av.codes blog

A single agent run generates roughly 40 decision points an hour.

Today, when you prompt a model, you write a paragraph of hedging: do this, not that, in this format, and please don't invent the numbers.

Dating the layers Most of what we call an agentic harness is a fossil.

I woke up at 2 a.m. because the fans were screaming.

You look at the Great Pyramid and the first thing that hits you is the physicality of their math. 2.3 million blocks, each weighing about 2.5 tons on average, quarried and moved into position over roughly 20 years.

Georgi Gerganov created llama.cpp on March 10, 2023, ten days after Meta released LLaMA, and showed that a 7-billion-parameter language model could run inference on a MacBook CPU without a GPU.