Temporal neurons not fitting a specific sub-type.
What It Does
Time.General neurons encode temporal meaning that blends or transcends the six specific sub-types. They activate on general temporal language: tense markers, temporal adverbs ('now', 'soon', 'recently', 'eventually'), aspect constructions (completed vs. ongoing), and temporal relationships that mix date, duration, sequence, and frequency in ways that resist clean sub-classification.
How It Behaves
Time.General neurons are the second-largest element in our corpus by count (behind Era) and are evenly distributed across all layers, reflecting the pervasiveness of temporal language. They show the most consistent cross-model proportions of any Time sub-type, appearing in reliable proportions across all 11 architectures. They represent the general temporal processing substrate that underlies all more specific Time sub-types.
Research Example
In cross-model analysis, Time.General neurons show extremely stable proportions across all 11 models tested, in contrast to Era neurons (which vary with historical training data) and Date neurons (which vary with news content). This stability suggests Time.General neurons encode something architecturally fundamental about how temporal language is processed in transformers — a universal temporal grammar rather than domain-specific temporal knowledge.