ChatGPT is a powerful artificial intelligence (AI) language model that demonstrates an understanding of human-like text and generates responses based on an extensive corpus of training data. Deep learning has recently witnessed widespread adoption in applications such as object classification, object detection, appearance inspection, and anomaly detection, achieving remarkable accuracy. However, it is essential to acknowledge that all deep learning techniques rely on the underlying computer architecture and necessitate hardware utilisation to execute applications. Consequently, resource-scarce environments, such as edge CPU, can significantly impact the performance of these applications. This paper addresses the challenges of designing AI models using ChatGPT and explores methods for accelerating these models on CPUs. Specifically, we endeavour to engage ChatGPT in discussions to assist in creating an accurate and personalised AI model. Subsequently, we aim to collaborate with ChatGPT to generate assembly code tailored for MIPS CPUs.
Su et al. (Thu,) studied this question.