RL, World Models, Entrepreneurship and What's Next

<21.08.25>

This is a bit of a life update of where I’m at and what I’m thinking. This is more just for me to keep track of my thoughts than anything else.

Finishing the Masters

I’ve just finished my Masters in Artificial Intelligence at the University of Edinburgh. The last year has been bittersweet to say the least.

I came into the course with extremely high hopes. I thought that there was an impending AI takeover of web development and my only saviour was going to be learning the foundations of AI, so that I could be the one building it. I still hold the view that AI is going to enable non-technical folk to build web applications—this is already happening—but I don’t think it’s going to be as sudden as I once thought. Maybe 3–5 years, not 1–2.

I met a lot of cool, very smart, driven people over the last year, but unfortunately I was really displeased with the teaching. I’ve taken some wonderful, extremely well-structured online courses over the last few years (shout out to Andrew Ng) and these were all considerably better than anything that was taught at the university.

Lecturers have to put together their own material, and only one of my courses was recorded ahead of time. This led to really disjointed, poorly articulated lectures and many of the lecturers were just really bad at lecturing. They might be good academics, but they aren’t good at conveying information. I’m not sure how many of these people have ever been taught how to teach.

Anyway, I think it just made me realise that for certain subjects that do not require practical teaching—like medicine or mechanical engineering—there is no reason why they can’t be taught for an affordable price online.

I feel that I didn’t get my £17k worth (plus opportunity cost of not working for a year) from the course in terms of teaching. It is what it is and I have to move on. It made me realise the importance of self-reliance, agency and showing your work.

If you have access to a computer then you can be a software engineer or give yourself a computer science degree with free or very affordable online resources without ever having to step foot in a university.

Build, show your work, make a blog, put videos on YouTube, post on LinkedIn, share your progress. These things are what really matter.

Doing this you might not get a job straight into some big corporation with strict hiring criteria, but startups and smaller companies LOVE high-agency people, who can self-learn. You’ll have no trouble getting work. This is kind of how I got my last job and it’s exactly how I aim to get my next.

Entrepreneurship & Web Development

I suppose I already am one. I have side projects, one of which—called Langchats—makes me a small amount of money each month, so I guess I’m a kind of side-preneur.

However, I know that at some point in the future I’d like to go all in with a few cool people and build something great together. I’m not sure if now is the right time. I don’t feel financially stable enough, nor do I feel like I have sufficient direction on what I want to create. I don’t want to necessarily be an entrepreneur of opportunity, doing it purely for the money.

Ideally, I want to have some deep fascination or interest in the space I am working in, with strong technical knowledge giving me an edge. I don’t feel like I have that yet.

I’m at the point with full-stack development where I feel I can take a swing at building almost any traditional full-stack application as I understand enough pieces of the puzzle. But, as I’ve mentioned many times, I don’t see this being a viable career path in the near future.

This does not mean I am giving up on web development—far from it (more on this later).

Entrepreneurship vs Employment

My views on entrepreneurship and being an employee have changed too. I constantly have to ask myself the question of what I want from my life.

I like to be in control, to have full autonomy over my actions and my time. I don’t like being told what to do. I get bored easily with things that I don’t find interesting or challenging.

Other important questions:

  • How much risk are you willing to take?
  • Are you willing to work 24/7 for an unknown period of time?
  • Are you ok with potentially coming out with nothing after X number of years? (The reality for most entrepreneurs).

I think over the years I have realised there are MANY variables to the entrepreneurship equation.

I’ve always had trouble with the idea that someone is potentially profiting from my labour. Why am I not profiting from my labour? Why do I not get to keep it all? It’s a hard pill to swallow, but you aren’t taking any risk if you’re doing a traditional 9–5, working 40 hours a week.

The founder of the company is on 24/7—there isn’t a moment they aren’t thinking about the business. It is all-consuming. It eats up every single minute of your day.

  • Say goodbye to close friendships.
  • Say goodbye to strong relationships.
  • Say goodbye to your health and mental wellbeing.

You can't have it all if you’re the boss— you have to make sacrifices.

You might even be smarter than your boss. Hell, I could do their job! But could you? Really? Do you want to be up until 2 a.m. fixing some broken part of your application so your customers don’t get angry? Do you want to deal with the person who never comes into work and is performing poorly?

As an employee you do your 40 hours a week and have a balanced life.

Realisations

I’ve finally come to the realisation of just how difficult it is to be a founder/CEO of a company, and what it takes to create something.

  • Many are driven by insecurities and fear.
  • Many are driven by the pursuit of wealth, status and power.
  • Some our driven by pure curiosity and desire for positive impact.

