Quick Questions with an AI Founder - Anudeep Yegireddi

Key Takeaways

following is the conversation with anudip yeri who is building an AI platform that automates the process of interviews for both companies and candidates anud has previously worked for companies like apple he has founded multiple companies in the past I met him at a co-working session in Bangalore and asked him a bunch of questions here's how it went what projects are you currently what you on so I'm working on AI agents as a general concept uh specifically I'm trying to build a mock interview agent that helps both interviewees interviewers and all the layers in between okay how do you think AI will change the world in the next decade um I think AI is the big change and we talking about this a little while ago is that it moves the branching from binary to gradients that previously it used to be if this do this else do this right and you to human to go in hard code what it had to do yeah with AI now because it is kind of a reasoning agent it can figure out the gradient of possibilities and you give it overarching instructions as such as like um only taking action when you have like higher than an x% probability where it is generating its own possible space like the action space it is generating and then it is choosing one action space based on the rules that you have given versus like if this then do this else do this else so you as a human have to figure out all the potential branches and then like direct the uh the the code to like go down a specific Branch with a specific set of instructions are verus with AI it can generate a much larger space of actions and if you code it well um it figures out which branches to go down by itself I feel like that's a big change that is going to come so how far AGI is according to you AGI is um it's a very uh uh seductive word but I don't engage with it too much because um we tend to look at AGI as AGI and also like human intelligence and like AGI is when this thing reaches human intelligence but human intelligence is like is like uh it's like gas and if you give it like an open space it'll fill it up yeah right but human intelligence is also good for another thing which is you can figure out new spaces can find new spaces that's an emergent idea where all of the current things hold but the jump is unobvious so for example uh before Newton came up with like the loss of gravitation we knew that when you throw things up they fall all of the knowledge was there the emergent idea is that there is a fundamental layer that is connecting everything that that that so that is the concept of creating new spaces and once a new space is created like human intelligence kind of goes and fills that space up nice artificial intelligence is better than humans at filling the space so if you give a new room it will fill that room up faster than a human but an AI at least in my current view of this is inherently incapable of finding new spaces I think that will always be something that humans do so um when will AGI come is like a complex question if um an AGI in the way that I look at it is in the space that we have defined of the rules it'll figure out more possibilities and in those possibilities there might be interesting uh things that we humans have not thought about because it's a search and find problem not like a figure out a new thing so that I think we can we should be able to start seeing some interesting applications of that kind where AGI in some ways is like the model kind of by itself is figuring out what to do where you're only giving the top level instructions not like okay create an agent that does this then also create an agent that does this then link these two agents together then link a third agent to do this which is doing this it's more human it's kind of like a branching logic that I said right versus like a gradient so I think that we should be able to we should start seeing some of those interesting applications what are some of the most amazing applications of a you've seen so far companies or so most of the innovation in AI right now is happening in the model space model Innovation space so training basically comp like anthropic or bigot yeah so I think most of the Innovation is concentrated there it's almost like they're taking out all the wi so everybody is now like thinking oh if I'm not able to train my own model then open has everything and like there's no point like innovating because they'll just launch a new feature and then the heating outome and that's what's happening that is what is happening because um this is maybe a hot take but I don't think AI has found a slam dunk use case for humans yet everybody's selling you PS and shels it's like better way of like putting it to the cloud better way of training better way of like creating the framework that orchestrates things but the end users the people who are using it there's only bits and pieces of use here and there so I would say that maybe co-pilot is one of the first like really good use cases but it's a it's for programmers and programmers anyway think it and the reason that GitHub Cal works well is because code is uh formal language yeah whereas human beings talking informal language and there's a whole like actual linguistic like chsky Frameworks and everything that goes into this but um informal language or natural language that that we call is inherently UNM modelable that's why you need freaking 175 billion parameters of like mixture of Experts of like so many things to do it whereas a code you can I can sit down and write a compiler fairly right by the way talking about copilot which editor do you use code editor so for large scale projects I use vs code but uh for small scale things I use Neo and I use wiim key bindings everywhere like right even in my browser I use WIS oh interesting yeah nice which llm do you find yourself using the most uh from prototyping perspective I use mostly open but I'm starting to feel like anthropic and all of these things are doing a really good job so 3.5 it yeah yeah so like the way that I'm building my stuff is it's configurable you can change one word from open a to anthropic and like okay ni yeah what's your programming language of choice um I think most deeply in Python most deeply in Python yeah yeah that's that's that's what I taught myself in and I would say I am an advanced developer in Python okay I'm an intermediate or slightly below intermediate developer in JavaScript um but the language that inspires me the most and something that I know I will have to learn um is lisp oh Paul gram Paul also mentions the same yeah so some of my architectures are recursive architectures okay so like things called things called things called things and I'm replicating that in Python and python is a good language to replicate things in because it's slow yeah but you understand what is happening very very well list is a perfect fit for that CU it's it's the functional uh Lambda calculus right so it's like recursive in nature so like list is a thing that I know I have to learn and I will slowly learn it uh but yeah good nice what drives you to do your job or whatever the tool that you're building so I I was born in a really small village okay apart from our Central Family like my dad and his two brothers outside of that everybody is really poor my dad was the first person in the family to educate himself well so he went and studied at I Amad the first he didn't know how like he didn't know to speak English or anything he really really pushed and he got that and because of that he became an inspiration for the younger two brothers they also doing decently well so the other two brothers are really but outside of our CORE family everybody else is like not they're like Po and I'm whenever I go visit my family I'm constantly reminded of the amount of privileges that I had in life due to circumstances that were completely out of my country right and in some ways I want to distribute this privilege to the people who could not have who did not get that because of circumstances I think in some Central Way like I that is the driver