a huge difference in how they are written.
The first one uses complex concepts, while the second one turns. . .
To analogies every time you want ChatGPT to explain something; or if you want to be particularly good at certain tasks, make it play a role. Those analogies can be pretty powerful too! Don't forget to ask for them when you need ChatGPT to teach you something.
To get even better results, I like to give examples for ChatGPT to follow. Again, the more information it has, the better the output. Typically, ChatGPT does a pretty average job when you don't give it anything to be guided by; the text it generates just screams AI and can really lag in structure and consistency.
So, if you have a particular text in mind that you want to get, make sure to include it in your prompt. For example, I first asked it to write me 10 email templates, and then I showed the example of the email contents. All 10 emails followed the structure I asked for.
You can also do the same with general writing. Check this out: I will first ask ChatGPT to create an outline for an article, then I will ask it to analyze my writing style. Remember this analysis part; we'll need it later when we get to custom instructions.
Now, I can just ask ChatGPT to write me an article following the outline from earlier and copying my writing style. I must say, the writing style in this output is spot on! Pretty much, the sentences are structured like I did.
The words it chose, with some minor tweaks, would be practically indistinguishable from my writing. Here, you have to be cautious; too many examples can confuse ChatGPT and lead it to severe hallucinations, repetitions, and so on. If the prompt itself gets too big, you can always divide it into smaller chunks, like I just did.
First, I asked it to read and analyze the example, and then to write its own version. If I had put everything in one prompt, this would have led to some crazy bad results. Dividing into chunks can be really effective, even within one single prompt.
Instead of giving a long and super detailed sentence, break it down into smaller ones. This will minimize the chance of hallucinations and any confusion. For example, if I have a small math problem to solve, I will just separate it into smaller sentences for better understanding.
Oh, and if your task is too big, ask ChatGPT to think step-by-step. This makes the output a bit better and more coherent, forcing ChatGPT to use reasoning better rather than its memory. ChatGPT usually knows when to do this by itself, but it doesn't hurt to edit manually.
Another thing that I like to do is reasoning. ChatGPT is really good at reasoning and giving arguments, but sometimes can get tangled in its own words. So here's what I usually do: I first ask it to convince me of something, and then I ask it to convince me of the opposite thing.
This gives me the best of both worlds; both sides of a coin get equal treatment. To make this even better, I just asked ChatGPT to create a comparison table for me and give an opinion based on the data. I wouldn't really use it for making life-changing decisions, but for some quick assessments, this can be super useful.
When prompting ChatGPT, you need to approach it like you are explaining something to a person who doesn't know anything. The things that are obvious to you, like the order of doing things, are not obvious to AI, so just spend an extra 30 seconds refining the prompt. Okay, oh!
And don't worry about correct spelling; ChatGPT can understand you even if the whole word is misspelled. Now, if you're ready to take your prompting game to the next level, you should try nested prompts. Nested prompts are kind of like the combination of everything from above, but in a very short form.
Nested prompts work particularly well when you want ChatGPT to talk about specific things. For example, I can ask for a story with a one-sentence prompt, but I will have no control over the content. Whereas, I can do a nested prompt and explain in detail what information I want to see.
I will just give four main points and let it fill in the rest. This somewhat falls under the umbrella of giving examples, but I'd say it is a part of being detailed. Remember, if you don't know how to prompt something, you can always ask ChatGPT for directions.
So, you've mastered nested prompts and can make ChatGPT write like you—what's next? How about customizing once and for all to be truly yours? That's my weird segue to custom instructions.
You set them up once, and they get applied to all of your new ChatGPT interactions, kind of like the baseline for ChatGPT to start with. To set them up, click on your profile, customize ChatGPT. Here, you can see two windows.
The top window is for the information you want ChatGPT to remember. It can be a description of you and your needs, your writing preferences, your background, etc. Just make sure not to include any data like credit card info, Social Security numbers, and so on—nothing sensitive, right?
Gosh! You can even paste a writing style analysis we did earlier here, and ChatGPT will use it as a reference for future outputs. The bottom window is for actual instructions.
