Understanding ChatGPT and The Future of AI/ML with Arnav Gupta

What if we told you that something similar to Tony Stark’s Jarvis existed in real life?

​Wouldn’t that completely blow your mind!

​Developed by OpenAI, ChatGPT is an AI chatbot that can answer just about any question you have. It’s as if Google were a person you could have a casual conversation with.

​ChatGPT has taken the tech world by storm, raising many questions about whether AI(artificial intelligence), robots, and ML(machine learning) tech will take over day jobs and replace humans, as well as whether these technologies are harmful in many ways in terms of privacy.

​Join Arnav Gupta as he explains what ChatGPT is and how it works, as well as the overall state of these revolutionary technologies and the future ahead!

​And, also, discuss how you can leverage this and mention the best ways to prepare for upcoming changes. In such times when everyone talks about AI making people jobless, its just the clever ones who actually understand and make the best use of these technologies.

​Suitable For Beginners:
​​The session will be taken by Arnav Gupta, former Android Lead at Zomato, who is not only an experienced coder but also a fabulous teacher. He joins us today at Code with Scaler and will be discussing ChatGPT and the Future of AI.

This will be an interactive session where you can ask all your questions/doubts directly to Arnav (and let ChatGPT rest for some time)

#Chatgpt #openai #artificialintelligence #codewithscaler #arnavgupta #chatgpt #futureofai #machinelearning #scaler

Tags
What is chatgpt?
Chatgpt explained
openai chatgpt bot
chatgpt and elon musk
how to use chatgpt
what actually is chatgpt
chatgpt
chatgpt and jobs
chatbot gpt openai


Uh hey uh hi everybody uh welcome to This live stream Uh see okay a lot of people already hate The live stream I think 30 people okay Uh so so we’ll discuss a lot of things Today uh chat GPT sort of title of the Video but but we’ll discuss more about Other things in in uh AI advances as Well which can as uh you know a software Engineer was a you know a programmer how That can possibly help you uh so that’s The idea just Here Um So how many of you have tried out used Chart GPT and then uh you know Tried to you know generate some text or Anything from that GPT at how many of You have done that Uh and then also what have you used it For if you have already used it like Have you used it to uh just have for fun Have you tried to you know uh tried Coding related stuff uh with it before Uh have you tried it to create any uh I Don’t know I have tried to create some LinkedIn posts out of it as well Um Draft templates for email and official Messages have a poetry code photos Awesome Um so so just to give you a quick uh Example for uh what chat GPT can do for The people who haven’t seen and then we

Will see uh what are sort of the uh Limitations also of it and after that uh In today’s video I would like to show You something Stick which is uh GitHub co-pilot how Many of you use GitHub copilot uh these Days so it’s a brilliant uh tool so you Can basically suggest you a lot of Things like it can suggest you uh code It can suggest you SQL queries uh and Then very smartly based on the context Of your code so I’ll also take a look at How I have been used utilizing uh GitHub Co-pilot uh update and you know where All in my workflow while writing code’s Been helping me the other thing very Recently that has uh come out uh by uh Copilot called GitHub co-pilot Labs Which has a very uh nice set of features Which is you can use it to translate Code from one language to another Language uh it can explain code that uh You know to you then it can also create Things like so actually there are custom Prompts that you can create so for Example uh Uh I do have that open so what you can Also do is uh take some piece of code And actually there are things like Debug the code fix a bug inside a code There are a lot of these things as well And actually you might think that okay That’s that’s really uh advanced stuff Like I mean uh can uh

Basically Copilot uh do things that There are a lot of languages where uh Typescript for example kotlin python Where data types from right hand side Are in from left front side of inferred From the right hand side but you can Have a way to add data types and all so We’ll see how those brushes can work I Have also personally not used all of Them yet I’ve just been checking them Out uh so we’ll see how those prompts Can work uh it’s a very new world I can Say where you can mix and match Different AI prompts and create a lot of New uh you know features and all [Music] Um So there’s some extension it’s suggested To install yeah ignore that so let’s Start with what all can chat GPT do and Then we will see okay [Music] Um So I think uh somebody has said uh chat GPT replaces uh Google uh chat GPT I am Not sure if chat gbt replaces Google uh That’s a very I think uh something that I’ve have been seeing on on Twitter and All people talking about the chat GPT uh Has replaced Google Um but let’s say uh you know if I some Of the from coding perspective I’m not Going into other things like okay I will Stick to technical topics let’s let’s

