EvolutionStrong architectural inferencev1.22.1

In plain English

This page explains how AI systems can change over time through updates, tests, retraining, memory, and approvals even when no single model rewrites itself.

  • Why this matters: AI risk can come from the whole arrangement, not one obvious model.
  • What to look for: data, memory, routes, adapters, tools, evaluators, updates, and rollback paths.
  • Technical version below: the expert terminology remains available and is linked through the glossary.

Model Breeding Operators and Selection

Evidence levelStrong architectural inferenceTechnical label: Architectural inference

The new reports add operator vocabulary for controlled model breeding. A behavior pattern that can survive, move, or reappear across a changing AI system. Open glossary definition uses those operators as review objects, not as instructions to build autonomous reproduction.

Operator map

OperatorReview question
A small add-on that changes or specializes model behavior. Open glossary definition mutationWhich adapter changed, what rank or density changed, and who approved it?
Multi-parent crossoverWhich parents contributed behavior, weights, prompts, or adapters?
SLERP / interpolationWhat layer or tensor range was blended, and what evidence says the blend is valid?
Task-vector mergeWhich task vector was added, removed, or scaled?
TIES / DARE-style sparsityWhat was trimmed, dropped, elected, or rescaled, and how was interference tested?
SpeciationWhich niche does the candidate belong to, and does it preserve useful diversity?
Novelty archiveIs the candidate behaviorally new or merely a duplicate with new packaging?

Selection rule

A The rule that decides what survives. Open glossary definition should never be only “highest score wins.” It should include utility, cost, latency, memory, novelty, evaluator disagreement, rollback completeness, source provenance, and no-op availability.

Failure modes

Defensive use

Use these operators to ask better questions about Creating a proposed new model, adapter, prompt, route, test, or policy. Open glossary definition. The site does not provide parameter recipes, exploit payloads, or model-construction instructions.

v1.22.0 risk-side boundary

Evidence levelStrong architectural inferenceTechnical label: Strong architectural inference

This page preserves possibility-side source material. The Cognivirus use is narrower: translate controlled-evolution vocabulary into risk review. For that translation, use ModelBreeder Risk Side, ModelBreeder Risk Translation, and Model Breeding Risk Boundaries.