Prompt Engineering Full Course | Telugu | Vamsi Bhavani

0:00 Hello everyone, welcome back to Vamsi. Bhavani will start from today. Prompt Engineering Challenge This is a 10 Days Challenge means daily for 10 days. A video will be coming regarding the Prompt Engineering Course Today In the first video, Vavi will discuss what is Proton Engineering and Why to Learn Prompt engineering makes sense, right? We have so many domains, Devops. It's called Android Development. Web development is called cloud. Why to learn by leaving all these domains behind? Pronto Engineering Vavil Discus Booth Of these aspects definitely me Saying that this course will help everyone I hope. Please, this course is Support the challenge with a like.

0:45 Comment a share for reach and engagement Do what you can. Okay now First of all what is prompt Engineering Prompt Engineering is a New Domain and Its New Domain Chat A sudden AI boom after the arrival of GPT You have arrived. Wherever you look, you are seeing. New AI tools and different Will AI enter our domains? AI will go into different domains Basically, what happens is different. Build AI tools in domains Doing. And EAI tools Using it, you can do things very easily. You can do it. Now you are my previous genai. If you watch the video, you will see that EAI tools are actually What are all the generative AI tools?

1:33 OK GenAI Tools GenAI to NormalAI What is the difference between Normal AI and our previous one? Models in AI are used to classify Use to predict so they are new to that Tools do not generate new data. Okay, they use to actually make sense out of it. Off the data means one data is one text If given, this text is positive or negative. Sentiment Analysis Clustering Prediction is for applications like this We used to. But what is GenAI? Generates new data I chat GBT Either go to Gemini and give me a Telugu The movie story needs to be in this genre. The climax should be like this If you say there should be an interval bang Writes. So it is generating new Story New Data So That's How This ZNAI If there are models, then what are these models?

2:20 What's missing are the AI ​​tools we make? The backend behind it is a Let's say model. Because it's a Mathematical model so all this is maths Models are just that, okay, these math models are Mathematical Models This Gen AI Models How are these made? What happens as a result of training? Training on large data sets Training is OK on large data. Data is not large, very very large. It's okay to train on amounts of data. So answer based on that. These models all say that ZENAI Models These ZENAI models have two There are aspects, the first aspect is Training aspect so good on data It's clean and OK on noise-free data. What are the biases on the data? Whether it is related to a political party or not

3:06 Caste is a race or nationality. Train on unbiased data Only then will you have unbiased models. Without any hallucinations They will tell you the answers. So that's one. Part of the Genia Models Part Two What is the input given to it? Because no matter how good the training is, The question you ask is the answer you get. If the input given is not correct, the output Very good for training too. There will be a contribution, but this training Why is it not in your hands, chat, zipty case? Training is done by OpenAI Gemini The case will make a logo. So they are in your hands. No, but the input is in your hands. So The better your input, the better your The output will be as good as it gets. Actually

3:52 We technically call these inputs That's what prompts are for. So prompt Engineering is a kind of AI. Good instructions, good questions, good Train people to give input. What is a trained AI good for you? To give output means that people are good. Input must be given. So that's good input. The training that humans give to AI is Prompt Engineering Okay So Prompt Engineering means we train for AI. We are becoming. Because we are good. AI is good for you only if you give it instructions. Gives output. And you If you look at it, every AI tool is a witch. Paid They Have Something Called Tokens That means your input length and all of that

4:40 Really matters so what is your input length? How many tokens are there in your input? How many times are there questions? Asking each of that has Actually a contribution to the cost So how many times do you do it? Ask a question and get the answer you want. If you bring it, it will cost you overall. No company will decrease that much. So going Forward Almost every company is AI Must use. So they are AI. More basic when using More questions using tokens If you ask, their overall budget. Overall, expenditure will increase. So you correctly guessed it. If you can ask questions in the right way, Supply prompts if you can provide input. If you can, you will get good outputs. They come fast, they come overall. Expenditure will decrease. Okay, so that's it. Prompt Engineering Now This Prompt

5:26 Why study engineering? Let me Answer that first reason prompt. Engineering is an evolving domain. This is how the syllabus will be. Its road map is this, if these skills come They will get jobs in this domain. There is no fixed set of definitions yet. Not defined. It's still a Evolving domain, whether you believe it or not Even if you don't believe it, you will have it in your entire lifetime. Such an evolving domain There are very few opportunities to learn. So that's why it's in the evolving face. If you learn while you are still alive, you will be able to Great help in your complete career. It will be. That's the number one reason. Number two is that there are also a lot of openings. Be there. So I mean definitely that. Forget the two crore package, but

6:13 In general, the openings are good. Because every company should try. She thinks, "Okay, let's hire some prompts." Engineers are prompt engineering Interns and Let's See What Worked Out At least every company is wondering if it will happen. She thinks let's try and overall IT Industry Down the Line 10 Years Then hire such specific roles. I can't say whether it will happen or not, but... This is a software engineering domain. The part is over. They will be tested. On-the-spot engineering skills also So learning for that Important. So you said the openings. Can you show me, brother? It's simple. Go. Currently on LinkedIn India 3163 for this role for prompt engineer The results are there, meaning all the openings.

7:00 There are. I think these are the core openings. Not even in domains, that's again one. The third reason of the plus points is that If you are into data science or AI Either into machine learning or the cloud Let's go to any domain like this I thought they would expect it. Experience is very important to you. You only get salaries if you have experience. Good, overall good profiles for you. There is a chance of getting it in big companies. But this is the reason why this is happening. In the meantime This prompt came recently. You obviously give engineering a five. Years of experience teach us, don't they? SR No, so all these openings Even freshers are okay, you really have the skill. If you have the talent to ask for referrals, then this

7:46 You can easily get hired in roles. And Early on in your career, it's okay. In domains evolving in domains you So if you don't have a hair salon today, maybe you can get 3 locks. Even though there is a per diem salary, it is a five-year Then you have about five years. Experience will be a new domain And that experience is definitely for you. These 3 tufas are made from 15 to 18 LPA. Has the ability. Okay, so that. The third one is real. Experience is not required. Okay, number four and the final one is This is my guess guys in the future. Let's see if the prediction comes true.

