Research project / ARMS

Can stored attention become reusable memory?

ARMS explores storing and retrieving model states through spatial organization. Its projection-based ARM prototype is an early experiment; the proposed native-state architecture has separate validation requirements.

Read the research Explore the code

The proposed spatial memory architecture. The paper distinguishes it from the evaluated prototype.

The public record

ARMS: A Proposed Spatial Architecture for Persistent AI Memory

Andrew Young · January 10, 2026 · Preprint

Scope: Proposed ARMS architecture and a separate learned-projection ARM prototype on 50-400-token contexts. The public ARMS repository describes small synthetic GPT-2 tests with fewer than 100 stored states.
Current limitation: The reported 5,372x ratio lacks a byte-level denominator, payload/weight accounting and a reconstruction criterion. It is not established as lossless compression.
Available artifacts: Public ARMS source repository linked. This record does not identify the exact ARM prototype run or a reproduction package for the compression claim.

Read methods and reported results ↗

Editorial correction dated 2026-09-22; no new experiment or independent reproduction. Reported implementation and verification gaps remain visible.