Here, you describe how ChatGPT should behave, how it approaches tasks, the formatting, anything. I, for example, like to include this set of commands: never mention that you are an AI. Without it, sometimes it can mess up, and if you're not proofreading the output, you can end up with "I am an AI" in the middle.
Of your performance report, then I asked to avoid any language constructs that could be interpreted as expressing remorse, apology, or regret, and to refrain from disclaimers about you not being a professional or expert. This way, if I asked for medical advice or dietary suggestions, I would get exactly what I'm looking for. I also asked Chad GPT to break down big tasks into smaller chunks to keep responses free of repetition, etc.
This is the basic set of commands I give it just to have the responses as good as possible. There is a much more interesting way to use custom instructions. Let's say I'm a business owner and I need some copy for my social media.
Creating a new chat each time and explaining what I want is not an option, so I just create custom instructions and type in all the information about the business in the first field, like the address, the name of the business, some selling points, etc. Just the basics I want to see in every response. Remember those email templates from before with custom instructions?
Those won't be templates; they will be full-on letters. And next up, in the bottom menu, I asked Chad GPT to play the role of social media manager, describing how it should write and all that. Now, if I ask for an Instagram post, it will not only look like something human-made but also have all the business details ready.
The best part is that you can do the same thing for pretty much anything: emails, blog posts, even image generation. But we'll get to that in a second; I promise. Everything I showed you before works with the existing training data, but this data is limited, and Chad GPT doesn't really know what's happening in the world.
To fix that, go to customize Chad GPT and enable browsing. Now your Chad GPT can not only use its own data but actively search for new stuff online. You can just ask Chad GPT something like, "Are there any plans to create a new version for Chad GPT?
" and it will search articles online and give an answer. Here it says GPT-5 is in development. So I can follow that up with a question about that.
I can ask it to compile all the recent rumors, and it will actually do that and even give links to the sources. The same way, you can search for the news, football matches, space, ball games, and so on. Chad GPT is really good at finding new information.
What's more, you can ask it for stock information, like, "Please find me five best performance stocks to buy right now. " I will also ask for projected earnings just to make it more interesting, and boom, here are the stocks. What's really helpful to know about web browsing is that prompts for it follow all the same rules as regular prompts, so you can specify the format of the response, the writing style, and so on.
Web browsing has many more use cases to it. For example, let's say you run a business, you have a website and a blog that you use to rank better in search. You can just feed Chad GPT the link to your blog and ask it to analyze the articles for SEO keywords, just like I did right now.
Then you can ask it to help improve the keywords, but be sure to mention how you want your site to be ranked. Like I said before, you need to give Chad GPT a direction, and boom, there you go—a list of better optimized SEO keywords. Of course, you can write an article with them for your business; this feature is a lifesaver.
Another thing you need to try is summarizing web pages. Though this part is a bit tricky; with some websites, it works like a charm, but other websites have some sort of protection in place, so Chad GPT can't read their contents. What you can do is either copy and paste the entire text of a web page or feed the PDF version of it as a file and get the summary.
This way, I will show you how to work with files a bit later. But right now, I want to show you a simpler solution: Custom GPTs. Custom GPTs are probably the closest thing to AI agents right now.
They're pretty much the same as Chad GPT but very skilled at one particular thing you train it on. For example, you can make Chad GPT interact with different platforms and APIs all under the hood, but I'm getting sidetracked. To find these GPTs, click on "Explore GPTs," and you will see the store with tons and tons of GPTs.
Some of them were created by big companies for easier interaction with their products, while others were made by regular people like you. When it comes to GPTs from big companies, I'd honestly just go for the real deal; not all of their GPTs have enough features. Again, that's why we've created aim master.
me. We add tools gradually as we test them. Some AIs are better at writing certain texts; others are great at generating specific images.
To find a GPT, open up search and type in, for example, "web browsing," then pick the one you like and start a chat. That's it; your custom GPT is ready to go. Now, with this GPT, I can paste the link that regular Chad GPT was unable to open, and boom, here's my summary.
Another cool GPT is Consensus; it's perfect for doing research. It was specifically trained to work with research papers. I can just ask it for something, and it will give me the correct answer with links to sources.
As you see, using these GPTs isn't all that different from using the regular Chad GPT. That's true, but what I haven't. .
