Induction Head

An attention circuit that completes prompt patterns by copying what followed a token's previous occurrence, identified as a core mechanism behind in-context learning.

grounded in: Trend theme 'Interpretability and causal reasoning about LLMs' — appetite for rigorous, mechanistic explanations of what models do inside; established interpretability finding.

Connected concepts

Circuit tracing, Superposition, Mechanistic interpretability, In-Context Learning