Causal circuit

A large model contains thousands of elements, but a smaller network of dependencies may lead to a particular answer. You want to check whether retaining only the relevant computational path can still reproduce the behavior under study.

A causal circuit is an arrangement of model elements and connections causally related to a particular result. “Circuit” is a metaphor for a computational circuit. Its elements may mix information between tokens or transform lists of numbers describing text.

For “John gave Mary a book”, researchers may seek the path responsible for choosing the recipient. Changing computations outside the selected arrangement tests whether the answer persists. The description's faithfulness is assessed against the full model and a specified intervention method.

Circuit Condensation, §2–3, examines concentrating behavior in smaller circuits. A small circuit need not be a simple list of independent parts: elements can interact. Nor does it prove the rest of the network is useless — it may be needed for other behaviors and inputs.