# How To Use Claude 3 Prompt Engineer Anthropic Console AutoPrompter (Custom Prompt Engineer) ## Метаданные - **Канал:** TheAIGRID - **YouTube:** https://www.youtube.com/watch?v=qr43Dc_d4DI - **Дата:** 12.05.2024 - **Длительность:** 15:29 - **Просмотры:** 24,365 ## Описание How To Use Claude 3 Prompt Engineer Anthropic Console AutoPrompter (Custom Prompt Engineer) Stay Up To Date With AI Job Market - https://www.youtube.com/@UCSPkiRjFYpz-8DY-aF_1wRg AI Tutorials - https://www.youtube.com/@TheAIGRIDAcademy/ 🐤 Follow Me on Twitter https://twitter.com/TheAiGrid 🌐 Checkout My website - https://theaigrid.com/ Links From Todays Video: https://console.anthropic.com/dashboard Welcome to my channel where i bring you the latest breakthroughs in AI. From deep learning to robotics, i cover it all. My videos offer valuable insights and perspectives that will expand your knowledge and understanding of this rapidly evolving field. Be sure to subscribe and stay updated on my latest videos. Was there anything i missed? (For Business Enquiries) contact@theaigrid.com #LLM #Largelanguagemodel #chatgpt #AI #ArtificialIntelligence #MachineLearning #DeepLearning #NeuralNetworks #Robotics #DataScience ## Содержание ### [0:00](https://www.youtube.com/watch?v=qr43Dc_d4DI) Intro So recently Anthropic actually released this which is their prompting tool for their AI system. Now essentially this is a really comprehensive LLM prompting tool and I'm going to teach you guys all how you can actually use this quite effectively cuz I think a lot of people did miss the mark with this because there are two uses. There are uses for this if you just want to use it very normally and there are uses for you if you want to use this in a sort of advanced way that I'm going to show you how to do. So uh hopefully you enjoy this tutorial and there is some value provided there. So first of all, what I want to say is that this is something ### [0:30](https://www.youtube.com/watch?v=qr43Dc_d4DI&t=30s) Free Credit that right now if you're trying to use this and you don't see this, it isn't free. So what I mean by that is that essentially some people when you go to the console. anthropic. com website, I'll leave a link in the description. If you see this and you don't see the prompt engineering button right here, all you need to do is you need to go into your settings and you need to essentially add some money to your account because this right now it doesn't seem that it's free. However, what Anthropic have done is that they've said get started testing Claude with $5 in free credit. So, what you want to do is you want to make sure that you click claim on this. Now, for me, I live in the UK, so I'm guessing that my number wasn't available for verification, but after some time, what I did actually do, I'm going to show you what I did now, just so that if you're in a region where this doesn't currently work, was I went ### [1:15](https://www.youtube.com/watch?v=qr43Dc_d4DI&t=75s) Adding Funds to the um site right here, and you can see that it shows my balance remaining of $24. So, this was where I decided to upgrade my plan and I decided to add funds. So, I decided to click this button. I added some funds and then it finally decided to work for me. So, if you don't see the button and you're wondering where it is, that's exactly where the button is. And I essentially just added a decent amount there. So, you can now see that this is what you'll see. You'll see this generate a prompt button right here. So, this is where you want to go and you want to click this one. You don't want to click any of these other ones. You just want to click ### [1:46](https://www.youtube.com/watch?v=qr43Dc_d4DI&t=106s) Generating Prompts this one. So, here we are in this generator. So [snorts] now you can see that essentially we have five different things right here and then I'm going to demo them show you guys how you can actually use these if you can use it in the basic way and then a really incredible way. So essentially what we have here are five different examples. So it says the experimental prompt generator can turn a task description into a high quality prompt. For the best results be sure to describe your task in as much detail as possible including what input data the prompt should expect as well as how much of the output should be formatted. Each generation will in each generation will consume a small number of Opus tokens. So with this, the reason you have to pay is because Opus is behind this. So when you generate this right here, it's actually using Claude Opus in the background. So that's why you need money on your account and that's why it's not essentially free. So just bear that in mind before you generate a different amount of prompts to use. So the first thing you need to understand about this and the first thing that I actually initially got wrong is that this is not like an LLM. What you're doing is you're just asking what you want in completely natural language. So you don't need to, you know, describe it in the most insane way. You just need to describe it as if you were talking to, you know, someone that was a prompt engineer. So this is essentially removes the role of prompt engineer. And you can see right here that all of