Grounded Content Token Insertion for Reliable AI Output

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Generative machine learning models face inefficiencies and inaccuracies due to high computational demands and probabilistic generation, leading to model hallucinations and reliance on outdated or inaccessible training data, which affects their reliability and stability in producing accurate outputs.

Innovation Solution

A computing system that integrates a model plugin with a machine learning model to identify provenance metadata, temporarily halt probabilistic generation, retrieve grounded content from a trusted data source, and insert it directly into the output sequence, ensuring accurate and efficient content delivery.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If generative machine learning models use token-wise probabilistic generation to produce output, then the model can generate diverse and flexible content, but the computational demands increase and accuracy decreases due to model hallucinations and reliance on outdated training data

Engineering Contradiction:

Improvecontent generation flexibilityVSAvoidoutput accuracy

Core Design Contradiction:

Adaptability or versatilityVSReliability

Solution Approach 1:

The patent introduces an intermediary component (grounded content module) that sits between the machine learning model and the output. This module intercepts the probabilistically generated tokens, identifies placeholders for grounded content, and replaces them with accurate information from trusted data sources. The intermediary thus preserves the model's generative flexibility while eliminating accuracy issues by substituting hallucinated content with verified information.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent segments the output generation process into two distinct phases: (1) the machine learning model generates a draft output with placeholders for grounded content, and (2) the grounded content module fills in the placeholders with accurate information. This segmentation allows the system to leverage both the creativity of probabilistic generation and the accuracy of grounded data without compromising either aspect.

Inventive Principle:

Principle #1Segmentation

2Stability of the object's composition

If generative machine learning models process complete output sequences through probabilistic generation, then the model maintains consistency in content style, but computational load increases and generation time extends

Engineering Contradiction:

Improvecontent style consistencyVSAvoidgeneration efficiency

Core Design Contradiction:

Stability of the object's compositionVSProductivity

Solution Approach 1:

The patent applies partial action by having the machine learning model generate only the portions of output that require creative generation, while using placeholders for content that should be retrieved from grounded data sources. This partial generation approach reduces computational load and generation time while maintaining style consistency in the model-generated portions. The grounded content module then efficiently fills in the remaining portions without requiring full probabilistic generation.

Inventive Principle:

Principle #16Partial or excessive action

3Ease of operation

If the model relies on training data for content generation, then the model can operate independently without external data sources, but the content becomes outdated and inaccessible when training data is not current

Engineering Contradiction:

Improvemodel independenceVSAvoiddata currency

Core Design Contradiction:

Ease of operationVSLoss of information

Solution Approach 1:

The patent applies preliminary action by having the system prepare and store grounded content from trusted data sources in advance. The grounded content module maintains connections to current data sources and retrieves up-to-date information before it is needed for output generation. This preliminary preparation ensures that when the model generates output, the most current accurate information is already available, eliminating the need for the model to rely on outdated training data while maintaining operational efficiency.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS20250356257A1Machine learning model with grounded content token insertion

Publication Date: 2025.11.20 MICROSOFT TECHNOLOGY LICENSING LLC

AI summary

A computing system is provided that receives a tokenized prompt at a machine learning model, generates a model-generated content portion of an output sequence of output tokens in response to the tokenized prompt, identifies provenance metadata for a grounded data source in the model-generated content portion of the output sequence. Upon identification of the provenance metadata, the computing system at least temporarily ceases token-wise probabilistic generation of the output sequence with the machine learning model, retrieves grounded content from the grounded data source using the provenance metadata, writes output tokens corresponding to the grounded content to a grounded content portion of the output sequence, and transmits the output sequence to an additional computing process, for display, storage, or additional downstream processing, for example.