Authors:
Bharath Kumar Nagaraj, R. Subhashni

Addresses:
1School of Artificial Intelligence Engineer, Digipulse Technologies INC., United States of America. 2Department of Computer Science and Applications, St. Peter's Institute of Higher Education and Research, Chennai, Tamil Nadu, India. bharath.kumar@revature.com1, subhashniraj2018@gmail.com2

Abstract:

Large Language Models (LLMs) have remarkably advanced in various natural language understanding tasks. However, their black-box nature often hinders interpretability and transparency, which are crucial for their ethical and practical applications. This research paper explores the development of LLM architectures that inherently produce more interpretable outputs. We examine techniques and strategies to make LLMs more transparent and understandable, considering both model architectures and post-processing methods. By focusing on the creation of inherently interpretable LLMs, we aim to address the challenge of reconciling the impressive capabilities of these models with the need for interpretable results. Large language models (LLMs) are a type of artificial intelligence (AI) model that has been trained on a massive dataset of text and code. LLMs can generate text, translate languages, write creative content, and answer your questions informally. However, LLMs are often criticized for being black boxes, meaning it is difficult to understand how they work and why they produce the outputs they do. This lack of interpretability can make it difficult to trust LLMs in sensitive applications, such as healthcare or finance.

Keywords: Large Language Models (LLMs); Model Interpretability; Explainable Artificial Intelligence (XAI); Development of LLM Architectures; Explainability in Deep Learning; Accountability in AI Systems; Human-Centric AI Design.

Received on: 12/01/2023, Revised on: 17/03/2023, Accepted on: 28/04/2023, Published on: 11/05/2023

FMDB Transactions on Sustainable Computer Letters, 2023 Vol. 1 No. 2, Pages: 115-129

• Views : 246
• Downloads : 10

Download PDF