Building systems that learn. Writing about the math behind it.https://hyerra.fyi/https://hyerra.fyi/writing/sft-ppo-dpo-preference-learning/https://hyerra.fyi/writing/sft-ppo-dpo-preference-learning/Large Language Models are trained on a large body of text on the internet. They aren't designed to be helpful assistants. This guide explains how to teach a language model to be helpful.Sun, 05 Jul 2026 00:00:00 GMThttps://hyerra.fyi/writing/self-attention/https://hyerra.fyi/writing/self-attention/Attention has many distinct advantages over its RNN predecessor. This article focuses on the intuition behind attention, so you can understand why it's so powerful and widely used.Sun, 21 Jun 2026 00:00:00 GMThttps://hyerra.fyi/writing/l1-l2-regularization-geometry/https://hyerra.fyi/writing/l1-l2-regularization-geometry/L1 prefers sparse weights while L2 prefers small weights. We'll explore why this is, and how circles and squares help answer this question.Sat, 20 Jun 2026 00:00:00 GMThttps://hyerra.fyi/writing/pca-lagrangian/https://hyerra.fyi/writing/pca-lagrangian/Eigenvectors of the covariance matrix aren't a coincidence. They fall out of maximizing variance under a unit-norm constraint.Fri, 08 May 2026 00:00:00 GMThttps://hyerra.fyi/writing/backpropagation/https://hyerra.fyi/writing/backpropagation/Hand-compute every gradient in a neural network. By the end, you'll know why we perform backpropagation to train a neural network.Mon, 04 May 2026 00:00:00 GMThttps://hyerra.fyi/writing/eigendecomposition-svd-pca/https://hyerra.fyi/writing/eigendecomposition-svd-pca/We'll discuss what an eigenvector is and then relate it to three common forms of matrix decomposition, showing how each form builds upon the previous.Wed, 08 Apr 2026 00:00:00 GMThttps://hyerra.fyi/writing/hello-world/https://hyerra.fyi/writing/hello-world/First post! Why I'm starting this blog, and what to expect.Tue, 17 Mar 2026 00:00:00 GMT