Transcription
For such a request, it will simply give you garbage, incredibly, practically at all our planning meetings, ChatGPT is present. It has upset you, it will not replace an expert, therefore everything I said before is definitely worth using, otherwise you will have contracts made of shit and sticks. It knows our referral program, it knows the effectiveness of our referral program, and TurboEra. Approximately this kind of copywriting, when you ask it to write your post, this video is for you, because today I will tell you how to train ChatGPT for work specifically with business. And there is a very important thing here that separates you from starting to get cool results. The problem with GPT is that you expect the final result from it, that is, a ready-made 100% result that you simply take, copy, paste, and it's supposedly ready. In my experience, even with all the incredible training, no matter how many files we uploaded, it's impossible to get a perfect final result from ChatGPT, because it can give, for example, 85% of cool copywriting words from the entire text, everything is clear, well done, good boy. And 15% is shit, and you will still have to finish this text a little. Here are a few examples of requests to ChatGPT, for which it will most likely give rather raw, superficial answers. For example, your task is to create interview questions for an SMM position. If you ask, "Make me 15 interview questions for an SMM position," it will give you super superficial questions. Another example: you need to write a regulation for a project manager. Similarly, you tell it, "I am the head of an agency, I have project managers, they do this and that. Write me a regulation for them." Most likely, the regulation it gives in the first iteration, yes, even in the fifth, with such a task statement, will be super raw and unsuitable for you. Therefore, be sure to watch this video to the end. At the end, I will show specific examples of how to set tasks for ChatGPT, how to give it tasks so that it gives you awesome, cool results. The next problem is that many people work with ChatGPT in one request, that is, one request is written, and you expect it to solve your task immediately. But it doesn't work like that. You need to divide your request, or rather your task, into several stages. For example, if you need to create a regulation, you first train ChatGPT, tell it about the position, for whom you are writing this regulation, what this person does, what your company is, what this person's responsibilities are, what their life consists of, what is included in their routine. After that, for example, you can tell it, "Now ask me questions and clarify, because you will have such a task in the future, what else do you need to know?" After that, you write, for example, "Create a structure for the future regulation for me and describe the key sections." Then you edit these sections, arriving at the desired result, or giving it additional edits, or simply rewriting the structure yourself based on what it sent. And then, for example, you can write the regulation only for the first section, then only for the second, only for the third, only for the fourth. A logical next step would be to create a custom GPT for yourself. There are videos and instructions on YouTube on how to do this. Just Google it, it's not difficult. I won't explain it in this video, but you will definitely figure out what it is. It's a bot into which all your files are already loaded. It is trained for a specific role it will perform, for specific tasks it will solve, and it already knows all your context that you initially loaded into it. That is, it's already a trained bot. I made such a bot for myself called "Marketing Director," and I uploaded everything, everything, everything, everything we have for marketing into it. And now this is a bot that I use almost every day, because with one click of the mouse, it studies everything that our marketing team knows, and you can communicate with it absolutely on equal terms, like with a team member. That is, I can even go to it and say, "Listen, give me 10 hypotheses on how we can improve our referral program, 5 ideas on what to improve on our website." It knows our website, it knows our referral program, it knows the effectiveness of our referral program, and from the reports that I feed it. It makes sense to use it for some discussions, idea generation, because it is incredibly trained for this. In principle, we do the same. We have a bot for the sales department, it's called "Head of Sales." Conversations of managers, scripts, a bunch of presentations that we made for clients, our entire sales structure, our sales training, which were filmed for us, into regulations. Again, we just take the regulations and feed them to GPT, and GPT becomes the head of sales. That is, it can now help our head of sales make scripts, because it already knows adjacent examples. It can help, for example, with auditing. Our custom bots solve this, or in general, ChatGPT for our project. So, to get cool results from ChatGPT, the first thing you need to do is start working with it step by step, that is, step by step, clarify the task, specify, correct, arrive at the next stage. The second