The Untold Story of Scott Wu, CEO of Devin AI

what is the value of 255 2ar minus 245 s answer me in 3 2 1 don't know the answer neither do I what is the value of 255 Scott 5,000 5,000 is the correct answer he didn't even finish the question oh my God look at the speed of that all right next question the digits 1 2 3 4 and five can be arranged to form many different five-digit positive integers with five distinct digits in how many such integers is the digit one to the left of the digit two two such integers to include are 14352 and 512 3 4 so the number one is to the left of the digit 2 and same thing you got to answer me in 3 2 1 I'm sure right now you are like totally baffled I don't even understand the freaking question and the next question is the digits 1 2 3 4 and five be arranged to Scott again he didn't even finish the whole question 60 is the correct answer the poor guy didn't even finish the first sentence he only read up to here and then Scott already knew the answer if the pattern shown continues what is the letter in the 2010th position Scott a a is the correct what all right well this time he actually waited for the host to finish reading the whole question but what is the letter in the 2010th position how on Earth okay I don't even know where to start to figure this out but anyways this kid that you see here is who we all know as Scott woo founder and CEO of cognition Labs which announced Devon a few days ago now Devon took the tech industry by storm it claims that it's the first AI software engineer and you just give it some instructions and it can build the whole project for you all autonomously it's able to debug and fix errors it's able to search the internet for more information and update the code accordingly to build out a finished polished product now really quickly let me play you a condensed version of their Showcase Video so you see what exactly Devon does hey I'm Scott from cognition Ai and today I'm really excited to introduce you to Devon the first AI software engineer let me show you an example of Devon in action I'm going to ask Devon to Benchmark the performance of llama and a couple different API providers from now on Devon is in the driver's seat first Devon makes a step-by-step plan of how to tackle the [Music] problem after that it builds a whole project using all the same tools that a human software engineer would use Devon has its own command line its own code editor and even its own browser in this case Devon decides to use the browser to pull up API documentation so that it can can read up and learn how to plug into each of these apis here Devon runs into an unexpected [Music] error Devon actually decides to add a debugging print statement reruns the code with the debugging print statement and then uses the error in the logs to figure out how to fix the bug finally Dev decides to build and deploy a website with full styling as the visualization you can see the website here all of this is possible today because of the advancements that we've made in both reasoning and long-term planning it's a really hard problem and we've only just started but we're super excited about the progress that we've made so far in the meantime if you'd like to try out Devon on your own real world tasks send us a request below and we'd be happy to forward it to Devon so Scott announces Devon a few days ago and before this the team was just building in stealth nobody has really heard of cognition Labs or Devon so who on Earth is this Scott woo it turns out he's an absolute killer in math and coding so you know I already showed you videos of him like winning all these math contests and then here he's competed in the international Olympiad in informatics yes there is an Olympics for informatics in fact I did a video a few weeks ago which introduces an AI which is able to solve The World's Hardest geometry problem so there's also an international Olympiad in Geometry right now this one is in statistics and Scott woo has won three gold medals in three consecutive years so in 2014 2013 and 2012 in fact this Beast of a human has gotten 100% on all questions in his final year reaching an unbeatable score of 600 or 100% if you're interested in seeing what kinds of questions this International Olympiad in formatics actually offers to the contestants they offer a few previous examples so just out of curiosity let's check out the first one so I'm going to click on this catfish farm and oh my goodness what is this what does this even mean so buun click owns a catfish farm the catfish farm is a pond consisting of n * n grid of cells okay so it's a square I guess The Columns of the grid are numbered from all right I'm just going to give up here but I hope you understand the complexity of these questions the these are the world's hardest math problems this is an international Olympics in basically informatics and statistics this was part of the 2011 rathon math counts national competition and there Scott was in eighth grader from Glasgow Middle School in Louisiana so at that time like you can see his speed in solving these very complex math problems already showed that he was a math prodigy at the age of 12 Scott started learning how to code how to program and then at the age of 14 he started programming competitively with his brother Neil wo who I'll talk about in a second Scott also enjoys a ton of other competitive strategy games such as smash poker chess and Tetris but I don't know if it'll be fun actually to play with him cuz he'll just destroy you so on this code Force's website which is a popular competitive programming platform which hosts online competitions you can see that Scott woo has achieved legendary Grandmaster status which basically means he's the best of the best out there he's gotten a 5-year badge 8-year badge and now a 10year badge so a math prodigy who learned coding at the age of 12 and then he's also a genius in coding as well so stalking Scott's linked in here it seems that after high school he went to Harvard University to study economics and then he worked at adapar hope I pronounce that correctly which is a wealth management platform for for registered investment advisors specializing in data analytics and portfolio reporting and then after that he was the co-founder and chief technology officer at lunch club and he did this from 2017 to 2022 lunch Club is a social platform that uses AI to connect users with common interests and objectives they received a $4 million seed funding from andrees and Horowitz if you haven't heard of andrees and Horowitz it's a very prominent venture Capital firm that's known to have invested in some of the biggest tech companies in the world so for example they were early investors in Facebook Twitter Airbnb GitHub and slack they were also early investors in coinbase open Ai