Transcription
You have probably already started using ChatGPT, right? It's not difficult and allows you to solve many tasks or get answers to complex questions in seconds. With ChatGPT, I write code without knowing programming and have created several plugins for the video editing program that simplify my work.
When I need to calculate complex numbers in my projects, I don't just ask the chat to do the calculations; I also ask it to write a program that will correctly handle complex mathematics. If I want to create a cool video on YouTube, I send the chat scripts from foreign authors who have gathered millions of views and ask it to rewrite them in clear Russian without the flaws of automatic translation. I extract the best parts from these videos and compile the perfect video. For another channel, I simply do a direct translation without recording the videos myself; a neural network reads them for me. By the way, that channel already has about 15,000 subscribers and attracted around 1,200 paid users to my neural network service. One paid user in this niche costs at least $10. So, I can say that ChatGPT helped me save about $12,000 in advertising just on this task.
But here’s the problem: now everyone says that there are no bad questions. However, in the world of ChatGPT, there are. The chat itself won't tell you this; it will just give you low-quality answers, and you'll waste a lot of time messaging it to get the right answer. This really annoyed me, so I decided to spend a lot of time learning how to maximize the benefits of using ChatGPT.
My first piece of advice is to use custom instructions, or simply put, assign it a role in the role field. Just describe who your interlocutor is, an expert in what field. If regionality is important—for example, answers to legal questions depend on your region—it's worth indicating that ChatGPT is a lawyer specializing in a specific area of law in a specific region, like Belarus. In the same window, let it know how you would like ChatGPT to respond. You can specify the format of the answer, for example, "provide the answer in the form of a table with the following columns" and list the names of those columns.
You can decide whether ChatGPT can express its opinion on topics or remain neutral, as well as whether its answers should be long or as short as possible. When I write scripts, I divide them into parts and say that each part should be no more than 300 words. Then, from about eight parts, I can assemble a decent video of 15-20 minutes. If I don't like a specific part of the video, I won't ask it to rewrite the entire section; I'll say, "this specific part is bad, rewrite just the third part." This way, it won't have problems with clutter and will handle this specific task well. I just gain more control.
Today, I will show you how this works, and it can really make a difference for you because you'll eliminate a lot of unnecessary back-and-forth and save time. I use this site specifically because it doesn't require a VPN, accepts Russian cards, and has the official ChatGPT connected. Also, because this is my site, and I gave myself a subscription for 2.87 days. I hope it will be enough.
Let's say you are good at Excel and use ChatGPT to help you when you get stuck. But you don't want to see and read all those explanations of solutions; you just want to see ready-to-use formulas. So, we can write to it in the role: "When I ask for formulas for Excel, just provide me with the most effective formula without any explanations." Then, when I ask for formulas, it doesn't explain anything. It just writes, "Price = formula," "Final price = formula," and I can copy and check.
If you were a programmer and didn't want to see all those explanations about the code, you could write, "When I ask you for code, just give me the most effective version without additional explanations."
Now, let's say I'm in Excel, and I have problems calculating something correctly. This is a simple example, but there can be much more complex tables with really complicated formulas. In one of my tasks, I needed to use geometric progression; in another, the factorial of a number over thirty iterations. This can really break your brain, especially when you're doing something late at night and your mind has already been overloaded all day.
When I need to calculate some mathematical examples, ChatGPT has saved me in those cases and always does. What I do is take a screenshot of the table along with the column headers and cell numbers and just throw it into ChatGPT, asking, "Write me formulas for Google Sheets."
Look at what it responds: "In column F, insert this formula; in column G, insert this formula." Let's copy the code and check. So, I inserted the formula, and the value worked immediately. I stretch it to others and get working values as well. This is a simple example, but imagine how it will help you when you have a really complex task.
And look, there are explanations here, but there are times when you need to create 10 formulas to fill in a large amount of data, and they might not even fit in one response from ChatGPT. So, we give it our hint, and now let's repeat our request. Look how much shorter its response becomes when you don't need to read a ton of extra text but just copy the ready result. This will also save you context in ChatGPT.
Many of you have noticed that for each new task, it's more effective to create a new chat because the old chat gets cluttered with information, and it can slow down or get confused. But with such a role, its context will get cluttered much less because it doesn't waste time reading the opinions it gave you in previous responses.
Another piece of advice you can include in the role is to ask ChatGPT to always inform you about its confidence level in its answer and, for example, how many characters it accepted. The thing is, when there are too many messages in the chat, it starts to throw out parts of your conversation from the context. You might notice this too late when half the work is done, and you'll have to delete the chat and start over.
When you write code, for example, you can send it 300 lines of code, but it might only accept 250, and you won't notice this until you see that it has lost part of the code during your work. So, just ask it at the end of the message with the code, "How many lines of code did you accept?"
