Prompt Engineering Crash Course 2023 - ChatGPT & Midjourney

AI tools are coming for our jobs so I  decided to put an end to it and create   the most detailed prompt engineering  guide to show it with the master tonight we're taking a deep dive  into the world of AI artificial   intelligence model developed by open AI chat GPT prompt engineering crash course is equally great  boot for beginners and advanced users and to my   knowledge the most detailed prompt engineering  crash course out there I'm going to show you the   following first explanations about prompts and  prompt engineering what are double and triple   prompts leveling up through open AI playground  biggest mistakes you can do how to master it and   lastly I will have a very big surprise regarding  the surprise at the end it will be worth almost   zero if you do not watch this in full because I  order information in a very very specific way and   now is the right time to subscribe and hit that  notification Bell because this is only the tip of   the iceberg I promise that this is not another one  of those videos that drags on for 20 or 30 minutes   without saying anything I will show you what is  basic and advanced prompt engineering using prompt   engineering with chatGPT midjourney and other  AI tools and step-by-step guides on making money   with these tools that were personally tested by me  additionally I will prepare you for my next three   videos will take you from learning this today  to making money online in different niches right   now as mentioned my goal is to cover everything in  the tiniest detail I'm leveraging this daily in my   seven figure design and development agency which  is called brightdock a two-time Deloitte fast 50   winner for fastest growing Tech firms and we work  with some massive Brands like Rita Ora Sky News   Airbus freelancer.com and many many more prompt  Engineers are currently one of the most sought   after people in it so if you are curious whether  or not you should do this you are definitely on   the right path before anything what are prompts  and how do they tie in with chatgpt MidJourney   stable diffusion and other AI tools prompts are  input data in most cases just plain text used to   guide the response of AI models so that you can  get the desired results let's say that you write   a prompt write me a tagline for my new fashion  product and you will get an output or a result   or let's say that you write to Midjourney design  a modern logo of an orange and voila logo it is   there are different prompts but I will mostly  focus on text to text like ChatGPT which is   an AI chatbot developed on large language models  and text to image like Midjourney which is an AI   program that creates brilliant images out of  simple text inputs there are also some other   models like text to sound or text to video but  more on that later on and let me show you one   insane result on the latest version of Midjourney  just so that you can see how far things went and no this is not a real photo this was composed  by a computer just out of a single prompt crazy   prompt engineering is the process of finding  the best way to ask a question or give a command   to an AI models that can give you the desired  response it requires creativity critical thinking   and understanding of how AI models work and to  repeat some of the famous AI that everyone is   talking about are ChatGPT which is a text to text  Ai and then Midjourney Dall-E stable diffusion and   other which can use textual prompts to create  beautiful images you can think here of logos   patterns illustrations and many many other things  so let me show you one before and after example of   prompt Engineering in action and before and after  tweaking it so the first thing is that we will   ask the tool to give us a tagline for a fashion  brand once we got that we will feed it with some   more details then we are going to ask the tool  to criticize it so that we can make it better   the tool asked us to provide some more Clarity  and details so this is exactly what we will do   now that we have a nicer quote let's  see what some famous designers could do   and let's try a bit of role play I will  ask you to pretend it's George Armani to   see based on the data it has about Giorgio  what Giorgio would do differently and we   will get those desired results our first  result was express your Unique Style with   our fashion which is quite basic and our  last is crafted with Heritage designed   for the future Louis Vuitton's sustainable  luxury for the modern world much better no? do you see how results are fundamentally  different just by doing better prompting   this is what prompt engineering is all about so  what did I do there I first gave the tool a basic   prompt and got basic results then I started  with basic prompt engineering and expanded on   my result with important additional information  later I did more and more information and asked   the tool to help me with prompts which I call  double prompt engineering with double prompt   engineering I actually asked the tool to help  me structure the very prompt itself and thus I   created doubled relationship sometimes you  will also see that these terms are being   interchangeably used with actually just adding  more than one prompt which is also called double   problem engineering if you want to add another  layer to it you can ask it to pretend it's Elon   Musk or any other well-known person and critique  the work of the prompt itself and what would Elon   do better I use Giorgio Armani as the famous  fashion Creator you can use any people as long   as there is enough info about them online so here  is the order that we have prompting writing the   query itself