# The Problem With Software Engineering Jobs in 2024 | This Week In Tech

## Метаданные

- **Канал:** Varun Mayya
- **YouTube:** https://www.youtube.com/watch?v=wN7J6eQkL9g
- **Дата:** 07.01.2024
- **Длительность:** 27:45
- **Просмотры:** 76,074
- **Источник:** https://ekstraktznaniy.ru/video/12390

## Описание

00:00 Intro
00:21 The problem with short-form content
02:45 Deepfakes posing issues for KYC
03:36 Oversaturation of developers in India
07:46 Problem With Indian Software Companies
08:58 The problem with agencies
09:54 Building apps is easier now than 10 years ago
11:00 Building app for performance vs. scale
11:43 Betting on future trends
12:27 Progress in open-source AI models
13:55 Tool to convert video to any style
15:47 Meeting Mythpat
16:37 Self-driving cars banned in India
16:50 Unfairness of protecting some jobs over others
18:06 Can AI handle Indian roads?
18:52 Varun's policy around AI taking over jobs
19:40 Using AI as a competitive advantage
20:07 Separate policy for internal vs. external jobs
20:58 Review of Humane AI pin
21:30 Real-time conversational AI now possible
22:50 Interactive paintings/toys now a reality?
23:40 Prototyping real-time conversational hardware
24:19 Why taking a hardware bet is risky?
24:44 AI as a Romantic Companion
25:44 Loneliness problem in India


## Транскрипт

### Intro []

so mpart is here and uh it's not an AI version sorry but he he's just here to say hi and we proceed in the video now bye you know the problem with doing a

### The problem with short-form content [0:21]

lot of shots and I've been right it's just very convenient it takes 5 six minutes a week the problem is you start irritating people after for a while because it's impossible to give too much context in a minute and I know this is what social media has become we all consume shots we're all absorbing so much of you know Instagram reals and YouTube shots that we've forgotten long form even exists but I think that going forward it's important for me to keep doing some long form even if it's once a week once in two weeks because I'm learning a lot more about the ecosystem now and the ecosystem in Tech has changed in India and all across the world and nobody's really speaking the truth because if you go on Twitter and this has started to happen to me on Twitter and many other Founders on Twitter where you sort of become Frozen you put two lines and in two lines you're going to say something and then it's going to piss off millions of people right and they're going to be people who don't even know you who come and start eating your head and the problem with Twitter is if you say something that's truthful but feels unfair you are going to get hated for example there was DD who went out and said that it's easier to have made servants in India than the us because M servants in India are paid much less compared to how much let's say a Family household makes and he really got lambasted for it even though it's the truth right this you could call it the negative of India or the positive India but he wasn't talking about that he was just like this is the truth in America it's very expensive to hire a maid servant or a driver and therefore it doesn't happen as often he just mentioned the truth and he got I mean I the comments were not very kind and I think uh it's also because on Twitter you can only put in so many words people are just going to see the first tweet they're not even rest of the tweets so you are going to get hate on that first tweet but with YouTube we have the opportunity to go long form I tell you over 20 minutes 30 minutes how long this takes what I really think and what the nuances are and I think the world needs more of nuance the word nuance means let me show you more about this what is the unique things about this that you can't just strip down into a sentence right so let's go deeper into everything and this show whatever I'm doing I just want to call it this week in tech I just want to go over give you my thoughts on it and I unfiltered thoughts on it right some of them will not be super kind but I will tell you anyway and I'll just tell you what we've been up to I can also discuss stuff that I've been up to which I have not done in a very long time because the business building side of thing has been just taking all my time but let's start with the first thing I

### Deepfakes posing issues for KYC [2:45]

think you might have all seen this deep fake of Nan Kut it was on money control and lots of different news and AI is sort of coming for kyc is know your customer whenever you sign up on a zeroda or platforms like that you need to sort of take a photo or a video of yourself saying it is me uh Varun actually here but now that you can create deep fakes of anybody right or AI avatars of anybody you sort of get into the situation where you can fool these things and you'll probably see this in the podcast I've done with nitin and we helped them with it we used hen for the video and I think we are now entering an era where it's going to be very hard to differentiate you can't easily tell between what's a real video and what's a fake video and it's definitely going to affect the elections and I think everyone watching this I really just hope you develop an immune system I'm going to keep reinforcing this but develop an immune system and the second

