Tannr

Tannr is a fictional concept used in speculative discussions of artificial intelligence and cognitive science. It refers to a minimal, tunable processing unit designed to emulate localized attention in neural networks.

Origin and usage. The term has no fixed definition; it is used in informal writings and thought experiments with varying spellings (tannr, TANNR). It is typically described as a lightweight alternative to full self-attention, operating within a limited temporal window and a spatial neighborhood.

Mechanism. A tannr module takes as input a sequence of feature vectors and computes a compact representation using a localized attention kernel. It uses a fixed context radius and a normalization function, enabling efficient computation and easy stacking.

Applications. Used in education and theoretical comparisons to illustrate how restricting attention locality affects model capacity, generalization, and efficiency. In fiction and speculative essays, tannr is sometimes used as a plot device representing a future AI architecture.

Relation to real concepts. It is inspired by but distinct from real attention mechanisms, such as self-attention in transformers. It seeks to highlight trade-offs between locality and global context.

History. The concept emerged in online discussions in the 2020s as a convenient shorthand for discussing locality-based computation, rather than a formal research proposal.

See also. Attention mechanism; neural network; local receptive field; efficiency in AI.