Real Time Retrieval Argumentation for Large Language Models

22 Pages Posted: 26 Apr 2024

See all articles by Udit Raj

Udit Raj

Indian Institute of Technology, Patna

Date Written: April 25, 2024


This groundbreaking research challenges the conventional approach of building large language models by scraping and storing massive internet datasets on servers - a slow, outdated process prone to hallucinations. We introduce the Real-Time Retrieval Argumentation (RTRA) architecture as a paradigm shift to develop more precise and efficient language models without relying on huge pre-trained datasets.

RTRA dynamically retrieves live web data as a secondary dataset during inference, enabling models trained on just 7 billion parameters to outperform those trained on hundreds of billions of parameters from static data.

Our novel approach marks a departure from the dependency on expensive computational resources for dataset curation and model efficiency. By seamlessly integrating real-world information, RTRA empowers language models to stay reliably up-to-date, mitigating hallucinations while significantly reducing resource requirements

Keywords: Artificial Intelligence, Machine Learning, Machine Learning Architecture, AI, ML, LLM, Large Language Model, Information Retrieval

Suggested Citation

Raj, Udit, Real Time Retrieval Argumentation for Large Language Models (April 25, 2024). Available at SSRN: or

Udit Raj (Contact Author)

Indian Institute of Technology, Patna ( email )

Kampa Road, Bihita, Patna, Bihar, India
Patna, PA Bihar 800020

HOME PAGE: http://

Do you have a job opening that you would like to promote on SSRN?

Paper statistics

Abstract Views
PlumX Metrics