and deep mind so they received a $4 million seed funding from entries and Horwitz and then after that they received a series a funding of 24 million also from a few well-known investors making the comp's total valuation at over 100 million so he was working on lunch club for 5 years and then in 2023 he created cognition Labs with two other co-founders Steven how who's the chief technology officer and then we have Scott woo who's the CEO and then Walden who's the chief product officer even though this was a very early startup they raised millions of dollars notably from Peter Teal's Founders fund but also other well-known investors such as former Twitter executive elad Gill now Peter Teal's Founders fund is also known for its great track record so Peter teal was one of the co-founders of PayPal along with Elon Musk and a few others and all of them they're known as the PayPal Mafia they've gone on to be very successful later in life as well so Peter teal has this Founders fund which is essentially a venture capital firm that invests in early Tech startups so some of their success stories was they invested in Facebook in its early days and also SP Spotify and Elon Musk SpaceX interestingly Peter teal also has this Peter teal Fellowship which basically he selects around 20 people every year and he Awards them with $100,000 to not go to college and instead pursue their own passion build out their own project and I think this is actually a really cool idea because you know after graduating from school I really didn't see the value of school and so a lot of successful people have come out of this Fellowship as well for example V who built ethereum actually was one of the recipients of the Peter teal Fellowship but anyways back to their story so these three had secured funding from Peter teal and a few other investors and they were working in stealth in 2023 so nobody outside really knew about it they were just building the product building the code building the AI secretly for months and then they've since expanded this team to 10 people one of them is actually Scott's brother so his brother is here he's called Neil wo and he also works at cognition I believe that your older brother Neil won the mathd Downs national competition several years ago yes so both Scott and Neil woo are well known for their coding and their math abilities they have achieved worldclass status the woo Brothers have been competing in and often they won these International coding competitions ever since they were teenagers so that's the woo Brothers which sounds like a nice name for a Asian gang in San Francisco but anyways the other you know eight members of the team are also also world class in math and statistics so this team is not your normal startup team with you know just a bunch of Indie hackers or coders and maybe a product guy maybe someone in sales or marketing this team has won a total of 10 gold medals at top International competitions and so Scott actually gives this really interesting Insight teaching AI to be a programmer is actually a very deep algorithmic problem that requires the system to make complex decisions and look a few steps into the future to decide what route it should take so programming the AI to be a programmer is more than just learning these basic coding skills and that's what gives this team an edge they're all worldclass in maths and statistics and coding which is the skill set that is required to basically Teach an AI how to program thanks to the sponsor of this video upix if you're feeling overwhelmed with mid-journey or stable diffusion you don't want to worry about prompting or learning all these different settings well upix has made it dead easy for you to generate highquality realistic images of yourself or anyone else in just one click it works on desktop as well as on your phone you don't need to install any apps or anything it just works straight from your internet browser simply select the template and then upload your photo and then click create it's as easy as that and look how realistic the results are there's many templates for you to choose from and more to come so check it out at up.app and you know when Devon was announced there was some criticism about its legitimacy about how good it is some people said it's just a marketing stunt there are plenty of Alternatives out there this just got the most press because it's funded by Venture capitalists or that it can only solve 14% of programming tasks so it's not even that good but keep in mind you are dealing with a team of worldclass gold medalists in informatics and coding do you really want to bet against this team and the Devon that we know today this is the worst you'll ever see of this this Devon is just their pre-launch showcase and they've already outperformed the best llms out there by a landslide and note that the blue bars are actually assisted that means humans had to guide the chatbot on you know how to fix errors in the code for example whereas green bar which is Devon this percentage of success is actually unassisted so it required no further guidance from humans to figure out the programming task again this is only going to get better and better with time so the team now consists of 10 people they were able to build something that significantly outperforms other coding agents so for example there's plenty of other ones out there such as GitHub co-pilot but I mean GitHub they have Microsoft backing them up they have unlimited funding like Microsoft is the highest market cap company in the world I think they have the world's top 10 tent they have Azure data centers so they have all the computing power they ever need but it's far inferior to Devon which is only Built by a team of 10 people so a few more notes on the other co-founders so this is Steven how the chief technology officer at cognition and how previously worked at scale AI this is a platform that helps label the data that's used to train AI systems so how was a top engineer there before moving on to create cognition Ai and then Walden Yan this is the guy on the right here he's the chief product officer at cognition he's still at Harvard University and then he requested that his status at the school be left ambiguous so he's kind of like he's not sure if he's still a student there or if he just wants to drop out but honestly if you're at Harvard or Stanford and you don't drop out you're kind of doing yourself a disfavor because slapping the word Harvard Dropout or Stanford Dropout on your resume looks so much more impressive than just saying I actually graduated from Harford or Stanford that's just my opinion so here's a demo of Devin from Walden hey I'm Walden one of the developers here at cognition AI we were playing around with whether or not Devon could start a side hustle on upwork so here's