Technical blog
Long-form articles on architectures, design decisions, and what we learn from measuring.
What a neural network actually is
Not a brain and not magic: a function built by composition, parameters tuned by measuring the error, and a rule for assigning blame backwards.
What a tensor actually is
Beyond "an N-dimensional array": how it is laid out in memory, why axis order decides throughput, and what changes when you move it to a GPU.
From transistor to instruction
What sits between a MOSFET switching and a line of code running: logic gates, the datapath, and why clock speeds stopped climbing in 2005.