How Sapience implements Complementary Learning Systems

Sapience Labs · August 2026 · a technical snapshot for collaborators · full paper: doi.org/10.5281/zenodo.21894568 (concept DOI; it resolves to the current version, v4.3, 10.5281/zenodo.21937795)
The founding argument. The brain doesn't have two learning systems by accident — it has them because one system can't do both jobs. McClelland, McNaughton & O'Reilly (1995) showed that a single network learning new facts fast overwrites its old knowledge: catastrophic interference, the same failure mode as fine-tuning an LLM. The brain's solution is a fast, sparse, episodic learner (hippocampus) feeding a slow statistical generalizer (neocortex), with a managed interaction between them — not a bigger single network. Sapience builds that division of labor as software around a frozen model.

The problem, in one paragraph

A neural network stores everything in shared weights, so there is no such thing as writing one memory without touching others: gradient descent on new data marches straight through the representations the old knowledge depends on. This is the stability–plasticity dilemma, and it is structural, not an engineering bug. The field's mitigations each concede part of it — regularization saturates after a few tasks; naive rehearsal means retraining on everything forever; adapters used as fact containers sidestep integration; bigger context windows avoid learning entirely, and collapse beyond the window (we measured this directly). The brain hit the same wall evolutionarily, and its answer was two systems with different learning rates and a managed interface. The 1995 CLS paper is explicit that the hippocampus exists because of catastrophic interference.

The blueprint: eight CLS mechanisms and their implementations

Brain mechanismWhat it doesSapience implementation
Sleep replayHippocampus re-fires the day's episodes at ~20× speed, teaching cortex offline; silence the ripples and consolidation fails (Girardeau 2009)Consolidation over the store; replay corpora rendered from the store for weight adaptation (measured below)
Memory typing (H.M.)Episodic, semantic, procedural memory are separable systems, visible in one patientTyped knowledge objects; episodic and semantic tiers with different decay
Consolidation gradientMemories migrate from hippocampal dependence to cortical storage over time (Ribot's law)Knowledge moves from raw capture to consolidated, schema-linked objects
ReconsolidationA retrieved memory becomes rewritable before re-storage — update, not duplicateSupersession edges: what replaced what, tracked at write time
Salience gatingThe amygdala tags what's worth keeping; most of experience is deliberately not keptThe write gate. Selection is the intelligence, not a compromise
Pattern separation / completionSimilar episodes stored as distinct codes; whole memories retrieved from partial cuesDistinct typed objects + hybrid associative retrieval
Schema-dependent consolidationNew information that fits existing structure consolidates far faster (Tse 2007)Structured store accelerates rather than archives
Working memory vs. the storesPrefrontal cortex holds a few items at high metabolic cost; the fix for its limit is routing, not enlargementThe context window is working memory; scaling it is the wrong axis
The meta-point. The brain runs at least five distinguishable learning systems — episodic, semantic, procedural, motor-calibration, emotional — in different tissue, with different learning rules and speeds, and evolution kept all of them interacting rather than converging on one. The current LLM stack has roughly two of the five (semantic ≈ pretraining, procedural ≈ RLHF) plus working memory (the context window). No hippocampus, no gate, no replay, no reconsolidation. That gap is what Sapience builds.

The implementation, measured

Reading: the store carries long-context reasoning

With identical model weights, a frontier reader answers 3-hop questions at 34.4% from a raw ~1M-token window and at 66.7% from ~700 tokens selected by the store (n=90 item-paired, five judges from three vendors agree on every row). The same system holds a flat ~82% band from 2M to 10M tokens of stored history — the reader's working set stays bounded while history grows 2,500-fold — while at the 10M tier the same reader over its raw window scores 0/20 with the answer verified present in what it reads. Beyond the window the mechanism is physical: truncated serving drops gold-passage availability to 2–3%, while iterative retrieval holds it at ~34%.

Updating: supersession beats relevance

On real Wikipedia revision histories, where facts change and the current value is defined by the edit stream, the store's supersession structure beats naive relevance retrieval by +40.4 points — the baseline retrieved both the stale and the current value and could not tell which was current. The reconsolidation analogue at the weight level: restricting replay to store-current memories removed every observed stale response (0/81 superseded probes) at non-inferior retention, and a volume-matched control does not reproduce the effect — it follows from which content is excluded, not from replaying less.

Learning: sleep replay in the weights

The base model stays frozen — it is the neocortex. Learning happens in a LoRA adapter, and the training data for new knowledge is interleaved with replay rows rendered from the episodic store: the store decides what gets rehearsed, as the hippocampus decides what the cortex hears at night. Sequential fine-tuning forgets ~85% of previously-correct items; store-generated replay cuts that to ~10% at formally matched acquisition — pooled protection of 47–75 points less forgetting, replicated across a second model family, a second domain pair, and a real git-history corpus. The controls make it CLS rather than generic rehearsal: matching the training dose or the optimizer steps without the replayed content restores full forgetting. Remove the replay, consolidation fails — Girardeau's experiment, run in the weights.

One substrate, two functions. The unifying object is the store's set of current knowledge: retrieval intersects with it when the system answers, and replay is rendered from it when the system learns. The same versioned state governs what the system treats as true when it reads and what it reinforces as true when it consolidates — which is the CLS claim, stated as an engineering invariant.

What ships today, what's in the lab

The store, the write gate, hybrid retrieval, supersession, and consolidation run in production over a frozen, swappable reader — the same system has run on frontier models and on an 8B open model. The weight-adaptation layer is measured at the mechanism level (≤8B open models under LoRA, with a qualified 72B extension) and is not yet in the shipped product. The paper reports negative results and falsified predictions alongside the positives, and every headline number passed an adversarial audit protocol before publication; scope statements travel with each claim.