prompt engineering which is finding  the ways to improve the prompt then we have double   prompting which is asking the tool to help you  with structuring the prompt and lastly triple   prompting asking the tool to criticize its own  results and then come up with even better prompts   reverse prompt engineering is a technique used  to create more specific and targeted prompts   for language models like GPT as the name says it  starts the opposite or reverse Way by adding the   desired output of the language model first and  then working backwards to create a prompt that   will generate that output the idea behind reverse  prompt engineering is to provide more context and   direction to the language model allowing it to  generate more accurate and relevant outputs by   creating problems that are tailored to specific  tasks or domains users can improve the performance   and efficiency of the language model and the  prompt itself and why do we do all of this   because great prompts lead to Great results right  so let me talk about some expert updates now   after watching the newest release of chatgpt4  they often ask the tool to act as an expert in   their respective field so I will do the same hey  act as a senior prompt engineer and write me a   very detailed prompt on how I can instruct you  to write brilliant taglines for a copywriting   contest I purposely went into the explanations and  Concepts first but there is one thing that is more   important than anything and that is setting a goal  the only way to get the desired result from an AI   at least for now is by having a very very specific  goal in mind and being really detailed about it   this can be done by questioning commanding  instructing or anything else but you still   need to know what you want right and let me just  quickly touch on some more advanced stuff one of   them is open AI playground which is ideal for  more experienced users who want to understand   the different options when it comes to large  language models how they work and so on and   so on open AI playground is an online platform  made by openai makers of ChatGPT and Dall-e The   Playground allows you to interact and experiment  with different AI models created by openai as well   with playground you can learn about AI develop  new AI models explore AI applications prototype   AI projects and collaborate on many different AI  projects with other genius people so let's go over   some key terms we have models a model refers to  a mathematical representation of a system or a   process that can be used to make predictions or  decisions based on input data the large language   model that is used here is called GPT and that  model has different specific models like DaVinci   Curie Babbage and other the main difference  between these models is the amount of information   speed and cost DaVinci is the model with the most  information around 175 billion parameters but at   the same time it is a bit slower and more costly  than the rest from there the models dropped in the   amount of information cost but increasing speed in  the following order curing Babbage and other and   how do models work for example if a user inputs a  text what is the weather like today ChatGPT would   break this down into individual tokens like what  is the weather like and today another example I   like and then the model would use the probability  of each possible next token to create a response   that is relevant and accurate based on the  context and patterns these models are built   using sophisticated algorithms that are trained  on vast amount of data such as books articles and   web pages they're designed to understand the  structure of language and generate human-like   text next one is tokens in gpt tokens are basic  units of language model used to understand and   generate responses to user inputs a token can be  avoided punctuation mark or any other element of   text that the model recognizes and suggests it  as the next possible token tokens are equally   divided between the prompt itself and the result  and one token is usually a four character word   even though answers vary and the length varies  from model to model many people claim that the   length is approximately 4 000 tokens or around  3000 Words which is actually divided between the   prompt and the result so if you write a prompt  that is too long you'll get results that might   be cut off temperature and top B temperature and  top B are parameters in the open AI playground   that affect the level of Randomness and creativity  in the generated text the main difference between   temperature and top B is that temperature affects  the entire distribution of probabilities while   topi only limits the selection of tokens to the  most probable ones this means that temperature   can produce more unexpected and surprisingly  outputs while topi produces more predictable   and consistent outputs and let me now show you a  simplified way on how we actually use playground   for our client we received a list of categories of  psychology problems and examples for the anxiety   disorders mood disorders and so on and so on  and we then combined NLP capabilities of the   playground and our fine-tuning efforts of the  mode to sort textual inputs by the users of   the app into pre-existing categories and later  based off of those categories the user will get   the appropriate solution and here's a quick simple  showcase on how the categorization could be set up   and just a quick note this is not the actual logic  behind it because we are forbidden from sharing it   it is just a simple rundown now this is covered  there are two big mistakes that people are often   making when putting together prompts they create  prompts