Forget uh you know social political Topics over and just chat GPT from a Technical coding programming software Development perspective as well does it You know replace uh Google search so so If I ask uh something like you know uh Best Monitoring tool for mobile apps Something like this 2J GPT to just answer the question Whether chat GPT actually replaces Google so first of all it takes a lot of Time Um And then it does come up with uh Some answers I personally don’t like This uh like typing setup because Uh the AI is not really typing it’s just A animation there it makes it very slow And all so if I search this thing and if I search the same thing basically on Google for example Um and then I get a little better things In terms of uh some of the articles that I get uh they have a lot of pros and Cons analysis done some of them are very SEO driven I get it but they have a lot Of pros and cons analysis done why this Particular uh metric monitoring tool is Going to be good or not Um right versus the chat GPT one it Gives me sort of uh Four answers to this uh You know most of them answers apart from

The name this part is actually just uh Keywords uh spread around right so you Know to take the argument that does chat GPD solve for uh you know Google does it Make a better Google search Um I have been saying that you know the Kind of things I do search on Google uh I don’t necessarily get that obviously Like I mean uh who was the 13. We saw something like a certain prime Minister of India might probably find an Answer there Um but then again in the same amount of Time if you go and search on actually Google you will get the same answer There as well So You’ll get an answer here I think It does give you the answer and it tells A little bit about uh you know UPA and One once again all of that as bad so Facts it will give an answer off uh but If you ask about something that’s Happening today uh you know Um I think you know who owns Twitter I’m Not sure if it might actually have the Correct answer because the model was Privately trained before the uh you know Elon Musk took over Twitter for example So that’s a disadvantage because this Model has been trained on a particular Date and the model can’t get updated Very fast on the fly so yeah it’s a

Public company in New York it does not Know that Elon Musk owns uh Twitter Right uh because the model was trained Before that so I think information Searching uh particularly uh I don’t uh You know See in fact if it’s before even Paragraph always made CEO so xcc or Twitter is Jack Dorsey uh that’s what it Says so uh that’s that’s sort of how it Uh works out with the you know um Google Thing but something good it can do is uh If you just start a new chat again Um From a coding perspective can it teach You how to code uh Using CSS uh how can I make a text And scrolling from Left to right okay so if I ask something Like this Now here I think it does actually Improve upon existing things like uh Let’s say uh you know Google also because if you ask that Exact same question I’ll have to shift Through a bunch of stack Overflow Answers so if I have this using CSS can I make text linking and scrolling from Left to right so I might find some CSS Tutorial right now I’m just going to the First answer okay uh which might show How to make it blinking so okay I found Out a blink then how to scroll I might Have to find it somewhere else as well

But mostly because Google picked up the Blinking part first most of the first Few answers are about blinking and then There’s some answer on scrolling how to Put them together uh but what does chat Gbt tell you that you know it will to Make a text Blakely you can use Animation property specify blink Um and uh then it also explains to you To use any this animation you can apply Blinking class to an element to make a Text scroll left to right you can use Animation property scroll left and in That same you know uh to make it Scrolling you can add the class calling Like this And whether it actually works or not if You want to just try it out we can check It uh quickly It answers are correct or not Um So in HTML uh so in CSS I will do two Things that GPT told me I will add the Blinking thing and I will add the Scrolling thing And I will add in HTML Scrolling and blinking Okay let’s see if that works or not Right Uh there’s a blinking animation does not Work uh let’s see why Okay uh Should have worked actually Rather than using the visibility maybe

You can just check it with a different Color uh Anyway uh I was a scoring thing at least Did work Um so the thing is another Advantage I Feel with chat GPT is that it has Context uh while you are asking Something so for example if I Um write something like uh how to make a Get request in node yes So this is something I ask if you give Me an answer on how to do it I can then Uh you know add on to it uh something Good somebody has pointed out that in The last question I had asked that this Text Will scroll from left to right but What GPT generated is the text that Scrolls from right to left so these Mistakes do happen with GPD and I will Talk about why these mistakes happen uh Right and then these mistakes might seem Very silly mistakes in a sense but why Silly mistakes happen with GPT uh we Will discuss about that as well uh Internally how GPT Works based on that We’ll be able to understand also why I Said silly mistakes happen Um so so okay so to make a get request Node yes you can use a built-in HTTP Module uh and it says okay and then Replace username with actual username Something something may be using Github’s API it shows okay and then it Also says that okay here’s an example

