In this media review, we leverage Hubbard's (2021) framework for computer-assisted language learning (CALL) to examine the use of generative artificial intelligence (AI), specifically OpenAI's GPT-4 (fourth-generation generative pre-trained transformer), as a tool for teachers to provide their students with feedback on writing. This tripartite framework consists of an operational description of the technology, and how it fits learners' and teachers' needs and abilities. This review is therefore divided into three main components: a description of generative AI and how GPT-4 could be used in an English as a second or foreign language (ESL/EFL) teaching context, an evaluation of how well this technology fits the teacher's needs in providing feedback, and an evaluation of how well it provides useful feedback to students in order to meet their needs.
| OPERATIONAL DESCRIPTIONGPT-4 is a large language model (LLM) that users can interface with through the chatbot ChatGPT (https://chat.openai.com/). GPT-4 is trained on massive amounts of text data and has been fine-tuned using reinforcement learning from human feedback (RLHF) (https://openai.com/resea rch/gpt-4). It can be accessed on mobile devices, PC, or Mac through a browser, as well as through an app for Android and iOS users. While there are LLMs that are free to use (GPT-3.5, PaLM 2), we prefer to use GPT-4 for generating feedback because it currently has superior reasoning capabilities and we have found its feedback to be more accurate, concise, and detailed. Use of GPT-4 does require a "Plus" membership (20 USD per month). Users can interact with GPT-4 by submitting a prompt (a question or request) to ChatGPT, to which the chatbot will respond. Because of high demand, prompts to GPT-4 are currently limited to 50 every 3 hours. To get the most from LLMs, effective prompt engineering is essential. Ingley and Pack (2023) offer a useful framework: assign a role or identity, define an objective, specify the context and constraints, and refine the output through continued conversation. Carefully crafted prompts will help the AI to generate more suitable and useful output. Prompt engineering is a process and a skill, and we underscore the need to refine prompts and output through continued conversation in order to increase the accuracy and appropriacy of AI-generated feedback.