Research of approaches to verification and correction of response content of a Large Language Model
Year: 2024
Type: dissertation
Abstract: The purpose of the work is to research and develop a system that implements automatic verification and correction of factual errors in text generated by a large language model, within limited computing resources. The object of study is large language models. The subject of the study is the verification and correction of text generated by a large language model. A study of existing approaches to correcting the response of a large language model was conducted, RAG method was recognized as the most appropriate method to enrich the knowledge of the model and prevent its hallucinations. The result of the work was the design and development of original RAG system. The system was evaluated for effectiveness in preventing hallucinations, determined that the system detects 93% of false facts in texts generated by Alpaca 13B, and 84% of false facts in a ChatGPT response.
Authors: Шатилов, М. Р., Shatilov, M. R.
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7070910892