-
What goes on inside an LLM
ISSUE 22.23 • 2025-06-09 AI
By Michael A. Covington
Researchers have figured out how to probe what goes on inside a large language model (LLM). There are some surprises.
LLMs are neural networks, represented as huge matrices of numbers, and the meaning of each individual number is nothing but its effect on the whole. Accordingly, it has long been impossible to assign a meaning to a neuron or any of its weights by itself. But now there’s a way.
Read the full story in our Plus Newsletter (22.23.0, 2025-06-09).
This story also appears in our public Newsletter.