Let's write a role for an example. Here’s the text I came up with: "Follow each point in every response." You can have five, six, or seven points; I wrote just two for example. "At the end of each response, inform me of your confidence level in your answer in percentage. At the end of each response, briefly list what other options were not included in your initial response, separated by commas."
Now you'll understand why this is needed. I click save.
Let's say we are making a YouTube video about Coca-Cola and want to gather facts. Then it writes five facts about Coca-Cola: the history of its creation, the secret formula, global popularity, and also states its confidence in the answer. This indicator might signal to you that the chat context is getting cluttered, and it might be time to create a new chat.
Now look at the other options that were not included in the answer: "Use in art, impact on health." When you create a video and select interesting facts, you sometimes understand better than ChatGPT what would suit you. So, let's write, "Let's talk about the impact on health," and it writes what it knows: all about sugar content, acidity, caffeine. But there are also other response options: "Impact on energy levels." It turns out Coca-Cola might drain our energy. This could make a great headline.
"Allergic reactions." In one of its other responses, it mentioned that there might be an addiction to this product. Let's explore the impact on energy levels, and it writes that first there is an energy boost, then an energy crash, then addiction, and finally long-term consequences. Imagine the suspense you can build in a video about Coca-Cola by exploring all these seemingly unimportant options that it didn't include in the initial response but did include in our little hint.
You could also say, "At the end of each response, briefly describe what options were not included," and you could write a minimum of 10 options, and then it will list at least 10 options separated by commas.
Now let's talk about the impact on sleep. It will write the main points that it thinks affect sleep: caffeine, reduced sleep quality, addiction, withdrawal effects, and influence on circadian rhythms. Here, it gave three options again, but that's because the chat was created with that role. You just need to recreate a new chat, and you can regulate your work with ChatGPT, gaining more control over it.
In the end, you are not the one being led; it does your work as you see fit. Now, let's move on to the main part of the video: the life hacks.
The first one is: if you ask ChatGPT to write a text, article, or script for you, the results will likely be quite mediocre and rather template-like, even if you give it custom instructions about tone, style, etc. This might not reflect your writing style, for example. But you can teach it to write in your style or in the style of your favorite blogger by providing just a couple of examples.
Let's go to YouTube and find any video from any blogger. For example, let's type "Fishing," and suppose we take this video where a guy talks about some fishing life hacks. Let's say we liked this video in English and want to make a similar one in Russian. We write in Google "grabber," go to the site anti.com, paste the link to this video, and click to get the text of the video.
But the problem with YouTube is that it doesn't place periods and commas. So, if we just copy this video and go to any online translator, like DeepL, we will get a complete mess because, firstly, automatic translation doesn't always work well and often makes mistakes, and secondly, we won't have any periods or commas. This text is completely unusable.
So, we open a new chat, paste the script we got, and write the following request: "Place periods and commas without changing the words, and for example, I only need part of this script." Look how in seconds our text turns into a ready-to-translate Russian text.
It still retains our condition about the probability of the answer, and so on. Don't forget to delete and clean the role. Now you can paste this text and write, "Now rewrite this text in Russian, adding meaning where it disappears due to translation."
Now look at the quality text you get: "Hello everyone, my name is so-and-so. For those who don't know me yet, welcome! For those who are already familiar with me, thank you for watching this video again. This video is based on a clip I shot in the spring about the tricks of storing fishing gear. It might have appealed to many, and this is a ready text for recording."
This is exactly how I translated scripts from these videos for this channel, one of which gathered over 100,000 views and brought in a decent amount of traffic. These videos were created almost entirely automatically.
Now let's create a new chat and write it: "Here's an example of a YouTube video script titled 'Types of Fishing' for my client." Suppose you have a client who orders from you, "I want you to keep the entire narrative style and manner of speech and use it when I ask you to write a script for me."
And here's an example. For now, don't respond. It will still respond because it has to write something, even if it's just a little. If before we figured out how to translate English videos into Russian, now suppose you have a reference and want to create a new video from scratch.
"I want to write a script for YouTube about the top five lures for pike. Write me a script structure and break it down into parts." In the end, it writes us what parts we have and what goes into those parts: a story or anecdote from personal experience related to that lure.
For example, I use the division of large texts into smaller parts to have more control over the situation. If I don't like a specific part, I will give it edits just for that part, not for the entire text.
Now let's write, "Write the script text for part one using the style and manner of the script for 'Types of Fishing.' This part should be no more than 300 words."
Now look, it greeted us exactly in the author's style. It writes in the same words as Benjamin Novak would. If this were your Russian client, they would be genuinely surprised at how accurately you write the script using their language, phrasing, and manner of speech.
It continues to develop the text: "Pike fishing is an exciting adventure. Why is it so important? The pike is a predator and requires a special approach." This really looks like an interesting start for fishermen.