8:33 Only time will tell, but I feel today. I mean, I have to search on Google. If you don't come, they say, "Ask around." Similarly, an eye I won't say specifically about the mean years. But in the future, everyone at least This field would be expected to know And understand basic engineering and That's what you do early on in your career. Being in nature helps a lot. And this This is not just for jobs, Not just for the sake of openings. To you Remember one of the previous videos? I said it in. Today you are from the EAI domain. If you want to earn money, you need AIA. No need to come. With AI AI tools If you have an idea, provide service with AI tool Even if you do it, you are as bold as anyone else.

9:20 They are making money and that's it. You also need prompt engineering. I Hope this is enough for you to get Motivated to learn Pronto Engineering As I told you, our complete effort What are we to make of large language models? The output depends on the input given. So we have good input. If we can give it, we will get a good output. So to give this good input There are various techniques, those are the techniques Prompt engineering, now you are doing this. LLM can be used in two ways. The first way is to give them directly. Using the UI means we are all As soon as we open the chat GBT, we This is how a U comes. Below here There will be a text bar here Questions

10:06 We will click Send and send it to you here. Responses will come, right? So this is basic. As OpenAI people chat for GPT You created it so you use it. We do what we want and we do what we need. You can do the necessary tasks. That's Way number one but let's say I'm a I am building a new application. Let's say I'm a food delivery guy. I am developing and delivering this food. I integrated this chat GPT in Yap. Let's say my chat support I think we should improve. Then You go directly to your customers. Use the chat box UI to ask questions there. Ask and get answers from it. I can't tell you because I want to. Keep our users on our platform So there is a way number for that.

10:54 ViaAPICE. Use soapy water. We interact with LLMAS by doing Yes. APIs What are APIs? Stand for Application Programming Interface for more technical details If you talk without going, API is this Use the backend of LLM Here in this U.S., there is a way to do it. Also here as soon as you click on send What's happening? Is this happening automatically? The question you asked is not magic, is it? Go to the backend and that backend That's where the response comes from. So we What do we do through APIs? This UI We can eliminate the layer directly. We will do it. Eliminate and directly Call the backend and respond. We will get it. What use is this to me?

11:40 Directly get my food delivered to your E-Apps. Integrate R Yup into the website I will. So that I can directly Support them with the chatGPT chat Facility Inside My Own Yep Or My Own Do you understand the website? So, this is how it works. You can use it in. Okay, great now. This UI exists, everyone can use it. Because everyone is already using it. There is no difficulty for users in this. No. But as you were told in the API Coding is involved and every For example, not everyone can code. I run a company. My The company has software developers. So if I really want to use APICE They will integrate it. But

12:25 Before I pay money to APIs, I I want to know if this will work. API testing, so I said, "Yes." For every LLA company playgrounds Will release. Let me show them to you. Yu. This interface you will use many times. You will see. This is the normal thing for Chat gbita. Similarly, this is also very Sir, you must be looking at this is the UI interface. For Gemini, these two are the first way into this. What will come is a second way into UI use. As mentioned for They Have Playgrounds So this is the playground for open ai So chat is a playground for GBI. Okay, so we'll discuss about what system. What is this user's temperature max? Let's discuss what all this means, length and topic.

13:11 Similarly, Let Me Show You the Gemini Playground Also So This Is The Gemini Playground okay so here is this chat GPT Playground and Gemini Playground If you look at both, here on the side There are settings with which you can Actually Configurations Change Can be done. I have written it down for you in advance. So these are all the different settings. That you can configure with in the Different LLMs are OK, first come, first served. To model is what you are Let's use the architecture model. Do you think this specific prompt For is defined. So far For example, here you have different models. Even in 3.5, there are different models. Similarly, if you have a Plus subscription [ __ ] t4 is also seen. Similarly here You will also have different models.

13:58 Gemini 1 Pro Gemini 1.5 Pro E First of all, what model are you in the setting? Do you want to interact with They will decide. Then we have Something called temperature is this temperature Basically, how creative is your model? It decides whether it can be. If If the temperature is low, the randomness is low. There will be. That means your output will be Creativity in responses It will decrease if it exists. Just for While training the example model Let's say this is followed by a specific word. It came mostly. Okay so your output This word also appears in . Let's say you If you increase creativity, you increase temperature. Obviously randomness increases. So its Why not try it with something else?

14:44 How deterministic is the word "OK"? Your model should be temperature. Okay, there is one more, along with the temperature. You know the parameter that you can play with. For similar requirement which is top That's p. This basically involves probability. A concept called mass is used. Now suppose the top P is higher. Will Actually Give You No Money Creative answers are the same as top p less Please provide more factual information. It will be relayed on. So this temperature and What if the top two P's are also low? If you need factual answers If you want, let me give you some information. I need the correct answer. Then this temperature and topic is low. Should be. Let's say I am a movie story. I am writing this because I am creative. I want more, I want more randomness, so

15:30 Note that the top P and temperature can be increased. What if you want C, top P and You can change both temperatures at the same time. But the recommendation is that the temperature Change R top P change Don't change both at once and then change the Have maximum length. This is your Restrict response length It is necessary to do. That means for me To say no more length We will use this. Similar stop The sequence is also for that. If you After the so-and-so stop sequence for me If you say no to output, then you will The output stops. You are on the list. Generate a list of items. Generate items and up to 10 Stop 11 saying generate

16:16 If you put it in sequence, it only comes to 10. It stops. That's how it is. Okay again These two are the same, but the other max length is Except using the stop sequence. Do not use both at the same time. Okay, great. Then we have frequency penalty and Presence penalty frequency What is the penalty, sometimes? What happens is that you get the same response. A token is the same word repeated over and over again. And It may irritate you, which means you You may not like it. So you frequency The penalty can be increased. Because of that What is it, let's say one word, let's say a If you think the word "Asthamanam" is coming, then Frequency penalty per word It's increasing, so let's say 10. If it happens several times, 10 * years penalty will be imposed. If Annamata Times comes, 2 * sum penalty so Obviously, the word that came up 10 times

