Dynamic Outer Layer Models for Generative AI Model Decay

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Generative AI models experience accuracy degradation due to information drift and concept drift over time, leading to reduced effectiveness without a clear method to prevent model decay other than complete rebuilding.

Innovation Solution

Implementing a nimble, dynamically updatable outer layer model that monitors and corrects responses from a foundational generative AI model, using machine learning to incorporate newer data and provide feedback for model reconfiguration.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Stability of the object's composition

If a foundational generative AI model is deployed as a closed loop model, then the model structure is stable and simple to manage, but the model accuracy degrades over time due to information drift and concept drift

Engineering Contradiction:

Improvemodel structure stabilityVSAvoidmodel accuracy

Core Design Contradiction:

Stability of the object's compositionVSReliability

Solution Approach 1:

The patent segments the monolithic foundational model into two distinct components: a stable foundational generative AI model and a dynamically updatable outer layer model. This segmentation allows the foundational model to maintain structural stability while the outer layer adapts to new information, resolving the contradiction between stability and accuracy.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The outer layer model serves as an intermediary between the foundational model and the external environment. It receives updated information and user inputs, processes them through the foundational model, and returns responses. This intermediary structure protects the foundational model from direct exposure to drifting data while maintaining accuracy.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Reliability

If the foundational generative AI model is dynamically updated with new information, then the model accuracy is maintained, but the complexity of model management increases

Engineering Contradiction:

Improvemodel accuracyVSAvoidmodel management complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

By segmenting the model into a static foundational model and a dynamic outer layer, the patent localizes the complexity of updates to only the outer layer. The foundational model remains simple and stable, while the outer layer handles all dynamic updates, reducing overall management complexity compared to updating the entire model.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The outer layer model is designed to be dynamically updatable with new information, allowing the system to adapt to changing conditions without modifying the foundational model. This dynamic component handles updates independently, simplifying the overall management structure.

Inventive Principle:

Principle #15Dynamics

3Duration of action of stationary object

If the outer layer model is trained on updated information, then the model can correct drift and extend model life, but the training time and computational resources increase

Engineering Contradiction:

Improvemodel functional lifeVSAvoidtraining time

Core Design Contradiction:

Duration of action of stationary objectVSLoss of time

Solution Approach 1:

Instead of training the entire foundational model, the patent applies partial action by training only the outer layer model with updated information. This approach extends the model's functional life by addressing only the specific component that experiences drift, significantly reducing training time and computational resources compared to full model retraining.

Inventive Principle:

Principle #16Partial or excessive action

4Reliability

If the outer layer model evaluates and modifies responses, then the accuracy is improved, but the response generation time increases

Engineering Contradiction:

Improveresponse accuracyVSAvoidresponse generation speed

Core Design Contradiction:

ReliabilityVSSpeed

Solution Approach 1:

The outer layer model implements a feedback mechanism where initial responses from the foundational model are evaluated and modified if necessary. This feedback loop improves response accuracy by correcting drift-related errors while maintaining the speed advantage of the foundational model for initial response generation.

Inventive Principle:

Principle #23Feedback

Data Source

PatentUS20250272567A1Nimble and modifiable outer layer for foundational generative artificial intelligence (AI) models to delay model decay and extend model life

Publication Date: 2025.08.28 BANK OF AMERICA CORP

AI summary

A computing platform may train, using historical information, a closed loop foundational generative AI model to generate responses to input prompts. The computing platform may receive, after training and deploying of the foundational generative AI model is complete, updated information that may be relevant to the generation of the responses. The computing platform may train, based on the updated information, an outer layer model that is dynamically updatable to evaluate the responses from the foundational generative AI model and to modify incorrect responses. The computing platform may input a first input prompt into the foundational generative AI model to produce an initial response. The computing platform may input the initial response into the outer layer model. Based on identifying, using the outer layer model, that the initial response is incorrect, the computing platform may modify the initial response to produce a modified response; and send the modified response for display.