### Oversaturation of developers in India [3:36]

thing I want to cover which I think it's just sad that this is happening but sometimes I go to Reddit and I will see these Indian developer groups and on the Indian developer groups I'm always seeing people who are like you know 19 20 year olds really hopeful remind me of myself when I was young I have so much empathy for these people and they will be posting things like I'm doing the right portfolio projects I've learned the right stack you know my interview skills are good uh I'm going to a decent College why am I not getting a job and this is a very common thing right and people think eventually that is their fault like these young kids think it's their fault it's sad that people are making it about that when I think the most obvious thing and this is you know from my learning because I empathize with these people because I was a young engineer learning skills in college but there is a very specific Insight I want to tell you which I hope might be useful to you and it's again my opinion when I was young and when I was in college what I was being taught was as asp. net and Java that's right I was studying computer science in manipal Institute of Technology and we were taught asp. net and Java which by the way at the time that I was learning them already old but my teacher said this is the market this is what the market wants and the truth is that it was what the market wanted but it was sort of on the tail end that was already saturated market and I taught myself Ruby on Rails because it just seemed cool at that time and Ruby on Rails just seemed simple and I had an internet connection so I said why not teach myself Ruby on Rails so I taught myself Ruby on Rails and the kind of opportunities I got in Ruby on Rails online freelancing was great and it was very different from the kind of opportunities you would get if you were doing esp. net and I feel we are exactly there now we have a bunch of Engineers who are super specialized in a framework react or flutter or whatever it is right and they're like this is the way this is where demand is but the problem is the demand supply curve constantly changes and I've been vocal about this at least with software that we're just producing too many software engineers and they're all doing the same thing see software is a vast field everything from AI to blockchain to game development it all comes on the software right and game development you can't use react to do to make games you probably have to use an Unreal Engine or Unity if you want to make serious games there's a plethora of options of what you can learn and what you can become but everyone wants to go and become a front-end engineer I agree it's very cool to be a front end engineer and all the memes and the content on the internet about it but also you have broken the supply demand curve I can tell you that for sure because when I put up postings to hire people uh even people in sales and things like that we'll have people from who have studied front-end engineering who know react or something like that and they're like no I want to be you know I want to get this role because I'm not getting a job otherwise we're producing too many of these people and the thing is there's no problem with producing too many of these people because if you're in the top 1% top 5% you'll still get a job you will be excellent your portfolio will be excellent we will still hire somebody in the top one 5 10% right but the problem is if you're are not that good then you have to look for other places where somebody will give you a chance an opportunity because sometimes you may be good but you may be undiscovered and therefore your CV will never say anything awesome because no reasonable company has taken a bet on you yet so what I found is maybe switching slightly like going from building apps to building games it's not such a big leap sure there are a lot of things you need to learn you need to understand the entire like 3D modeling and design side but look for empty niches this a war Buffet story right like when people are being fearful be greedy greedy be fearful and I think we've just been extremely greedy all of us about oh we want to get into a particular front-end job and you oversaturated the market and not to say that there are a plethora of colleges right now that are coming out just you know it's all the same degree and it's all they're just over producing people and you're seeing that in the effects on the downstream effect of placements because also remember you need enough people on the other end right who are working at a company that's hired a TCS you need enough client demand also and there simply isn't enough client demand especially in India we've learned a lot about software which hopefully I can come to uh you know as we keep chatting

