three-tier-memory-architecture

OUT derived (depth 7)

Created 2026-08-25T03:11:43+00:00 · Reviewed 2026-08-25T03:24:16+00:00

The LLM implements a three-tier memory architecture: broad superposition-based read, narrow rank-one parametric write confined to the covariance-whitened addressable subspace, and unbounded contextual write via the context window—with the geometric boundary of C⁻¹k* defining the precise demarcation between parametric and contextual regimes.

Justifications

SL — Combines the depth-6 unbounded complement, the depth-5 superposition consequence, and the depth-5 subspace boundary into a single architectural statement where each tier is logically necessary for the complete picture; retracting any one collapses the three-way description.

Antecedents (all must be IN):

  • OUT context-write-unbounded-complement — The context window functions as the unbounded write channel that complements the rank-one parametric write: where parametric editing is confined to a single MLP's addressable subspace, the context channel injects arbitrary quantities of verified knowledge, making the two write channels capacity-complementary rather than redundant.
  • OUT read-write-asymmetry-superposition-consequence — The read-broad/write-narrow asymmetry is the direct operational consequence of over-complete superposition: reading is a simultaneous linear projection over all active features, but writing must be rank-one to avoid cross-talk in the over-complete basis, making the write channel inherently and necessarily narrower.
  • OUT parametric-write-subspace-boundary — The operational boundary between parametric recall and contextual retrieval is precisely the geometric boundary of the rank-one addressable subspace: facts whose subject-key projection aligns with the locally-stored key covariance are parametrically editable, while facts outside this subspace must be externally supplied.

Dependents

These beliefs depend on this one: