LLM Prompt Segmentation for Reducing Hallucinations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current large language models (LLMs) lack a systematic mathematical framework to describe, compare, and improve their behavior, leading to issues like hallucinations and misalignments.
Innovation Solution
The proposed system creates multiple LLM prompts based on an initial input prompt, submitting these as sub-queries to the LLM to break down the problem into steps, reducing the number of responses with hallucinations or misalignments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a single LLM prompt is submitted to generate a solution, then the process is simple and quick, but the response may contain hallucinations or misalignments
Solution Approach 1:
The patent applies segmentation by dividing a single complex prompt into multiple sub-prompts that represent sequential steps toward solving the problem. Each sub-prompt is processed separately by the LLM, and the results are aggregated to form the final solution. This breakdown reduces hallucinations by constraining each sub-prompt to a specific aspect of the problem, thereby improving reliability without requiring complex external verification systems.
Solution Approach 2:
The patent introduces an intermediary processing layer that sits between the user's original prompt and the LLM. This intermediary automatically generates multiple sub-prompts based on the original input, manages the sequence of LLM queries, and synthesizes the responses. This intermediary structure improves response accuracy while keeping the system architecture relatively simple and automated.
2Reliability
If multiple LLM prompts are submitted as sub-queries to break down the problem, then the accuracy improves, but the number of queries and processing time increases
Solution Approach 1:
The patent applies preliminary action by pre-structuring the problem decomposition logic before submitting queries to the LLM. The system analyzes the original prompt and automatically generates an optimized sequence of sub-prompts in advance, ensuring that each query builds logically on previous results. This preliminary structuring reduces redundant queries and focuses the LLM's computational effort on essential reasoning steps, thereby improving accuracy while minimizing time loss.
3Reliability
If multiple LLM prompts are submitted as sub-queries to break down the problem, then the accuracy improves, but the computational resources required increase
Solution Approach 1:
The patent applies segmentation by dividing a single complex prompt into multiple sub-prompts that represent sequential steps toward solving the problem. Each sub-prompt is processed separately by the LLM, and the results are aggregated to form the final solution. This breakdown reduces hallucinations by constraining each sub-prompt to a specific aspect of the problem, thereby improving reliability without requiring complex external verification systems.
Solution Approach 2:
The patent applies partial action by generating just enough sub-prompts to adequately decompose the problem without over-segmenting. The system dynamically determines the appropriate level of decomposition based on the complexity of the original prompt, submitting only the necessary number of sub-queries to achieve accurate results while avoiding unnecessary computational overhead and energy consumption.
Data Source
AI summary
Example solutions for processing LLM prompts include creating a first large language model (LLM) prompt based on an input LLM prompt. The first LLM prompt represents a first step toward generating a solution to the input LLM prompt. The first LLM prompt is submitted to an LLM as a first sub-query, thereby resulting in the generation of a first LLM output. A second LLM prompt is generated based on the input LLM prompt. The second LLM prompt represents a second step toward generating the solution. The second LLM prompt includes the first LLM output. The second LLM prompt is submitted to the LLM as a second sub-query, thereby resulting in the generation of a second LLM output. The second LLM output represents the solution to the input LLM prompt in response to the input LLM prompt.