thing you need, and without which it is really harder to work, is role prompts. You can probably Google what they are. I won't go into their construction deeply now. In short, what you need to know is that it's great at the beginning of the prompt to assign a role that it will perform. For example, "You are the head of sales." This is how it's usually explained in most videos about role prompts. I suggest you go a little further and describe this role, what exactly ChatGPT should be able to do. For example, not just give a role prompt "You are the head of marketing, act from this role," but add further: "You can create awesome marketing strategies for this type of business. You are excellent at understanding its target audience, you can get into clients' heads, you can create creatives, you can generate hypotheses for website improvement." And then you just list what else you want to do with your ChatGPT. That is, essentially, you pack the tasks into your GPT's skills, and you get such a role prompt: "You are this and that role, you can do this and that," and this is your role prompt. And the final stage is to package all of this into a custom GPT that already knows everything, knows its role, its tasks, all the tables, all the documents are fed into the custom GPT, you save it, and use it as a full-fledged team member. So, let's go through 14 cases of how I use ChatGPT in my work. Screenshots will appear on your screen of what it looks like specifically in ChatGPT, what the request looks like, what its answer looks like. Also, a QR code will appear somewhere on the screen now. Scan it if you want to get all these prompts in our Telegram channel. There you can just click on the link and copy them all for yourself. This is how they look for us, how we use them in the team daily, without changing anything at all. We give them to you. Scan the QR code or the link in the description, go, subscribe, there will also be a lot of interesting stuff there. Let's probably start with the first task. The first task for which I use it most often is some kind of brainstorming. That is, when I don't understand what to latch onto, I need some idea, I need to brainstorm something, think something through. And the first prompt I use here is "What could go wrong?" I use it, for example, when I need to evaluate something. I've come up with something, described it, and now I need to think about various scenarios. Most often, I use this, for example, in HR, when I have a job profile. That is, I described what the person will look like, their gender, age, tasks, responsibilities, requirements. I give this job profile to ChatGPT, say "What could go wrong here?" and it tells me, for example, "Here you are missing functions, here the person is overloaded, here are functions that are incompatible with the person's psychotype." For example, you expect analysis and creativity simultaneously. Perhaps one of the functions will prevail, and it will highlight to me what could go wrong. Or, for example, we were organizing an event. I tell it, "We want to organize such an event, we will have advertising placements here, here, here. Here there will be a banner, here there will be something else, some advertising. I want to get as many clients as possible from this event. What could go wrong?" And it tells me, "Listen, these formats might be unnoticeable. Make sure they are like this." It starts drawing a small apocalypse scenario for me. And this is convenient because you sit and think, "Ah, the robot knows something better than me because it generates all possible scenarios that could go wrong simultaneously." A very frequently used function by me. The next function is similar to the previous one – it's criticism. When I've done something, and I think it's the final version, I want to send it to ChatGPT and say, "Criticize it." And it simply criticizes it, says, "This is wrong here, this is not taken into account here, this needs to be thought through here," and so on. So, I use criticism for almost everything I do before I put the final dot and say that I'm done with it. I give ChatGPT a screenshot or copy something, say, "Criticize it, criticize my post, criticize my reel, criticize my script, criticize my vacancy, criticize anything," and it criticizes it awesomely, especially if, again, all of this is trained, role prompts, information is loaded. It says, "Your vacancy is too stuffy, it's too stuffy to read, it's not interesting. You should sell your company with vacancies, you have some stuffiness, redo it." For example, how to redo it. Anyway, a cool prompt. The third prompt is decision-making from the user's perspective. That is, we often don't understand what a particular process looks like from the user's perspective. And, for example, we can say, "We want to organize an event like this. Tell me what this event will look like from the user's perspective." And it describes: "I come here, I go through here, I want to do this and that." And at this moment, we can think, for example, "For example, it says that the person goes through the corridor and hangs their jacket in the cloakroom. We think, damn, we don't have a cloakroom at the event." Well, that is, it's not an obvious thought when you