these different ones, they are very, very simple. The thing with this that you want to understand is that Claude Opus is going to be doing the hard work, not you. This essentially just removes the entire amount of the hard work and ideally you want to enter as much information as possible but Claude Opus is going to be doing the heavy lifting. So don't make the mistake I made that like making this super formatted and super crazy because that's not what you're supposed to do. So for example, let's try it with something super basic. Okay. So if we can say um draft an email responding to a customer complaint and offer a solution, we can see generate a prompt there. Now, as the video progresses, I'm going to just increase the uh I guess you could say difficulty in terms of what people might want to use this for, cuz I think that this is really cool. And I'm pretty sure that, you know, these things are going to be really useful in the future. So, you can see here, this is how it's written. And when this is written, you can see that there are certain details. Okay? So, let me just show you guys what's going on. So, right here, it says, "You'll be drafting an email to respond to a customer complaint. I will provide the original complaint email, the customer's name, and the name of the product they complained about. your goal is to write professional email yada yada. So essentially what you want to do with this is a tool where you can use a prompt to use at scale. It's not really for something that you want to use for a one-time use. This is for something if you have a recurring problem and then you want to switch out variables in that and then that is something where this is going to be most effective. So if you constantly get emails and you want to be able to, you know, respond to a certain type of email, this is the type of tool that you would use in order to do that. Okay. So, for example, I'm just going to use this and I'm going to show you guys a real world example that I can. So, I'm going to discard this because I'm going to show you guys a way that makes sense, but I wanted to show you how it's formatted because it has all those variables. So, I would say draft an email responding to a YouTube sponsor that is offering, and I'm just going to write this down literally as it would be for my channel right now, is offering to sponsor my YouTube channel called the AI Grid with 215K subscribers. Um, then I can try to include as much detail here. And the reason this is good is because if you get a lot of these kind of emails, which I currently do, it's able to classify them in many different ways. So, I'm just going to try to include as much detail as possible. I offer I'm just going to say I offer, you know, different things. And there's different ways to do this, but I'm just going to do it this way at first, and I'm going to show you guys after. I offer two packages. One being a 60 second ad placement for, I don't know, let's just say uh $1,000, and let's say a full and separately, I offer a dedicated video for uh let's just say $2,000. Okay, those aren't actual prices that I currently have. I honestly don't know. I have a brand manager that does that stuff, but let's just say that's there. Um, and I'm just trying to think what other information I could include. The channel averages between, you know, uh, 20K to 40K views per video, okay? And this video might not even get that much, okay? Cuz this is just a really niche tutorial. But we're going to click generate a response. Um, and the reason I've done this is so that it shows you guys with a basic example. So, it clicks generating response. And then when I edit this, I show you guys how someone can actually use this on a basis. So remember this is what I copy here. This is what I'm going to be using in Claude 3 Opus. Okay, so right here. And okay, once we click start editing, this is where we go to this one right here. And then right now, okay, before we want to click run, we want to So now you can see that we're on this menu. You can see right here we click run. Okay, and then of course you can see here it says sponsor offer details and then it says channel details. So with this one, what we can do here is we can say the sponsor offer details. This is where you'd put like their email. Hi, I'm from OpenAI. Of course, this would never happen. I want to sponsor your channel with a dedicated shoutout for $3,000 in a single video where you talk about chat GPT. And then the channel, you can just put uh the AI grid. And then there we go. Okay. So now I'm going to go and click run. Okay. And then this is where we're going to get the response. And it says here is my draft email. There we can say this is where we are getting all of the kind of information that we currently need to do. So you can see right here essentially what it's doing now is it's managed to give me this information in a very good way and I'm able to use this and just simply respond to this. So you can see it's in markdown and it's in text. So you can choose whichever formatting way that you want. But you can see right here that like literally with these variables once I change them I don't need to do a lot more anymore because I have this prompt here. So you can