### Problem With Indian Software Companies [7:46]

I don't think you building a pure software company in India will work you can export it outside yes uh and there are some b2c use cases where it'll work but most of it needs a mo that's not really software zomato has a Mot in logistics they have the one of the best Logistics networks out there same with Amazon so I think you really need some advantage and software is a good scaler or an enabler like what does software really help you do if you're building a food delivery company because people can pick up the phone and they can also dial let's say a central operator and the operator says okay what food do you want and I say Chinese and then the person on the other end says well here are the five Chinese options five Chinese restaurants and here are the best options and you could respond you say I want option one or option two or whatever right why app because app makes it frictionless it makes it faster and most importantly it allows you to really onboard people at scale you can bring on as many restaurants all in one view you don't need the operator but what this really unlocks for you is scale it's distribution it's the ability to go much faster and that's why Tech valuations was so high tech growth was so high because it's just that you know if this thing works it goes from 0 to 1 very quickly then you know it goes from 1 to n very quickly 1 to 100 very quickly

### The problem with agencies [8:58]

this is not true for an agency let's say you start a marketing agency tomorrow like every kid wants to do it I'll tell you the problem with an agency if even if you do it well you get stuck after your third or fourth client right I've seen this so many times in my life that I just know it's like this is evident right because at your third and fourth client all the very motivated founders of your agency all the people at the top are like I love doing this is great I'm getting clients I'm making money but the minute you start hiring other people they're simply not as motivated they simply don't care about the thing as much so uh keeping their motivation many of them will keep quitting you have to maintain their morale and getting new clients is now a sales thing it's not just your network you need to go out and start selling to clients starts becoming complicated to scale and it's always hard to scale things when every new client you bring on you need to hire a new person right it's just hard and that's the reason that people choose to go the product rout because it's it scales much faster it's a little bit easier to scale at least Engineers on

### Building apps is easier now than 10 years ago [9:54]

the internet in India forgot that they thought it wasn't about that it's about building pretty websites that load really fast there are incremental benefits to it to having a website that load slightly faster than somebody else's website but it's not as high when people have blazing fast internets on their phones because remember most of the thumb rules around building fast pages and fast websites and fast apps were in an era where everybody had terrible internet and where we had to build everything from scratch 10 years ago building a simple app where you just like I built a jobboard 10 years ago I built a company called job Spire 10 years ago where it's like no. com right you could go if you're an employee you can hire candidates on one side and on the opposite side you know if you're a candidate you could find jobs at that time we didn't have the support of many of the Frameworks that most of you have today it's a different world we had to re-engineer some of it that looking today I could be like I could build a job board in like 3 days today but 10 years ago was not that easy so the point is and this is not we're not even talking about AI it's going to change after AI it's just going to get even easier after AI but the point is

### Building app for performance vs. scale [11:00]

it's getting a lot easier now and it's no longer about incredible performance very few people like I know people who working at Viacom who where I know it's about performance where they'll have some really shitty phone and they have to stream the match uh the cricket match and they have to do it in certain benchmarks for themselves I get it and they'll probably have like cror of people watching most of you and I've seen some of your CVS right because on those redit groups you'll put your CV and then you'll ask somebody else to say why isn't this working and I can tell you it's you're not working on problems of scale yet you're working in some real estate company that needs you know a simple app for some backend stuff there like five people using it right so you're bothering yourself with the wrong things learning the wrong things hoping that a large company who needs the scale will hire you but actually what you need

### Betting on future trends [11:43]

to do if you want to build your own Niche is to do something that looks like it's slightly less popular right now you need to bet 6 months in the future or maybe a year in the future and hope that's right and for me this year a lot of things I have done this year are exactly that I bet 6 months a year in the future and every time I bet on the current there's a there's things that in the middle of the pandemic I said okay this I'm going to bet on the pandemic now already too late so you need 6 months of time to actually build into that role and hope that grows over time right and people forget that it's not you guys and girls who are thinking it's about you learning another framework or you doing this it's the supply demand C it's over the edge find another Niche software is so vast there are infinite niches go find one of those niches instead of just being the next front end guy all right moving on

