The Neo Show: Founder Fridays #8 – Yuzheng Sun on AI Education with Hosts Howie Xu & Axel Pilette

Thank you. Hi thank you for having me.

Yeah, I left Stac and I didn't expect it to be actually a big news this week. But after I left Stac, it was acquired by OpenAI. I think the largest uh acquisition on Open history. But I mean what I'm doing is unrelated to that. Uh what I'm going to do next is continue building my AI builder community and AI course. I started doing that um since last May uh May last year and uh we built the community around uh like later that year. Uh, and it actually took off even with my part-time participation. Like I have a full-time job and I have a kid and I'm I have been able to give the community like a few hours per week. But uh we already have 2,000 students in the community and people launch projects there every week. So I feel like this is an exciting opportunity and I see tremendous value from this community. So I decided to uh go all in as my next journey. Yeah, it's all because of AI, right? You know, because of AI, we have new show. Because of AI, we have a new AI native browser. Because of AI, you are on the new journey. So, AI is really making an impact to any everyone at this point.

First impression is is very new and uh like and I I especially like where we can have the search bar and the URL and the chat all in one place and I feel like that is actually the destination I'm looking for when I searched when I go to or open any browser. I remember like Chrome when they did it uh I forgot how many years ago probably over 10 years

Yeah. Omniar uh to blend search and uh URL and I think this is the destination uh that we actually want wanted to get to. Uh I want an answer. I want a solution. I don't want to go to a particular path. The path is only a path, right? It's the destination that matters. So that to me feels like the future but I also feel like we are still very early in this journey. I actually watched uh your podcast with guru and you talk about there are three stages in AI browser and I agree with that. uh I think the third stage the second stage is uh like uh browser proactively give you new things give you the information or cater uh to your demand and the third the third stage is agentic uh uh browser right the browser uh do things for you and I look forward to that future

How phrasing you

Yeah and I see a question from uh chat as well what about today's guest makes them particularly qualified in the teaching of AI so I guess I I didn't do introduction of myself and maybe so uh I like I I didn't study AI in school uh my PhD was in economics uh from Cornell but uh when I was doing economics it was actually the kind of the down of machine learning in social science like uh I graduated in 2017. Uh computer science folks are already doing uh AI uh so so to speak machine learning and uh deep neuron neuronet and uh I just started to learn that uh in PhD. So instead of going into academia I went to Amazon as an economist and uh experienced machine learning firsthand. I learned all the machine learning courses at Amazon and I applied machine learning in my daily work and then I was actually a little bit disappointed by kind of the sitting of machine learning uh back then because the whole thing was doing correspondence of of data. It is very good at doing a particular task but not very good at doing a general task. Um and uh so I went to Facebook as a data scientist. I went to Tensent as a data scientist. I was doing data analysis and doing some part of some machine learning. Actually uh the way I met Howie was end of uh or later uh 2020 22 right the down of charge before the launch of chibbt it was September

22 yeah 2022 and uh we we did an episode and uh back then I was uh seeking wisdom of you know what is the next big thing uh in tech industry from Holly and uh we were talking about is AR gonna be a big thing. Is crypto going to be a big thing? And how he basically said no, but generative AI is going to be the big thing. And this remember this was before the launch of CHPT. How he was uh experiencing GPT3 and he saw kind of the glimpse of the future. And of course neither of us actually predicted how big it is going to be. Uh maybe how it did. Uh but uh certainly I didn't. But uh from that uh conversation I started to pay attention to generative AI and uh at Tensent I was kind of in a unique position because part of my job is to explore new tech for the company leadership. So I actually spent a lot of time both during work and uh uh you know in my part-time reading paper and studying generative AI and after the launch of CHP I wrote an article about you know the five most important questions of chaptt I can share the article here but it basically predicted a lot of things that is going to happen in the generative AI field. So the article was over three years ago, two years ago, over two years ago and a lot of predictions still hold today. Uh so and after that I have been you know uh forming a community at at the beginning it was a WeChat community uh with the Chinese social media and uh I get to talk to a lot of uh investors and but mostly builders, practitioners and experts. On my channel, I interviewed with a lot of uh AI startup founders and uh the researchers. For example, Shucha uh was one of my guest and he was recently pushed by Meta and joined Meta MSL. Uh so anyway like through my own research, my own application and uh the community and the conversations I got a lot of understanding and I feel like every day I'm I feel painful that a lot of people say wrong things or when I look at the titles uh from social media or from news articles uh I I feel angry. I feel angry about how wrong they are. So I I I actually found my friend who is a 10x engineer and also the AI researcher uh to start a course and uh he will teach the technical side. I'll teach kind of the manset and the insight frontier research part and uh through that collaboration I also learned a lot uh uh by teaching by engaging with the community. So that's kind of my background. I don't think uh you know my credential naturally speak to uh you know why I'm qualified in teaching AI but uh I try to do my best in learning and uh in figure out what is the bottleneck of learning what is the gap in learning and try to help people bridge the gap. Yep.