see right here it says here is the draft email response. And with this response, I didn't mean to click get run again, but with this response, you can see that it's able to just have all the details from your channel and it's able to give it a really nice response. Best regard. So you could probably um you could say the sponsor offer details, you know, in that sponsor offer details. Of course, you'd want to include their name, all that stuff. Essentially, you just paste the email right there. You know, I don't actually have any emails because I don't want to show what people are offering and stuff like that because I don't know, might be a little bit unfair, but um yeah, you would just paste that in there. That's what I would do. And then essentially, you just use that. And you don't need a system prompt, but this is something that you would literally use. Okay? And you would just simply use this as you would in your day-to-day basis. So, I'm going to rename this prompt. And then I'm just going to save this uh YouTube uh sponsorship. Okay? And then uh I can't even spell right now, but if I click save, that's there. So, that is something that's pretty useful. So, if I click plus, we could get a new prompt. Um and of course, this is where you can write your own prompts and you could add ### [8:30](https://www.youtube.com/watch?v=qr43Dc_d4DI&t=510s) Examples your own variables and stuff like that. But this is why this is really useful if you want to do certain things at scale. So for example, let's say you want to summarize a document. Okay, let's say summarize documents into 10 bullet points max. Uh but with this one, I'm going to say these documents are from Axive and this is what I could do cuz this is something that I do and I'm going to show you guys another example because this is just summarizing documents. Okay, so these documents are from arxive um and they are mainly focused on AI. I have an audience of 200k subs on YouTube. And I'm only writing that because I'm trying to give this um the most amount of details as possible. Okay. And then what I'm going to do as well is I'm going to basically say right here is that my audience wants a layman's terms explanation for these papers. So no fancy talk, please. And let them know how it impacts them. Okay. So I can say that. Okay. And I'm going to generate a prompt here. And remember, you want to keep this part as completely as casual as possible. So now that we've done that, we can use that right there. So now we're just going to wait for it to generate a prompt. And then we can see your document is your task is to summarize the document about a research into concise, easy to understand information. [snorts] Firstly, read through it. Uh yada yada, write okay yada yada. Okay, so it's going and going and going. And we're actually going to see how this works cuz uh there was a paper that I recently do want to cover. I've been meaning to cover it for ages, but I just haven't been able to. So, now we can always just copy this and we can like use this for other systems. Okay, but you know, you don't really want to do that. But if you do want to use this for other systems, I'll show you guys how to do that in a minute. But anyways, here you can see is this is our prompt. Okay, so with this, we're going to copy this. We're going to start editing. Let me just paste this somewhere else. Um, so here you can see is here are the documents to summarize. So essentially what you want to do is when you click run, this will pop up. The variables will be here. And also you can see right here, this is where we can enter the custom value. Now, of course, with the max tokens to sample, this is basically how long you want the response to be, and the temperature is basically how, you know, how varied you want your response to be. So, lower variable just means it's going to be super predictable. But with this is basically a c creativity bar, if you uh want to call it that. So, that's what it is. So, with here, what I'm going to do is I'm going to paste in um some information from a recent research paper. I'm just going to go get it right now. So, I've just literally just hit copy and paste, and I've just literally pasted this all in here. Uh this is about CAN networks. This is a paper that I'm going to be doing a little bit on recently. So, uh, stay up to date for that. Honestly, my speaking is awful. But anyways, I typed in CAN networks. Uh, and now we just wait. Okay, so it's going to take a little bit of a while because this is, uh, quite long. But remember, this is useful because this is going to be something that you can use all the time and it's something that is very, very useful for you if you're trying to use this all the time. So, we're going to wait and it says here is the condensed. It says result. Here we go. It's giving me bullet points and yeah, it's still going. It is still going. And it's given me a lot of bullet points here, which is really good cuz I got to be honest with you guys, when I usually ask Claw to summarize a paper, it doesn't usually do this well. So, for example, the reason this is good, and I'm going to show you guys uh a difference. So, I'm going to basically uh just do this, okay, on a different one without this