I don’t want to play games of status and power, but I do want wealth and to make the world a better place. I’ve realised that there are other forms of creating wealth without taking on huge amounts of risk and stress, but you do have to sacrifice more autonomy and time.

For example: if you can become a highly paid tech worker in the West you can do incredibly well, working a relatively low-stress job, doing interesting technical work, and if you aggressively save you can build wealth moderately fast.

I would bet that the expected return (in terms of wealth generation) of someone taking this path is probably higher than someone taking the startup path.

Some people choose the startup path for the thrill. They thrive on the autonomy and the unknowns. For these people it’s not really about the money—they just want to build something really cool and work on it every day. I have utmost respect for these people.

High-Value Skills

I’ve also come to realise the importance of becoming highly skilled at something valuable, useful and ideally helpful to others.

A motto I’ve come to adopt is: Do what contributes: master a high-value skill that helps others.

If you can master a high-value skill then you will have no problems in terms of career or finances.

Additionally, I’ve come to realise most of the things I enjoy doing don’t really cost that much and therefore I shouldn’t really have to worry about money.

The only financial thing that I have to worry about is being able to afford somewhere central in a small city.

What’s Next?

I have a number of things that are interesting me at the moment:

  • Reinforcement learning
  • World models
  • A new wave of web technologies that bring things like video editors and 3D/4D renderers to the browser

Web Development Directions

I think the web is still going to be a medium for information exchange for the next 5–10 years, before we start to see the introduction of XR/VR technology.

However, to stay in web development I want to work on more technically challenging problems, especially things that AI will struggle to produce.

I’m becoming increasingly interested in things like PlayCanvas and SparkJS that are building graphics engines for the web. Things that would have traditionally been made using desktop applications are now being made in web apps operating right in the browser.

Examples:

  • Spline.design

     : web-based platform for collaborative and real-time 3D designs and interactive experiences

  • SparkJS

     : creating an advanced 3D Gaussian splat renderer for Three.js

  • PlayCanvas

     : web-first game and graphics engine

Advances in web technology, browser technology and computing power (particularly GPUs) mean that these things are now possible. I think there is a lot of opportunity here.

WebAssembly and Low-Level Languages

I’ve also come to find out that you can take Rust, C, C++ and other low-level languages and compile them to WebAssembly so that they can be used in combination with JavaScript code.

This means compute-intensive mathematical operations (like creating millions of Gaussian splats in SparkJS) can be done in a low-level language like Rust, compiled to JavaScript, and then you can make a package compatible with Three.js. Amazing.

Immediate Plans

I have to find a job in the next 2–3 months before my money runs out.

I’m still very interested in world models and spatial intelligence, but I have to be realistic and pragmatic: I know I’m not getting an applied AI/engineering job at some AI lab anytime soon.

Therefore, the web technology side of rendering and interacting with 3D/4D worlds is something I could definitely work on and contribute to in the next 3 months.

It’ll be a grind, but at least I know that this is possible and it’s still related to world models.

I’m also interested in building web-based simulation environments. We are going to see a new wave of AI being trained in these 4D worlds and environments generated by Gaussian splats. Someone has to build the infrastructure for this. This could be something super interesting to work on.

Reinforcement Learning

Another area of interest of mine is reinforcement learning.

I’m convinced the era of experience is upon us after listening to David Silver on the DeepMind podcast and reading his paper on the topic. RL is going to be incredibly important for training robotics systems, and it was a module at university that I found quite interesting.

I’ve also recently been following the work of Joseph Suarez of Puffer.ai. He, along with his team, are doing really cool work on creating extremely fast RL training environments (low-level C work).

I don’t have the skills right now, but I’d love to take a stab at creating Entropy (Hyle 7), a board game that Sir Demis Hassabis is the five-time world champion at, using a Puffer.ai environment, and then seeing where that takes me.

Narrowing Focus

Unfortunately, I don’t have the time to do all of this before I need to get a job. Therefore, I think that the advanced web technologies path is the most promising and the most likely to land me a job I will enjoy.

Therefore, I’m going to spend the next few days exploring and researching, before embarking on some new projects and/or contributing to open source, along with documenting my journey.

I want to display my progress and learnings as I go so I can prove—unequivocally—to a potential employer that I can build.

I’m going to keep Entropy as an interesting side project post-job.

AI

AI (Claude/ChatGPT) from now on is to only act as a teacher. It is never to autocomplete or write my code.

Some Quick To-Dos

  1. Learn about Gaussian splats and their rendering
  2. Learn how to do simple simulations in either C or Rust and then how to compile them to WASM for the web
  3. Think about cool 2–4 week projects showcasing skills in the above
  4. Look into PlayCanvas / SparkJS / other open-source projects to see if I can contribute
  5. Show all work and progress