Only 1% Use ChatGPT Correctly (This is How)

Video Statistics and Information

Video
Captions Word Cloud
Reddit Comments
Captions
if you want to know how to use chargeability and avoid wrongdance first same question to completely different answers first you need to understand how it works and what are the fundamental principles plus limitations we will get to all of that but first let me ask you a question how do you know if a result you get from chatgpt is even good or correct if you're an expert in a topic you're asking about you will very quickly notice that it is very often wrong but what if you are not an expert most people are at this stage they don't know what they don't know and more scientific terms it is called unconscious incompetence my goal on this channel is to get us at least to level three combine unconscious incompetence with media sensationalism of AI and you get a ticking bomb a dangerous idea growing in Public's mind that AI is smarter than us and therefore whatever it says it is correct we will talk how consumer products like chai GPT works but let me just jump ahead and say you should not trust it blindly and I understand any result you get now feels like magic but one day and I think very very soon we'll have many products like chai GPT design for a specific purpose or a use case or even natively integrated into software that you are already using once the hype dies out it will be just normal famous using a computer became normal or having access to Internet became normal or even access to their knowledge like Google I want to focus on the fact that people who understood how a computer or the internet works and learn how to use it those people created value and jobs not just for themselves but for other people too and later in this video we will cover exactly that how to use it correctly first we need to look what is child ability charger PT is just a user-friendly extremely limited interface under the hood is a very powerful AI model called GPT 3.5 GPT stands for generative pre-trained Transformer actually GPT 3.5 is not just one model but the series of three and that's why it can work with code and text let's talk about how chatgpt works we covered with chatubity is just web interface on a browser so the real question is not how Chad GPT works but how does GPT 3.5 model work can a robot write a symphony can a robot turn a canvas into a beautiful masterpiece can you I will link all the resources and one specific article for those of you who want to go down the rabbit hole but in this video I will attempt to simplify it and I want you to be critical and check resources for your own interpretation if you have a better explanation than my simplification please call me out and share it with the community in the comments below I have no intentions to be the smartest person in the room and I'm here to learn together with all of you okay so here we go child GPT models are trained on massive amounts of text Data like books websites social media and even Reddit they are trying to learn language rules and how words are used together in a sentences this set of training data goes all the way up to 2021 and the most common language in GPT models is English there are other language models that cover way broader spectrum of languages than GPT models in English language there are 40 000 commonly used words and if you look at the large part of English text altogether a few billion words we can get an estimate how commonly each word is used if you ever Googled the meaning of a word you might have noticed or not this option so this graph right here shows how the word has been used in text or literature over the time and if you ever clicked on this it actually shows you way more data on the usage of the word even though we know that chargpt data goes all the way up to 2021 I would speculate that the starting point could be as early as 1800s now I wanted to think of two dices and the probability distribution table we learned in school while there is only one way to roll 2 or 12 there are six ways to roll seven probability of getting a sum of seven is just simply higher same way it is with word combinations word crocodile and alligator will more commonly appear in text by an alligator and Cat we have alphabet and individual letters that turn into words however with GPT models it is tokens and fractions of words that turn into words not individual letters this is open Ai tokenizer and if we take an example of alligator I want to show you what happens so the word alligator breaks down into two tokens of all and I Gator I guess and you can see here it is expressed in text but you can also have token IDs and generally the higher the number the higher probability but for example if we do cat it's just one token and it's one word I want you to pay attention to one interesting thing if I do alligator cat we get these numbers but if I do cat alligator and numbers completely change remember I asked you how do you know transfer is good or correct in that dice example there is one universal truth that probability of of getting a sum of seven is higher than 12 because it's based on math right charity picks words and word fragments based on probability as well but there is a trick here just think about if you and I ask exactly the same question and charge a BTS signs words based on the highest probability on the huge data set to you and I it would provide the same mishmash of the highest probability words so we would just simply get the same answer but that's not really a case and there is a secret