During the access you know axios is a Library uh to to do this request okay [Music] Um So both requests to access the popular Libraries uh to send HTTP Um Uh can you tell me in your value using Been very interested if I just ask this The thing is that this question that I Asked can I can you tell me using jQuery Instead I did not specify tell me what Exactly right tell me what nothing what We’ll do is because I have just talked About a get request uh most likely you Will be able to understand this so let’s Try so if I can tell I’m using jQuery User instead uh I I I’m expecting it Should be able to tell me using [Music] I’ll do the same thing So another thing that I like about how Chart GPT answers is it is very human uh Humane in nature like certainly to make A get request it does not just start off With the response it tries to talk to You that’s part of the way the model Behaves the model has been made Conversational in nature so unless You ask it to do something uh specific Like not be conversation it will try to Be conversational in nature Okay Um it does I mean I feel uh it’s uh very

Good at teaching basics of uh certain Things Um the downside obviously is that just In case if there was something wrong in What it taught me right uh if I say it Here and think that GPT actually knows Everything and then I just take what it Is teaching uh as the truth and go on Problem is there is no no checks and Balances around what text it generates Is correct or not there is no unit test Written around what GPT generates nobody Is monitoring what it generates as well It’s just you sitting and generating and If like a bunch of people to different Place in the world generating GPT Answers in different places nobody’s Really monitoring if the cards is coming Out or correct or not and there is a Fair bit of scope of doing mistakes and The kind of mistake that GPD does are Also usually very minor small mistakes But can change the entire meaning for Example the left to right uh issue for Example so it does not make a mistake in Getting the text to scroll so instead of Scrolling uh making it I don’t know uh Go big and small in size which is like Absolutely different logic it won’t make That kind of a mistake but will make a Much more severe mistake that when I Asked you to go from left to right it Does right to left that’s because uh a Lot of uh how these AI models actually

Generate uh text or generate any uh sort Of responses are a lot like pattern Matching where it has been trained on a Lot of data where it has seen CSS code For scrolling and it has seen 50 Examples or 100 examples of it and based On that when you ask it to write its Calling on CSS it tries to generate text Which is similar to that Um it does have there are two models uh Of uh GPT by the way there is a codex Model and there is a DaVinci model The DaVinci model generates text and the Codex model generates code on chat GPT Both the models are available so it can Actually show you code and then explain It to you uh and then GitHub copilot Uses only the Codex model we will go and Talk about it also Um one of the things that GPT is Definitely good at is that if there is a Particular type of problem that has been Solved a lot of times already right Which GPT actually solves better because There has been more data for it to train On so for example if there are lead code Questions that a lot of people have Solved uh before right Um and that if it tries to solve it it Will actually be able to solve it a Little much more easier just to give you An example I’ll go to uh you know GitHub Co-pilot and show that as well Um one more thing that I think is also

Good is that it you can actually give it Code and it will explain it to you as Well so if I just for example take this Coded GPT itself uh has in it or maybe You know uh let’s say uh An alternative to access maybe okay Fetch apis there okay so fetch API Examples JS so maybe so Um okay this is a fetch example so I Will take this I will tell uh Or this code does and I ask Okay I don’t know what happened Uh okay I think maybe it’s not because JavaScript is trying to get executed uh So I think we’ll start a new thread and see So let’s see Sometimes some server issues like this Happen Um but but we can I think Those are more not AI bugs but basic Software engineering bugs I think they Had just released it a few days back and It has uh you know immediately gotten uh I think more than one million users Within the first five days Um One of the things you can do is like you Took it and it asked me asked I asked it To explain and it gave me a very Detailed explanations uh you can re-ask GPT to explain it more concisely rather Than uh you know uh to to sort of expand On status codes and all of that

So now it’s it’s much more concise Explanation you can you can do those Kind of things you can optimize that as Well which means it’s good to generate Actually documentation uh for source Code as well that’s also something that Uh GPT can definitely do Um one of the things I have been uh just Trying out GPT for is also that you know Um So if I write something like Right uh 2000 Character LinkedIn post Explaining Distributed okay So a more theoretical topic for example Distributed caching is a high level Design topic if I ask to generate a LinkedIn post explaining distributed Caching for example Okay So it does generate the sort of LinkedIn Post uh you know uh that you might uh Find on LinkedIn sometimes people just you know Trying to Garner engagement by writing Posts like this and I have been trying That so there have been people who have Engaged with those posts as well so Um if if I just show you uh some of my LinkedIn posts that I recently wrote Those are generated by uh GPT so for Example I think