actual real job from upwork where the client wants to set up this computer vision model which actually looks quite interesting seems very difficult to set up um I'm not sure how I would start doing this but you know you give the task to Devon and ask Devon to figure it out and things just kick off Devon immediately goes ahead and you can see it sort of starts setting up the repo it actually runs into some issues here with the versioning so if you watch how Devon deals with it deon's actually updating the code to make these things work he continues with this loading and importing packages you can see that actually downloads images from the internet to run through the model but you can see here that there are actually some issues that come across however Devon knows how to handle these things Devon kind of pushes through and if you look closely Devon's actually doing print line debugging here where Devon is adding these statements to track where the data flows and Devon continues to do this until Devon understands how everything's working and actually then updates the code with the fixes after removing print line statements de continues this pattern of fixing code and running it again until it runs the image model across all these roads across the world and we can ask for a report from Devon at which point Devon sends over some sample images of roads with damage marked out and a nice txt file explaining Devon's work and the different kinds of outputs of the model good job Devon imagine you are a freelance software engineer you build websites you build apps you build plugins for people through upwork or Fiverr so you might either look for jobs yourself on upwork or if you have people contacting you you'll scan through your messages and select the jobs that you choose to take and then it takes you a few days or a few weeks to build out the actual project and finish the job for your client plus of course there's going to be back and forth you're going to ask your client is this correct do you want any revisions here's my hourly rate or here's how much I charge for the project etc etc with Devon and these other autonomous AI a it could do everything for you imagine an AI that can just Auto Select tasks for you autor respond and message your clients back and forth based on what you require and then we have Devon here who actually builds the project for you in a matter of minutes not days or weeks and all this time you just sit in your rocking chair enjoying the sunset now I'm not trying to overhype anything here but you can see how revolutionary or helpful this technology is I also want to end this video by saying that there are plenty of Alternatives as well they're all working towards the same thing which is an autonomous agent that can kind of work across different platforms and autoc complete complex tasks for you based on logic and reasoning this technology isn't new so last year we had autog GPT and baby AGI and both of these are all built by very small teams similar to Devon however what I found after using these two is that they always get caught in an infinite Loop and it's not really possible for them to complete more complex tasks more recently we have a few more of these that are a lot better at actually you know completing the task for you so we have OS co-pilot it's free and open source you can install and run this locally for example here's how you would just talk to the AI to complete certain tasks for you for this Excel spreadsheet so you can see you're prompting the AI on what to do and it's able to carry it out for you it breaks it down into various subtasks and then completes the one by one based on logic and reasoning here's another example where you're prompting it to play music so just a very simple prompt play music it kind of needs to understand well what are the steps involved to play the music so right now it's opening iTunes and then finding music to play for you another very similar alternative to OS co-pilot is multi-on it's literally the same thing so you can see in this example you're prompting it to order some books on Amazon so you can see it's navigating to Amazon now it's searching for the book that you stated it's clicking on the book now it's clicking add to cart and now it's searching for the second book clicking on that and then also clicking on add to cart and then it clicks on proceed to check out and more recently after Devon has been announced there was this medium post which kind of explains how you can build your own Devon this paper released earlier this year also introduces an AI coding assistant that uses multiple agents to help you write and test codee another really new AI is called Mesa they call themselves a knowledge processing unit they say that the kpu is the most advanced reasoning system for llms so here's an example of Mesa in actions so here you can see this AI agent is helping a customer with a question about an order that did not arrive it's doing all these steps autonomously you can see it's looking for I'm assuming the customer's information it's now thinking of what steps to do next based on logic and reasoning and it's explaining what it does in each step and then it automatically sends an email out to the customer thank you for reaching out to us we apologize for XYZ and we did XYZ to you know address your concern so you can see more and more of these autonomous AI agents that understand how to do complex tasks across different platforms more and more of these are coming out and interestingly they're all formed by these small teams of like Indie hackers it's not necessarily from these big Tech giants like Microsoft or Google which have has a lot of funding and a lot of computing power and again this is going to be the worst we'll ever see the technology is only going to get better and smarter with time and the success rate will only go up with time so I went on this tangent to say that well Devon is only one of many autonomous agents that are being developed right now Devon has a killer team of worldclass experts in statistics and informatics and math and coding but can they pull this off can they build the best autonomous AI agent out there or or will they just be average or will they just be outc competed by all these other teams that are also building AI agents whatever the case the future is going to be wild we're going to have so much new and revolutionary technology that can really help us automate a lot of tasks so anyways that's the background of the founding team of Devon let me know what you think in the comments below if you enjoyed this video remember to like share subscribe and stay tuned for more content I'll see you in the next one

Channel: AI Search

Views: 331,566

Keywords:

Id: EOampe5_ONM

Length: 21min 26sec (1286 seconds)

Published: Mon Mar 18 2024