Net impact of large language models trained on code

Abstract

Natural language processing has seen many improvements in recent years, particularly driven by machine learning models such as OpenAI’s GPT-3. This paper aims to present the various language models, as well as OpenAI Codex, which is considered to be an AI revolution in the field of programming. This system has been trained on Python code from more than 50 million GitHub repositories and is capable of generating and explaining code, translating it between various programming languages and more. In addition, the benefits and potential dangers of its use will be analysed and presented.