force that I had a lot of privileges in life I'm not especially special everybody's capable of what I'm doing it's about like making those opportunities available in more places and especially in a place like India like being able to distribute that privilege is is one of the core driving forces man that's that's a really good answer nice resonate with you on uh many levels there cool uh what was the first program that you ever wrote I the first it's a very canonical thing but I uh had um so the initially I was trying to find a way of scheduling things and like doing my things in a way that works for me and just started with too lists and everything and was not really working for me so one of the first things that I built for myself was uh task manager task manager yeah and it was completely in the command line okay cuz if you're a python programmer you try not to build anything with their interface for as long as possible unless you dip into streamlet and everything uh so it was completely command line but it really well what's one book that you would recommend everyone should read at least once in their life I'll tell you the book that shaped my perspective um there is this book called The Emperor's New mind emperor's new mind okay by Roger Penrose okay I'll put it in the show notes cool and what is one piece of advice you would give yourself 10 years ago nothing I you done I mean you if you're happy with who you are today then you're happy every past version of you needed to exist in that exact way y so Matrix yeah just continue what's the best advice you've ever received it's not an advice it's more like something that I derived from it a philosophy class that I took in school and it'll fundamentally change the track of my uh life but one of the central things that I took from that was a new idea is only truly understood if you connect it to something that already exists in your brain if not it is a fact and facts are forgot oh okay this is a lot to do with how aens are also working with cont cool nice uh what is a topic in AI that you would like to explore more reinforcement learning with human feedback reinforcement learning I feel like is where decentralization of AI is going to happen and right now it is something that only the really big players like open a and everybody everybody does it is what is going to decentralize the so I want to explore that a lot more than I have what is a risk that comes with the ey what is a risk yeah to humanity so necessarily there is going to be a transition period for jobs M uh there's going to be a transition period for what humans do and a lot of people it's easy for me to say because I the benefiting end of all of this but um when um the printing press or like the when Renaissance happened and Ving like became automated people were like this is the job that humans do like you're taking away all the jobs we will be completely jobless and like we won't have anything to do but what happened was once it came people figured out new jobs yeah much higher uh level jobs that humans are better suited to right so there is going to be a transitionary period where some jobs are going to go true and especially people who are in the business of um resynthesizing other people's work that used to be the way that they earned money and fed their families so middlemen are basically going to get eradicated what's your favorite operating system I mostly work on but but almost everything I deploy is on Unix so some something like that your favorite car my favorite car yeah uh that you have or may wish to have a Nissan GTR Nissan GTR nice your favorite exercise if you work out my favorite exercise is uh a deadlift dead nice a website that you like to waste your time on um so I removed myself from social media while one of the social media uh but a website that I do spend a lot of time is um wait but why wait but why yeah Vlog nice uh okay what's the bravest thing you've ever done I think the most recent one um 3 years ago I was living in San Francisco I was earning uh about 200k I had a job lined up that was going to pay me 300K and I dropped all of that decided to move back to India and uh told nobody because if I told anybody they would tell me they would tell me all the reasons why I was stupid which I was like unequivocally was a stupid decision but a good stupid for me um so I knew that everybody would like talk me out of it so I decided and a week later I was in India uh what's one thing you afraid of becoming complacent okay what's one thing that excites you uh knowledge like learning new things the further away they are from what I already know the better your favorite browser I use Brave okay you get an all expenses paid trip to one city where do you go somewhere in roome like history okay um and I haven't explored Europe at all there a lot of history yeah probably that and everything I se what is one thing you can't live without my laptop okay uh how you describe your experience at uh you know the school NBA School met fantastic it it it I was not a Sho in by any measure it was a really good school and uh up until then I thought I was uh arrogantly smart then I went to that and I realized oh I'm the dumbest person in the room was a very freeing thing I realize you want to be the dumbest person in the room because that means there's that much space to grow so yeah okay who's the one person that you really admire a few but of the people who's alive um Roger penro is one person who I really admire or like I'm inspired by um the people who are I I tend to go a little bit more towards scientists okay um so other people who have inspired me include people like Charles Darwin Maxwell okay like these kind of people philosophers well who is one person you looked up to when you were in college Steve Jobs and is that true today is that I don't want look up to anybody I'm inspired by okay what is your good place to learn AI or engineering Concepts it could be like if it's books just so I have a framework for learning that is that I have developed over many many years that works well for me so for me to understand something hard that is new I like to follow the human history of it okay so like for example there was some time ago I was like I realized I don't really understand electricity well I understand what it is but like I don't understand it at the level of like the depth that first principle that I wanted so the best way to understand that was to find um the people who were like the key contributors to us two main people of and Max and read their stories follow that you've realized why is that why did these people feel the urge from within to work on something so so fundamentally new and like and when you follow that you contextualize it with human history and that leads to a much deeper understanding I have found so like usually biographies are like good starting points for me if I want to understand something deeply okay nice and like you and those books hard to read like if you're going back and reading their stories but it's a biography so it's much more we humans love reading about humans it's boring to read about a concept like oh like when you do a matrix multiplication like happens and then the dimension gets like wared in this that's boring if you're not like specifically interested in that read about human their problems their like uh troubles that they had to overcome the things that they did and the experiences that they live it's much easier to anybody it if somebody's building an AI today uh where should they start the deeper you understand the word embedding okay the deeper your understanding of artificial intelligence as a whole okay understand that word go deeper understand it as deeply as possible and know that there will come a point like okay I understand it that's your starting point okay okay interesting I I spent 3 years doing this um and I understand things from first principles up still I cannot say I understand embeding completely okay nice cool and thank you so much for your time and it was very fascinating to hear your experiences and uh what you have to share thank you so much