### Progress in open-source AI models [12:27]

now one big news over the last few weeks is that Mixr just got dropped which is an open source model that you can run on your computer that has GPD 3. 5 like capability it is wonderful a because I really respect the team the last time they dropped a model they said this should be open source and they just dropped a torrent link seriously they on Twitter and the second time they dropped a model which is now this new mixed drill which is a mixture of experts they dropped torrent again they just said take it should be free and uh the founder went on uh radio recently and said they're going to have an open- Source gbd4 level model available by next year and that is wonderful because you can actually run this right now for free you can download an app on Windows called LM Studio it's a completely free app just search for the mixl 7B model it might not run on some of your computers but there are also many other models you can try on it like any model that gets popular on hugging face eventually makes it to LM Studio you can just search the name download it gives you a chat interface ask it any question you want and I think this dependence on paying $20 for GPT or signing in for Bard and all that is no longer necessary with the open source model and mix is even though I don't personally believe it's GPD 3. 5 level I did use it does make it has some Chain of Thought errors but I still think the fact that there's so much progress there is amazing and LM studio is sort of like uh you know an app on your computer for you to trial out any other model similar to LM Studio we also have comy UI which many of you have used

### Tool to convert video to any style [13:55]

heard of which is a node based editor for stable diffusion well we made the anime in this tool and this tool is basically composed of many different nodes and these nodes itself are different repositories that people come up with for example IP adapters to capture my likeness without training a model you can actually do that you can capture my likeness feed into a image or a video without actually training a model on me which we used to do in the past you can now do that and every new repository that comes out comy will have an update and that repository and the models available with it will be available so you can put it together you can make an anime in fact there's a workflow okay and we kind of figured it out ourselves independently from anybody else although we did take a lot of help from the internet and now a lot of other people have figured out that workflow it's the combination of Ip adapters for maintaining likeness and then the combination of anime diff which you know allows you to produce multiple frames of motion that combination is really good for converting video into anime but it's also great 3D for example take a look at this is a video of PR that we had shot we had made this anime scene for example check out this to/ Pixar Style reimitz character in 3D or somebody draws this character by hand and the new flow is you actually shoot the footage first and then put it through one of these AIS that give you either the Pixar style or an animated style etc and you'll have different models for each and that's the world we're entering and you can do it right now hello hi come

### Meeting Mythpat [15:47]

say hi come say hi uh this is our screen yeah come say hi so I'm shooting a video I'm not live okay doing a this week in AI whilea thing earlier I used to do shorts getting really irritated with shorts so I have to do one long form now and they'll clip it out we're getting a much bigger one come say hi I'm say hi so by the way we have uh I'm no we're not live but it it'll be recorded okay so uh so I'm coming in your video so mpart is here and it's not AI version sorry but uh he's he's just here to say hi and we proceed in the video now bye all right what's the next thing the

### Self-driving cars banned in India [16:37]

union Minister for Road Transport and highways nitin gadkari has said today publicly that self-driving cars will never be introduced in India the reason cited by the minister is to safeguard jobs of drivers now that's going to

### Unfairness of protecting some jobs over others [16:50]

cause a problem because I can already sense that there's going to be a section of society that's going to find this unfair for example why are the jobs of drivers being protected when let's say a software engineer who is slowly seeing GP chip away at his Advantage why is he not protected and I understand the math I know that India really relies on delivery boys and drivers and truck drivers and that makes a huge chunk of the GDP plus so many jobs are given right you don't want to take away those jobs and compared to that software engineering looks really tiny but software engineering design art till now they're all very um Niche job roles right certain Elite have had access to those job roles so nobody really cares about that falling but the truth with AI is eventually that's going to pervade everywhere right eventually it is going to be most generic jobs done on a computer and we're fairly certain we're going to get there I still feel like that software engineer or that Designer is still going to feel unfair why is the government protecting one class of jobs and not the other is it just sheer number at some point there's going to be jobs that are very critical to the economy which create a lot of jobs in India but we'll still have to choose the government will still you have to choose is this AI protected or is this not AI protected it's a very dangerous precedent by saying we will ban automated cars forever and I know a