Hey Eugene, there is a question in the chat room about is the target audience for like a computer science professionals or for like a you know it says regular people I'm assuming people without computer science background.

Yeah, that's a good question. Uh so when we designed the course uh our target audience was the I guess coders or software developer uh software developer tech roles or data an analyst or data scientist who have some basic coding skill at least but uh surprisingly a lot of nontech people for example our latest cohort uh if you go to our website actually I I'll type the website right uh and you can check the student review from the website and the latest uh review was from a sales manager from uh Splunk. So he didn't have a lot of coding experience but uh in this course I I feel like he was the one he was one of the students that got the most out of the course because like obviously he was the expert in many things such as go to market sales uh relationship uh but uh he always felt like uh it's hard for him to learn to start building and during the course he actually did uh like you I can probably like if you join our course we have the recording of uh the office hour of that uh cohort and you can see this sales manager actually did all the projects and ask a lot of uh thoughtful questions and he actually was able to overcome this uh I don't know fear of coding and he can get comfortable in cursor for example and start building and start seeing actually oh this works uh to him that was a big uh breakthrough so going that's a long answer to the question I guess the answer is uh no matter where you start uh by just getting comfortable in building with AI you will progress a lot and uh the growth and progression and uh how you compared to you know uh yourself two week from two weeks ago uh that is what matters it's not you know how you compare to uh like if you are not a coder uh you won't you wouldn't compare yourself to another proficient coder. So that salesperson what level like I know he he he must be able to do some coding or not not really at all. I think through my teaching uh I would summarize this as you can download Python and you can install Python and that's good enough and why why why this particular requirement it means uh you actually need to use compute and that is one of the key insights from our course the revolution of AI it's not about producing generating image or generating video uh that is useful but not the most useful the most useful the the real revolution the real paradigm shift is how we access compute and I think how you mentioned that very early on right you mentioned that beginning of 2023 um so before that before AI we mostly have two ways of using compute and if you look at all the tech companies it's mostly about how human interact with compute but before we can either use GUI or we can use programming GUI someone have to build the GUI for you like for example If I if I want to automate email, I can buy a service but someone needs to build the service or I can do programming but it's hard to learn. It's very hard for ordinary people to use compute in this way and uh but with AI uh we can use natural language to inter interact with compute or interface with compute and that gives ordinary people who people who do not know how to code kind of a superpower and uh they can solve their real life problem by building by building with natural language.

Uh so that is true but uh the unsaid part is this person needs to be highly motivated. I think as long as uh that person has the motivation then yes.

Oh please say hi to him I uh yeah I I need to thank him for uh building labor for us.

Yeah, so we have a couple more questions from the chat. So, someone was wondering uh do you think the current academia so sort of higher institutions are adapting to AI effectively? So, in the future, you know, to what extent is learning AI going to involve numerous years studying uh in university?

