7 Strategies To Learn Anything Using ChatGPT

Key Takeaways

at this point you probably heard of Chad GPT the large language model released by openai that can generate human-like responses to text input it's been Turning Heads all over the world because of its amazing results what I would like to focus in this video is how you can use Chad GPT for learning complex subjects in essence what I want to discuss is seven strategies you can use to leverage chege PT for learning complex topics the first strategy that I would like to discuss is interactive conversations let's say you have a question like what is a neural network right I can get this question and I can paste it in chargept so what's a neural network and then it gives me a response in this case it said a neural network is a type of machine learning algorithm model after the stretch and function of the human brain all right it consists of layers of interconnected neurons that process and transmit information neural networks are used for a variety of tasks such as image and speech recognition natural language processing and decision making they're particularly useful for identifying patterns and features in complex data the thing I think is really interesting about the workflow of learning which HPT is that I can get the response that it gives me and then iterate over that response so I could say something like how does a neural network learn patterns from the data and then from the response that church pity is going to give me I'm going to continue to iterate over by asking questions and taking notes and making sure that that conversation is being fruitful for me that I'm actually engaging with the content that I'm interested so the answer that I got was another learns patterns from data by adjusting strengths or weights of the connections between the neurons the process is known as training the network during training the network is presented with a set of input output pairs called the training set the network uses these pairs to adjust the weights of its Connections in order to produce the correct output for a given input so far it's a pretty you know boilerplate explanation of how neural networks learn uh the process of adjusting rates is them by an algorithm called back propagation back propagation calculates error between Network's output and desired output and propagates this error back to the network adjusting the weights in a way that reduces the error process is repeated all right and then it mentions gradient descent at the end okay so I have two things here that I'm interested in back propagation and gradient descent so now I can say can you explain to me the back propagation algorithm and then it goes you know the back propagation is used to train a neural network by adjusting the weights of its connections the algorithm is called back propagation because it propagates the error from the network output layer through the network adjusting the weights of these connections as it goes and then it gives me a detailed explanation of the process so there's a feed forward pass where the input is passed through the network and the output is produced then there's a step of calculating the error that's correct and then you back propagate the error back and you adjust the weights and then you repeat what I would continue to do here is continue to iterate over my questions and have a convers conversation getting those responses and then you know sometimes I would come here and say okay so there are a few things that are important to know so there's like back propagation rated scent there's this idea of adjusting the weights layers Etc what I think is interesting about this strategy is that one of the main drivers of learning is feedback and I'm not talking feedback in the sense of you know you ask a question you get a response or you get the correct solution for a problem or whatever whatever simplified definition of feedback when I say feedback in this case I mean continuous feedback you know as you learn and go through a subject you want to you want to have the capability of having feedback on what you understand and by having the possibility of engaging with an agent of some sort that can give you concrete feedback in a topic that you're learning this can really boost your learning in general because you don't have to have these delays say if you search for something then you gotta find a good source for that particular thing that you're looking for and then from there you you know you read through don't know how much of text in order to get to the thing that you want when you can just ask and get a conversation out of your when you can turn learning into an engaging conversation where you have feedback I think that boosts learning a whole bunch so now the second point I would like to make so without the second point I would like to make is about content summarization I'm going to be writing the strategy you're so interactive conversations and now content summarization the idea with content summarization is when you're learning a new topic you're going to have to go through textbooks and papers and web articles and videos Etc at least in my experience one of the ways that I waste a lot of time when I'm learning something new is going through a bunch of materials in order to get to that one resource or to that one material that has the content that I was looking for and one of the ways that having a tool like jagi PT although this this idea is not new at all is to be able to summarize content that you want to engage with but you're not sure if that content really has the um ideas and discusses the topics that you're interested in and demonstrate that let's um let's take a paper like attention is all you need which is the paper that started this whole Transformer Revolution where using Transformers models for pretty much everything so I get this paper I get the abstract which is already a summarized version of the paper so I'm not really interested in getting a summary of a summary but let's say this paper and let's say I copy the introduction right and I'm feeling lucky so I copied introduction and the background and I paste that into chargept and then I say write a summary and now what I get in return is just a summary of that piece of text from that paper now of course um it's not ideal to be copying and pasting uh text from a paper into chatgpt because that can be you know Troublesome in itself but the main idea here being that instead of having to go through an entire paper you can get a summary like I just did here Transformers neural