AI Prompts you MUST start using | Prompt Design Tutorial

[Music] welcome back to Dynamics dojo in this video you will learn about prompt design for AI and how to get the best responses from large language models by writing effective prompts you'll learn what a prompt is how to write better prompts and how to get better responses I'll show you some patterns you can start using now in chat Bots like co-pilot for the web and chat GPT so let's get started [Music] a prompt is a text that instructs AI to generate a desired response a simple prompt example is what is a computer you can see the response I get back is relatively lengthy and detailed with useful information but what if it was a more complex prompt an example of a complex prompt is act as an IT consultant and explain what data verse is assume I am a 10-year-old you can see the response I get back is very different I get the information presented to me in a different way as if I was a [Music] [Applause] child you want to understand how to design or write the best prompt possible to get the best response from AI to help you with any given scenario a well- constructed prompt can help the model generate top quality accurate and relevant respons answers for you in this video we will look at some different prompt designs to help you get the most from AI models and help you in your day-to-day [Music] [Applause] tasks to begin with let's look at some essential prompt Design Concepts a root prompt is the first prompt the initial instruction you provide to the model to start your conversation this helps set the interaction context for the conversation for example explain what Microsoft copile is conversation context generative AI models retain context of your prompt which allows you to have a conversation with the model and refine answers just like you were chatting to a friend or colleague unlike in the real world AI models have context limits for co-pilot it's 30 aerations in a single conversation here's an example of conversation context first of all I'll ask is there a studio where a build my own co-pilot once I get the response back I can then say how do I get access to Microsoft co-pilot Studio I'm now interested in finding out how I get access to Microsoft co-pilot Studio I get a response back telling me how I can access Microsoft co-pilot studio and based on that I can ask more follow-up questions what are the system requirements Microsoft co-pilot Studio delusion or hallucinations a delusion or Hallucination is when the AI model produces false content it can happen when the input is not understood by the model it may also happen if the model's training data is inaccurate or incomplete the key to a delusion is not to treat the model's response as 100% accurate key takeaway here is you must always verify and validate all generative AI model outputs you must ensure the outputs are correct and appropriate to use non-determinism when using generative AI the model can produce different responses when provided with the same input this is non-determinism this can be confusing because it means there is no consistency in the responses however this is also a good thing as it provides variety diversity and creativity in those responses let's look at Microsoft designer we'll use the same prompt and keep generating a different response here you can see we get the same cute little dog with a different variation of a jum jper and a crown fantastic how sweet if you like what you're seeing in this video and you're learning something new please subscribe to the channel prompt examples in co-pilot so let's take a look at some prompt examples in co-pilot and we'll use Edge for this formerly Bing chat Enterprise now co-pilot it's highly recommended you do this logged in with your work account so that your conversation both personal and Company data are protected you'll know it's protected because you'll see the green protected signs when choosing a conversation style you can choose between more creative more balanced more precise depending on what your use cases and what you're looking to generate if you hover over each of these you'll get a quick overview of what they're going to produce I'm going to go with more precise because I'm looking for a more precise accurate response but as mentioned you'll still want to review and validate the outputs so let's look at some prompt examples first of all we'll start off with instruction prompting the idea behind instruction prompting is to provide a clear and succinct prompt to the model and we want to be specific with this type of prompt so for my example generate a list of Workshop techniques to elicit requirements over a 3-day Workshop here you can see I get really good and detailed response which helps me generate a lot of ideas day one suggests I do ice spray Breakers so I'll ask co-pilot to provide me with some ice breaker ideas I can use as I'm not very good with these I don't actually need to add in use on day one at the end of my prompt uh but I'm getting into the conversation swing of things here I get some helpful ice breaker ideas and I can continue to iterate through the responses asking copilot about other elements it's suggested for me zero one or few shot examples here are some definitions of these terms zero shot refers to a prompt with no examples one shot refers to a prompt with one example and fot refers to a prompt with two or more examples here I'm actually using co-pilot to explain to us how these work and what some examples of these prompts are so here I'm saying