to this or at least most people don't know this GPT models pick words based not purely on probabilities but on temperature which is known as the most important AI setting and I have good and bad news for you good news you can control the setting with GPT models this is open AI playground and this is what people use to interact with GPT models before Chad GPT was launched chatgpt is a free and available to everybody chai gbt plus you pay monthly fee and you can use it as much as you want and with this finger paper usage in a playground you get way more control and these are all the settings here we have temperature if we set it to zero it will look at the all the words and pick the highest probability word and the temperatures 1 means it's way more creative there is more Randomness and less probability and I want to show you how temperature impacts your results I see in YouTube analytics that everyone wants to know how to get rich with share GPT how to get rich it needs transfer in one sentence in one sentence the best way to get rich is to save and invest your money wisely I really disagree because I think the way to get rich is to earn money first because if you don't earn money you don't have anything to save does it make sense let's try again and this time we set temperature at one I I get a bit speechless because actually I agree with this one way more work hard be disciplined and invest in yourself and your future same question to completely different schools of fought bad news with charging PT you can't control the settings and I couldn't find information to which settings is charging PT set to now that you know how charged APK works and that it just predicts word but this prediction sometimes is random to make answers more interesting and therefore you and I will get different answers let's talk how you can use child GPT correctly and avoid wrong answers first to interact with AI use prompts there is the whole art and a new profession called prompt engineering just started learning about Prime's visit learnprompting.org child LGBT is extremely limited every time you see this message as a large language model I can't la la la it means you hit the limitations or rules or restrictions which are said by open AI onto chargept be aware that if you ask about restricted topic you are going to get very generic wrong or not answer at all and there's a lot of political and bias issues emerging that comes with these limitations I am a crazy person and I read privacy policies and whenever I hear that people copy their resumes and place it in charge between just gives me very unsettling feeling please never provide your personal information or personal identifiable information to any of these models or chat Bots this video is going to be a beast but I just have to share this with you so I thought I'm going to be smart and instead of Google Googling I would just ask charging PT to List personal information types it did but because I worked with this I was able to spot it that there are some things missing so I regenerated and I got different information so I asked to compare this list this just shows to you that even with information which is specific but it still gives wrong answers in general I have trust issues with internet and I would really really encourage you use critical thinking with any tool you use you probably don't know but it's real humans who are used in a testing process they are reading and rating how accurately AI provides the answers and therefore you also can rate answers if you're worried that you provided open AI with sensitive personal data you can use this form to submit and get your data removed so with this knowledge when you use charge GPD ask yourself am I seeking facts or I'm seeking ideas and operation in general charge if it is not going to make you rich is not going to fix all your life problems but it's really good with the specific tasks like content generation summarization classification categorization and sentiment analysis data extraction and translations so instead of copying and paying for prompt guides from experts online you can actually go to the playground you don't have to pay anything you can check research-packed prompt structures here and you copy them and you go to back to chat GPT if you want to explore even more structures you can navigate to example page you can use search if you click on any of examples you get a prompt and you can even see what is the temperature of this for example grammar there is not much of creativity needed for grammar so naturally the setting is zero how good is your answer depends how good is your question learn fundamental methodologies behind prompts and how to talk with AI which I cover in this video and check out my personal method criticize me mode
Info
Channel: Goda Go
Views: 55,442
Rating: undefined out of 5
Keywords: chatgpt, what is chatgpt, chatgpt explained, how to use chatgpt, chatgpt tutorial, chatgpt examples, chat gpt, artificial intelligence, ai, openai, goda go, how chatgpt works, how does chatgpt work, how to chatgpt, what can chatgpt do, how to train chatgpt, prompt engineering, chat gpt prompt engineering, chatgpt prompt engineering, chatGPT, chat GPT, learn prompt engineering, prompt engineering chatgpt, openAI, GPT3, study prompt engineering, prompt, chatgpt prompt
Id: cWil0mqdXRY
Channel Id: undefined
Length: 11min 43sec (703 seconds)
Published: Sat Feb 25 2023
Related Videos
Note
Please note that this website is currently a work in progress! Lots of interesting data and statistics to come.