[Music] So this is something I did using GPT Which is generating uh Project ideas uh for people similarly uh This was a post where I generated like a Post on how uuid works okay The thing is uh that if I do know what Uuid is and I ask GPU to generate I can Check whether the explanation is correct Or not before posting it Um some people had guessed that whether It was generated with GPT and I said That yeah this is generated with GPT but These are good like I mean if you just Want to write random posts on LinkedIn And grow a following and all of that Stuff uh this is actually uh makes it Easy to do that it increases the amount Of spam on LinkedIn that uh day by day Is anyway already increasing uh but that Aside if somebody really wants to Generate limit post it can do that I Think I did one for load balancing as Well Um again I think there were people who Actually engaged with that uh like I did Answers strategy PD yes but there were People who actually answered uh the the Debated stuff and they were actually Basically debating versus chat GPT not Versus B because this text is not even Generated by me Um the thing is uh that if I now go back To the answer that GPT has generated

Here Um a 2000 characters in post explaining Distributed caching Um This uh Answered Um is not actually about distributed Caching but just about basic caching so It says the dispocational technique uh Used computer to improve performance and Scalability application by storing Frequently accessed data in a cache this Allows the application to retrieve the Data from the cache and slow out more Resource intelligence so this answer Even if it was not distribute cache if I Just asked explain HTTP caching this Answer would have still been the same Right the distributed part uh you know It does not particularly Expand on that much so I might not use This answer a lot but then I know what Distributed caching is but if I just go And ask you know [Music] Foreign Between distributed caching and normal Caching I can get some more data points On that right again that also to be able To validate whether GPT gave the correct Answer or not I should be someone who Knows the answer already Um but if you are somebody who does not Know this answer and if you’re thinking

That I can ask GPT and learn from it uh You might end up with wrong answers you Might end up with partially correct Answers you might end up with answers Which uh don’t go beyond the surface Like there are a lot of people Influencers on Instagram talking about Tech topics and they really don’t go Deep but they’re just talking on the Surface about okay these are some Keywords uh GPT can generate a lot of Text like that which is at the surface Level but not going deep So here I think the answer gets a little Bit more uh significant uh because it’s Better because it’s now actually talking About uh you know what’s the difference Between normal caching and distribute Caching so actually double down on that So so the thing is that this answer the First one that I got if I’m somebody who Uh already knows caching then this Answer is not a good answer for me Because I know caching and disaster does Not tell me a lot about what distributed Caching is additionally on top of normal Caching but if I don’t know caching uh At all then this answer is even worse Because I feel that this answer is about Distributed caching uh While I mean I don’t really know the Difference between normal caching and Distribute caching and I think take this Answer I think that this is what

Distributed caching is it does not Really go deep on the other hand this Answer actually goes a little more Deeper into uh why distribute caching is Different from just caching on a single Server okay So anybody else has any suggestions that That golden credit sessions if you think That we can ask some GPT about Um one of the things that I do uh find Interesting is uh I teach Um you know people one of the things That I do uh find interesting is Sorry there’s a bit of Eco there so Um when I teach people to build projects In my classes and all after teaching People to build one or two projects I Asked them that okay you go ahead and Build a few more projects on your own uh Make uh you know uh 10 15 projects and Two three of them would be good so over The next one year spend working on Projects and people listen to that and Then they say that how can I come up With 10 15 project ideas I can’t even Come up with two project ideas I’m not I Can’t think of what project I can build Right Um that is something where GPT can Actually help you so if I say that Uh you know Uh Suggest some Second API project ideas for someone who

Has recently Learned to use uh Java Input The project should take One or two weeks to complete and should It was Concepts like Databases uh rest apis Authentication Um so if this is the prompt it’s a Fairly complex prompt though so imagine Who can answer these questions somebody Who knows bring boot well has built Projects themselves also uh such a Person can answer this question right This is the kind of question that I Think a lot of you people sometimes when I’m doing live streams you do ask me Right you ask that okay here what kind Of projects I can make I have learned Back in development so and so right this Is not somebody who does not know Development somebody who does not know Um API database authentication has not Built their own projects they can’t Answer this question which means this is An expertise question this is not a Question that a Layman can answer if Somebody is experienced an expert they Can answer this question and this is Something where I think GPD shines a lot Where you can ask this question and you Get good answers out of it and you can Actually pick any one of these

Um you know project ideas and actually Start working on that so it gives me Task management app recipe database Budget tracking API real estate listing API weather API and it tells you know Things in details okay [Music] A weather API you mentioned uh Point Number five uh Give me detailed [Music] Project Requirements and specs for building it Uh What Features What features it should have uh Which API and points to build What Is Should we create in the I’m getting a little deeper now at this Point I’m saying that okay I want to Pick up project number five can you give Me detailed requirements for it which API it wants to build uh okay and what Entities uh in the schema should I build Uh so I like the answer maybe number Five I want to pick and actually build It Um so it’s giving me some detail uh okay So now it’s saying features the API Should allow for so and so things so This now starts looking like uh the

