A double helix
Learning Blog
One switch after softmax
Every trick is a trade
Shape, not the KV cache
Reading the signals
Where the time really goes
Experts are computation paths
Same output, much faster
How LLM sampling really works
From regression to GPT
Q,K,V, with code snippets
Inner Mind
find your place
Navigating Change and Connections Along the Journey
fear and courage
about words and soul
self-forgiveness and self-acceptance