17:01 Next because the penalty will be higher You will get that word in that response. It's okay to say that the chances are low. Then the second one is presence. Penalty is also similar but in this case Let's say the word x 10 times. The word "vaalivai" has come 20 times. Penalty for y in frequency penalty It takes more than x, it takes less than x. But if present penalty is used then x and v The same penalty will be imposed. Repetition It means it reduces overall. Single word repetition does not reduce Presence penalty is okay again similar Here too, the frequency of use is different. Penalty or presence penalty Do not use both together. Okay then we This is the output in json You will be particularly lucky if you are in Gemini. Using this, I get the output JavaScript Object Notation

17:47 You said you want it in the format. You can say. This option is also available in Chargebit. Even if you don't have it, you can give me the prompt. You can say that the output is in json. Okay You can include that in the prompt. Directly okay so this is different Settings that you can play with if you Are you interacting via the app? Then you If you look here you will see the system user and Here is another roll. Witch Can Be Assistant, okay, what does this mean, user? The output you give. What is an assistant? The output given by the model. The system is you. A specific model for that model in advance A character used to give a characterization. Configuration Let's say I'm here. An experienced software engineer That means it will think in terms of Software Engineer Only Legal Terms

18:34 Doesn't think. Okay and here is the user and If you want an assistant, then this is what you need. If you get a question, answer it like this. You are the model answer, too. What will happen if you give it away? Model Will understand that okay like this If a question comes up, answer it like this: Saying that, "So these are the three." Personas that you can use with the system User and Assistant So One More Yano Key The point in this video is about Prompt elements So in your prompt Whenever these four elements are present It will always be better first. Instruction according to the context Third the input data Fourth the An example of an output indicator Let's say I'm a four-line. Okay, summarize the existing paragraph. So in this case my instruction would be to

19:22 Summarize OK Context Would Be Basic Let's summarize this in some context. Who do you think this summarization is? A software engineer uses a Is a product designer used in the legal team? What happens if someone uses it? Summarization is also such a technique. You can do it with or without using terms. So That's the context input data. These four lines are also what? Input data output indicator means As mentioned earlier, Jason needs R. Do you want comma separated values ​​in CSV? Do you want values ​​or do you want a summary? You can choose whether you want the output in points. It can be mentioned as an indicator. So this case In one good morning will be the summaries. Following

20:08 Text following Text for that Legal Team output The Response In pointers Saying that data equals to three like this Yano Open Balance Four Lines and Three closing balances this way you This means you can highlight specific data. This is the opening and closing of the three yanos. Rotations are okay so definitely this is it. Just start and let's discuss more. About her techniques and all of that And coming videos will be available directly

20:53 Start using the chat gpita ui Let's chat about this instead of Writing it on the white board so stay Tuned Actual Prompt Engineering Let's discuss the techniques. First If you are new to the channel, please subscribe. Like the video for your friends. Share how you like this series. Comment if you like. Because of this Engagement reach increases and I Motivation to make videos like this Increases. So please help in that. Okay, so what we're going to discuss today is What is Techniques? The first one is Zero short prompting is okay. Zero short Prompting is the second technique. Short prompting

21:39 Okay and third technique is a bit short. These three are the words that are used to prompt today. We will discuss in this video. So this Before I explain the techniques, watch these videos. Watch the video and let's give it to you. There is a lot to discuss. There will be similarity. What is happening here? The hero first threw a stone, but it did not fall. Say "Sir" and try again with another stone. He does. How long has that stone actually existed? Try until the cone falls into that bottle and into that cup. We will do it. Okay, so if it's the first attempt If it falls within, we call it a zero short. Prompting, if the stone goes, If it falls into the cup, there are still others. Attempts are necessary or not because Since the stone has already fallen,

22:25 Not outside, but inside, so similarly you The first time the prompt was given, the first time The input given is good enough for you. If you get a response, you can still Don't try again and again. You don't have to give a prompt. If you answer at the first prompt, it's The best case, but most likely it won't happen. But if it happens, it's called a zero short. Prompting means you have to use these stones. If you want prompts, the first one you gave Go to the prompt and you will get a good output. If it comes, it's a zero shot, okay? About One Shot What is the first in One Shot? A prompt and its corresponding message to you Give the two required responses and that Then respond to the province. They are expected correctly. That is

23:13 If you look at this, in this raw case you Already gone and tried it first with my own hands. It's okay to go near and say it. That might give you some hands-on practice. Saying it will happen, the second time again They write. If it falls, it is one. If you miss the shot, you have to try again. Then what will you do? You will use few. Short prompting means prompt one I need response one. Province I need a response. To prompt 3 I need a response woman. So this way The prompt I gave when there was a pattern What is your response to 4? So you are a R4 is expected to Expected value is expected. If you get the answer as, you are successful and Go directly and print this picture. You can use it in that cup beforehand. Go and throw some stones directly.

23:58 It's okay to say that you'll do it soon, then Will try. Why practice already? So may that be the length of breath. Try saying you can judge correctly. Will do. If it falls, it's great, if not, it's good again. Okay, I'll try the shorts a few more times. Go closer and then practice. Must do. Similarly, here you are Before the model, before the LLM model If prompted like this: There will be a response. Prompt like this If so, the response would be like this Tell me and give some examples. How to answer the examples Next is coming in the same way. Saying "Give answers to the questions" I'll tell you. That's all about you, you know. Short prompting techniques. This short Because of prompting techniques Advantages: If you have zero Uno is correct in short prompting. The response you get means you are the same. No matter how many times you try the prompt

24:45 Differently, mostly you are bigger. There is no difference. A prompting shot How do we say it was successful? This is the prompting shot if you ever Even if you try opening it in a new chat If the output is the same, you are Successful. What does it mean if it doesn't come? Whatever you give, whenever you give, every time. What is new is not what is new. Deterministically deterministically When you are not in actual applications You can't embed this. Expect the output to be the same every time. You said you would come and expect it. Can't. As expected, if Only if you have behavior can you get that shot. You can say it's successful. So If something is shorted, that short is shorted. How many times can you repeat the prompt? Even if you try, you get almost the same result. It comes so you can embed that into it.

