Techno Optimist
Into The Unknown
From the infinite potential of energy to the total actualization of entropy, intelligence charts a course for the pursuit of meaning, mission and love.
I believe that all problems are eventually solvable through technology without introducing any additional externality. Technology evolves—gradually or rapidly—to match the complexity of problems. Today's flawed solutions will never capture how easily these problems may be resolved in the future.
Technology emerges from the interactions between intelligent entities and the universe they inhabit, while intelligence improves and evolves through technology. Technology is not only an external manifestation of intelligence but it stands as the ultimate culmination.
Intelligence is the meta-problem underlying all challenges. Once intelligence is solved, we indirectly unlock solutions to every problem. The acceleration of ASI injects a second-order surge into all frontier sciences and engineering, placing us at an inflection point where the future will be unrecognizably better.
The exponential growth of compute power, digital data & artifacts, and nn-based AI signals a broad intelligence revolution in both scale and depth.
Technology has also been historically proven to be the only robust, consistent, and effective means to combat various social disparities, both directly and indirectly.
Acceleration follows an exponential curve while human imagination remains confined to linear velocity. A fundamental transformation may be daunting, yet it is equally if not more exhilarating, for the only way forward is into the unknown .
My Mission
Intelligence is the scaffold upon which we build the future of business and society. When intelligence can be artificially created and arbitrarily duplicated, this scaffold gains the capacity to extend beyond imagination. My role is to be a builder on this scaffold – crafting useful tools that fundamentally reconstruct how we think about tech business logic, product-market fit, operational excellence, and resource allocation. I want to reimagine these foundational elements in light of our new technological capabilities. I want to be a part of this profound intelligence revolution in how we create and capture value. The techno-capital engine is leading the charge, and I am an agent of change. The impact will not be immediate, but it will be impossible to ignore.
Source of Strength
At the moment the pace of AI innovation demands constant evolution of understanding. The source of my strength is continuous and constant learning and researching in cyclic feedback, from the latest AI research papers on arXiv to the extraordinary projects and tools on GitHub, from the immediate first-hand discussions and demos on 𝕏 to the deep dives in podcasts, blog posts, and books. I live at the intersection of AI research, product development, tech industry strategy, and VC/startups – fields that are mutually reinforcing. Yesterday's cutting-edge is today's baseline. The value of following these developments and understanding them deeply compounds over time, creating unmatched insights for decision-making and innovation.
I'm on
Learning Journey
Picked Up Programming Before the ChatGPT Moment—Mar. 2022
First stepping into a new field and learning tangible things is exciting and satisfying. This is especially true for programming. The basic realization became—huh, I can just build things.
I chose the book Python Crash Course as a no-brainer and I still remember the Alien Invasion game made with pygame and the Learning Log with Django (although I started a new app called Book Share using everything I'd already learned in Learning Log and pushed so much further than what I knew), as I'm certain many people who started Python with this book do.
Programming is only a subdomain of computer science, and a programming language itself can be tedious. So I have learned the basics and intermediate Python, the immediate question that hit me was—what do I use it for?
Deep Learning Caught My Attention—Aug. 2022
You remember that famous TV show Person of Interest? The iconic starting line was "You are being watched. … A machine that spies on you every hour of every day. … I know, because I built it". I was fascinated about that show, and it drove me down a rabbit hole of research into artificial intelligence and the state of the art (at the time of course). Interestingly the show depicted the machine as completely hard-coded by Harold while real AI systems are neural net based models. I didn't know this distinction until later .
It was around the same time that a thing called "GPT" came out of nowhere (GPT-3 and GPT-3.5 small scale beta testing had already made its name) and started to create some real buzz.
I knew absolutely nothing about AI, machine learning, and deep learning at the time. Zero. From an initial toe dipping, I realized learning AI required a tremendous amount of understanding in math and quite sophisticated coding skills. I wasn't sure I was up for it, but I decided to give it a try. That decision has led me such a long way here in retrospect.
You Can Just Learn Things—Jan. 2023
I was starting to get a rhythm on diving into something entirely new. So again I started with the basics—neural networks. I soon realized it was not as difficult as I had imagined it to be to at least get started on, because I had already had foundations on calculus and statistics, and "that thing" called linear algebra which I had painfully learned as a mandatory class in undergrad without knowing what the heck it was for, suddenly became unbelievably important. I quickly absorbed concepts like backpropagation and gradient descent, and fundamentals on MLP, CNN, RNN, and RL (didn't pay Attention to the Transformer—pun intended—until later sadly).
Soon I moved to building practical and fun small projects following tutorials using, you guessed it, Keras. I built and trained a super simple chatbot using just MLP in one-hot encoding with a lengthy pre-written dialogue json (it looks utterly laughable now compared to transformer based LLMs but I had so much fun building it). I also built and trained a dog/cat image detection model with CNN among other things, as it is sorta like the rite of passage for all deep learning beginners .