So, that so no, the answer is no. And uh I don't think going to university is the right way of learning. And that is our core proposition in the course. The course is this is a new paradigm. And what do you need? How how do you learn new paradigm? There is only one way that is to build because there's no textbook. There's no right answer. Like uh if you hear from the top uh experts in the field, right? Lequen would tell you something entirely different from uh uh other other uh you know big name professors. So uh how like which textbook would you trust and by the time a textbook is written it's outdated and uh if you want to apply the old uh way of learning one of the examples would be learning through frameworks it will be wrong and it will be uh inefficient. So the only way to learn effectively is to build and through building you actually get to learn you know what works for you or not. So that is kind of the first principle approach. Um and I actually have some concrete anecdote from university. So I'm a close friend with uh Jojo who is CS AI professor at Colombia. So she studied uh AI for over 20 years and we chat we had like two uh two podcast together and he came to my uh Seattle uh meetup when she visited Seattle. So she actually wanted to uh get me to teach at Colombia because uh as a adjunk professor because she believes that uh university is not offering a course like what we are offering right now and uh she feels like uh the education is actually lacking. So Uel is actually special because she's not only a researcher top researcher and a professor she actually uh does her own startup. She won the Alexa Amazon Alexa award competition in 2018 and uh she has her own agent AI uh startup uh called uh actually artlex a r a rk klx.ai AI. So like she she's very hands-on and then she can see the gap and I I think I also did a seminar to kind of a statistician conference quality control and productivity conference and uh I teach to mostly professors and postdocs and uh they were like wow I never heard of this before and this is so useful uh can can we uh invite you to teach in our university. So I don't think the curriculum is catching up. I don't think professors have the time or the motivation to actually get down to uh you know uh learning AI and teaching uh efficiently. I'm not saying like none of the useful courses exist. I do remember that uh there are a few Stanford classes that are very good. I actually wanted to go to the class but I didn't have time. I'm considering enrolling in those classes myself. But uh for most universities I think uh they don't have the resource or the people to teach AI properly.

So you mentioned uh that you had various obligations obviously as you developed your course and so we have someone wondering sort of how did you manage that and uh growing your channel as well as your uh advertising your course.

Yeah. So luckily not I mean I developed the course mostly during our parental my parental leave and my co-instructor uh happened to have his uh second child uh almost the same time as my first child. So we had overlap our parental leave and we spent a lot of time developing the course uh combined I think we wrote 300 pages of material from scratch like we are not copying and pasting from uh anywhere else we are writing like not using AI to write the material and there are over 300 pages uh so but uh yeah if I was just doing my full-time job I wouldn't be able to sit down and produce that much material once I have that material uh selling I feel like yeah if you spend more time setting the course you'll get more results but uh because the quality like my understanding is because of the quality of the course is so high it doesn't take too much effort to sell the course like I can just show people what the course is about and people like okay this is what what I'm looking for

So you mentioned like obviously clearly it took you a while to develop the course so it must consist of a vast array of of items and knowledge to acquire. So someone is wondering sort of still for the non-coder could you elaborate um as to how how they would leverage your course um you know say HR or sales right you mentioned downloading u um the Python package but could you elaborate?

Yes. Uh so like just speaking to a non tech person um we have our course is actually designed in a way that you don't have to consume the 300 pages. We have uh about five hours of video. Um the cohort version is of course we do the cohort live video and live demo and the self-paced version is the recorded version of that. But uh when we when we do the video my part is basically correcting what limits you from learning. Like for example uh when I like when you when you made the post right I reposted and I point to one thing that is the the importance of unlearn. uh what I observe is a lot of new grads, high school students, they're actually very effective in learning AI but a lot of experienced people even developers they're not effective in using AI and why is that is because the new paradigm actually bring some identity challenge like you cannot learn new paradigm using the old way so you have to unlearn the old way like for example you identify yourself as a non tech person. That is kind of the definition from the old paradigm. A high school student wouldn't self-identify as I'm a HR, I'm a sales, I'm a non-coder, right? So, and so the self high school student doesn't have this kind of uh self-imposed limitation to say, oh, I'm not going to be able to learn this. So, you need to unlearn a lot of things. And then I teach you how to learn effectively. And we teach you for example how to dissect the problem uh when should you use AI when should you not use AI and uh when you decide okay I should use AI for this to build this then we give you projects to try or you know how to do effectively effective context window management how to hand off knowledge between different models and how to uh take advantage of the existing APIs and existing ecosystem. So it's a progression like uh kind of learn as you build. Uh that is the way of teaching. So if you don't believe like like everyone learned coding uh gradually right nobody was born to know how to code. So if you if you believe that you can do you can build things then I mean of course you can because um if or if someone just look at our demo for example we have a half hour demo of how to use cursor to do complex data analysis or write a business uh weekly business uh review or build some small uh projects. Um we designed in a way that if you look at the task it will feel daunting but we'll tell you that you can do this within five minutes and once you actually do that uh most people get it done within 10 minutes. So that is kind of the magic of wind clicks that oh I can do this.