that are either being too simple or too   vague so that the tool cannot understand what athe  user is asking like yellow lemon logo or the one   we started with which is a tagline for a fashion  brand or on the other hand prompts that are being   too complicated and thus the tool gets confused  and what do you need to do to become a master of   prompt engineering learn the basics what types  of prompts are there and how to construct them   practice with variations start with simple prompts  and experiment with variations then analyze and   adjust review generated output and adjust the  prompting approach next on try Advanced Techniques   explore with more advanced prompt engineering  techniques like triple prompts or reverse prompts   collaborate and seek feedback this can be working  with other users and seeking feedback to improve   use prompt engineering tools here you can take  advantage of prompt engineering tools like the   openai playground or sandbox or any other next on  keep practicing and refining continue to practice   and refine The Prompt engineering approach over  time and just see what works for you as well   let me guide you with step-by-step explanations  on how I structure my prompts the tricks that I   use and how I'm leveraging what exists to  be a better prompter is that even a term   our goal is to make money with these tools so  we need to see what are the different options   for us and how that can bring us to actually  make money online right and in this video we   cover the importance on prompt engineering for  text to text AI tools and how it can be used to   generate high quality and engaging content we  also export basic and Advanced Techniques for   prompt engineering including single prompts double  prompts triple prompts reverse prompts and many   many more we also discussed some common mistakes  to avoid when working with prompts such as using   vague or misleading prompts that can often produce  irrelevant or low quality output there are some   tricks that you can use and that are quite helpful  you have a tool called airprm that can assist you   with thousands and thousands of prompt generation  templates and those templates can be used for   generative a category like chatgpt mid-journey  stable diffusion Dall-e and other integrates   directly into Chrome and chatgpt so that you  can easily use it within the interface and   within your browser if you do not want to create  that integration there are tools called prompt   hero and prompt vibes which are classic search  tools with numerous different prompts and many   prompts are added on daily basis so even if you're  unsure on how to write them they can greatly help   as well we have Future tools .IO that has many  many cool features in case that you just want to   take things further let's now see how we can  put chagpt into good use with gpt we have so   many different and advanced methods like Chain  of Thought which is cot as well its connected   methods like zero shot and self-consistency  least to most and so on and so on but as this   is a crash course I won't go into that many  details today but I will definitely invite   you to check out learnprompting.org which  is an incredible site packed with free value   the main idea of CLT is that by showing the llm  which is large language model some few short   examples where the reasoning processes explained  the llm will also show the reasoning process when   answering the prompt this explanation often leads  to more accurate results however let's simplify   things and see how you can start creating great  prompts right now you need to give a tool a role   you will act as senior prompt engineer then  instruction or a task and you will create me   a prompt for a specific task question please  ask me as many questions as possible to Define   this context prompts will be about creating  a Harry Potter character in a cyberpunk world   or something crazy examples which is called few  shots create an illustration of Hogwarts schools   of Witchcraft and wizardy with its dollar  interest Gothic architecture and magical   atmosphere generate an image of a quidditch  match with players soaring through the air on   broomsticks chasing the golden snitch and casting  spells to gain the upper hand and lastly you need   to iterate change correct give feedback and in  simple words iterate until you are happy with the   results additionally we can use something which  is called knowledge generation the idea behind   knowledge generation is that you first as the  tool to generate information about a certain topic   and then use it for any purpose like creating  ads or writing blog posts or anything else   let's say that we want to create a blog post about  the history of website development we will ask the   tool to First provide us with unique facts about  website development and that is called knowledge   generation and then we use that information to  create a post ads or anything else there are   many different ways to use these tools and many  many different commands but some of the popular   ones are role prompting you will instruct the tool  to act as an expert and create different outputs   structuring data or text summary different  applications where you can use it to structure   the data or create summaries of the of the text  for a specific purpose like tweet or anything   else emailing in which you can use different  style modifiers for example you can ask it   to create a cold email which is very humorous to  check up on a client Block store content writing   blogs with some limitations because Google can  detect and doesn't like AI content ads you can   really become a famous Advertiser without being  too creative just by