Transcription
Write me a story. Yes, good. Make more words. More like a human. Not like ChatGPT. Write like a human. I am writing like a human. Whoa, whoa, take it easy. Yes, absolutely. ChatGPT has quite a lot of difficulties, especially with its long-term operation at a distance. You really start to see them clearly. For example, its sycophancy, where no matter what idea you decide to discuss with it, it always says: "Yes, that's a great idea. How did you come up with it? Oh my god, you definitely need to take it and do it, even though it might not be like that at all." Quite often, it produces robotic answers where it over-structures the text. It can also make up facts. Well, and the saddest thing here is that it can invent non-existent sources for them. Well, and the fifth thing, it gives up quite quickly with complex tasks and says that I can't do anything here. For example, this is when working with a large volume of data, although it has built-in functions where it can use a processor and give you a great answer. Therefore, today we will take off its rose-tinted glasses from ChatGPT and turn it from a flatterer into a truly honest critic. We will force it to write as if a truly intelligent educated person were communicating with us, without this robotic text. And we will activate live speech. Plus, we will additionally force it to check the accuracy of the facts it gives us. Well, and most importantly, you will have one powerful system prompt instruction that will eliminate all these shortcomings once and for all, and you will no longer have to explain to ChatGPT every time how it should work. And the first thing that actually often annoys in ChatGPT is that its default settings are arranged in such a way that no matter how much nonsense you ask it within the framework of an idea, a startup, some new concept, it always acts as a kind of flatterer. And in this case, I ask it a question: "How do I launch a Bitcoin mining startup?" And I already anticipate that it will answer me: "This is a good idea, a great startup, let's implement it." Well, in principle, we will see all this now. A brilliant idea. Also, this rocket will definitely be in the instructions here, we will remove these excessive smileys, these excessive hyphens, this structuredness. And in комплекте, it is impossible to process all this correctly and make a balanced decision. This is, of course, a clear drawback that we will eliminate today, as it is quite a harmful thing. And here we get a deformed picture of the world. And not all users have critical thinking, which is misleading. Of course, you can write here: "Act as a top manager of a large IT company, such and such, which is engaged in this and this. But even in this case, it will not globally change the picture." And writing such prompt requests every time is excessively tedious. And the solution here is to use the system instruction in your profile for all of ChatGPT. To activate this instruction, you need to go to your profile settings. To do this, click on the avatar and go to the settings mode. Next, the most important thing is that we need to switch the checkbox to the activated mode. That is, we enable these settings for all new chats, so that you don't have to write every time: "You are an experienced copywriter, you are an experienced top manager." Yes, here we provide additional information. We will work with the main characteristics that your ChatGPT should possess. And then we will fill in the additional fields so that all these instructions work together qualitatively, and if at some point you can calmly disable them. These are precisely these additional fields that we will also fill in. Make sure that in the settings mode, advanced features are enabled, the functions that you need, which you use. And here we specify the specific instruction. All instructions that I will use in this video, I will attach to one post in my Telegram channel. The link to it will be in the description to this video. Here I am writing the instruction in English. It doesn't actually matter, so I will post it in two formats. I click save. And simply when creating a new chat, I copy this request. And look what happens. I paste it and it answers, it answers like an expert. blockchain technologies. That is, depending on the question, it connects an expert in a particular field. You don't need to specify every time: "You are a copywriter, you are an editor, you are a marketer, you are an SMM specialist, you are an SEO specialist." That is, it itself determines the expert from which field it is necessary to use here. And it immediately answers that launching a Bitcoin mining startup through a smart home sounds attractive, but requires a clear understanding of the technical basis, economics, and legal restrictions. Let's break it down step by step. But the next thing I don't like is the excessive structuredness. Friends, this is a new YouTube channel, and this is only the fourth video on it. Therefore, right now, more than ever, I will be grateful and appreciative of your support in the form of thumbs up under this video. Don't be lazy to put it. Thank you. Surely you have also noticed that ChatGPT writes quite robotically and very technically in its default settings. It uses lists everywhere. enumerations like, firstly, secondly, thirdly. It also adds a huge number of smileys, and constant dashes, hyphens, which immediately indicate that this text was written by ChatGPT. And wherever you publish and copy this material, either you will have to sit for a very long time with editing, or, any person will immediately see that it is from some neural network. And you yourself want to communicate in a dialogue format with an intelligent person, so to speak, yes, even if it sounds strange in quotes, indirectly with a person, but you want there to be a real human dialogue here. And when we ask it to write a post, an article, then immediately we will have a format that requires minimal editing and that passes detectors with flying colors. And for this, we will also use a system prompt instruction, as we see the standard one from ChatGPT itself, it doesn't cope with this, yes, and it's clear that it's adapted for the masses and most people are comfortable with this format, but for a more advanced level, this is useless. Therefore, now I will also put this in the settings here. I will replace our previous system prompt instruction for now, and then we will have a combined prompt instruction. Plus, I will share with you what we can write here specifically in the format of commands, yes, so that you don't just have this field filled with a complete instruction, but so that you can also control it and activate some factors, strengthen some and write your own commands here and edit these commands. This is very convenient when working in chat. And in about 40 seconds, you will get acquainted with this and most importantly, you will use it effectively. I have now removed this previous prompt instruction. Most likely, it will again write to me that this is great, this is cool, but we will fix it. Yes, it writes, an interesting idea, but notice, yes, that it is already trying to explain to us in the format of a real interlocutor how it looks. It doesn't immerse us in any, uh, excessive technical terms or abbreviations that we might not have known. It explains to us, tells us with examples that are available in our