Questions that people actually get and Take home interviews uh so if you apply For startups and all right uh those kind Of companies generally give you a Take-home uh project like this so I see That there are two advantages like if I’m actually hiring people I can use GPT To generate assignments and then if I am You know uh you know learning let’s say A particular Tech stack I can use GPT to Suggest me some ideas and then actually Give me very specific requirements so I Can actually build the Project based on This It still means I mean you have to go and Build the application yourself uh right But But the idea and the requirements Somebody has given okay so I mean it Also means that you know you have turned Chat GPT into your product manager uh And they are creating these requirements That you have to build on but it’s good I think I mean uh this endpoint should Accept query parameters for location and Uh you know units of management if I Take just this part of the text from Here And you show me some sample code for This Um this I am not expecting GPU to Actually be able to give a good answer To it but let’s see how far uh GPT can Be pushed

Foreign The questions become more specific you Will also notice that the model takes More time to respond Um okay and these kind of things can Happen the request timed out uh okay so It’s most likely it will time out Because it’s a very specific question And the model might not be able to Generate something for this uh very Easily You can maybe make this a little more Specific uh Okay it’s giving some answers So it’s making a rest controller in Java Which is fairly good actually I would Say it is calling a Weather Service and It would actually call that so this is Actually creating your boilerplate code For this controller as well Right so your boilerplate controller Code is actually made out of this Um it’s a good place to start from right And giving these explanations which I Think are not required I did not ask it But it’s giving these explanations I Mean that’s fine Um But it’s a good companion at least a Learning level that I can see that okay Nice learning companion not bad I would Say okay Um So by the way uh you can actually use

This as an API in your own project as Well so there is the open AI API where Text completion image generation uh you Know and code completion kind of uh apis Are there so these are actually apis so You can actually link to this API uh From from somewhere and you can actually Use this there is API docs also Available for the documentation uh how To use you know Basically uh how to use the API this is Uh open the open API project download There is a they have I think sdks in Node.js and flask so you can download That and use that there is the Playground so this is where you can try These things out in more detail this is The same back end actually uh what this Uses and what chat GPT uses very similar Packets uh here you can Give more contact complex prompts as Well and you can select which model to Be used The Codex models are text based the GPT Models are sorry the codus models are Code base the GPT models are text based The bigger dots means those models are More complex as in bigger models which Means they have lot more data set Trained but those models are costlier to Be used as well you can select certain Things like what’s the length of the Prompt you can select like some Frequencies and all all of these things

You can hover over them and tells them What exactly it means in terms of what Is getting Uh generated there okay If you have text S have evolved very recently like Tech Stacks or programming languages which Had changed uh since 2020 a lot you Won’t get very good answers because this Models uh has a cut off in 2021 for Example which means that uh things which Have recently happened uh let’s say React latest version released a few Months back if you want some example out Of that you might not get that okay uh Now coming to actually GitHub co-pilot Which is uh something that actually does Change the game when you’re working with Code and this is something that I have Been using a lot Um so on on GitHub co-pilot uh you know Uh and I’ll show you some of the few Things that it can do And then there is GitHub copilot prompts What it can do so there are two things I Think uh you know uh and I’ll show you In two examples one is like if I do lead Code right Um so if I just Open uh Okay Another belief called uh answered and I’m not very good at lead code I will Pick up something very easy uh I’m not a

Good competitive programmer by the way Just just uh you know I don’t want to Embarrass myself on the live stream so I’ll pick up some easy question Um so let’s pick up some some random Easy question uh what is the plus one Question Um Let’s say Roman 20j right so this is a Question we have to uh I think input uh Is this take an input and convert right Now this is something that a lot of People would have done already right Um Everyone to uh integer conversion so I’m Using GitHub copilot here I have get a Copilot installed as a plugin here and I’m using GitHub copilot so if I uh Start doing this uh project In what language I can pick up let’s say Typescript Foreign So it says okay if This is uh split and accumulator and Current so it splits it into this create That is something like this okay uh if I Just I mean do this uh okay But it will be zero Um Foreign This is not actually correct though I Don’t think it handles the case where There has been a number before it like IX IB those kind of cases are not