are organizing an event that a cloakroom is needed in the venue in general. But when you look at the picture from the user's perspective, you understand that the user will behave not as we thought about it. And, you know, this thing with the user's view is very cool. For example, "Describe how a user will make a decision to buy my product," and it will describe it awesomely, especially if there are case studies, especially if there is everything we listed before, it's a treat. Another one of the frequently used functions is "Suggest more ideas." This is when we have a list of something that we have already generated, for example, three or four items in some list. We say, "Here are some functions or items. I'm doing this kind of document. Suggest 10 more similar ideas," and it suggests. And again, we remember, this is the level of GPT, maximum GPT-plus. That is, it won't offer you brilliant ideas, but out of ten ideas it offers, one or two will be cool, and then they will need to be refined. But it generates three or four good ideas that you then refine and they go into your marketing, sales, or something like that. That is, approximately every third idea that is generated at our planning meetings is generated with the participation of ChatGPT, specifically in the prompt "Suggest more ideas" based on what you already see on our screen. The next big block is HR. What do we do here? Preparing interview questions is super cool, but you must feed it the job profile, 100%. Because if GPT doesn't know what this person does for you, what you want from them, what are the requirements for them, what are their tasks, and so on, it simply won't be able to make good verification questions. And the second point is, I recommend highlighting to it what exactly you want to check. This is also one of my favorite functions. For example, I want to check an SMM specialist's ability to write creative scripts. "Suggest questions." I want to check a project manager's level of responsibility and work under stress. "Suggest questions." GPT can even give recommendations on whether to hire or not hire a particular person. We don't use this function in this way, but we feed it a recording of our interview. That is, first we conduct the interview, translate it into text using services, there are plenty of them on the internet, I'm sure you'll manage. And we send the finished text file to GPT and say, "Highlight the areas where the person has strong opportunities for growth, for strengthening us, where they will reveal themselves." From this interview, it extracts meanings and highlights opportunities for us. Moreover, it actually does this much better than an average HR person or even an entrepreneur who simply interviews a person, because GPT sees the entire text at once, and it sees where certain words, for example, don't match each other. For example, at the beginning, he said he has problems with management, but at the same time, in his tasks, he has tasks related to project management or managing other people. It can see contradictions. It can see, on the contrary, things that complement each other. It can even see things where he slightly distorts the context, that it doesn't fit with the previous picture the candidate painted. That is, GPT sees the entire transcript at once, while we go through it sequentially. Another area is the use of GPT in content. I must honestly say, we don't use it for writing posts, because in my observations, no matter how we trained it, no matter how many examples we fed it, no matter how much we told it, "copy the author's style," all of this, it writes in a way that it's obvious that ChatGPT wrote it. All the times someone told me, "You won't notice that this post was written by ChatGPT," it writes exactly like ChatGPT, like a person. You can see that it's ChatGPT. Well, it writes like a junior plus level copywriter, that is, a copywriter with one and a half to two years of experience at most. Yes, you can see all of that. Therefore, unfortunately, we haven't managed to create a completely GPT writer that you just throw meanings at and it turns it into a ready post. It doesn't work. That is, it will upset you, it won't replace an expert, but it can do, for example, some repetitive posts. Suppose you have some weekly post that comes out with approximately the same structure and always has more or less different meanings, but in the same format. For example, event announcements. We have such posts, and for this, we created a custom ChatGPT, into which we uploaded the structure, examples, screenshots of good posts. Now we just give it the meanings and say, "Put these meanings into this structure and give it to us." And it gives us weekly event announcements equally well, stably, with normal headlines, normal everything. It just extracts the information we gave it and creates identical posts from them. But as soon as it needs to step left, step right, a little bit of creativity, headlines, post structure, all of this collapses. That is, there are so many edits that it's easier to actually get a normal copywriter, and they will write it all themselves. But, for example, for us bloggers, with Natasha, my partner, GPT helps us write blog posts, but it doesn't write