25:32 Your application is valid. Expecting the result so how? You will know in advance what is coming. Okay, so let's go into the chat. Try this live so I can Try some examples first. I made it and kept it so that it will take. Less time for us to go through it so here If you see me without any other shots Zero Shot Sun Rises in the East I gave it. So in response to that, S Saying that's correct, we are told this way A response was received. But let's say I want it this way. So I am here. Trying to shoot Sunrise in the The answer to the East is true, it is universal. It comes saying "Truth". That way Acidic substances when taken care of What does it mean when you say "R Sour"? So here it is, from that's correct. If you look at it, why did he come to True?

26:19 So here is True but still Here's to giving you a lot of extra hammering It happened. So I'm Yuno again. I tried a few shots, but more than one is true. It is a characteristic property. After saying that, another question was asked. I'm asking, "North and South." Poles of a magnet repel each other. Actually, it's not attractive, so here it is. I am a false positive. I am doing it. So if you look at this, here you go. The ans did not come as a dot. But few shorting Because of that, we also have ans dot formatting. Also followed by ans dot false it is Tell me what characteristic property it is. We got the answer. Then again I think the cheetah is the fastest animal. I gave it. The name ASA is a good start. It's true, it's true, but we're screwed. I'm so happy that more will come.

27:04 Trying a few shots again here. I say, Sunrise in the East again. True Sun Rises in the East So Me What I'm saying is, if it's correct, then so be it. If you put it wrong, give the correct answer. I'm saying that, so that's correct. So we wrote the same thing again because it's correct. We wrote the same thing again, so it's false. What exactly are the North and South Poles? Attracted, so we wrote it correctly. Then This is correct, so we wrote the same thing again. Then the Sun is the largest planet. I said "In the Solar System." Witch It's false, so it's false. We need to get it right. So it is not that. Large Planet Largest Planet Is It said Jupiter. So if you If we watch Sea Now from the first here Complete this journey of propelling Few short last whatever is prompt. It looks good to you because we Expected execution behavior

27:51 Our LLM replicates it. So A simple example of that is: Now I have the same problem. It was given in a new chat in a new window. Should work similarly if you look at it. It is working more similarly. There is no difference, it's the same as saying The answer came. So anytime you are okay. When writing the tests, they should become Deterministic means that you can predict that response. If you expect it, you will be able to If the short is not successful It's not because of the shorting principle. That's what you expect when you say it. That's when the output is generated. Why your short is successful Because you can do this in your application. When embedded, after this again Code is written, then logic is written. They write. That logic is the previous output.

28:36 It depends. In that case, this Every time it comes to its liking It's difficult, so you will have To maintain pattern which you can Expect Before the Prompt Itself Okay, so this is how you short. Principles can be used so mostly In most of the cases you have zero shot. There is never a correct answer in One Shot. In rare cases, mostly few short You will get the correct answer in the prompting itself. Comes. Now this is the advantage. I told you, you can expect it. Will become a little deterministic but Disadvantage is logical questions. This is a short prompt. Let's just say it doesn't work sometimes. So what does logical mean? Let's say a There is an integration question, okay? Complex Integration Question So Me

29:23 An integration question followed by the answer. Then follow up with the integration question. Answer then integration question It's okay to say "follow by answer" The fourth question is an integration question. There is no correct answer because it is just Not direct statements Directly it's not you know it From the previous training data, it Directly its information I can't get it, I have to apply logic. So this Mathematical Logical Reasoning Based Reasoning Based on Examples for UNO This zero is short for questions, one is short. Few Short Pronouncing Techniques It doesn't work, you have no chain for that. Of thought and all that which we will Discuss in the upcoming videos Some people watched Day One and stopped watching Day Two. Some people saw Day One Day Two Day Three

30:11 Stopped. Some people go from Day One to Day Three Seeing this, they will stop today, Day Four. That's for Sure That's How It Works But That's All About your commitment guys, because you Committed to 10 days, 10 videos only If you can see, you can prompt engineering. Comes. No, we'll stop in the middle. It's all up to you today we will discuss. Technique to be performed The ESCO statement was short yesterday, but today it is cut. Okay so this cat stands for Chain off Thought Prompting is okay, so stop. Pronouncing Stands for Chain of Thought Pronouncing Today's Discussion Let's do it. It may be a bit short, but

30:58 Important concepts first, please. Hit that Like Share and Subscribe button Comment how you feel. In you Many people have asked me, "You are the one who prompted this?" Where did you learn engineering? You are asking me, aren't you? What courses didn't guys do online Articles, blogs and some research I learned to read papers. You too That's what I'm teaching. Okay so first off Let's recap everything about the short. Pratting is okay, had one shot, okay zero. What did we do with this shot and few shots? What did we do before? Question answer Give the question answer sequence and We will answer the next question accordingly. In a way that the answer is expected. Let's see. We are trying to make The model is deterministic, so this process

31:45 What we thought in There Are Few Disadvantages and those disadvantages What if the questions are for reasoning? Related or mathematical Are related or not logically If our model is directly this short May not work from prompting. We thought so, that's why we are here. Introducing the So This Chain of Thought What do we do when we pronounce? Basically, we If a man could solve this question, How to do it step by step Think about that step-by-step idea. We will inform it. Don't you understand? Let me tell you. Let me tell you. I'm using a QWERTY keyboard, okay? QWERTY keyboard means you can type normally

32:32 Use it on your laptop with your mobile phones. The keyboard is the same as the one on this keyboard. What do you have next to you? Right. What's next to the side? Okay, what's next to x? Okay, what's next to x? What's next to you? So I'm here. Can I ask you a question? The question is, let's Service Stands FurbyC You for example okay in that case next I will ask the question, what am I saying? DSVAMSI Stand for So you are the answer to this question. What do you do logically? Okay, right? There is a side B so again this is it.