. shown you is how you can create your own custom GPT just for your needs. You can make it part of your business or just automate certain boring tasks that involve a lot of prompting.
To create a custom GPT, find "Create" in the top right corner. Now you'll see the interface divided into two halves. In the left one, you give all the instructions and do all the tweaking, while on the right, you test everything.
Creating a GPT is really simple; just answer the questions that ChatGPT asks you—that's it. It will ask for the name, the icon, and a few more clarifying questions, like, "What do you want this GPT to do? " Answer however you like, but remember the prompts and tricks from earlier: be direct and clear in your instructions.
The better you explain what you want, the better the results will be. I recommend testing the GPT right when you're creating it and making all the adjustments on the go. Give more instructions, check the results, give even more instructions, check again—do it until you're satisfied.
It all sounds more complicated than it really is. In just over a minute, I created my own GPT for Crea and Comics; how cool is that? There’s also a second way that you need to learn: switch to "Configure.
" Here, you have more precise control of what you want the GPT to do. You can give it direct instructions and even upload your own data. First, you type in the name and description, and then switch to the instructions window.
Here, you need to tell your GPT what to do. For example, if you often do repetitive prompting for work, you can paste all the prompts here into the instructions, where you can make it access an external server connected to apps like Zapier and so on. Here you have almost endless possibilities, and in the knowledge section, you can upload all the responses ChatGPT gave you, where in my case, it was the script for a couple dozen videos on this channel.
I recommend uploading all the text information in the simplest TXT format just to make it easier for ChatGPT to read the files, but you can also upload PDFs, images, Excel tables, custom data sets, and so on. At the bottom, you can also enable or disable web browsing and image generation, but I recommend leaving them enabled unless you want your GPT to use only the data you uploaded. What you must turn off 100% is hidden under the additional settings: "Use conversation data" and "Your GPT to improve" or turn this off if you don't want your company data or personal info to become publicly available.
Now, with custom GPTs mastered, there’s one more thing about them you need to learn: calling them out while chatting with ChatGPT. You can call out a specific custom GPT to do something. For example, I can call out that research GPT from earlier and ask for some data.
Then I can call out another GPT, for example, this one that draws diagrams and does visualizations. This GPT will use the data from the previous one and generate this mind map. This way, you can create really complex pipelines for your workflow.
Let's say you first need to analyze the document a certain way and pick up specific keywords. You first create a custom GPT just for this task and then call it out—summon it like a wizard. Then, if you need to take that data and turn it into a blog post, you can ask regular ChatGPT to do that, and for images, you can straight up summon DALL-E.
All these GPTs will work together, share information, and not interfere. That’s probably the one thing that really can take you from a noob to a pro. All I showed you before revolved around text inputs, and that’s how 90% of people use ChatGPT, but if you want to be a top-tier user, you have to learn how to use the vision feature.
ChatGPT can actually recognize objects in photos accurately and work with that data. For example, you can do simple stuff like asking it to describe an image. Here, I have a photo of a cat, and the description is spot on.
Or you can take it one step further and introduce some text with this image. It shows the water cycle, and ChatGPT had zero issues seeing that and accurately describing it—it recognized all the text, all the pointers, and pieced it all together the right way. But all this is pretty simple; ChatGPT is actually incredibly good at text recognition.
You can feed it a full-on screenshot, and it will extract the text word by word. Check this out: here’s the image with a ton of words. All the words have different sizes, slightly different fonts, and it all just looks like a mess, but ChatGPT, once again, accurately extracts every single word and arranges it in alphabetical order, like I asked.
If you are a student, this would be indispensable—really think of all the lecture concepts and how easy it would be to transform that into plain text. Another thing Vision can help with is turning your handwritten scribbles into code. Here, I have a poorly made diagram for a simple program.
I will upload it and ask ChatGPT to write me code for it, and it did. It also wrote a detailed explanation of how this program works. This might seem like magic, but the magic here is quite limited, and you need to remember two things: first, it can create a similar diagram without you calling out a custom GPT; secondly, even though ChatGPT can hear you, the web version cannot do anything with audio.
If I upload a simple voice recording and ask ChatGPT to transcribe it, it will get nothing. The same goes for. .