### Can AI handle Indian roads? [18:06]

lot of you will argue whether okay is automation even going to happen is can they drive on Indian roads and first of all a lot of that is a myth like a lot of people say Oh India you can never drive a driverless car in India but what we have learned especially this year is that computer vision is getting really good better than humanized and more importantly a computer is just far more reactive so paradoxically a computer might be better suited to drive indan roads than human beings now till we see a working prototype I'm going to reserve I'm going to measure I'm not going to say anything that is too forward looking but it is possible that at some point in the US we already have driverless cars and they work to be able to combat Indian roads eventually somebody will build a custom model it's only a matter of time and the fact that the government is protecting one set of jobs and not the others it confuses me a little and you need to have a very clear policy

### Varun's policy around AI taking over jobs [18:52]

around this and I have a policy around this I've thought it through right so I've spent some time thinking through I'm not saying I'm right but this is a policy I think all jobs that are internal to the country where the competition is internal then you can choose to prevent AI from playing there but all jobs that are dependent on export where we're competing with all the other countries in the world there you should allow the use of AI let me explain to you why let's say there's a project available in some country the US that company can go to many different providers right and they end up coming let's say to TCS but just like TCS I'm sure there are people from other countries competing bidding for this project now if TCS goes in and it's banned from using AI then the cost will obviously be very high because you need a lot of talent there right so using AI

### Using AI as a competitive advantage [19:40]

is actually a competitive Advantage for TCS like think about us we use AI to make content for different companies and we will use every Advantage we can to reduce the price of producing that content because we are also competing with other companies across the world that are generating content and we need to be competitive because I know that if there's a US client or a Dubai client coming and spending on me versus some other company across the world I am bringing some revenue for India and I will pay tax on it right that's why

### Separate policy for internal vs. external jobs [20:07]

exports is the Holy Grail we are always counting our GDP by how much export can we actually create as a country and there is competition for this because if I lose the deal some guy in the US will win the deal and that Revenue goes to the US so it's fair for me to use any advantage I can if I'm competing on the global scale but if it's internal if it's only inside India if every group of Mumbai taxi cab drivers is only competing against the others then it's fine they can compete on price quality of services etc but it's very difficult for them to compete with AI so in this internal rule set of India you can choose to say okay this competition is only going to happen internally then well there we can prevent the users of AI this is one policy and we already do this for a lot of things we definitely do this for tax where if you are selling a service abroad you get lots of tax rebates all

### Review of Humane AI pin [20:58]

right last thing for today um I did a quick review of The Humane pin and I said while it looks good I'm not going to buy it because I thought the latency is just not going to be good enough because we had run some projects and the latency of you speaking to the machine understanding what you're saying which is transcribing whatever you're saying into text and then you know sending that to GPT and then getting a response the entire round trip takes 3 to 4 seconds probably longer but the thing about AI is the astound in Pace of innovation is

### Real-time conversational AI now possible [21:30]

mental there's a company called Deep gram and they've been working on text to speech models for a long time and they finally have a realtime model now this is not a gimmick I actually tried it it's really good I can speak to it not just in English but also in Hindi the model is called Nova 2 it's very fast and we are actively using it in production and now that this thing is there this real time ability to speak back to you things are going to change now you can actually do things like virtual girlfriends action figures that talk to you because until now you speaking to an Alexa waiting a few seconds then getting a response that would bore you out but if it's nearly instant where your speech is being transcribed instantly which in simple language is you sending a text to GPT you sending a prompt to GPT and getting a response that speed is also much faster a because GPT has come out with GPT turbo that's faster and all of that but also there are plenty of smaller models like stable Zephyr is a three Bill model there's a f model that's also a small model Google Gemini also has a model called Gemini Nano these are all very small models that can work extremely fast and get you the results very quickly so combine realtime text to speech with these fast models and I think you have as close to real-time latency as possible and one of the cool things with having this realtime technology is now I can put one