33:19 The right side is B, so B is written OK. The right side is new, so we have it. So what is the right side of A? Okay, what is on the right side of m? We have M. There is actually nothing on the right side of the It's a comma so comma comes.S. What is on the right side? There is a d. I have a right side. What's up? Okay, not O, this is D and O. Yes, so we have the corresponding with Vamsi. Answer Is ESBS D? This is logical for you. But you got it from answering. LMS may not get this logical reasoning. So what we do is step by step. What does it mean to say "I will tell you step by step"? The reason for the B is the right side of the V.

34:06 Because there is a lobby The reason for t to get y is the right side of the Our logical conclusion is that it is because it is low. We will also give the steps in the answer. Just say okay and answer next time. What do you mean, try to tell us that? Does. Okay, so we have one in this too. There will be a short cut and a few short ones. There will be a cat. Okay, obviously zero shot. It doesn't work, okay, so one short cut and Too short I mean too short a cat There will be. This means that in this case you only have one Give an example and give an answer. They are asking the question next time. So they only tried one shot. It means they only tried one shot cut. Chain of Thought in One Short Explain and answer the second question.

34:52 You are asking directly. Work If it's done, it will be great, it will be one short one. If the cat doesn't work, you can go further. Let me give some more examples. Its Then you will get the correct answer. So it will become a few short cuts. Obviously you mean Dayfor you. You should have seen this cat three times. This is called Chain of Thought Prompting. Gagu introduced the technique. Research team Annamata Okegagu Research A research published by Team OK This technique is the result of the paper. That research paper directly to you. I'll show you, okay, this is the actual one. Research Paper This paper is for 10th Jan. Released in 2023. So this is it.

35:40 The people behind this technique are okay. Okay, so the Google Research Brain Team There's a twist here guys, that twist I'll tell you at the end of the video. But for now Let's go through this so you know what I'm saying. What do we do in standard print? Standard Yano Short Prom in Ragar He has five tennis balls, he is the best. Cons of tennis balls each with a handle Three tennis balls how many tennis Balls does he have now answer is 11 that Next time we will ask one more question. It says wrong. Because it So what is this, since actual reasoning doesn't come? Actually this is a one shot. Prompting so one shot prompting this The case failed. So that's why They are using chain of thought prompting. In this, Roger started with five balls. 2 Cons of 3 Tennis Balls Each Is

36:26 6x tennis balls so so 5+ 6 Is 11 so in the first case it got 11 That said, that step-by-step logic I will also explain. After that Basically the cafeteria had 23 apples okay If they used 20, they said it again. If you ask the next question, this time it was Able to actually get the right answer Because of this chain of thought, okay? Hope you understand this name. I think the name is Chain of Thought. That is, we can use sequences of thoughts We will give it in the morning. One man sequence If it tells it what off thoughts are like That too to replicate it Will try. Okay, so this is complete. Read the research paper if you are interested. Okay, actually I have read this. So if you are interested, go through this one more.

37:12 The thing is so this cat is whatever it is. Prompting technique This technique How big is your model size actually? It works as well as it can. Okay so Here are some Uno Multiple Examples for you. There are. You can go through them. Interesting paper so if you look here you will When they were in the hospital and Trying to check this black line is Standard prompting and this circle So, what is the meaning of the chain of thought? Almost a chain of thought in every case Pronouncing performed better when the Number of parameters OK model scale It's okay so that's one more thing to do. The note says it all, okay, so just go. Through this I will share the link so you Look at all the examples given here. I mean, it's not too difficult to see. As a child, in the third and fourth grades at school

37:58 We will have solved it, probably today. It will work in 2024. Let's check that. Okay, it didn't work in 2023, but let's see how. It works today so if you see the correct The answer is 9. The correct answer is 9. But in 2024 without using the chain of thought It's Able to Solve Just with the Short writing can help you with this. Do you understand what AI models improve? They continue to happen. Okay but even then Any complex logical problems Solve these models today. If you can't, the COOT will still help, okay? So in the same way, this quiver given above I also tried the keyboard example. So Here I am without actually giving anything. This is an example of what it actually is.

38:46 Is Zero Shot Pratting So Zero Shot It worked. So actually this We understand that the day AI models are improving every day. Because of the data training and All of that twist is in this paper. Many examples today directly They work without a vote. Okay However, that's a valid technique. Learn but try, okay just be Sure of that great yesterday There is a doubt in the video, please clarify it. Let's do it. So doubt us Satwik asked Anna, now you are basic to me. It seems like you are explaining. If I want to learn in deeper how can I like today's video, take an example. Only You Said About How Cat Works But how can Kat still work?

39:34 A little bit of clarity is missing for many people. Doubt is Anna's 10-day challenge. And a 10 to 12 minute video a day We are uploading the duration, right? So 10*12 at most gives us a 120 Minutes which is approximately 2 The hour is coming, isn't it? It's time for the hour. Is prompt engineering enough for us? Many of you have doubts. The answer is Enough. So whatever is same for Learning the example full stack Okay, full stack, 30 hours of content. Learning. Those 30 hours off Even after learning the content, you will still be able to What about this full stack? If you can't learn, that's not how it is. This is the place to start with. Enough. Similarly prompt Engineering to start with these hours Or whatever duration of content

40:20 Enough. Also I am happy for you. Almost from yesterday in the video too Research papers are also linked. I am doing it. So you can go further. If you are interested, read those research papers. Read. Because you already have first. I said it in the video. Prompt for now A well-defined term for engineering What's missing is structure and that's the plus The point is that you have this limited awareness. Learn things fast in the domain Okay, let's move on fast. Definitely don't worry about it. You are learning well. Okay, got it. Great, in today's video we will discuss What is going to be done is a self-consistency Technique is self-consistency. Okay, so we discussed that in the first video. About I mean the first technique was It was all around shorts so it was zero one.