giving and creating great   prompts and additionally we have various different  purposes like study buddy assistance recording   contract clarifications not actual writing of  the contract different writing styles and lastly   amazing automatic Integrations with tools like  zapier similarly we have different ways on how we   can Master text to image AI tools like me Journey  dolly or stable diffusion first you go to Google   then Midjourney and it will take you to a landing  page from there you will be taken to this part on   Discord you can browse one of the public rooms  for inspiration called newbies and work there   or you can start your own midjourney bot which  is basically a private room before everything   initiate settings and in settings you can pick  from different midJourney versions different   art styles quality and something which is super  important called remix and let me show you why   with all midjourney images we have something which  is U one two three four and v1234 U is an upscaler   which simply recreates the image in a larger  resolution and V is a variant of that specific   image top left is one then we have two three and  four and now the remix if you have a remix option   on it will also allow you to include additional  details when recreating your image you will   often see some weird characters or numbers behind  prompts in midjourney and maybe you were wondering   what those are they are called parameters on their  websites docs.midjourney.com you can find all of   their parameters they are divided into basic model  version upscaler and then other I will cover up a   couple of basic ones but I definitely recommend  that you check all of them in the documentation   itself we have aspect ratio which can change the  ratio of the produced image quality which is the   level of detail and Chaos which allows the tool  to create a more freely interpretation and lastly   no which basically removes a certain object shape  or color from that image next on we have commands   according to learnprompting.org style modifiers  are simply descriptors which consistently   produce certain Styles they can be combined  together to produce even more specific styles   they can include information about art periods  schools and styles but also art materials and   media techniques and artists there are various  examples that you can use like photorealistic   by Christopher Nolan painting digital painting  concept art octane hyper realistic natural light   and so on and so on next on we have quality  boosters and I think that is quite descriptive   on its own you can see that results are a bit  better and a bit more clear and some examples of   usage can be high resolution 2K 4K 8K clear good  lighting detailed extremely detailed shark Focus   etc etc and let me show you one example of  repetition for some reason if you say that you   want a beautiful flower it will give you a certain  result but if you write it you want a very very   beautiful flower it will give you just something  that is a bit better now let's say that we want   to emphasize something or simply remove elements  of that flower we can do the following we have   the initial initial image of Brad Pitt and you see  the distorted with weighted terms which is simply   giving importance to certain elements in an image  we can fix that larger number means that element   is more important and lower or minus number mean  that number is less important right now that all   of these bad elements are marked with -5 the image  actually looks a lot better for some reason this   work better in older version before and with new  one you can simply explain to the tool and you   will get equal results please note that you can  go a lot deeper with all of this and I invite you   to check the documentation and my favorite tools  you should start with learnprompting.org which   is a great starting point and then use prompt  hero prompt vibes flow GPT future tools.io and   airprm to assist you and lastly and maybe most  importantly if you want to grow lifearchitect.ai   the memo by live architect is a Weekly Newsletter  about emerging breakthroughs in Ai and it does   come with a price of 15 dollars per month but  the value that you get is simply replaceable   first I want to thank you for staying all  until this moment and you definitely deserved   my surprise I went on and personally tested a few  ways to easily make money online with some very   very simple methods all from the comfort of your  home I created a whole video on how you can make   money freelancing for designers with MidJourney  here for writers with ChatGpt here and for data   entry Specialists some cool tips and tricks here  it took me around 20 minutes to make 150 dollars   on design contest without any design skills so  I strongly recommend that you check all three   of them they are short free super detailed and  efficient and lastly I hope that you will be   able to use this information and I would really  appreciate a subscribe and that you hit that   notification Bell as the content will just keep  getting better and better see you in the next one

Channel: Matt Starky

Views: 6,457

Keywords: prompt engineering course, prompt engineering tutorial, how to prompt engineering, how to become prompt engineer, basic prompt engineering, prompt engineering tools, chatgpt prompt engineering, midjourney prompt engineering, how to write chatgpt prompts, how to write prompts, prompt engineering, learn prompting, prompt engineer, chatgpt prompts, midjourney prompts, knowledge generation, text2image, prompt job, text2text, chatgpt, midjourney ai tutorial, how to use midjourney

Id: Fqsh_kqGi2M

Length: 22min 1sec (1321 seconds)

Published: Thu Apr 20 2023