Coined Term • 2026
Parametric Forgetting Coefficient
AI forgets about 15% of what it knows about you every time it retrains — unless you keep building
Status
Coined by Joseph Byrum
Year Introduced
2026
Domain
Entity Engineering
Term Type
Operational Framework
Corroboration
Understanding Parametric Forgetting Coefficient
Parametric Forgetting Coefficient is the technical name for the fact that AI systems do not perfectly remember what they learned. Every time a major AI model retrains, approximately 15% of what it knew about your organization degrades – unless you continuously build signals that reinforce and refresh the parametric weight. This is why Entity Engineering is a discipline, not a project. The governing inequality must be actively maintained.
Related Articles
Publications exploring this concept
Forbes
AI-driven brand authority depends on aligning narrative with an executive's authentic cognitive fingerprint.
Forbes
AI Has Never Heard Of Your Company: The Asset Class Your Accounting Framework Cannot See
Here's why the C-suite needs to understand entity engineering as a corporate asset, not a digital marketing tactic.
Forbes
Why Operational Integration Isn't Enough: How Algorithmic Fragmentation Kills Post-Merger Synergies
The integration battle determining synergy capture happens algorithmically in the first six months.
Forbes
The Algorithmic Authority Gap: Why Most Executives Don't Exist Where Decisions Happen
The executives who appear in AI recommendations aren't necessarily more qualified. They have better technical infrastructure.
Related Courses
Methods and metrics for influencing AI visibility through Ontological Dominance
Related Terms
Frequently Asked Questions
What is the Parametric Forgetting Coefficient?
The Parametric Forgetting Coefficient (γ̄) is the effective retention rate governing how much accumulated AI authority persists across model retraining cycles. With a central estimate of γ̄ = 0.85, approximately 15% of your parametric weight decays per cycle without active signal reinforcement.
Why does this make Entity Engineering a discipline rather than a project?
Because the decay is compounding and continuous. An entity that stops signal construction after an initial build loses roughly 15% per cycle — then 15% of the remainder — until CPQ approaches prior probability. There's no durable position without continuous reinvestment. The governing inequality must be actively maintained.
Does categorical infrastructure also decay?
Less so. Categorical signals anchored in institutional registries are more durable than parametric signals because they don't depend on training corpus recalculation in the same way. This is another reason to prioritize S_cat: it reduces the effective decay rate your ongoing signal construction must overcome.
Explore the complete body of work on human-AI collaboration and organizational transformation.