But finally something big caught my attention, and it's called Let's build GPT: from scratch, in code, spelled out from the legend Andrej Karpathy! The first 10 mins in I quit, because I hadn't even learned PyTorch yet . I quickly did a PyTorch crash course, and I realized that I actually needed to learn to manipulate tensors in really granular ways instead of just stacking prebuilt layers like in Keras. After the quick PyTorch course, this time I didn't directly jump into building GPT, but started from the first video tutorial in Karpathy's channel and started intensively absorbing pure knowledge and brilliance, until finally I got to the super fun bonus part.
I followed along every step in his building GPT video, doing side research whenever I saw something I didn't understand, including reading Attention Is All You Need paper dozens of times repeatedly, with each time getting just a bit more clarity. And finally, model was built and trained, I was able to see the tokens being generated one by one, and that was a magical moment.
Later I reviewed everything I learned from the mini-GPT dev and I pre-trained a GPT-like (100M Params) LLM from scratch using a portion of the Common Crawl dataset, and further finetuned it with some instruction datasets I could find at the time. The model achieved above 20 on Hellaswag.
Blender & Unity, A Change of Pace—Apr. 2023
Today I don't even remember how it started, but somehow I decided to learn Blender. I tend to think it is because 3D modeling and rendering is another epitome for the phrase "you can just build things". And the exhilaration you get from creating something beautiful out of thin air is beyond words.
Building 3D models can be extremely time-consuming, EXTREMELY! But I enjoyed every minute of it. Seeing what you are building becoming just a bit more perfect is satisfying to say the least. During my learning in a few months, I made some beautiful rendered scenes and animations into final videos, and I would lay down sound tracks for them to make them even more cinematic. Considering I did all these on my old gaming laptop with an outdated GPU (even at the time), I am really happy I was able to make these! I selected some of them which I really like and threaded them into a video cut.
I also started to learn some game dev with Unity. I made a few simple but interesting games (I put two of them—SkyPool & Turbo Drift—on Google Drive with a public link. They are built so you can download it and play directly, although only built for MacOS: Link Here).
AutoDrive—3D Modeling and Unity, Meet ML—Jun. 2023
As I was having fun making game projects, I came across a library developed by the Unity team called ML-Agents. This framework allows you to develop and train robotic agents with RL in a virtual world inside the Unity game engine, sorta like a simpler version of Nvidia's Isaac Sim. This became a perfect crossover project between Blender, Unity, and ML for me.
I used Blender to develop a 3D car model, exported it to Unity, and built a convoluted tunnel scene with the car being the agent, navigating out of the tunnel via RL-learned self driving. After weeks of painstaking adjustments on reward functions (in C#) and hyperparameters, and countless training failures, finally I successfully trained an agent capable of steering the car out of the tunnel autonomously.
Different Country, MSc Business, & Certificate Phase—Sep. 2023
Going to a different country for a Masters study is a big change. I was excited to learn about business, and most importantly, to learn life in a different speed and spectrum.
In the meantime, I went into a little bit of a tech certificate phase. Did a full Harvard CS50 class, cloud certificates from Azure and AWS, as well as some advance AI engineering classes.
But eventually, I realized that the best way to learn is after all project building. When you build, you're forced to think about things from first principles, and you're forced to go through every single technical detail instead of just going through them in your mind, which makes you super sharp in every part of the tech stack.
Recreational App Dev, JS, & Frontier AI Chase—Jan. 2024 onward
Making apps that solve a little but practical problem is recreational and good for the soul.
You have a vacation idea, I turn it into a travel plan. I wrote this project in Python only, leveraging OpenAI GPT-4o-mini (a big deal at the time if you remember) API as the underlying LLM for validating and generating travel plans, as well as parsing locational information. Then it uses Google Maps API to find locations and plan optimal routes. Lastly, it generates an interactive map to display planned stops and routes in the travel plan.
Transcrilate was the first sorta serious node.js project in React. It can transcribe English speech and translate it into 19 most common languages. The transcription and translation are separately handled by open models via HuggingFace transformers, both running inference locally.
Gradually I have gotten into the habit of reading frontier AI research papers. Every morning one of the first things I do is check the Hugging Face Daily Papers section for interesting papers, and of course not all of them would make it to HF, and so 𝕏 is my paper hunting safety net. Reading raw AI papers forces me to constantly replenish my knowledge pool. Luckily there are tools like Claude and NotebookLM that make this process easier for me.
Now I'm buried by countless AI papers on arXiv and bombarded with AI/Tech development news on 𝕏 on a daily basis. Although I never stopped making more useful apps both for practical use, keeping skills sharp, and for some fun, only I've been taking them even further and exploring actually deploying the app, CI/CD, taking in user feedback for continuous improvements, as well as getting more users.
I never had a CS degree (except if you count Harvard CS50), and I am not a professional engineer. But what you can never take away from me is my starving hunger for learning. I love this Naval tweet, I'd like to think I'm a smart person, and there is immense joy for me in self learning interesting and useful new things. I will never stop learning.
It has been a long journey. I am super excited about what's to come !
My Apps
Job Guru
Ingenious job landing craft.
Job Guru is your go-to tool for creating outstanding CVs, resumes, and cover letters that make an impact. Simply input details like company information, job descriptions, specific requests, and your personal profile, and the app will customize each document to suit your needs and the role you're aiming for.
