An Overview of LLM-Based AI Chatbots
Year: 2026
Type: conference-paper
Abstract: Large Language Models (LLMs) have greatly advanced the field of AI chatbots, making multimodal generation, long-context reasoning, complex task solving, and automated workflows possible. This article examines the progression of chatbot technologies, from early rule-based and retrieval-oriented models to contemporary generative and multimodal LLM-based systems. Models are examined in terms of architectural design, performance evaluation, applications, limitations, strengths, safety, and ethical considerations, based on the analysis of technical reports, benchmark results, and analyses reported in the literature. From this perspective, a classification of them across domains has been done, highlighting that different chatbot models have different strengths and weaknesses across applications, and model selection should balance performance, security, and user requirements. The results indicate that autoregressive Transformer-based architecture demonstrates great performance in generative and open-domain chatbots, while encoder and task-oriented architecture models are suitable for specific Natural Language Processing (NLP) tasks. Integration of Mixture-of-Experts (MoE) approaches with multimodal and long-context shows exceptional performance in long-context understanding, complex task solution, and automation workflow. This paper presents a valuable contribution, offering guidelines for users, developers, and researchers.
Authors: Amarildo Rista, Edmira Xhaferra, Lindita Mukli
Institution: Aleksandër Moisiu University
Language: English
Topic: Speech and dialogue systems
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7167705294