write me a zero shot prompt example and it provides back you are a highly intelligent AI model Write a brief summary of the novel Pride and Prejudice by Jane Austin then I've said write me a one-hot prompt example given the following example input she was walking a dog in the park output she was strolling with her Kine in the recreation area now paraphrase the following sentence he was driving in his car on the highway now I'm going to ask write me a few shot prompt example given the following examples example one input she was walking a dog in the park output she was strolling with her canine in the recreation area example two input he was driving his car on the highway output he was operating his automobile on the freeway now paraphrase the following sentence they were playing soccer in the field and now I'll get back this response they were engaging in football in the open area so by using the zero shot one shot few shot examples you can basically describe to the AI model the format in which you want the response by providing it the different examples have a play around with it yourself and see what you can come up with leave any examples you can come up with in the comments below now let's take a look at some prompt patterns prompt patterns are instructions guiding AI responses for specific tasks and are defined by core contextual statements that enhance the Precision and relevancy of an output from a large language model let's take a look at some prompt patterns now that you can start using first one is flipped interaction pattern if you have an outcome in mind but you don't know the information needed by the model to achieve it the flipped interaction pattern is perfect to help this means the model asks you questions instead of you asking questions you don't know to the model here's an example I need to plan an in-person requirements Gathering Workshop but I don't know how to structure the agenda ask me questions until you have enough information to create me the perfect in-person Workshop agenda including the requirements Gathering techniques I should use a bit wordy I know but it actually generates a really good response what you'll see here is the response comes back and it starts asking you questions of information it wants to know in order to provide you with the answer or perfect response you're looking for another example here could be I want to improve my knowledge of World Geography ask me trivia questions about this topic until I've correctly answered 10 this is quite a fun one actually and what you'll be able to see here is you'll get questions asked to you you can then attempt to provide the answer and the AI model will tell you if you're right or wrong and then move on to the next question quite a fun way to learn something cognitive verifier pattern you can ask the model if it has enough information to provide the best output AI models will work better if problems are broken down you'll get better results if you ask the model to ask you for more inputs here's an example provide me with the steps to build a PC from scratch that will be the quickest and cheapest ask me any additional questions that will help you more accurately answer this question here you can see the response comes back and says Please provide answers to the following questions this will help me provide more information but at the same time it also gives a breakdown of the different steps that you'll need to build the PC from scratch Persona pattern this is one of my favorites by asking the model to act as a particular Persona we get a much better response here's an example act as a personal trainer and explain the five key things I should do each day to stay healthy no prizes for guessing what my New Year's resolutions are building on this we can move to the audience Persona pattern by telling the model what audience they dealing with the output can be expressed in different ways to help with understanding or context here's a couple of examples act as a personal trainer and explain the five key things I should do each day to stay healthy assume that I am someone in their 20s who trains four times a week and here we get a great response breaking it down over the five key things and it assumes a response that I'm in my 20s and I'm training four times a week and it takes that context into account now if we look at another example act as a personal trainer and explain five key things I should do each day to stay healthy assume that I'm someone in their 70s who exercises lightly five times a week here we get a slightly different response of course but also in the context of a person in their 70s who's also doing light exercise I can also ask the model to outline key differences between the two different responses which is really helpful the alternative approaches pattern is a structured format used to produce a variety of responses or solutions to a given problem or question it's useful when you want to explore different perspectives possibilities or strategies here's an example when I run workshops remotely I like to try and get all participants active however not all participants do join in and I can't always see their reactions to the workshops if there are better approaches to involve participants in remote workshops list them out and compare them and the responses I get back are really helpful and again I can have that contextual conversation with the model and start drilling down into some of those responses if I want to find out more ask different questions and get some different