41:05 And few shot technique then Discussed about this again in this thread We can do one shot cat two shot cat There is no shortcut, so this was it. Second technique so today third The technique is self-consistent so What is happening to us in this cart? Normal human thinking process You think about replicating Reasoning or logical steps They tell the LLM that Steps are also further away. Just like those logical steps for shorts It will think and give you the right answer. But What did you observe? What does ECOOT mean? One disadvantage when using is The model scale should definitely be higher. I told you, so number one. Parameters It Should Be More So Model

41:50 The scale should be higher. Disadvantage number one and two is that Common Sense and Symbolic Reasoning Use based questions more often. You also solved it correctly. They're not going to do it. So today we have Self-consistency. So this self What does consistency mean in CO? Explain the logical steps. Those logical steps were taken. Explain and give it to LLM Ask the question, "What is LLM?" The response received directly The answer is fixed, right? What is self-consistency? You tell logical steps. Okay, similarly, you ask a question you don't know. But you are not a single response. Discovering the various reasoning parts

42:37 The most consistent answer is It will be considered as the final answer. Do you understand? So again this is also from This is what the Google Research Brain Team says Also research paper based Uno Technique is so I will take it the right way. Paper is this, paper is that again, I am like that There is a link in the description. If you Are interested read the paper and I Would highly recommend OK really to you There is interest in this field, that is, in this Read the papers. So the paper title is Self-consistency improves chain of Thought is not reasoning in language. Models Again This Is From Research Brian's team is going down, okay? This is how it is, so we like it. Overall, what is the summary of this model? Technique means chain of thought.

43:24 Giving a gift in a gift shop From the language model to the You answer the answer directly. Considering it. But this self From the language model in consistency You have a diverse set of reasoning parts. Sampling is being done. And in that The most consistent answer in this case is 18. 26 18 Very consistent 18 Final Answer 18 and that's the right answer to OK. So I will talk about how you will figure it out. Out of the diverse set of reasoning Parts But Before That Here We Go What is the logic in the technique used? Simple Guys in the middle of a group of people Are you a common sense person like this? When asked reasoning based questions Everyone has a different perspective. Maybe even among humans, okay?

44:10 But in the end, what do most of us do? Whatever the majority says is correct. We hope and mostly let's say to you There is KBC, so which one of you is a millionaire? Okay, so we'll have an audience poll. Why the Okay and Mostly Audience Poll? It will be correct again the same thing. Self-consistency is okay so a group When is the majority when there are people of They say it's right because the majority of the Other perspectives are always right in The group of people so similar approaches It's okay to say that we're doing it here. This diverse set of reasoning paths How do you sample? How do you find out? That means you will also have it in the same paper. Okay so if you scroll down there will be There is something called sampling. Where is it that a sampling scheme is okay to

44:55 Sample of various reasoning parts Follow Similar Settings to Dos Suggested in these two papers okay this The two papers are referenced below. You can read them okay but overall What they said was this temperature It's okay to say I'm using sampling. Temperature sampling with t= 0.5 5 and Top k = 40, right? They are sampling. If we say we are doing it, then we are already ready. Discuss in the very first or second video We can watch that again. This is the initial, just like yesterday's vote. When is 2023? So in March 2023. Now without such a question Self-consistency means without Sample Divers Set of Reasoning Pass We don't always get the correct answer. 18 once 26 thus different answer

45:43 The ones that come are okay, but today Uno is even without Doing all of this directly to us The correct answers are coming. Why? Because obviously training has increased. And the model scale is also quite a bit. Increased. But this technique in the future If it doesn't work, that's how it is now. Based on how complex the question is You can use these techniques as your prompt. It can be used in the engineering journey, even used. Got it great I hope it was clear Thank you very much to all my friends. Share Like Subscribe Do it and many of you are asking. Can we write this question in our resume? Skill means you can write. You are On the right path. Someday in the intro Ignore him by saying, "That's what I'm saying." Don't hit. You support this series. If you do, you will get more free series like this. The motivation to take comes. So Hope

46:30 You understand and support the series. The technique we are going to discuss today is What is out-of-date learning? That's it. Okay out of date learning Let's discuss the technique. Butt Ya once again gentle request like Share and subscribe so this is out of Let me tell you what is the point of learning a date. Let this chat be one thing, GPT. Let it be Gemini or whatever today. These are also available for any LLM you have. Not in continuous training. That is, C is our LLMs are any large language. The models are completely based on them. Now, if we were to use the trained data You go to Chat Zipty and watch the 2011 World If you ask about the cup, that's your answer.

47:17 It says. Because that 2011 World Cup Know about the cup. But the recent IPL When asked about matches, he didn't say anything. Because its training is there. It's not over yet. So every large language There is something called a cut for the model. Off date okay this cut off date what is this date Even the amount of knowledge that is previous to They have already trained on that model. But this After the date, knowledge is also a model. Unknown. So many times What happens is that the obviously recent data The model is unknown. And many times we You will have to work on recent data. So what do we do then? So basically Already in these LLM modules As many learning parameters as needed

48:02 They are already tuned. So you Give new data and do what you want on it. If you ask me again, it will still be there. Work that is what is out of date Learning is for example SRH. Was there a match vs RR? Chat GBT doesn't know about the match. But give this commentary and that commentary If you ask questions based on that, it will Be able to answer OK and this technique Many times we need a lot. Okay So to give an example let's say I am ESPNinfo.com From ESR vs. R Much Commentary First of all, I copied it. Let's understand what the cut off is. Date for chat gbt free means chat gbt 3.5

48:47 So we can ask it directly what it is. Your knowledge cut off date is my last. Training data includes information Available until October 2023. Means I May Not Have the Latest News R Updates Beyond That Time Let Me Know How can I help you so October What data will be available after 2023? Chat gbti 3.5 I don't know. So obviously our previous I don't even know the RR-SR match. So What I did was summarize a cricket match. Based on the following commentary So, give me three crores like this and that Copy and paste the commentary directly. I did it. Okay If you look at the response let's

49:32 Scroll down and you will see the response. In an IPL cricket match, SRH Faced RR in a thriller EncounterRR was in a good position. With the game in the grasp of us Jaiswal and Parag were playing well. After Early Departures of Butler Anderson Howeever Borrissed in Quick Succession putting or under pressure so That way it gave some summary. Can Ask Father what could Have RR Done better to win this. Match is okay match

50:19 The Rajasthan Royals were close to Winning the match against SRh But there are few areas where they could Have performed better to secure a victory Having said that, here are some points: Gave. So you saw it that way, right? You can also talk about things that you don't have information about. Input your own data and click on it. You can ask questions. That's what it is. Called Yes Out of Date Learning That's it. Okay, so yesterday there was a video. One comment. What is that comment? We already talked about it in previous videos. Chain of Thoughts and Self We learned about consistency. Okay this. Both are actually research papers. It's okay to say based techniques. Important techniques are present outside. As soon as Uno Prompt Engineering said Mostly covering this, okay?