. Videos: it will try to extract the audio from a video but then fail at transcribing. So right now, you'd still need to use custom GPTs for transcribing your audios or videos.
In this case, I'd still recommend finding a standalone AI tool. Again, aim. me is a good place to start; here we handpick and personally test the best AI tools.
Okay, ChatGPT Vision is cool and all, but guess what? It's even cooler if I whip out my phone. The ChatGPT app on the phone has pretty much the same feature set and even more.
This app is seriously bonkers, and I urge you to install it right now. In this app, you can access all the same chats from the menu on the left, open up custom GPTs, and interact with them, even generate images. But all that is easy to learn—I'm sure you'll get a hold of it in a minute.
What's really cool about this app is how simple it gets to use Vision. You hit the plus sign in the bottom left corner and choose camera. Then you just take a photo.
If you want the app to pay extra attention to certain things, you circle them, which can be super useful if there are a lot of things in one photo. Then you just ask ChatGPT what it sees. You can just type your prompt or hit the little microphone icon and dictate it.
The quality of translation is actually amazing—super accurate! Vision with the mobile app gives you so much power; it can recognize landmarks from unusual angles, foot dishes, and give recipes for them. It can even read medical images like X-rays.
Even memes, it gets them. It's like having a personal, all-knowing, all-seeing helper. But what makes this app a must-have is the voice chat feature, and that is something you absolutely have to use.
There is a headphone icon in the bottom right corner, and if you click on it and wait for ChatGPT to connect to servers, you will see a big wide circle, and then you just talk like you would with a regular person. For example: "What's the weather like in Austin, Texas? " "I don't have the latest weather report for Austin, but it tends to be pretty hot in the summer, often with temperatures in the 90s F.
" "Tell me about the upcoming baseball game in Austin. " "Austin doesn't have a major league baseball team, but they do have a minor league team called the Round Rock Express. Their games are a lot of fun.
" "How much are the tickets? " "Ticket prices for the Round Rock Express can vary based on seating and the game itself, but generally, you can find tickets ranging from about $10 to $30, based on the layout of the stadium. " "What are the best seats I can get, and how much would that cost?
" "The best seats at Dell Diamond, where the Round Rock Express play, are typically behind home plate and along the baselines. These seats offer a great view of the action and range from about $20 to $30, depending on the game and seating section. " "Are you looking for something right up close to the action?
" With this voice chat, you're not limited to anything; you can talk about anything, ask any question, and so on. Just keep in mind that the responses will be verbal, so if you ask for images, you will see them only when you go back to the chat view. I already showed you basic prompting, web browsing, image recognition, and custom GPTs.
By now, if you've been following exactly what I did, you already know more than 90% of ChatGPT users. So let me teach you image generation because if you master this, you will be unstoppable. To generate images, you don't need any special GPTs; you just need to check if you have DALL-E enabled and then just type what you want in a chat.
If you've ever tried tools like MidJourney, then you know that correctly picked parameters result in a good image. With DALL-E, the situation is slightly different. DALL-E actually uses the same text recognition as ChatGPT, so you can actually just ask directly what you want.
For example: "Generate four images in various styles depicting the green apple falling on Isaac Newton. " As you can see, we don't get four separate images—we get one image with four panels. When writing DALL-E prompts, try to be as clear as possible.
If you want four separate images, directly state that you expect four separate images. The more descriptive you are, the better. I recommend describing each element in detail.
There are no particular parameters that you need to use like in MidJourney; just be as descriptive as possible—specify the aspect ratio, the orientation, how wide the angle should be, and pay extreme attention to describing each element in the shot. Objectives help in refining the image. For example, instead of saying "a dog," say "a fluffy small brown dog.
" DALL-E will actually autocomplete the prompt for you, even if you type in only a few words, and if you struggle with that, just ask ChatGPT for help—it will fill the prompt with tons of objectives and additional descriptions. Now, let's say you have an image and you want to change a part of it. You can do it in one of two ways: one, write "replace something with something.
" In this case, you can only hope that the image will look similar to the original one because DALL-E will actually regenerate the image from scratch. Sometimes this works well, but sometimes the images it generates are only remotely similar to the ones you tried to edit. A better way would be to open the image and at the top click on the brush icon.