### Interactive paintings/toys now a reality? [22:50]

of my dreams into reality like one of the dreams I had was I don't know if you've seen Harry Potter there these paintings on the walls that can actually talk to you they can look at you they can maintain eye contact with you they can talk to you they can respond to what you're thinking they have personalities inside of them and we already know how to do this with the models fine tune models we already do that for certain companies so I run a company called AOS under that we have Labs that purely creates fine tune models for companies across the world we work with a lot of companies in India doing it so we're fairly confident on the software side of things but we've always been Keen to do it on the hardware side of things and until this latency problem got solved if it was 6 seconds if you talk to it and it respond 6 seconds later that is boring so we waited for the perfect moment and now that deep gram is out now that it seems like it's possible to do it real time we

### Prototyping real-time conversational hardware [23:40]

are now working on the hardware so let me show you something so this is our base prototyping kit it's based on the Raspberry Pi and there's a lot of interesting stuff in here but for now our painting is going to be very tiny it's going to be a small little character and the use cases for this are incredible right like I can have chanakya uh available that you can talk to at any time uh on a painting you can have a painting of the smartest I don't know script writer in India that you can just ask while writing a script or a very smart maths teacher who can help your kids actually learn something throughout the day so I think toys are about to get much smarter and we definitely want to take a hardware bet but I kept saying that Hardware bets

### Why taking a hardware bet is risky? [24:19]

are risky because the margins are low you probably end up making no money and Samsung might come tomorrow and kick your ass otherwise there's really no point doing this if you're not trying harder and harder things over time so wish us luck there's a very high chance we'd fail and we'll keep you in the loop of this project like as we learn more things probably do a small series of videos on this channel about what we're building and hopefully you get to build it with us in a way digitally virtually

### AI as a Romantic Companion [24:44]

and just on Q one of the things that's taken the Internet by storm is this thing called Digi which is a romantic partner that you can have available on your phone anytime you want and there's been lot said about this people are like no I don't want a partner that uh is completely digital says yes to everything I want no I'm not going to pay for a partner that's perfect and I agree with all this and the people who make this are going to be smart about this and they're going to make you pay in very unconventional ways which means that they're going to purposely make that digital partner be upset with you and I saw this on Twitter where somebody said you're going to pay to roll back your conversation let's say you screwed up the conversation and we've all had that in our life where we've known somebody and we've ruined the conversation and we go back in our head and we were like maybe we could have had this better now you will be able to pay for it you can screw up the conversation with the bot and it will be part of the model to get upset with you and if you want to roll back the conversation to a previous point you pay it to blank out its memory for that period what a dystopian world to live in but also I know the

### Loneliness problem in India [25:44]

loneliness problem at least in India is insane I'm not going to say whether it's a good thing or bad thing but I think having some hope for the loneliest of us is necessary and maybe this will help solve that but I also think for the people who are not already lonely who have friends they're now going to spend a lot more time with these things and therefore spend less time with the real people in their life one last thing

### ByteDance banned for training on GPT outputs [26:04]

promise H okay so bite dance has been caught for taking the outputs from the GPD model and then using that to train their own model which is something you can do you can actually do that you can take the outputs from one model as long as you have the input prompt and the output you can train a new model to perform in a similar fashion so how did openi respond to this they found it unfair and they immediately banned bite Dan's account because open explicitly States you cannot train on our output see I'm not the biggest fan of China but

### Hypocrisy of OpenAI [26:29]

there is a little bit of dolap in this because look open I never asked any of us the writers on the internet before using their work they just scraped the entire open internet I'm sure they scraped some of the content created by Chinese individuals too and they never give them any permission or compensation they train technically on their outputs right so if that's fair and if you're anyway creating a model on some other human outputs your outputs are fair game if somebody else wants to use it to train their own model now I'm not going to get in the ethics of whether openi should or should not have trained on the public Internet it's already public people want Google to Archive it I get all of that but to ban people for training on your outputs when you guilty of the same mistake in my opinion that's the laaper

### Outro [27:15]

and with that we end this episode I hope you guys learned something I like having this long form set of conversations with you may not agree with all my thoughts and I get it like we are living in a more and more polarized internet you will not agree with most of my thoughts and that's okay we should agree that at the end of every conversation 10 20% of what I said you might not need to agree with and we move on with that right and I feel long form is the best format to do that if you like this please subscribe I'm going to keep putting out more stuff bye