amazing prompt that we just did because you can see right here. And the reason that this is a lot better than the standard version of um Claude because essentially there's some internal stuff going on. You can see that essentially it has its internal scratch pad where it writes out the summary of the document in claude's own words. Then it has a draft summary. Then it refineses and condenses it down into 10 bullet points. Then you can see it go throughs adds one two brief sentences. And then the lastly it compiles the bullet point summary into the final result. So it goes through three stages of iterations before we get this amazing thing. And that's why I'm going to be using this a lot more than just standard claude because standard version of Claude is pretty good. But with all of this stuff, it's just a whole new level. I guess this is kind of like a system that's around Claude, which is like a kind of anthropic rapper that's on top of this and it just makes everything a lot more effective. And I know I've seen people state that this isn't that good, but I think it is. I think it is like once you understand that this is really cool. And right here you can see I essentially asked Claude the same question here and it's just not that good. It's only given me five basic but here because I actually used the initial prompt thing uh I can really exactly understand what's going on with the paper and I'm going to essentially save this prompt as claude oxive understand. There we go. So that I really like this. Like I really do like this. And of course we can go ahead and here and we can click your prompts. So I can go ahead and click YouTube sponsorship. Uh and of course you can see here there's a different YouTube sponsorship. But essentially what you want to do as well cuz I actually just realized I made this mistake is for example if we're going to um do a response for example like this and we actually want to save it iterate it. Just firstly click uh generate response or generate your prompt whatever. Um and I think maybe I'm missing something here but once you generate your response I'm just going to click start editing. Uh there we go. Okay, let's just wait for this. And like I said, the reason this is cool is because, you know, it Claude has, you know, like some internal stuff where it's writing out a bunch of code. Not code, but like tokens and outputs that you just don't see. So, I'm guessing this is, you know, using some advanced stuff. But here, uh, I can go ahead and this is where you actually want to save this. So, uh, Claude demo prompt. Okay, this should work. Okay, if I put demo prompt here and then, um, essentially if you go to a random prompt, I don't know what this one is. Uh, but if I go to claw demo prompt here, before I click run, I can always just change this. this yada yada. I can put um screw you guys uh AGI handbook. I'm just going to put it's awful. And then run cuz I just want to test something real quick. So um this is not going to be like an actual response. So it says the dank out yada yada. And of course that's how you edit it. But of course essentially what you want to do is if you want to do this again and you say okay we just I want to try this again. What you can do is go to your prompts, then of course go to um back to this the demo manage to rename your prompts or save your prompts. If you ever want to come back in and get a new uh variable value, all you do is click this button right here. And this is where you paste your information in. So, for example, if I want to go to my YouTube sponsorship, all I want to do is paste in the sponsor offer details or the channel details right here. And then I can literally click run and it's going to give me a very, very comprehensive thing. So, with this, I would say that this probably bumps up Claude's accuracy maybe about 10 to 15%. I know that might be just like an eyeball figure, but trust me guys, this is something that you want to be using. I know that, you know, the next version of GPT5 is probably going to, you know, probably stun people and stuff like that, but I think this right now, the way how we currently use LMS, thank god that they actually made this cuz this isn't something that's actually hard to do at all. And as someone who understands that like a lot of the non-technical stuff is just, you know, impossible for people to even want to get to grips with. I think as long as you just click this button here, you can enter your information, then you can just click run, you know, you can see channel details or whatever. You can literally just click anything and go to, you know, another one. For example, if I want to just put in another paper here, I can just click this. I can just delete this and then I can put in another paper and it works. So, I think this is really useful. I'm glad that Anthropic really did this. So, um, yeah, if you guys, you know, enjoyed this video, if there was anything I missed, I'll make a follow-up video. But, um, hopefully this tutorial was very valuable. It did get a little bit longer than I did initially think, but this is a tutorial where I really wanted to cover all of the, uh, most comprehensive things. Now --- *Источник: https://ekstraktznaniy.ru/video/14325*