network architecture for sequence modeling transduction problems such as language modeling machine translation I can get this summary and then make a decision whether or not I want to continue reading and going through that paper and I can do that with book chapters and I can do that with web articles and I can do that with YouTube videos for example if I go to YouTube what is a neural network so I can come over to show transcript and then I can literally just I can just copy the transcript of that video oh sorry let's go back and I can copy the summary of that video and I can come here on to judge BT paste that here and one of the things with doing this is that sometimes you will get that the text is too long the message you submitted was too long one little hack that I do although I'm sure there's a better way there's a quicker way but this is the this is my hack for doing this I paste it into something like vs code I then delete the timestamps so I select all the timestamps from that video I then delete the timestamps so now the text is all together and it's a bit shorter because now it doesn't have the timestamps and what I will do also is uh after doing that I will select the points and then I'll use vs code shortcut Ctrl D to access all the moments where uh in the transcript where it had a point to do a pause and what I'll do is I'll return one line per point and then give a space yeah I think that's okay I mean it's not perfect but I think it will work and since I still think it's too big let's take a look at the I can use vs code's capabilities to see how many words were selected so kind of like too many so what I'll do is I'll come here at the end and I'll remove the end part of the video where you know usually in YouTube video at the end people are saying you know subscribe like Etc so I'm removing a bit of the text and finally I get that text I pasted into chargept so let's paste this and I'll say write a video summary and what I get in return is a summary of that video let's switch this up there we go what I can return is a summary of that video neural network is the type of machine learning algorithm inspired by the structure of the brain consists of layers interconnected neurons and then it gives discusses the fact that in that particular video the um the author talks about you know recognizing handwritten digits and this is a great summary it's a competent summary of a Content that usually would take 20 minutes to watch that doesn't mean that this replaces watching the video but now I can make a more informed decision of whether or not I want to watch that video hopefully the point drives across that content summarization is quite a useful feature to have and it's quite useful when that feature can be accessed quickly that can be helpful in a variety of scenarios when you're learning a new topic so the third strategy that I think is useful for using Chad GPT for learning is getting detailed explanations of complicated topics you know complicated slash complex topics that seems a bit obvious but um one of the things that I think is just really Paramount is your ability to just something that looks complex in the surface and be able to quickly deconstruct it to Showcase that I want to show you an example so I'm trying to learn how to program better right right now I'm trying to study a little bit more about object oriented programming in the Python programming language and let's say I want to learn what is polymorphism now in the context of object oriented programming using python so I'll say even better I'll say explain to me what is polymorphism in the context of object oriented programming using Python and what I get in response is exactly what I asked for a detailed explanation of a topic that's you know relatively complex in this case polymorphism so he goes polymorphism is a concept in object-oriented programming that refers to the ability of a single function or method to work with multiple types of objects that's true in Python polymorphism allows a function or method to work with different types of objects even if they have a different different internal structures or methods this is achieved through a mechanism called duct typing where the function or method checks that the object passed to it and has the necessary attributes or methods to perform its tasks rather than checking the type of the object for example consider the following code in this case since I got a code sample let's copy that code and test it on VSCO so I'm pasting on my text editor I'm saving as share GPT polymorphism and now I'm going to go to my terminal and I'm going to run this code chat GPT and as you can see incredibly enough the code runs now I'm not really defending that you just blindly believe um the outputs of chargept because one it doesn't reference its sources so that's just just that's the first red flag that you should not trust blindly whatever you're getting out of chpt however um for topics that are very easily checkable like you know programming Concepts pragmatic definitions of you know things in science chemistry mathematics programming Etc GPT is absolutely incredible because it will explain to you what something is and it will break it down it will give examples they can form analogies it can give you concrete examples of whatever you're looking for so the ability to the ability to quickly get detailed explanations about a topic is just incredible okay so for the fourth strategy that I think jagpt is quite useful is writing procedures for stuff so writing procedures now what I mean by that is um a lot of the times learning is not just about you know learning just general concepts that you can apply to a lot of things a lot of times mostly in programming I mean in programming that's definitely true you're learning procedures which is just a sequence of predefined steps to do to do something for example if I'm learning how to write a an algorithm I'm learning how to write a set of steps to solve a problem right and that's another thing that you can use Chad GPT for that's quite incredible you can use GPT to write procedures for pretty much you know anything you can think of now I'm not saying that those will be necessarily correct and you should never blindly believe whatever you're getting out of chapter however for again things that are either easily checkable or um for things that lack ambiguity like ambiguity ambiguity and be whatever ambiguity things that are not too ambiguous you can catch up it is incredible so let's let's do an example let's I'm gonna ask a question I'm gonna