Abstract
This paper reviews the current role of Large Language Models (LLMs) in the treatment and support of people with autism spectrum disorder (ASD). We explore applications of LLM-based systems designed to improve communication, social skills, and emotional learning for individuals with ASD, highlighting their ability to generate personalized conversational interactions and simulate social scenarios. Current research demonstrates promising results in different domains, such as dialogue interventions, emotional recognition training, and work-related communication assistance. However, significant challenges remain, including ethical concerns about overreliance on AI, personalization, and privacy. This review synthesizes recent findings, highlights gaps in the existing literature, and proposes directions for future investigations. It particularly underscores the need for real-world applications and long-term studies to assess the efficacy of LLM efficacy in interventions for ASD.