All terms
Safety & Alignment
Circuit
A small group of parts inside a model that work together to carry out one identifiable step.
Definition
In interpretability research, a circuit is a small set of connected pieces inside a trained model — particular features, attention heads, or groups of neurons — that together perform one recognizable job, such as tracking which name was mentioned earlier or deciding whether a sentence is a question. The name borrows from electronics: rather than treating the network as one undifferentiated blob, researchers look for repeatable wiring that explains a specific behavior. Finding circuits is the main goal of mechanistic interpretability, because a behavior you can point to inside the model is one you can test, predict, and potentially fix.