ask a classic traditional question in programming um write the steps for the binary search algorithm now this is a boilerplate question for code interviews and I'll admit that's a simple question but it the point here is to drive across the idea of getting procedures out of chatgpt so I ask writing the steps for the binary search algorithm and what I get in return is literally the steps for the binary search algorithm set the lower and upper bounds of the search range star that end and sort of Rave up boom those look correct but I can even go further into the procedures and ask for code samples so write the code in Python for the binary search algorithm and this is what I get now we all know that chbt is incredible for writing code but the but the point I'm trying to drive across here is not just that you can write complex procedures but the fact that you can get step-by-step descriptions of something that's you know relatively pragmatic something that you know it feels unintuitive to have to search for and find it on the internet when you know it's much easier if you can like access a database of some sort tragedy is great for that as well we can now test this code to see if it works so I'm just gonna replace I'm just gonna create another file here and it's going to be binary search I already tried that with charging 50 but let's let's try the second version so I'm gonna paste it here in my editor and I'm going to say binary search and well found it now again and this is something that I've been thinking about a lot because uh although writing code which HPD might feel incredible you should never take it as you should you still have to know how to write code you still have to understand what's being written because your debugging is going to take forever if you just blindly trust the outputs of the model I think I'm trying to drive across is not to you know give free code from chatgpt but you can use this feature of writing procedures to speed up processes and things that you're trying to learn by getting the procedures out of pragmatic things that are involved in the in this thing for example let's get something more fun like uh write the procedure for cooking rice and what I get in return it's quite awesome rinse the rice in a five match trainer and they're to measure out the desired amount of rice and add it to a pot or something I mean this is just insane uh when you're what you're getting is just a description I'm not sure I'm not a great cook so I'm not 100 sure this is correct but to drive the point across you can get procedures out of chpt okay so Point number five now Point number five is creating exercises slash challenges for learning topics no what do I mean say you're learning how to code right and you already learned some of the syntax and you want to work on some small projects you can look it up you know projects to do with python you can do something like that for sure but sometimes you can you can you can have these I don't know how to describe it's kind of like a decision friction thing that happens where you're like it's hard to start it's hard to just begin what should I do what should I work on by having a tool like that you can just quickly get to what you need by just asking for you know exercises with Solutions and in this case I have to like reiterate that um when we're asking for exercises with answers which is something you can do with chpd remember don't blindly believe will you get out of it and just use it for things that you can quickly check you know by Googling or whatever uh and you can quickly check the validity of the information that you're getting because the um the room for misinformation is huge so you have to be careful however for practicing topics like programming for example like write me five simple python exercises with corresponding Solutions and now he goes right up item function that takes a list of numbers as input and Returns the sum of all even numbers in the list and then he writes the code for the thing so what I could do is I could just not look at the code I could pay copy this copy this question I could come to vs code I could I could do something like create a new Jupiter notebook and now I'm going to call this Chad GPT python exercises I can paste the question here write a python function takes a list of numbers as input and Returns the sum of all even numbers in the list right and now I can write the code for this description thank you okay so uh some even numbers and just uh you know let's say it will have numb lists and then four and in um list so the number the number has to be even so if divided by two okay right so zero I'm gonna say total equals zero total plus equal n and then return total and now I can write some tests like some even numbers uh let's give it a list uh I have GitHub co-pilot on this vs code editor so write tests for the below for the above function is actually super cool and then now I can check the answer from cha GPT and look at that boom it's pretty much what I just wrote now I can do that iteratively this is the thing it's engaging and it's just immediate feedback you can write a python fresh that takes a string as input and then I can come here and check the answer but it's my belief that at least for things that are pragmatic enough like programming mathematics chemistry you can trust it enough that it will allow you to at least get started and get past that friction of you know learning something new or learning a complex topic um okay but beyond just creating simple exercises sometimes you want to brainstorm quick challenge projects that you can do when you're learning stuff I can come here and say right five fun challenges to Learn Python Programming and it will say I create a program that generates random quotes from famous people and displays them on the screen codes should be stored in a file database and programs should be able to read and display them at random this is great this is a great example this is a great example of a little fun little challenge project that I can do create a simple game that allows users to guess a randomly generated number with a certain range the user should be able to input their guesses and programs should give them another great example these are all amazing simple challenge ideas that you can do to work on your you know in this case on your programming skills strategy number six it's something that I call Explain domain specific materials now what I mean by that is I could come here to this binary search algorithm that charging wrote and I'm just using this example