With this brush, you can simply select the object you want to change and. . .
The prompt window tells Dolly exactly what to do: replace with a basketball ball. Now Dolly will work only on the selected area without touching anything else. It's great for masking small adjustments here and there.
Another thing you can sort of do is copy the images you upload. If you upload an image and ask Dolly to replicate it exactly, it will actually turn it into a prompt and generate its own version of it. This, by the way, can be a great way to show Dolly what you want to get in the output.
After a few iterations, you can actually get a pretty similar image. Unfortunately, there is no way to just edit an uploaded image; you'll first need to recreate it, and only then edit. There's a standalone Dolly website, but it works with the older model.
However, there are plenty of alternatives that let you generate content more easily. For instance, an AI platform for faster visual content generation and realistic AI dubbing in different languages, which you can find on aim. me.
By default, all Dolly images have a pretty low resolution, but you can just ask Dolly to upscale it to any resolution you want, and then you'll get a link to download a file. The same way, Dolly can downscale images, turn them grayscale, and make all sorts of adjustments. But the top level of ChatGPT is data analysis.
Many people don't know this, but ChatGPT has a special code interpreter built in, which is insanely good at working with different data sets: big data sets, small data sets, CSV files, archives—ChatGPT does it all. Here's how it works: you start by uploading the data set. I upload mine in an archive, and ChatGPT will unarchive it by itself.
Then I always recommend asking it to analyze the data set just so it fully understands the contents. Usually, this leads to a small overview of the data set. Sometimes data sets come with useless data or just poor formatting, so next up, you have to ask ChatGPT to clean and format the data set.
After the cleanup, ChatGPT will suggest a couple of options on how to proceed, like visualizing something or analyzing trends in something. If you're new to data analysis, these suggestions are a lifesaver, but even without them, you can just tell it what you want to see. Now, I will upload an archive with two data sets and show how ChatGPT can cross-reference them.
The overview says one data set is oil consumption, and another one is oil reserves. Can you cross-reference these two data sets and find any correlations? This takes time to process, but check this out: ChatGPT analyzed the data from both files and gave a short explanation.
ChatGPT can also create graphs using data from multiple data sets. I want to see a visualization in the same graph of how reserves and consumption change over time, and here's the graph. But it's not your usual graph; it's interactive!
I can hover over parts of it, and it will display the corresponding data. Unfortunately, not all types of charts are supported in this interactive mode, but it's still super useful. But I'm still not done, because ChatGPT can predict trends based on the data.
Here it says that oil consumption will grow at least until 2031 and even gives a graph. Data analysis with ChatGPT is the top level you want to reach. Really, if you can master data analysis here, you're pretty much a god of ChatGPT.
Now I want to give you a few more simple tricks that can make your life easier. ChatGPT might be amazing at understanding what you want, but it still operates on math, and math has parameters. So here are a couple that you should remember: to make the text more or less random, use the temperature parameter; type in 'temp = value' (where lower values give more focused outputs while higher ones give more creative and varied texts).
To stop ChatGPT from using tokens that were previously used, use the diversity penalty; again, higher values increase diversity while lower values make text more repetitive. If you want ChatGPT to use less common words, you need to set a frequency penalty: positive values lead to ChatGPT using rare words, while negative values give more common language. And like a cherry on top, I really recommend you learn keyboard shortcuts!
You can find them by clicking on the question mark icon. I often use them for creating new chats, deleting chats with poor responses, or quickly accessing custom instructions. And if you want to really know your AI, you'll need to master other AI tools as well.
In the last years, I've tested hundreds, if not thousands, of AI tools and put the best ones on our website, aim. me. It's not that difficult to master ChatGPT, especially given how good it is at understanding; the main thing you need to remember is to be as descriptive and direct as possible.
Treat ChatGPT like a person who doesn't understand how you think but follows the instructions perfectly. And of course, just practice! Go into the rabbit hole of custom GPTs, find the ones you like the most, create your own GPTs, and now you know how to do it: where to click and what to type.
Give it some time, and you will become an AI master yourself. Thanks for watching, and see you in the next one!