perspectives question refinement pattern this allows you to say when you ask a question suggest a better version of the question to use instead so an example would be when I ask questions about business analysis is tell me if there is a better question to ask the responses to get back are really interesting and provide different questions to ask which I can then ask the model so let's try that so I'll try what are the key principles and techniques used in business analysis now I get a really helpful response listing out techniques that I may not have got had I not told the model to suggest a better question so this is a really good way of going about it and being able to really refine and get better responses from your questions the template pattern allows you to provide the model with a template of how you want the output or response to be structured here's an example list out the top five selling movies of 2021 using this format movie name director name movie genre here I get a really good output but it's actually not that easy to read so I can refine the output further by asking the model to put the output into a table here provide a table with the following headers movie name director name movie genre for the top five movies of 2021 and now it puts into a table for me and if I want to I can export this into different formats Chain of Thought The Chain of Thought prompt is a way to ask the model to explain its reasoning this is really useful if you want to understand how the model came up with its response you can do this by adding let's think step by step to the end of your prompt so an example would be if I have a single stakeholder to gather requirements from what requirement solicitation technique should I use so the response is use but let's try breaking that down and get the model to give us a more detailed explanation if I try if I have a single stakeholder to gather requirements from what requirements of licitation technique should I use let's think step by step this time I get a really detailed breakdown over the different steps of how I can go about single stakeholder requirement solicitation so that's really [Music] helpful responsible AI let's take a moment to talk about responsible AI I put a link in the description for you to read in more detail Microsoft have set out six responsible AI principles they are fairness reliability and safety privacy and security inclusiveness transparency and accountability you should be aware of these principles and make sure that you keep them in mind when working with [Music] AI here ATT 10 prompt best practices be clear and specific clearly communicate your request or question to the model to guide its response specific prompts tend to generate more focused and relevant responses provide context give the model relevant information about the topic or scenario you want to discuss this helps the model generate responses that align with the given context use complete sentences frame your prompts as complete sentences to ensure the model understands the structure and context of the conversation begin your prompt with setting a persona begin the conversation by setting a Persona that sets the behavior or role for the AI model to follow experiment and iterate the model's responses may vary experiment with different phrasing or variation of prompts to achieve the desired outcome iterate and refine your prompts based on the responses you receive incorporate previous answers referencing the model's previous responses and subsequent prompts can help maintain context and build a coherent conversation limit the response length if you want concise answers specify the desired length or ask the model to provide a specific number of key points this helps in getting more focused and concise responses use explicit instructions if you have specific requirements for the response such as asking the model to list examples provide step-by-step instructions or compare options clearly State those instructions in the prompt verify and fact check while AI models can provide helpful information it's important important to fact check the responses for accuracy the model generates responses based on patterns and the data it was trained on but it doesn't have real time access to the latest information review and refine after receiving responses review them critically and refine your prompts if necessary adjust your prompts based on the model's strengths and weaknesses to get more accurate and relevant [Music] [Applause] answers [Applause] and here's what co-pilot said for prompt best [Music] [Applause] [Music] [Applause] [Music] practices and to finish here's a useful prompt formula to use you can find it in the description too you can modify this with some of the patterns I've discussed enjoy testing these out if you like this video please give it a thumbs up and subscribe to the Dynamics Dojo Channel thanks for watching and see you next time

Channel: Dynamics Dojo

Views: 345

Keywords: AI Prompt, artificial intelligence, best chatgpt prompts, chatgpt examples, chatgpt prompts, chatgpt tutorial, open ai, Microsoft Copilot, Prompt Design, AI Prompt Design, LLM, LLM Prompt Design, Generate best LLM Responses, AI Responses, How to get the best AI responses, ai prompt generator, ai prompts explained, ai prompt tutorial, Microsoft, Bing Chat Enterprise, How to use Microsoft Copilot, Prompt Patterns, template, perfect prompt, llama, lamda, bert, bard, bing chat, GPT

Id: Tsw7YHrPxgo

Length: 17min 15sec (1035 seconds)

Published: Mon Jan 08 2024