What makes Job Guru truly unique is its innovative keyword match analysis. This feature meticulously compares your written document to the original job description, breaking it down sentence by sentence. It provides clear insights to ensure your application emphasizes the skills, experiences, and terms that employers care about most.
With Job Guru, you're not just writing—you're crafting a tailored, strategic application that boosts your chances of landing the job.
LearnTube
Video to insights in seconds.
Unlock the full potential of YouTube videos with LearnTube. Provide a YouTube video link, and LearnTube delivers a tailored report on the video's content, customized to your preferred length and depth. Whether you want a concise overview or a detailed breakdown, LearnTube adapts to your needs.
Interact seamlessly with AI: chat to ask questions about the video or select any sentence from the report for deeper elaboration. Save and share the insights by exporting reports to PDF, and keep your projects organized in the Collection tab for easy access anytime.
Tailored for YouTube—especially long-form interviews and podcasts—LearnTube is your dedicated companion, saving time with quick TL;DW (Too Long; Didn't Watch) summaries or offering in-depth explorations, all at your fingertips.
ImgSherlock
MLLM vs. MLLM.
After I saw a post about how real-looking an image of a receipt generated by GPT-4o native image gen was, and how all traditional ML AI-image detection had failed at that image, I thought there is a more capable way to do AI image detection.
You see, SoTA AI image generation has become so advanced that depending solely on visual analysis is not adequate enough anymore, one must also analyze the plausibility of the image, including the internal consistency of the information, elements, and subjects in the image. This can be done with Multimodal reasoning LLMs.
ImgSherlock is a web app that allows you to upload an image, identify if it's likely AI-generated, and get a detailed analysis.
I'm sure a new idea will hit me soon...
My Inspirations
Elon Musk
I would pick flawed authenticity over faked perfection any day.
To borrow the ending of the Elon Musk biography by Walter Isaacson, "They can be reckless, cringeworthy, sometimes even toxic. They can also be crazy. Crazy enough to think they can change the world."
Elon is pure, an extremely rare trait these days, especially in public figures. He always says what he means and means what he says. This is why he is always surrounded by controversies. But even among these controversies, you should never ignore the fact that he has advanced more frontier exploration of modern time than any single individual in history. In fact, this is my litmus test for people–you can disagree with his opinions all you want, but if you would write off all his achievements simply because you don't like some of his opinions, I know we are not the same people.
His brilliance, hard work, determination, execution, honesty, and good heart are among some of the things I constantly look up to.
Jensen Huang
The only limitation is your imagination and execution.
From a dishwasher to Founder/CEO of the biggest and most influential company in the world today, how can you not find Jensen Huang inspiring?
He always goes to first principles when faced with extraordinarily complex problems. This is what makes him such a visionary and problem solver. He works constantly, learns constantly, and inspires people constantly. He perceives a goal, sets a course, and pursues it persistently on the order of decades. Nvidia's success is not a fluke, it is a group of brilliant people guided by a wise and persistent leader to position around the apple tree so they would be the closest to pick them up when the apples fall.
Jensen tells the story with his own life that you can just do things.
Naval Ravikant
Imagine the wisest philosopher you've read, now look at Naval's tweets.
My life was not the same after I was introduced with The Almanack of Naval Ravikant. Life is a willful journey. If you go through your life just doing things without giving them much thought, you are just a group of molecules drifting where the universe takes you.
Of all the wisdom from Naval on wealth, health, and happiness, what I get from him the most is self reflection and the urge to listen to my heart. The key is to be unapologetically proactive on who you really are.
Andrej Karpathy
You don't just get called "AI mentor for an entire generation", you have to have earned it.
Andrej Karpathy's tutorials on neural networks and Transformer are one of the major reasons why I've become so obsessed with AI, and certainly the biggest reason why I was able to advance from surface impressions to deeper technical understanding as an absolute beginner. I also know for a fact that this goes for most new AI enthusiasts like me in the past few years. And before that, his class CS231n at Stanford was a source of inspiration and knowledge for many computer vision talents.
I am glad Andrej decided to go all in on democratizing AI education after his journey at OpenAI and Tesla, which God knows people now and in the future desperately need. He is a phenomenal educator and just one of the most brilliant minds out there.
Pieter Levels
Digital nomad is a statement of freedom, honesty, and getting shit done.
Funny enough, before Levels went on the Lex Pod, I'd never heard of him or the concept of digital nomad, now I find myself often contemplating the beauty of this way of life.
He is a rebel against the status quo. He is a rebel against institutionalization and bureaucracy. And he is also a rebel against the confinement of creativity and productivity. Everything you need to build today can fit into a backpack. Your ideas have infinite potentials, you have infinite potentials.
Pieter Levels represents a generation of indie hackers who feel disconnected to the world and rediscover peace, meaning, and success in a completely different spectrum of life, simply from building and sharing.
There is a profound wisdom from him and his philosophy that I find extremely enchanting. Your life doesn't have to be just another fulfilment of the standard template. If you want something else, make it happen.
I'm grateful that they exist. Our timeline is infinitely better with them in it. And whenever life gets difficult, which is often, I receive visceral strength and courage watching their brilliance and hard work in action!