Me
Thirty Million by Thirty
I turn 30 next September, and I’ve set myself a goal.
Build and train a 30-million-parameter language model from scratch before my birthday. Not fine-tune an existing model. Actually build one, initialise its weights randomly, train it, and see what it can do.
The plan is to focus entirely on TypeScript code generation. Right now, I’m still working through the fundamentals of machine learning. Gradients, loss functions, optimisation, and the maths that makes it all work. So 30 million parameters feels like a pretty ambitious leap.
I’m not expecting to compete with frontier models. I want to understand how these systems work, build something real, and hopefully release something useful as open source.
30M is the minimum target. If I can push towards 100M or beyond, even better. I’ll be documenting the journey, including everything that goes wrong. Let’s see how far I get.