Prompt Generation for LLM Fact Determination Accuracy

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional methods for fact determination using large language models (LLMs) suffer from hallucination issues and inefficiencies, particularly in black-box models lacking internal information access, necessitating improved accuracy and efficiency in fact determination.

Innovation Solution

A prompt generation model is trained to generate optimized prompts for a specific language model, utilizing a training process that filters and refines data pairs to enhance the model's ability to accurately determine factual information.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If conventional fact determination methods using language models are used, then the system can process information quickly, but hallucination occurs and factual accuracy deteriorates

Engineering Contradiction:

Improvefactual accuracyVSAvoidhallucination

Core Design Contradiction:

ReliabilityVSObject-generated harmful factors

Solution Approach 1:

The patent applies preliminary action by generating multiple candidate answers before final fact determination. The system creates several potential responses using different prompts and models, then evaluates them against evidence sources to select the most accurate answer, preventing hallucination before it occurs

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent implements feedback mechanisms by evaluating generated answers against external evidence sources and using uncertainty measurements. The system receives feedback on answer accuracy and adjusts its fact determination process accordingly, improving reliability while reducing hallucination

Inventive Principle:

Principle #23Feedback

2Reliability

If external information is used for fact determination, then factual accuracy improves, but data requirements and system complexity increase

Engineering Contradiction:

Improvefactual accuracyVSAvoidexternal data requirements

Core Design Contradiction:

ReliabilityVSQuantity of substance

Solution Approach 1:

The patent applies universality by creating a multi-functional system that can determine facts across different domains using the same core architecture. The fact determination system handles various types of questions and evidence sources through unified prompts and evaluation mechanisms, reducing the need for domain-specific data

Inventive Principle:

Principle #6Universality (Multi-functionality)

3Reliability

If uncertainty measurement methods are used to identify hallucination, then factual accuracy improves, but applicability to black-box models deteriorates

Engineering Contradiction:

Improvefactual accuracyVSAvoidapplicability to black-box models

Core Design Contradiction:

ReliabilityVSAdaptability or versatility

Solution Approach 1:

The patent uses an intermediary approach by introducing external evidence sources and evaluation mechanisms that work independently of the language model's internal workings. The system mediates between the black-box model's outputs and factual verification through prompts and evidence comparison, making uncertainty measurement applicable to any model regardless of internal accessibility

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS20260072903A1Method and system for fact determination

Publication Date: 2026.03.12 SAMSUNG SDS CO LTD

AI summary

A fact determination method is provided, the method comprising receiving target text, generating one or more question prompts using a prompt generation model to induce extraction of information associated with the target text, obtaining answers to the respective question prompts by inputting the question prompts into a language model and outputting a result of determining whether the target text is factual using the language model.