51:06 But this is research paper based. Techniques So therefore this research The papers are old, so it's 2023. What are the examples they gave? They are now directly straight away. Working. So that's what I have for you. I explained. So the question was If they work directly around Anna If you are still doing it, then that Techniques Why if techniques Please give real time if you want. They asked for examples. So The point is that there can be instances Where can you still use this in Future then those examples Can you give me some examples? Now I have to search for it separately. Because research is not possible again I can't do much, so what can I do? What you did was at least aware of it.

51:52 Explain those papers saying that there will be I did it. So that you can do anything in the future. When going for job interviews, ECOT and You probably know what SC means. So That you will be confident with yourself Interviews That's the Whole Point of It Okay, that's it, but something to increase the content. This is not to confuse you. Important yes technique okay great me A suggestion for everyone: If in 2024 If you want to get jobs, you need to go on campus. Decreased directly on off-campus Applications have decreased, so you have a referral. It has to come through and referrals are very important. Very powerful tool, okay. Referrals are people you know. They want to give, but there is nothing. So you Go directly to someone on LinkedIn. Even if you ask, if you have the skills, they will give you Referral will be given. I have said it many times.

52:37 Because if you don't select They also have a bonus. Got it So what did you learn from this? Prompt is not just a wish, it's okay to learn something. Posting your skills on LinkedIn Stay. That will build your profile feed. Yes. You came to me today and said, First off, if you need a referee in an attribution I will definitely follow your LinkedIn feed. I will check. Not me, but anyone. Let's say you're getting married in July. Go and tell me, sir, what is this in your company? Another referral for the prompt engineering role If you want them to open your feed, If you look, it will already be in your field in April or May. Posts about this prompt engineering There are. Okay, that will give him. Confidence is okay, it's something that comes to him already.

53:24 A person who does not try to interfere is called Confidence comes. Because of that you The chances of getting a referral increase. So what will you do from today? I'm learning, okay, please post that. On LinkedIn if you want you can tag I like this post for you, okay? People on the network are my posts. You can get it because you saw the likes. Opportunities too okay so do this a lot It will help you, okay, so today we The technique we are going to discuss is the roll. Playing technique is a role play. The role playing technique is so simple. Very this is okay chat GBT R R Gemini R Claude R Any LLM Okay A Even if it's an LLM, it's super super, another one.

54:10 Two or three super intelligent people. Okay, so you know about coding. Know about the medical field Know about manufacturing Everything you need to know about rocket science I know, so how can you be so super intelligent? One problem is that you, let's say, They are asking a question called ABC C. Asking a question. That question I don't know what domain you are asking about. Let's say you're talking about gravity. Let's find out. Okay, simple. Let's learn what gravity is. They think. But this simple gravity What does that mean? From what perspective are you? Let's find out, you might be wondering. Are you in 10th, intermediate, or JE? Whether it is for mains or advance or you Rocket scientist, that's what space is for.

54:57 What do you need it for, for research? Based on the internet data all this From Perspectives to Answer LLM I don't know, but which one do you like? So that's why role playing is needed. It gets important so role playing What do you do? It's a three-step process. Process and first aid Role assignment is okay, second step is Actual Question Third Is output instruction I will discuss that first of all role. Assignment means you are doing an LLM. They ask you to think about it and answer. A two year old kid sings a 20 year old

55:44 UNO Fresher Graduate Gana A 20 Software with years of experience Engineer Gana R May Be New Now A UNO who joined as a scientist How do you answer as an employee? What is expected is the role. Assignment. So what are you doing here? Say you are and roll so and so They give. You are that experienced. Are you a software engineer? You are an experienced script writer. That experienced rocket scientist So that after that you will need the actual Question: What is this output instruction? That is, let's say it's a programming question. If so, in which language do you want it? Any specific package in the language Is there a requirement for all that or just one?

56:29 Think of it as a rocket science question and answer. How do you want you know S R no type Do you need an explanation? Raw data is separated by commas in CSP. How do I get it in values? I want it in JSN format, so that's it. You can feed the Indian system, that's all. You mentioned in the output instruction Will do. Okay, let's take some examples. This is a straight away prompt. Write Code to Generate QR Code for The website is obviously well-trained. So directly you can code in Python He gave the OK and said what was in the script? He explained it. Butt Definitely what you are looking for in this. You don't know if you're using the function, but Let's say let's try with you. Experienced Software Engineer I Want to generate again same instruction

57:15 Same you know question provide the code InPython with Step by Step I didn't say that Python was an explanation. Let's say Python took it by default. I need JavaScript then so that's it. Output Instructions Important Okay, if you compare this with the answer above, it looks like this. Is more detailed actually okay right? Script for YouTube video on ChatGBT OK Instructions No role given, no role given, random It was written. Okay, but when you say you are Experienced Script Writer I Want a script for So and So Make It Engaging with hooks and interesting Elements So depending on that you have the How should hooks also be visual? Everything is clear when you say it. Gave. So, guys, today is what I said. After this video just try them with

58:01 Your own examples are okay with roll and Without roll and you really want this Unlock the power of LLAMUS Let's do it. Prompt Let's use engineering to the right power. If you think so, you can discuss with us. Intercombine the techniques being performed You should keep doing it and using it. That means only Role Playing Only Shots Only Not that way okay you shorts and roll Even better if you mix playing The results will come. That means it's your first time. It didn't come out right. So maybe you can say I want it this way. Okay you are so and So this is the question, this is the answer, let's say Again one shot two shot try that way You will get better answers if you do. Okay, got it, great, I hope you like it. Definitely how the video felt.