Handled by this uh but it gets me to Like a fairly close by uh what I feel is That with questions like Roman twin Teaser which is fairly common a lot of People have solved it uh what will Happen uh you know is uh That If you actually know the answer if you Have thought of the answer you can Generate a lot of the boilerplate by Copilot it might not give you the Exactly correct answer here okay Um [Music] So if I I mean just submit this uh Answer here okay so what will happen in Lead code is uh Just a second So if I test it on uh Default test cases There’s some problem in my lead code Extension apparently Um So that’s something that happens with I Mean I’m just trying to show that what Happens with questions which are very Commonly solved uh right Um particularly this now if I uh try to Take a look at uh something that I was Doing recently which is advent of code Which is uh every year this thing Happened which is advent of code is a Foreign Questions that happen every December

There is this guy Eric wastel who does This right Um so For this kind of question what happens Is every question is fairly unique here Now I think the first question there was Um Something around there are some Santa’s Reindeers are delivering something and You have these numbers what I have seen With these questions is that a GitHub Copilot does not actually work very well Because the format of the question is Fairly uh you know uh the question it Has not been trained on this particular Model because this question actually was Released in uh you know like December of 2022 and the model has never been Trained on any Solution on this Particular question but within that Question uh if I take a look at uh some Part of the code for example I was just Writing this code and uh the part that Got generated to me very easily was I Think Yeah so if I’m writing something uh like Split a line into two parts at midpoint Uh it generated these two lines for me Automatically if I say that you know Group lines into uh groups of three it Generated this for Loop for me uh very Easily uh I can I can do this uh You know just one more time as well so If I do four and then

I wait for the answers These uh you know uh things that are Generated they are fairly fast now let’s Say the next thing I need to do here is That find elements that are present in All three parts you can actually write a Comment Uh find elements that are Present in all three lines So it actually does this so what I have To do is I have to split uh The first string into like characters And then for every character I need to Find out characters that are included in The second and the third or not okay And it gives me priority and then stuff Like that and varieties and plus variety Okay so this is generates well because I’m not talking about like particularly If I look at my question number three What it was about was some people Packing bags and all of these things I Did not talk to get a co-pilot about These things I read the question I Understood it and I understood that what I have to do is I have to split the String into two halves and I have to Find characters which are in those two Parts and for the second part I have to Do is I had to take three strings at a Time and I have to find out characters Which are common in all three so now if I understand what I have to do like Translate the you know complex story in

The question down to actual logic I do Myself then I can you know rely on Copilot to actually generate the code For me which I feel is is good like uh You know for example if I was actually Taking interview of somebody and if they Use GPT for actually generating these Lines of code while they have already Figured out how to solve the answer I Think that’s good because there’s this Famous saying which says right uh a Mathematician knows how to solve the Problem not uh he does not sit down and Actually solve it so this is fine like If you have figured out the logic in Your head and then you basically prompt Uh GitHub copilot into generating the Lines for you This I think is a good superpower to Have as a software engineer it makes you More productive it makes you very fast Like since I’ve been using GitHub pilot What I’ve been seeing is that I have Been getting faster and faster in in uh Code that I write so this is something Good I believe Um And it was generated this way right So that’s there the next thing which I Was more interested in was uh for Example if this piece of the code and That’s where I was talking about the Brushes so here if I talk about uh GitHub brushes language translation and

Explain so if I take this piece of line Code Here and I asked it to explain [Music] It actually gives you a good explanation I would say uh we iterate over lines in Steps of three so that uses the same Logic the first line is same as before The second line is same F4 this is is Wrong these three these Point number two And three are not so useful this and This these are not so useful uh I’ll so I’ll just generate that again sorry Uh Um we are using Speedmaster convert Second RF characters using the filter Node to find the elements that are in All three lines uh we are using index of Method to find the priority it’s a good Explanation I would say uh so if Somebody looks at this code for the First time and does not understand what It is they can ask that and code does The following as copies is another way To solve it as well so this is more line By line Um it gives an explanation Um even more beautifully than that is What I feel is the brush part which is Uh The Brushes where I mean if I take this set Of lines right and I say Okay add types It adds that this is of type string this

Is our type string it added these data Types which was not there like just Before that these data types were not Present if I’m using a language like Typescript or python where I can add Types if I want to optionally it can Generate types for me those are I think Very supremely useful features and I Think these will become part of the Standard uh toolkit and if GitHub Co Pilot is a paid service I think most uh Professional software Engineers who are Writing code for a living they would Start to pay for GitHub for pilot and They’ll be using this because it’s a Very important feature I feel Um I think you can do a bunch of other Things as well like you know uh if I ask It to Um sort of uh document this code Um It will turn it into a function and then Document it basically So all of that is something that it can Do Um if I want to make it more readable so It tried something uh what it did is I Think I don’t know Hi Made it first line Second Line third Line stuff like that so if the variables Were something like x y z here right Um Foreign