posts for us. It helps us come up with ideas, because, you know, anyone who keeps a blog knows that at some point you face the situation where you just don't know what to write, you've run out of ideas, everything, your brain has shrunk, you have absolutely no ideas what to write. We came up with a cool prompt that analyzes a large correspondence in a chat and generates ideas and theses from it about what we can write in our blog. But we don't write according to this structure. For us, it's an opportunity to latch onto fresh ideas. That is, we go into a specialized chat where our colleagues discuss work issues. Well, everyone probably has some specialized chats. We copy 200 random messages from there, throw them into GPT, it extracts meanings, theses from there, and suggests, "You can write a post about this because people were discussing it, people are interested in it, and so on." And we just choose what's closer to us, like, "Oh, this topic can be revealed well, we have a lot of experience and history here, and so on. Let's reveal this topic," and that's it, you go write the post. Another awesome task for which ChatGPT is suitable is, for example, to create timecodes for videos. It's very simple, just one prompt and the transcription of the video. We combine them, tell GPT, "Make timecodes." It first gives us the first version. We can tell it, for example, "Make them more clickbaity, make them more motivational, make them more, I don't know, inspiring." Or we can even give it an example, say, "I like these kinds of headlines in timecodes or this structure." We tell it, "Here, here we are talking about this, and here about that. Do the same for the rest." It makes them for you. The sales department. In the sales department, we use it most often for creating scripts. That is, we already have scripts. For example, we want to create scripts for another product or for another position. We give them to it, not always perfectly, but generally as a starting scenario that needs to be refined by hand. It's okay. Another cool thing: sales managers write their conversations with clients. We give all of this to ChatGPT and say, "Analyze this as a Head of Sales." Again, the Head of Sales is already trained on other managers' calls, on everything, on our product, blah blah blah. That is, it's already trained. It's not raw GPT. We give it the transcript of the managers' conversations, say, "Dear, please give your audit, highlight the strong and weak sides, where the manager pushed, where they didn't." We describe to it what we want to get from it, and it gives us an excellent audit of the manager's call. And a bit of assistant tasks. An assistant regularly uses ChatGPT in their work, for example, for such tasks as checking a contract. We send it a contract, say, "There's a lot of complex, unclear stuff here. Please highlight the key points and risks that I need to know." It immediately tells you, "Here are the deadlines, here are the risks, here are the responsibilities, you need to do this." It translates your contract into a short, human-readable format. The opposite is also true: "Draft a contract for me. These are my tasks. This is what I want to achieve. Within these deadlines, create an adequate contract." Not always in the first iteration, but it creates a great contract. Many contracts that we make now are made through ChatGPT with subsequent refinements. By the way, I had a cool case, well, not cool for me, when my assistant, who hadn't worked with ChatGPT before, I told her, "Make a contract through GPT." She went and did it, exactly in one iteration, without dividing by tasks, without role prompts, just said, "Make me a car purchase agreement." And then I came with this contract to the traffic police. I say, "I want to sell a car." The traffic police looked at me and said, "Well, your contract is quite a mess." Well, like, I don't read it, I just signed it, my assistant gave it to me, I just signed it without thinking. I started reading it, and there were really some crazy points. That is, if you don't give it context, it really goes off on a tangent. There was a contract like, "I, Anton Vlasov, decide to sell the car, the buyer decides to buy the car, we agreed to buy and sell the car. Signature." Something like that was the contract. Well, the traffic police laughed at me and said, "Well, okay, let's sell it legally." Of course, it has legal force, but it's a terrible joke. That's why everything I said before is definitely worth using, otherwise you will have contracts made of shit and sticks, where you saved on shit and sticks. Therefore, I remind you that to get all these prompts, scan the QR code on the screen and follow the link in the video description. There is a post where all these prompts are collected. You can take them, use them, copy them directly for yourself. I strongly recommend you to refine them a little and adapt them to your tasks, because these prompts are made specifically for us, and yours may differ slightly. Never take ready-made prompts from the internet, adapt them for yourself. That's all from me. I hope that GPT will become your loyal assistant in business, you will unload many of your employees' tasks onto robots. See you. All the best.