because it's quick and I have it and easily accessible here and I can just paste it and say explain to me this code step by step let's say even better line by line and now what I get in return is a literally an explanation of everything that's happening in this function step by step now I'm calling this Explain domain specific materials and I'm not calling it explaining code because it's not just for explaining code can get you can my point to drive across the point I want to make here is that you can explain complex topics using this tool which for me is just absolutely mind-boggling when you think about like what's when you think about the reach of this kind of capability if used you know in a loop for learning all sorts of topics it's just a super powerful learning assistant that you can access through text and get explanations and just you know everything you can think of the fact that you can get it explain in detail step by step is just mind-boggling to be honest is just so cool the seven in uh last strategy I'd like to discuss here is what I call uh exploring personal interests and ideas now the the idea here is that say you have a topic you have an interest in but you know you you want to kind of brainstorm it a little bit you dive into that topic how you want which topics which subtopics are interest within that topic that you that you like say something like for example uh I'm really interested in what makes a much what makes a representation inside of a neural network transferable right so I can say something like uh can you give me a few ideas on how to explore the subject of transferability in deep learning and then I can get suggestions for how I can go on I can get suggestions for how I can go through that topic he goes on saying you know one way to explore is to train a model on a test and they have value performance on another that's kind of boilerplate for example training model and image classification and testing performance and object detection we can help to determine whether model has learned General features another way is to compare performance of a model has been pre-trained on a large data set and from another that has been trained from scratch on a smaller one you can explore transferability by comparing performance of a model that's being trained on a specific domain and then okay so another investigate is fine-tuning pretend on a different data set transfer comparative performance all right so these examples are a bit obvious for the topic however they do touch upon One Core topic within transferability in deep learning which is this idea of you know assessed performance of a model train on one test in another task right but I can go I can go deeper I can same what makes what makes a representation learned by a simple multi-layer perceptron transferable and now I'm getting an answer for a more specific question within that first initial topic right I want to understand a bit more about transferability and deep learning machine learning and then he gave me an explanation he gave me a few ideas suggesting that I would you know look into this idea of training a model that's trained on one task and then applied on another task and now I'm going deeper I'm saying okay but what makes a representation uh transferable and I'm getting an explanation now I won't trust this explanation in um in its core I won't trust it with all my heart right because right now at least we don't have access to sources so I wouldn't be able to trust it uh completely but I would use it as a kind of trigger it's uh maybe another question that I have or another path I can pick to go down the road of exploring the subject you know now he's saying representation learn can be transferable if it has learned features they're useful across different desks sure this can occur if it's trained on diverse represent data set pop up okay so but I was looking for something more statistics and you know the the math behind it so maybe I can say mathematically speak what makes a representation transferable because I wanted a more let's say technical explanation of transferability transfer we can use to make accurate predictions uh there you go okay so here it does fail a little bit one way to measures to calculate correlation between features learned by the model and Source test and features important for the Target task and then by comparing future importance activation patterns all right is to Value performance so this this paragraph here gave somehow something a little bit more technical saying that I could calculate correlation between features learned by the model and features they're important for Target test go do this still kind of vague um overall the the point here is to you can use uh you can use jgpt to kind of like go down the rabbit hole of an ID you had or a topic you're trying to learn and I think this is really powerful because it kind of ties up to the initial to the you know number one strategy that I discussed in the beginning which was uh interactive conversations you can use chat GPT interactively and you can get content summarization and detailed explanations and procedures and suggestions for exercising challenges you can get explanations for domain specific materials like you can explain code and you can explore your own interests and ideas so you can you can create this very minimal workflow with this super powerful model and I really think that this can boost learning in good ways because uh one of the fears right now I think with models with large language models like chbt is that they won't help necessarily learning in in itself because now people can get answers for things but when we pass the when we just for a second assume that people will when we overcome the assumption that people will always act on Bad Intentions and we assume that they will act on Good Intentions right that they do want to learn and develop their interests having a powerful tool with which you can get immediate feedback engage ask questions get challenges from it and you don't have to go around the entire internet looking for that one source that has that one thing you don't have to waste time uh searching when you can just get what you need the I really think that this just drives deep learning of things and just drives learning itself you can just really there's no limit to what you can do when you can quick iterate over problems ideas things you're interested in I guess that's that's the point I wanted to make with this video thanks for watching don't forget to like And subscribe and see you next time cheers