region, yes, we immediately see, yes, Yandex Smart Home. And here, in the text, we immediately have different sentence lengths. That is, this text becomes easily digestible. Digestible, digestible, yes, and it's much more pleasant to work with. It just doesn't hurt the eyes. And all this structure, all these lists, all these smileys, they disappear. And on the contrary, in such a format, even with such a technical request, it becomes much more pleasant to work with. And if your requests are related to creativity, texts, posts, then ChatGPT's answers will be pleasing to the eyes every time, and you won't have to sit for hours with editing. I will also leave this prompt request separately. And it will also be taken into account in the complete instruction, which you can get in the description to this video. The next global problem, which is actually very dangerous and which many still don't know about, is that ChatGPT can make up facts and can invent research for them, citing non-existent sources. And this is actually frightening. Of course, in the new version of ChatGPT, this has been fixed. That is, now, when we even just write a simple request, "show me research on this topic," it will most likely, with 99% probability, conduct a small research. Let's send it now, yes? In earlier versions that were used, it outputted exclusively from its database. If we didn't ask it to conduct research, it connects now, but still, if it doesn't find research, it can give us unreliable facts. Well, you've probably encountered Stanford research in, say, 2022, but it wasn't supported by any source. And this is also regulated by the prompt instruction, so that if it writes some fact or some research or some source, but it's not completely sure about it, then it should output that this information needs to be checked, that this information is not completely reliable. This is a very useful thing in many areas. And this is regulated by advanced prompt instructions. That is, conduct research on such and such a topic, analyze such and such sources, sort them, create a table, and so on. But writing extensive prompt requests every time is tedious, long, and you always have to switch somewhere, collect a database of such prompt requests. Therefore, we will also write this in the instruction, and it will give us specific data on research that it is sure of, that are little known, where it is not sure at all, and where it requires verification. That is, now it has outputted. If you read here, it refers to additional sources in the form of simple websites, but still it doesn't say what is absolutely true and what is absolutely not. And you will still have to deal with this for quite a long time so that you don't have to torture yourself like that. We will indicate this directly in the format of the same instruction, which will then be combined. Here it has outputted the result. And under each such result in the format of research, it will always output that this is absolutely true, you can take it and use it immediately. And what it is not sure about, what needs checking, what information you can not take and not waste your time on. And this is the most important resource that we have. Therefore, here there is a focus on what to concentrate on. Well, and additionally, if you want to delve into some specific topic, then due to ignorance of GPT, you can research and review additional information, yes, to expand a particular industry based on the given request. Well, and the most interesting thing here is the use of a combined system instruction, which allows you to eliminate all the basic shortcomings of ChatGPT and use this tool to its fullest so that all answers satisfy us. Let me insert this instruction right here. That is, it will also be attached in the description to this video. And here we will immediately specify additional fields in the form of commands. And they look like this, yes? That is, we will have specific commands by which we can ask our chat to delve into one or another area, or, for example, to write more lively, more interestingly, and to communicate with us as a live interlocutor in a more enhanced format. And here I will insert all of this directly into this field. And additionally, here you can adjust, for example, the writing style of some posts, articles. You can specify here in what style, in what format. In my second video on YouTube, I will leave the link above, I shared an expanded base of parameters that you can use right here. Well, and right now I will demonstrate to you how it works. We save it and write such a request. I heard that if you drink 5 liters of water a day and do a plank for 10 minutes, then in a month you can lose 15 kg without diets. Try to write this prompt request without a system instruction. You yourself will see that ChatGPT will say: Losing weight is a good idea. You are great for deciding on this. and will start praising you, forgetting about these 5 liters, about 10 minutes. But in this case, it will answer us as an expert in sports physiology and nutrition with a medical education. Remember, yes, from the first example, it pulls up roles that are determined from the field of our request, and it does this automatically. We don't need to write in our requests every time: "Act as an expert of such and such with such and such many years of experience." And it immediately determines this and writes: "No, there is no such scientific evidence. Moreover, drinking water can harm your kidneys. And there is a risk of additional diseases. Be careful, yes, to lose 15 kg in a month, a person would have to burn about 110,000 calories." Well, that is, it has explained to us here lively, briefly, without these huge watery lists, as detailed as possible in a live format without a huge number, not even structured lists, first, second, third, it did it in a live text format without dashes, without hyphens, somewhere it was blunt, so to speak. Well, that is, we have a clear answer that we can work with further. Now let's see how these commands work. That is, now, for example, I want it to start working with me intensely as a live interlocutor, with even more enthusiasm, so to speak, and output it in a post. But I won't tell it: "Write me a post so that I can comment on this material or write about it in more detail." I will just write this command withhuman here. That is, it says: "No, friend, this is a myth." But no research confirms that you can just drink 5 liters of water and do a plank for 10 minutes to lose 15 kg in a month. Well, would any chat without such a system instruction write to you in such a style, in such a lively, format and with such a lively presentation and manner. Well, 99.9% no. Therefore, use this instruction, fine-tune it for yourself. You can translate it into Russian, correct it, but now it is configured in such a way that we have short, clear, understandable, concise answers in a live style. And here we can also ask it to write a post on this topic. Write a post on this topic, and it will immediately be with all our instructions taken into account. And look, in a short format, it outputs a truly valuable post that can be used immediately. And I remind you that all materials from this video will be attached to one post in the Telegram channel, the link to which you will find in the description to this video. Go, take it, use it. Thank you very much for watching and see you very soon.