58:46 Comment what our technique is today. That is RAG prompting is OK, ERAG. Prompt as a prompting technique For those who want to use engineering It will help a lot. So it's in the field. ERAG prompting is more for those who have It will help. If you are in non-IT Alternative to this is non-technical What is the solution? It's already day five. I will discuss in six. It is out. Remember the out of date learning? Date learning prototype is Non-technical for ERAG Pronouncing There is actually no solution. The alternative is OK, so ERAG. LORAG stands for Retrieval

59:33 Augmented Generation Pronouncing Annamata SORAG stands for Prompting Retrieval Augmented Generation Prompting Actually, these three words are for you. So basically, what does this RAG mean? Understand. First of all generation means What? We have this artificial intelligence. What do all the LLM models mean to us? Generate responses They will. That's why these are called ZenAI models. It is also called. JanAI already on day one I gave a brief about it. Watch That If You haven't had any but what are you all doing? Why are generative models new? Generates data. So Generation So prompting that helps Here comes G. Now what is Augmented means augmented on top of what is there. That's why AR adds extras.

1:00:18 What is augmented reality? Those non-real 3D objects in the world Adding an augment from outside That's why augmented reality is being done. So augment means extra with what is there. Adding something additional. What is retrieval? Retrieval means to Actually fetch something to actually get Something from somewhere so that's why you are DB If we have done side work, we can do DB retrieval. What do we mean by databases in basic terms? Retrieving data from Retrieval means fetching what is there. Whatever you have to do, wherever you are. We call taking it as retrieval. So this If you understand three words, you have RAG. I understand. Let me explain that. Yu. Okay, so we have all the LLMs.

1:01:05 That's why they work on context. You need the context length token length. Parameters are also common in LLAMs. So what does this context mean to you in general? What is the LLM train like? Feed complete data directly Mom doesn't say "get over it, train yourself" on this. From this you can actually multiply Chunks are divided. Okay, that. Embeddings from multiple chunks They say they will create and train. Okay So this context is basically that chunk The maximum size of this context Let me give you something very important. For example, if you tell me Famous I named it Biryani Restaurants.

1:01:56 So if you look here you will see Across the world from multiple places Telling you that there are many in the world Info from multiple places It was given. But let's say you were Let's say in Vijayawada. So this doesn't matter. Make sense of your context in Vijayawada Is an important keyword so Context is so now I'm here Same in the same prompt. In What is happening here in Vijayawada? Actually, some of Vijayawada's The restaurant names came up. Now the Basic The point is that now in this prompt You are Vijayawada in their context. By adding in the prompt context The response you need to be excited about It arrived. So this context is good. Important. So in this case Vijayawada

1:02:41 There is info online about it. It is working right away so context It's already on its train so it It's working right now, but lately you've been If you look at what AI is developing, what AI is doing? It went into the domain. One of it Is Actually Chat Bots Company Chat Bots are used in many companies' online What are customer service chatbots? They are all AI models, okay, so AI models. Isn't it very difficult to train on your own? It takes a lot of effort, a lot of effort. Data requires a lot of processing power. Instead of that, it's already trained. Let's say you use LL.M.S. The chart is GBT. This is definitely good. I can't go by train, so use this. If you do this, you can do it on your own. Just add the company data.

1:03:26 Make it into a different argument. If you find the responses Even easier and better with less effort. It comes up mostly lately. The company is offering additional training on this LLM. They input their data and create their own Create customized chat bots They are doing it. Okay, so in this case Context does not exist on the internet. Yes or no, it's already on the internet. Biryani hotels in Vijayawada are located in the vicinity. So we got a response. Butt Their company specific policies and LLM doesn't have anything to do with data. If You are using chat gbiti ui If you want, as I said before, that out of Along with using the learning base technique Give your additional data and answer the questions. You can ask. But here you are using chat.

1:04:13 Bots are good so build this. You should be able to provide data while doing so. In the coding, as a technical term, Then we have this retrieval augmented The generation technique will help. What happens in this is that first of all you Custom your company's own data first off Retrieval for All Context It happens. Okay, so additional data. Retrieve. After that, your Retrieve at the actual prompt The data obtained will be augmented. That means they will add. And finally this ad This augmentation made a new prompt. The output is generated with. That's What is Retrieval Augmented Generation? That's it. Again this retrieval straight That won't happen. First off in this retrieval All you need to do is fix the data source. Let's say it's a document of yours.

1:04:59 Company internal source is okay then you It should be indexed. Okay, indexing. After that, query it. Okay, querying. Once done, it should be ranked. Ranking Even after it's done, you finally get it. Augment and then generate That's how it is. It's involved, so what are you doing? First retrieval will be done. Includes all of this plus augmentation So augmentation is simply a prompt in Uno. User question so and so okay then Context is you. No additional knowledge so and so Tell me and give me this retrieved data. And finally you have the underlying chat. Let's use the GBT architecture If you want, they can do it simply. Using APIs to get that specific

1:05:45 They hit with a bang and that's it. Generates a response and That's your required response. It's so this is an exercise for many of you. I know I want to see what the code looks like. I like it with a long chain and without. How to do it with a long chain Next Explain in the video, part one. I will. Okay, I hope you understand this. I think I understand. Obviously this The video is so helpful for non-technical people. It may not be possible, but technically you This is it, if you are in the IT field. Definitely a good point to you. Learn I Think This Way on YouTube Practically implemented I can't see any YouTube videos. Not Just Telugu Guys Hindi English There is no one else so it is definitely for all of you. I'm trying this. Help please.

1:06:30 I think so. How many of you support? I don't know if you'll do it, but if you like it, just like it. Do share the channel with your friends. Share this playlist Do share, okay and new. Subscribe okay so just Prep Guys Rag is Retrieval Augmented Generation So What's In It? What we do is we use an external data source. Retrieve data from and Augment with a prompt and that prompt Use LLMs. We will generate responses. This What additional data is needed? The context is not available at LLM. You provide it directly. Can be done. Actually using you. This is what you call out-of-date learning. You can use the technique, but You can build your own applications.

1:07:15 If you want to do it, that chat is GPT U. Can't use. So applications To build, you need to use APIs. This retrieval is also automatic. Must do. And the augment is the same.

…