Programming kind of code where people Just write XYZ variables let’s see if it Can make that more readable or not if I Click on readable Uh uh it did not it will need more Context actually so if I give the entire Code LLC I think Foreign [Music] So it does work sometimes it does not Really work that well sometimes but I Think it’s improving uh slowly slowly Um And then finally another thing that I Was doing recently is that these Questions from Advent of code that I was Solving I was trying to solve them in Multiple languages to learn uh rust and Go myself so I am very good at Typescript and kotlin but I don’t know Uh rust and go so what is doing is Solving the question first in typescript Or in kotlin which is on my first Language in that sense and then I am uh You know trying to write it in go and Press to learn that language so here There is another thing which I found out Is that if I sort of Do is create a new Um Kotlin file for example and if I ask It to take this typescript code and Convert it into a different language so If I take all of this code ask it to

Convert it into kotlin Okay How close would that be so I’ll just Copy this code Here I had to make a few changes So to run the kotlin code I need to include the comments And usually like I don’t need these Arguments actually here and I’ll read Lines Yeah I do this the rest of the code uh you Know Um is it actually usable code or not um So if I take the typescript code I just Run it Um this is the answer seven seven six Six two four one five I have not even Looking at the rest of the code I’m just Trying to see if it can run it or not Uh I think it did not print because it Has not called the main function I’ll Just Call that Foreign [Music] If you don’t know that language then you Will actually have to go and debug that Um so the problem that would have [Music] Uh have had is I think here Index out of bound would have happened Somewhere but if I just compare it for a Second

Um It’s it’s actually a fairly good uh Capability I feel because there aren’t a Lot of online services for example where You can paste typescript code get kotlin Code in return Um so you know it does some lines dot For each and for every line it’s doing Line dot slice is doing the second line Which is taking from uh here till there Um right I think this until uh setup is Not exactly correct uh line dot slice Does not take this as an argument it Will have to take a comma there so there Are silly mistakes again like I said This until here is correct for in zero Until lines dot side this is correct so Until is used in kotlin to do for Loops It’s not used in slice so that has to be Fixed Um right Um I think the rest of the code is is Looking pretty good I would say I mean Uh it does uh things like take the first Second third is correctly it does the Intersection is uh takes all the items And then sees if it contains uh in the Second and third Um so in that sense I think if it’s Uh not bad to actually translate bunch Of code from one language another Language which is something that people Might need if uh you are converting a

Project let’s say which was originally a Node.js project to a kotlin or a Java Project in Future these kind of things Can be fairly useful uh in my opinion So Um I think that’s one of the state of How things are getting used today I’ve Been using uh GitHub copilot as well as Chat GPT a lot [Music] Um The the question does uh you know uh beg To be asked that how much of tools like Gpt3 chat GPT githubilant replace uh the Jobs of human beings Um I think it’s a tricky question uh I Would say somebody can look at what These things are doing and think that oh Uh whatever I am doing these tools are Able to do that exactly same thing If you are feeling like that I think uh You need to learn and grow a lot because Uh these tools are not able to do very Complex things and they do make a lot of Mistakes as at least as of today so if What they do today if you feel like this Is something that threatens your job Already then you do need to do some Upskilling because probably you are Doing very machine like work you are Probably not doing stuff that actually Requires original thought Um because if I tell somebody to build a Software make an app it’s not like

Copilot can generate the entire app for Them but obviously there will be places Where they have let’s say thought out And I’m thinking it out loud like let’s Say I’m making an Android app and inside That I have to create a You know video player inside it if I Write a comment uh you know uh generate A video player and uh fetch the video From uh URL and play it and if GPT can Generate I mean a GitHub Pro pilot in This sense can generate at least a basic Skeleton of that code and then I will Make the fixes like maybe change the url Correctly maybe buffer it or stuff like That that’s actually a superpower uh but That does not mean that if a manager who Does not know how to code comes and Tells GitHub copilot that create a video Player app for me it won’t be able to Automatically make it but then me who as A software developer is making the video Player app myself for me it’s useful Because it can come in and do those Things which are boilerplate work or you Know more tedious work and all of that So what do you think uh any any thoughts Any questions people have I uh we’ll Just uh you know stop speaking here just Stop the demo of the things that I was Showing and I’d love to pick up some Questions any questions that you have About these AI tools uh what things that You might want to use these for in the

