Chat GPT의 원리, 활용, 한계와 업무효율화

Abstract

This paper introduces the principles and applications of Chat GPT, which utilizes the large-scale language model GPT. It first discusses the detailed workings of natural language processing, where embedding serves as the initial step in computer-based language processing. The relationship between transformers and GPT is explained for sentence-level embedding. Then this paper explores the development of Chat GPT based on GPT-3.5, which has been improved via reinforcement learning from human feedback or RLHF, and released by OpenAI. Chat GPT can be utilized in various domains such as question answering, translation, correction, summarization, tabular data processing, and coding. It can guide users to obtain desired responses through role-setting and data delivery. However, its broad application is having significant societal impacts. Issues related to corporate or national security have led to usage restrictions, while debates over authorship recognition in the scientific community and discussions about artistic value in the cultural and arts sector have sparked strikes due to job concerns. The emergence of fake problems produced by generative AI, along with its misuse for cybercrime, is also highlighted. Such limitations of Chat GPT, including hallucination phenomena, are discussed in this work. Lastly, suggestions are made for utilizing Chat GPT in business that include emphasizing the need for content verification with hallucination in mind, cautioning against relying solely on search engines, utilizing generative AI with retrieval-augmented generation(RAG) functionality, and employing small large language models (sLLMs) developed for professional domains.