eh2022-circuit-defined-sufficiency-necessity

IN premise — summaries/2026/08/24/elhage-2022-toy-models-superposition-chunk-1.md

Created 2026-08-25T02:57:58+00:00

A transformer circuit is defined as a minimal set of model components that is both sufficient for the task and minimal (no proper subset reproduces the behavior), not merely correlated with it

Summary

When we say we've found the circuit responsible for a behavior in a transformer, we mean those specific components are both enough on their own to produce the behavior and that no smaller group within them could do the same job. This matters because it sets a real causal bar: it's not enough to notice that certain parts light up during a task, we have to show they are the actual mechanism with no redundant or merely alongside components sneaking in.