Future uh who would open the floor to The questions for everybody If you have questions which are not Particularly GPT related but some other Questions as well happy to uh answer Them as well Uh Sameer uh you asked like how did you Get a collection of lead code question Set so actually there’s a lead code Extension Um on on vs code uh I’m using this extension there’s this Early code problem extension it’s a very Popular one uh with that you can do Um lead code actually from your browser Itself you just open see the list of Questions and start solving them right From here Well this would I think [Music] Foreign Was actually correct I believe I think priorities sums uh Foreign ’s not disclosing the source of their Data can lead to policy violations and Copyright issues so I think that’s one Of the questions which are more about uh Law catching up to technology so the Technology was there and it got built And I think the data that it trained on Uh the data was publicly available so it Got trained now after training that People are making these AI models and if

As long as they’re free that’s fine but As long as you start charging money for This Right which is where People have to see that how do we Interpret the law of Copyrights and all right now the current Law has fairly gray areas because it’s Not like I’m copying the art directly Right Um but but nevertheless uh let me show You something Um in Um Okay open air itself like if I go to the Image generation one right Uh sorry it’ll have to be So if I open Dali and here if I do Something like Uh upload an image to start with right So let’s say I upload my image It can straight up generate a bunch of Variations for my image I can do things Like uh let’s say uh It generates a bunch of variations of This image right uh some of them Obviously don’t look like real human Faces but for example this one is can Definitely be a real human face for Example Um the other thing is uh for example if I take and try to edit it right And this is sort of my Uh I think oh sorry

So if I give it a prompt uh That if I say this is my generation Frame And let me say I give a prompt Oh sorry I think I have to Um We draw this image in Mango’s Uh self Portrait Style So Now the thing is that if I take one of These images and if I actually see Uh back till Vanguard self portrait Right Um so is this copyright this is not Copyright becomes a little bit gray area Because obviously it has learned from That and it’s generating things which Are very similar to that uh right so If I take this image again I say that okay Let me draw this in monolitha style Uh do we have things like charged I Think uh uh so but to the answer the Question both are uh free actually uh So you can do a bunch of funny stuff of Course uh the thing is like when it Generates things like this and then if I Frame it and I probably sell this image And if I make money then whether it’s Copyright or not I think laws of Different countries will catch up to This they will

Create a mechanism which has already Existed always like on YouTube if you Use background audio from some singer Then you can’t make money from that Video if I do something like this maybe It will say that this if I make money Out of this some loyalty loyalty might Need to go to monalisa trust in Paris in Luday or something like that right those Things can happen but it’s very cloudy Uh so far about the free version uh I Think crazy Studios you’ve asked so Right now all the tools I showed chat Gbd gpt3 daily GitHub copilot they’re All free to be used actually They’re not going to be free permanently They are they are very at Alpha Beta Stages so that’s why they’re free but They won’t be free forever Um To answer your question about being an Android developer with eight plus years Of experience and whether DSA is Important to join or not it depends Company to company for example you want To apply to Google DStv for sure if you Apply to Facebook or I mean Twitter has Changed a lot but I knew for sure that You know before this year on Mustang When Twitter used to hire Android Engineers they would not ask DSA Facebook also does hire mobile Developers without DSA questions uh There are many companies who do

Um So my target where I used to work at uh They they asked system design mv’s MVB Mmvei kind of things clean code Architecture ask you to build an Android App take home project but they won’t ask You DSA questions for a senior Android Engineer roles so it depends when Comparative company that’s the situation Uh so it depends [Music] Uh cool I think uh if there are no other Questions uh from anybody uh uh we will Wrap up the live stream then Um thanks everybody for tuning in to This Um and and you know feel free to uh you Know Connect with me on on Twitter Etc and share your thoughts on what all You are using uh AI based tools for what Do you think that direction is going Towards uh if you find out some very new Use case I think people have done AI Based uh uh I think image generation There’s a lens app those kind of things I think there’s also a lot of scope for People today to build uh products which Use these AI tools and become the first Player in the market and gain a lot of Uh you know attraction so there’s been I Think uh this uh Avatar AI uh Kind of thing where uh I think You upload a photo and it generates a

Lot of AI uh images and all and you have To uh it will take 30 minutes or so to Generate all of these and you can you Have to pay money to do that so they are Making a lot of money right now tomorrow Some large company like Google will Start doing this and you won’t have an Advantage but uh today probably you can Do that so that’s something that you can Probably look at and build Um Foreign Thanks everybody uh have a good day uh And uh bye bye we’ll see you soon in

You May Also Like

Leave a Reply

Your email address will not be published. Required fields are marked *