ml blogpost series overview

jan 19 2026

This is a short series of blogposts intended to work up to modern language and image models from ‘scratch’.

I’m mostly writing this for myself- I tend to learn best when i’m forced to explain things. But others may find it useful.

chapters

• ml1- intro theory
  • theoretical learning background. skippable if you understand the motivation behind ‘optimize a loss function over a training set’ or just buy that that’s a sensible thing to do.
• ml2- optimization
  • basic optimization stuff, gradient descent
• ml3- architecture
  • universal approximation, why MLPs are a reasonable functional form
• ml4- loss functions
  • softmax, cross-entropy loss, perplexity