Production scale memory writes store precise facts but not reasoning steps
Atoms, Not Knowledge: What Production-Scale Addressed-Memory Writes Deposit
Computation and Language
Summary
This paper studies how memory writes in large language models hold information and whether they can combine pieces to reason. The authors test very large models and find that stored facts can be retrieved exactly, but combining those facts to form new conclusions does not happen naturally. They show that memory writes behave like separate puzzle pieces, not like steps in a reasoning process. This suggests that AI systems need explicit procedures to put pieces together rather than expecting memory by itself to create complex answers.
What this means in practice
- •For machine learning engineers: Optimize language model memory modules by measuring exact fact recall and designing explicit reasoning chains for compositional tasks.
- •For software architects: Design AI systems that use memory write coverage metrics to evaluate knowledge storage and require system-level orchestration for reasoning steps.
Authors
Wei Gao, Cheng Tian, Duan Yang, Jianqing Yin
Abstract
Deterministic hash-addressed conditional memory permits local post-hoc knowledge injection. Extending arXiv:2610.21553's bounded-feasibility audit rejecting under-specified generality claims with repaired probes and training-dose/capacity axes, we give, to our knowledge, the first production-scale write-content verdict. We test 124M-7B self-trained/publicly pretrained beds, Qwen3.8-Flash-Next-FP8 (176.95B; 320,001,536 FP8 rows), and DeepSeek-V4.1-Flash. Write coverage requires a sufficiently trained backbone prior, full-width readout, trainable gates; undertrained-bed intervention confirms a threshold, not a slope. At fixed budget, coverage falls from 98% (50 facts) to 12.7% (1,000); table size is not binding. Two-hop composition does not exceed the independent single-hop baseline (26% observed vs 25% expected; 15% lexical-distortion floor). Synergy remains absent across dose/load in two independent bed families. A 28-percentage-point in-context gap (given vs recalled intermediates) localizes failure to retrieval-for-reasoning, not gate access. Writes are independently retrievable, precisely addressable, reversible atoms, not composable into reasoning. A layered protocol, the first, to our knowledge, to cover compositionality in behavioral memory evaluation, separates baseline, synergy and distortion floor. Any addressed/key-value memory must report coverage, synergy ratio and distortion rate before deployment claims. Composition must be system-orchestrated; multi-constraint queries should explicitly chain written atoms. In-context controls show given intermediate entities are exploited at twice the implicit rate. Production load fits exponential interference ($R^2=0.98$); its mechanism, between budget dilution and write interference, remains open. All numbers are recomputed from archived raw logs/JSON; a fresh 52-target production rerun gives 52/52, exactly matching archived baselines.