Encodes absolute positions — here, there, center, top, edge.
What It Does
Space.Position neurons activate on absolute spatial location references: 'here', 'there', 'at the top', 'in the center', 'on the left', 'at the corner'. They encode a specific spatial address rather than a relationship between two locations. They fire on both precise positional descriptions ('at coordinates 40°N, 74°W') and imprecise but definite references ('somewhere in the middle').
How It Behaves
Position neurons are evenly distributed across early and middle layers, reflecting that positional parsing begins early and is integrated with semantic context through the middle of the network. They co-activate with Identity.Place neurons (named locations have positions) and Space.Region neurons (positions are often within regions). Position neurons are particularly active in models used for navigation, spatial description, or technical documentation tasks.
Research Example
In Gemma 2B, Space.Position neurons fire on 'the nucleus is at the center of the atom' and 'the button is in the top-right corner of the screen' with similar firing signatures — both reference a position within a defined spatial domain. The neuron encodes positional reference independent of domain (atomic physics vs. user interface design).