Symbolic neurons not fitting a specific sub-type.
What It Does
Symbol.General neurons form the largest single sub-type in our entire corpus. They encode the general symbolic character of text — the fact that language is a system of arbitrary signs with conventional meanings — without specializing in any particular kind of symbol. They activate across all symbolic contexts as a baseline signal of 'this is language, proceeding token by token.'
How It Behaves
Symbol.General neurons are distributed across all layers and all models with remarkable consistency. Their proportion does vary significantly across architectures (25–55 percent of total features), tracking training data composition. They represent the residual symbolic processing capacity after all specialized Symbol sub-types have been identified — the general-purpose symbol processing substrate of the language model.
Research Example
In cross-model analysis, Symbol.General neurons are the single most consistent predictor of 'this token is part of a natural language passage' vs. 'this token is part of structured data.' When a model is generating a response, Symbol.General neurons activate throughout, providing the symbolic scaffolding on which semantic content is hung. Suppressing these neurons produces responses that are semantically coherent but syntactically disorganized.