Definition
Plain language
A specialist AI model built to work on data shaped like a network of connected things, like papers linked by citations.
As stated in the literature
GNN — a neural architecture that computes node representations by aggregating information from a node's graph neighbors; used in this corpus as a frozen tool that an LLM agent defers to for node classification.
Also called: GNN, GNNs
Why it matters: It lets models make predictions about connected data by using the relationships between items, not just each item on its own.
For example, to guess a research paper's topic, it looks at the topics of the papers that cite it and that it cites.
Heard on the show
“The tool they hand the agent is a graph neural network.”Episode 144 — When an AI Agent Just Copies Its Tool — And Bigger Models Copy More