The next life hack is self-criticism. You can ask ChatGPT to look at your own text and leave feedback on it. It sounds strange, but it really works. For example, I asked it to give me a brief description of why Python and Excel can work well together. Its answer again looks quite template-like and standard.
Now we write, "Be ruthless with the text and tell me how to make it better." Look, when you don't ask it to actually change the text but just say, "Give me recommendations on how to improve it," it will give you a lot of recommendations.
At this moment, it doesn't expect you to make it do this, so it will write as much as possible. If you immediately say, "Make this text better," it will give you one recommendation at most two. But now we have six recommendations.
Now we can write, "Apply all six recommendations and give me the perfect text." Now look at what it does: it gives an updated text that takes into account all six recommendations it wrote above.
The same works when, for example, I write HTML games. If I just ask it, "Let's make this game better," it will make some minor changes. But if I say, "Write a list of points on how to make this game better, what further steps I need to take," it will give you an impressive list.
Then you just catch it on that same word and say, "Now do all six points." The main thing is that at this moment, its context window doesn't run out because it might start ignoring some points simply because they don't fit into its so-called operational memory.
Then you just copy those points, go to a new chat, paste the code, and throw in those points, and it will start implementing them.
The next piece of advice is to let ChatGPT ask you questions. How about using ChatGPT to optimize your own prompt for ChatGPT? It turns out you don't even have to do this yourself.
Now people think that whoever writes good prompts is the king of the planet, but ChatGPT can do this too. Here’s how you can do it: suppose I want to create a content plan for reels for a week to get the perfect prompt. I can ask ChatGPT, "Write me three ideal prompt plans for reels for a week."
Before writing anything, "Ask me questions until you are sure you can create an optimal prompt for me." Ask questions one by one, gradually correcting them.
Look, I didn't specify what my blog's theme is, so it asks. I write that it's about neural networks. "Who is your target audience?" "Beginner specialists in technology or just people interested in new technologies?" I write, "Just people using neural networks."
Next, it asks, "What goals are you pursuing?" I say, "I need views on Instagram so that people learn about me." It writes, "Here are educational videos, practical cases, examples of using neural networks, trends and news, interactive elements, success stories, behind-the-scenes, myths and reality."
Now, suppose I want educational videos. I say, "Give me a content plan for educational videos," and now it outlines what videos should be released each week and each day. If you want a specific number of videos on a specific day, you can give it that information so it can prepare a content plan using those data.
Ideally, before writing any content plan, you should teach it which videos have gathered views on Instagram. You can find five viral videos on Instagram, download their text using similar tools to those we looked at for YouTube, but for Instagram. I'm sure something like that exists.
Show it the texts and phrasing of popular videos and ask it to write ideas for videos in the style of those viral videos, as well as ask for scripts for those videos using examples. In other words, teach the chat using examples of successful videos.
Now, without training, let's write it a simple request: "Give me a table with columns: day of the week, topic, idea." I want to see seven videos for a week, a table for two weeks.
Now look, you get a table in seconds: "Monday: Introduction to neural networks; Tuesday: History of neural networks; Thursday: How neural networks are trained." I saw a video with such a title that gathered quite a few views because people are genuinely interested in how computers understand what to respond with and how to formulate those answers.
Such tables: "Neural networks in medicine: How Elon Musk is restoring vision to people using chips that also utilize neural network training." "Neural networks in art: How much were the most expensive images generated by neural networks sold for?"
When you read the content plan, you might come up with a new idea, like "How much were the most expensive images generated by neural networks sold for?" This is exactly the case where the context of ChatGPT has become cluttered, and I forgot why I initially created the prompt.
I asked it to write me three ideal prompts for creating a content plan for reels for a week. Knowing all the previous information, it now creates requests that will likely lead to the desired result.
Your topic, for example, is artificial intelligence. The theme ideas for videos that will be interesting for my audience: "Specify the topic of each video, a brief idea, and the main essence that should be conveyed."
If you don't know how to write a prompt for ChatGPT, you can ask it to train you in that regard.
The next piece of advice is to shorten often. When you ask ChatGPT to write something, the response can be quite long and tedious. So, one way to avoid this is to first ask it to adhere to a certain word count, as we did when creating the script.
But you can also literally write it a phrase: "Rewrite this text, shortening it by half and leaving only the most important parts," or "Leave only the most interesting parts for a certain target audience."
For example, going back to the text about fishing, we got four paragraphs. "Rewrite this text, shortening it and leaving only the most important parts." We got the same four paragraphs, but they became much shorter and now fit on one screen.
Although, as I see it, it didn't delete any unnecessary information, and the main essence was conveyed correctly.
I hope you found this video useful. If you want to get a bunch of neural networks for free, like the latest version of ChatGPT, Sunet, Haiku, and also stock resources like Iwata Elements and Freepik, subscribe to my Telegram channel, where I give away subscriptions to all these neural networks every week. The link is in the description. Good luck!