Self-Teaching Large Language Model Reducing Hallucinations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Large language models (LLMs) face challenges in generating accurate and reliable outputs, often producing false or incorrect information, known as hallucinations, which affects their reliability and trustworthiness.
Innovation Solution
The method involves a self-learning framework with two phases: a diversification phase that generates a diverse set of outputs using techniques like few-shot learning and context manipulation, and an aggregation phase that combines these outputs to produce a single, more accurate result through filtering and majority vote strategies, iteratively improving the accuracy of LLMs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If LLMs generate outputs using standard prompting techniques, then the model can respond to various input prompts, but the outputs contain hallucinations and are unreliable
Solution Approach 1:
The patent implements a self-learning framework where the LLM uses its own previous outputs as training data. The model generates outputs, stores them, and then re-trains itself on this stored data, creating a feedback loop that progressively improves reliability and reduces hallucinations over multiple phases
Solution Approach 2:
The system performs preliminary actions by generating and storing outputs in advance during training phases. These pre-generated outputs are then used as training data for subsequent phases, allowing the model to learn from its own performance before facing new questions
2Manufacturing precision
If the LLM operates in a single phase without self-learning, then the process is simple and fast, but the accuracy and reliability of outputs remain limited
Solution Approach 1:
The training process is segmented into multiple phases (first phase, second phase, third phase, etc.), where each phase involves generating outputs, storing them, and re-training the model. This segmentation allows the system to progressively improve accuracy while managing complexity through structured iterations
Solution Approach 2:
The framework maintains continuous improvement by repeatedly cycling through generation, storage, and re-training phases. The model continuously learns from its own outputs, ensuring that accuracy improvements are sustained and built upon across multiple phases
3Reliability
If the LLM uses self-learning with multiple phases, then the accuracy and reliability improve over time, but the computational cost and time required increase
Solution Approach 1:
The LLM performs self-learning by using its own outputs as training data without requiring external intervention. The model autonomously generates outputs, stores them, re-trains itself, and improves its performance, eliminating the need for manual data collection and annotation processes
Data Source
AI summary
The present disclosure relates to methods and systems for self-teaching a large language model (LLM). The methods and systems use a self-learning framework with a plurality of phases. In each phase of the self-learning framework, the LLM generates a diverse set of outputs for a question and an aggregation is performed on the diverse set of outputs generate a phase output. The phase output from a previous phase is used as an input to the LLM in a next phase.